From b1e512c15297ad20d2d41e7a79ef9a0fe32fac4b Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Tue, 15 Sep 2026 21:10:36 -0400 Subject: [PATCH 01/69] Some cleanups --- .../features/settings/model/BooleanSetting.kt | 1 - .../features/settings/model/IntSetting.kt | 3 - .../model/view/GpuUnswizzleSetting.kt | 89 ---- .../settings/model/view/SettingsItem.kt | 41 -- .../settings/ui/GpuUnswizzleDialogFragment.kt | 206 -------- .../features/settings/ui/SettingsAdapter.kt | 12 - .../settings/ui/SettingsFragmentPresenter.kt | 1 - .../ui/viewholder/GpuUnswizzleViewHolder.kt | 71 --- .../main/res/layout/dialog_gpu_unswizzle.xml | 96 ---- .../app/src/main/res/values-ar/strings.xml | 26 - .../app/src/main/res/values-de/strings.xml | 16 - .../app/src/main/res/values-es/strings.xml | 26 - .../app/src/main/res/values-fr/strings.xml | 17 - .../app/src/main/res/values-ru/strings.xml | 26 - .../app/src/main/res/values-uk/strings.xml | 26 - .../src/main/res/values-zh-rCN/strings.xml | 26 - .../src/main/res/values-zh-rTW/strings.xml | 11 - .../app/src/main/res/values/arrays.xml | 48 -- .../app/src/main/res/values/strings.xml | 26 - src/common/settings.h | 21 - src/common/settings_enums.h | 3 - src/qt_common/config/shared_translation.cpp | 41 -- src/video_core/host_shaders/CMakeLists.txt | 1 - src/video_core/host_shaders/astc_decoder.comp | 118 +++-- .../block_linear_unswizzle_2d.comp | 16 +- .../block_linear_unswizzle_3d.comp | 16 +- .../block_linear_unswizzle_3d_bcn.comp | 164 ------ .../host_shaders/pitch_unswizzle.comp | 22 +- .../renderer_opengl/gl_texture_cache.cpp | 5 +- .../renderer_opengl/gl_texture_cache.h | 7 +- .../renderer_opengl/util_shaders.cpp | 11 +- .../renderer_vulkan/vk_compute_pass.cpp | 483 +++++++++--------- .../renderer_vulkan/vk_compute_pass.h | 61 ++- .../renderer_vulkan/vk_texture_cache.cpp | 175 ++++--- .../renderer_vulkan/vk_texture_cache.h | 22 +- src/video_core/texture_cache/texture_cache.h | 145 +----- .../texture_cache/texture_cache_base.h | 20 - 37 files changed, 509 insertions(+), 1590 deletions(-) delete mode 100644 src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/GpuUnswizzleSetting.kt delete mode 100644 src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/GpuUnswizzleDialogFragment.kt delete mode 100644 src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/viewholder/GpuUnswizzleViewHolder.kt delete mode 100644 src/android/app/src/main/res/layout/dialog_gpu_unswizzle.xml delete mode 100644 src/video_core/host_shaders/block_linear_unswizzle_3d_bcn.comp diff --git a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/BooleanSetting.kt b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/BooleanSetting.kt index ca3309245e..c0f85d0f60 100644 --- a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/BooleanSetting.kt +++ b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/BooleanSetting.kt @@ -39,7 +39,6 @@ enum class BooleanSetting(override val key: String) : AbstractBooleanSetting { RENDERER_SAMPLE_SHADING("sample_shading"), RENDERER_FRAME_GEN("frame_gen"), RENDERER_FRAME_GEN_FLOW_SCALE_AUTO("frame_gen_flow_scale_auto"), - GPU_UNSWIZZLE_ENABLED("gpu_unswizzle_enabled"), PICTURE_IN_PICTURE("picture_in_picture"), USE_CUSTOM_RTC("custom_rtc_enabled"), BLACK_BACKGROUNDS("black_backgrounds"), diff --git a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/IntSetting.kt b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/IntSetting.kt index ea31692fbf..7f1b49a94a 100644 --- a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/IntSetting.kt +++ b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/IntSetting.kt @@ -51,9 +51,6 @@ enum class IntSetting(override val key: String) : AbstractIntSetting { FAST_CPU_TIME("fast_cpu_time"), CPU_TICKS("cpu_ticks"), FAST_GPU_TIME("fast_gpu_time"), - GPU_UNSWIZZLE_TEXTURE_SIZE("gpu_unswizzle_texture_size"), - GPU_UNSWIZZLE_STREAM_SIZE("gpu_unswizzle_stream_size"), - GPU_UNSWIZZLE_CHUNK_SIZE("gpu_unswizzle_chunk_size"), BAT_TEMPERATURE_UNIT("bat_temperature_unit"), CABINET_APPLET("cabinet_applet_mode"), CONTROLLER_APPLET("controller_applet_mode"), diff --git a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/GpuUnswizzleSetting.kt b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/GpuUnswizzleSetting.kt deleted file mode 100644 index a9a40f6903..0000000000 --- a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/GpuUnswizzleSetting.kt +++ /dev/null @@ -1,89 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project -// SPDX-License-Identifier: GPL-3.0-or-later - -package org.yuzu.yuzu_emu.features.settings.model.view - -import androidx.annotation.ArrayRes -import androidx.annotation.StringRes -import org.yuzu.yuzu_emu.features.settings.model.AbstractSetting -import org.yuzu.yuzu_emu.features.settings.model.BooleanSetting -import org.yuzu.yuzu_emu.features.settings.model.IntSetting - -class GpuUnswizzleSetting( - @StringRes titleId: Int = 0, - titleString: String = "", - @StringRes descriptionId: Int = 0, - descriptionString: String = "", - @ArrayRes val textureSizeChoicesId: Int, - @ArrayRes val textureSizeValuesId: Int, - @ArrayRes val streamSizeChoicesId: Int, - @ArrayRes val streamSizeValuesId: Int, - @ArrayRes val chunkSizeChoicesId: Int, - @ArrayRes val chunkSizeValuesId: Int -) : SettingsItem( - object : AbstractSetting { - override val key: String = SettingsItem.GPU_UNSWIZZLE_COMBINED - override val defaultValue: Any = false - override val isSaveable = true - override val isRuntimeModifiable = true - override val isSwitchable = true - override val pairedSettingKey: String = "" - override var global: Boolean - get() { - return BooleanSetting.GPU_UNSWIZZLE_ENABLED.global && - IntSetting.GPU_UNSWIZZLE_TEXTURE_SIZE.global && - IntSetting.GPU_UNSWIZZLE_STREAM_SIZE.global && - IntSetting.GPU_UNSWIZZLE_CHUNK_SIZE.global - } - set(value) { - BooleanSetting.GPU_UNSWIZZLE_ENABLED.global = value - IntSetting.GPU_UNSWIZZLE_TEXTURE_SIZE.global = value - IntSetting.GPU_UNSWIZZLE_STREAM_SIZE.global = value - IntSetting.GPU_UNSWIZZLE_CHUNK_SIZE.global = value - } - override fun getValueAsString(needsGlobal: Boolean): String = "combined" - override fun reset() { - BooleanSetting.GPU_UNSWIZZLE_ENABLED.reset() - IntSetting.GPU_UNSWIZZLE_TEXTURE_SIZE.reset() - IntSetting.GPU_UNSWIZZLE_STREAM_SIZE.reset() - IntSetting.GPU_UNSWIZZLE_CHUNK_SIZE.reset() - } - }, - titleId, - titleString, - descriptionId, - descriptionString -) { - override val type = SettingsItem.TYPE_GPU_UNSWIZZLE - - // Check if GPU unswizzle is enabled via the dedicated boolean setting - fun isEnabled(needsGlobal: Boolean = false): Boolean = - BooleanSetting.GPU_UNSWIZZLE_ENABLED.getBoolean(needsGlobal) - - fun setEnabled(value: Boolean) = - BooleanSetting.GPU_UNSWIZZLE_ENABLED.setBoolean(value) - - fun enable() = setEnabled(true) - - fun disable() = setEnabled(false) - - fun getTextureSize(needsGlobal: Boolean = false): Int = - IntSetting.GPU_UNSWIZZLE_TEXTURE_SIZE.getInt(needsGlobal) - - fun setTextureSize(value: Int) = - IntSetting.GPU_UNSWIZZLE_TEXTURE_SIZE.setInt(value) - - fun getStreamSize(needsGlobal: Boolean = false): Int = - IntSetting.GPU_UNSWIZZLE_STREAM_SIZE.getInt(needsGlobal) - - fun setStreamSize(value: Int) = - IntSetting.GPU_UNSWIZZLE_STREAM_SIZE.setInt(value) - - fun getChunkSize(needsGlobal: Boolean = false): Int = - IntSetting.GPU_UNSWIZZLE_CHUNK_SIZE.getInt(needsGlobal) - - fun setChunkSize(value: Int) = - IntSetting.GPU_UNSWIZZLE_CHUNK_SIZE.setInt(value) - - fun reset() = setting.reset() -} diff --git a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt index 3f38831a17..8c07d394b0 100644 --- a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt +++ b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt @@ -141,14 +141,12 @@ abstract class SettingsItem( const val TYPE_SPINBOX = 12 const val TYPE_LAUNCHABLE = 13 const val TYPE_PATH = 14 - const val TYPE_GPU_UNSWIZZLE = 15 const val TYPE_FX_TOOLBAR = 16 const val TYPE_FX_PRESET = 17 const val TYPE_FX_SHADER = 18 const val TYPE_FX_BUTTON = 19 const val FASTMEM_COMBINED = "fastmem_combined" - const val GPU_UNSWIZZLE_COMBINED = "gpu_unswizzle_combined" val emptySetting = object : AbstractSetting { override val key: String = "" @@ -794,45 +792,6 @@ abstract class SettingsItem( valuesId = R.array.gpuValues ) ) - put( - SingleChoiceSetting( - IntSetting.GPU_UNSWIZZLE_TEXTURE_SIZE, - titleId = R.string.gpu_unswizzle_texture_size, - descriptionId = R.string.gpu_unswizzle_texture_size_description, - choicesId = R.array.gpuTextureSizeSwizzleEntries, - valuesId = R.array.gpuTextureSizeSwizzleValues - ) - ) - put( - SingleChoiceSetting( - IntSetting.GPU_UNSWIZZLE_STREAM_SIZE, - titleId = R.string.gpu_unswizzle_stream_size, - descriptionId = R.string.gpu_unswizzle_stream_size_description, - choicesId = R.array.gpuSwizzleEntries, - valuesId = R.array.gpuSwizzleValues - ) - ) - put( - SingleChoiceSetting( - IntSetting.GPU_UNSWIZZLE_CHUNK_SIZE, - titleId = R.string.gpu_unswizzle_chunk_size, - descriptionId = R.string.gpu_unswizzle_chunk_size_description, - choicesId = R.array.gpuSwizzleChunkEntries, - valuesId = R.array.gpuSwizzleChunkValues - ) - ) - put( - GpuUnswizzleSetting( - titleId = R.string.gpu_unswizzle_settings, - descriptionId = R.string.gpu_unswizzle_settings_description, - textureSizeChoicesId = R.array.gpuTextureSizeSwizzleEntries, - textureSizeValuesId = R.array.gpuTextureSizeSwizzleValues, - streamSizeChoicesId = R.array.gpuSwizzleEntries, - streamSizeValuesId = R.array.gpuSwizzleValues, - chunkSizeChoicesId = R.array.gpuSwizzleChunkEntries, - chunkSizeValuesId = R.array.gpuSwizzleChunkValues - ) - ) put( SingleChoiceSetting( IntSetting.FAST_CPU_TIME, diff --git a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/GpuUnswizzleDialogFragment.kt b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/GpuUnswizzleDialogFragment.kt deleted file mode 100644 index e14bc7639e..0000000000 --- a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/GpuUnswizzleDialogFragment.kt +++ /dev/null @@ -1,206 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project -// SPDX-License-Identifier: GPL-3.0-or-later - -package org.yuzu.yuzu_emu.features.settings.ui - -import android.app.Dialog -import android.content.DialogInterface -import android.os.Bundle -import android.view.LayoutInflater -import android.widget.ArrayAdapter -import androidx.fragment.app.DialogFragment -import androidx.fragment.app.activityViewModels -import com.google.android.material.dialog.MaterialAlertDialogBuilder -import org.yuzu.yuzu_emu.R -import org.yuzu.yuzu_emu.databinding.DialogGpuUnswizzleBinding -import org.yuzu.yuzu_emu.features.settings.model.view.GpuUnswizzleSetting - -class GpuUnswizzleDialogFragment : DialogFragment() { - private var position = 0 - private val settingsViewModel: SettingsViewModel by activityViewModels() - private lateinit var binding: DialogGpuUnswizzleBinding - - override fun onCreate(savedInstanceState: Bundle?) { - super.onCreate(savedInstanceState) - position = requireArguments().getInt(POSITION) - - if (settingsViewModel.clickedItem == null) dismiss() - } - - override fun onCreateDialog(savedInstanceState: Bundle?): Dialog { - binding = DialogGpuUnswizzleBinding.inflate(LayoutInflater.from(requireContext())) - val item = settingsViewModel.clickedItem as GpuUnswizzleSetting - - // Setup texture size dropdown - val textureSizeEntries = resources.getStringArray(item.textureSizeChoicesId) - val textureSizeValues = resources.getIntArray(item.textureSizeValuesId) - val textureSizeAdapter = ArrayAdapter( - requireContext(), - android.R.layout.simple_dropdown_item_1line, - textureSizeEntries.toMutableList() - ) - binding.dropdownTextureSize.setAdapter(textureSizeAdapter) - - // Setup stream size dropdown - val streamSizeEntries = resources.getStringArray(item.streamSizeChoicesId) - val streamSizeValues = resources.getIntArray(item.streamSizeValuesId) - val streamSizeAdapter = ArrayAdapter( - requireContext(), - android.R.layout.simple_dropdown_item_1line, - streamSizeEntries.toMutableList() - ) - binding.dropdownStreamSize.setAdapter(streamSizeAdapter) - - // Setup chunk size dropdown - val chunkSizeEntries = resources.getStringArray(item.chunkSizeChoicesId) - val chunkSizeValues = resources.getIntArray(item.chunkSizeValuesId) - val chunkSizeAdapter = ArrayAdapter( - requireContext(), - android.R.layout.simple_dropdown_item_1line, - chunkSizeEntries.toMutableList() - ) - binding.dropdownChunkSize.setAdapter(chunkSizeAdapter) - - // Load current values - val isEnabled = item.isEnabled() - binding.switchEnable.isChecked = isEnabled - - if (isEnabled) { - val textureSizeIndex = textureSizeValues.indexOf(item.getTextureSize()) - if (textureSizeIndex >= 0) { - binding.dropdownTextureSize.setText(textureSizeEntries[textureSizeIndex], false) - } - - val streamSizeIndex = streamSizeValues.indexOf(item.getStreamSize()) - if (streamSizeIndex >= 0) { - binding.dropdownStreamSize.setText(streamSizeEntries[streamSizeIndex], false) - } - - val chunkSizeIndex = chunkSizeValues.indexOf(item.getChunkSize()) - if (chunkSizeIndex >= 0) { - binding.dropdownChunkSize.setText(chunkSizeEntries[chunkSizeIndex], false) - } - } else { - // Set default/recommended values when disabling - binding.dropdownTextureSize.setText(textureSizeEntries[3], false) - binding.dropdownStreamSize.setText(streamSizeEntries[3], false) - binding.dropdownChunkSize.setText(chunkSizeEntries[3], false) - } - - // Clear adapter filters after setText to fix rotation bug - textureSizeAdapter.filter.filter(null) - streamSizeAdapter.filter.filter(null) - chunkSizeAdapter.filter.filter(null) - - // Enable/disable dropdowns based on switch state - updateDropdownsState(isEnabled) - binding.switchEnable.setOnCheckedChangeListener { _, checked -> - updateDropdownsState(checked) - } - - val dialog = MaterialAlertDialogBuilder(requireContext()) - .setTitle(item.title) - .setView(binding.root) - .create() - - // Setup button listeners - binding.btnDefault.setOnClickListener { - // Reset to defaults - item.reset() - // Refresh values with adapters reset - val textureSizeIndex = textureSizeValues.indexOf(item.getTextureSize()) - if (textureSizeIndex >= 0) { - binding.dropdownTextureSize.setText(textureSizeEntries[textureSizeIndex], false) - } - val streamSizeIndex = streamSizeValues.indexOf(item.getStreamSize()) - if (streamSizeIndex >= 0) { - binding.dropdownStreamSize.setText(streamSizeEntries[streamSizeIndex], false) - } - val chunkSizeIndex = chunkSizeValues.indexOf(item.getChunkSize()) - if (chunkSizeIndex >= 0) { - binding.dropdownChunkSize.setText(chunkSizeEntries[chunkSizeIndex], false) - } - // Clear filters - textureSizeAdapter.filter.filter(null) - streamSizeAdapter.filter.filter(null) - chunkSizeAdapter.filter.filter(null) - - settingsViewModel.setAdapterItemChanged(position) - settingsViewModel.setShouldReloadSettingsList(true) - } - - binding.btnCancel.setOnClickListener { - dialog.dismiss() - } - - binding.btnOk.setOnClickListener { - if (binding.switchEnable.isChecked) { - item.enable() - // Save the selected values - val selectedTextureIndex = textureSizeEntries.indexOf( - binding.dropdownTextureSize.text.toString() - ) - if (selectedTextureIndex >= 0) { - item.setTextureSize(textureSizeValues[selectedTextureIndex]) - } - - val selectedStreamIndex = streamSizeEntries.indexOf( - binding.dropdownStreamSize.text.toString() - ) - if (selectedStreamIndex >= 0) { - item.setStreamSize(streamSizeValues[selectedStreamIndex]) - } - - val selectedChunkIndex = chunkSizeEntries.indexOf( - binding.dropdownChunkSize.text.toString() - ) - if (selectedChunkIndex >= 0) { - item.setChunkSize(chunkSizeValues[selectedChunkIndex]) - } - } else { - // Disable GPU unswizzle - item.disable() - } - - settingsViewModel.setAdapterItemChanged(position) - settingsViewModel.setShouldReloadSettingsList(true) - dialog.dismiss() - } - - // Ensure filters are cleared after dialog is shown - binding.root.post { - textureSizeAdapter.filter.filter(null) - streamSizeAdapter.filter.filter(null) - chunkSizeAdapter.filter.filter(null) - } - - return dialog - } - - private fun updateDropdownsState(enabled: Boolean) { - binding.layoutTextureSize.isEnabled = enabled - binding.dropdownTextureSize.isEnabled = enabled - binding.layoutStreamSize.isEnabled = enabled - binding.dropdownStreamSize.isEnabled = enabled - binding.layoutChunkSize.isEnabled = enabled - binding.dropdownChunkSize.isEnabled = enabled - } - - companion object { - const val TAG = "GpuUnswizzleDialogFragment" - const val POSITION = "Position" - - fun newInstance( - settingsViewModel: SettingsViewModel, - item: GpuUnswizzleSetting, - position: Int - ): GpuUnswizzleDialogFragment { - val dialog = GpuUnswizzleDialogFragment() - val args = Bundle() - args.putInt(POSITION, position) - dialog.arguments = args - settingsViewModel.clickedItem = item - return dialog - } - } -} diff --git a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsAdapter.kt b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsAdapter.kt index 8d47db94e0..c84d0160f5 100644 --- a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsAdapter.kt +++ b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsAdapter.kt @@ -106,10 +106,6 @@ class SettingsAdapter( PathViewHolder(ListItemSettingBinding.inflate(inflater), this) } - SettingsItem.TYPE_GPU_UNSWIZZLE -> { - GpuUnswizzleViewHolder(ListItemSettingBinding.inflate(inflater), this) - } - SettingsItem.TYPE_FX_TOOLBAR -> { FxToolbarViewHolder( ListItemSettingFxToolbarBinding.inflate(inflater, parent, false), @@ -511,14 +507,6 @@ class SettingsAdapter( settingsViewModel.setShouldShowPathResetDialog(true) } - fun onGpuUnswizzleClick(item: GpuUnswizzleSetting, position: Int) { - GpuUnswizzleDialogFragment.newInstance( - settingsViewModel, - item, - position - ).show(fragment.childFragmentManager, GpuUnswizzleDialogFragment.TAG) - } - private class DiffCallback : DiffUtil.ItemCallback() { override fun areItemsTheSame(oldItem: SettingsItem, newItem: SettingsItem): Boolean { return oldItem.setting.key == newItem.setting.key diff --git a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsFragmentPresenter.kt b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsFragmentPresenter.kt index ca282ba159..53e0e3a3d7 100644 --- a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsFragmentPresenter.kt +++ b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsFragmentPresenter.kt @@ -561,7 +561,6 @@ class SettingsFragmentPresenter( add(IntSetting.ANDROID_PIPELINE_WORKERS.key) add(BooleanSetting.RENDERER_ASYNCHRONOUS_GPU_EMULATION.key) add(BooleanSetting.RENDERER_ASYNC_PRESENTATION.key) - add(SettingsItem.GPU_UNSWIZZLE_COMBINED) add(HeaderSetting(R.string.extensions)) diff --git a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/viewholder/GpuUnswizzleViewHolder.kt b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/viewholder/GpuUnswizzleViewHolder.kt deleted file mode 100644 index 06701eb077..0000000000 --- a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/viewholder/GpuUnswizzleViewHolder.kt +++ /dev/null @@ -1,71 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project -// SPDX-License-Identifier: GPL-3.0-or-later - -package org.yuzu.yuzu_emu.features.settings.ui.viewholder - -import android.view.View -import org.yuzu.yuzu_emu.R -import org.yuzu.yuzu_emu.databinding.ListItemSettingBinding -import org.yuzu.yuzu_emu.features.settings.model.view.GpuUnswizzleSetting -import org.yuzu.yuzu_emu.features.settings.model.view.SettingsItem -import org.yuzu.yuzu_emu.features.settings.ui.SettingsAdapter -import org.yuzu.yuzu_emu.utils.ViewUtils.setVisible - -class GpuUnswizzleViewHolder(val binding: ListItemSettingBinding, adapter: SettingsAdapter) : - SettingViewHolder(binding.root, adapter) { - private lateinit var setting: GpuUnswizzleSetting - - override fun bind(item: SettingsItem) { - setting = item as GpuUnswizzleSetting - binding.textSettingName.text = setting.title - binding.textSettingDescription.setVisible(item.description.isNotEmpty()) - binding.textSettingDescription.text = item.description - - binding.textSettingValue.setVisible(true) - val resMgr = binding.root.context.resources - - if (setting.isEnabled()) { - // Show a summary of current settings - val textureSizeEntries = resMgr.getStringArray(setting.textureSizeChoicesId) - val textureSizeValues = resMgr.getIntArray(setting.textureSizeValuesId) - val textureSizeIndex = textureSizeValues.indexOf(setting.getTextureSize()) - val textureSizeLabel = if (textureSizeIndex >= 0) textureSizeEntries[textureSizeIndex] else "?" - - val streamSizeEntries = resMgr.getStringArray(setting.streamSizeChoicesId) - val streamSizeValues = resMgr.getIntArray(setting.streamSizeValuesId) - val streamSizeIndex = streamSizeValues.indexOf(setting.getStreamSize()) - val streamSizeLabel = if (streamSizeIndex >= 0) streamSizeEntries[streamSizeIndex] else "?" - - val chunkSizeEntries = resMgr.getStringArray(setting.chunkSizeChoicesId) - val chunkSizeValues = resMgr.getIntArray(setting.chunkSizeValuesId) - val chunkSizeIndex = chunkSizeValues.indexOf(setting.getChunkSize()) - val chunkSizeLabel = if (chunkSizeIndex >= 0) chunkSizeEntries[chunkSizeIndex] else "?" - - binding.textSettingValue.text = "$textureSizeLabel ⋅ $streamSizeLabel ⋅ $chunkSizeLabel" - } else { - binding.textSettingValue.text = resMgr.getString(R.string.gpu_unswizzle_disabled) - } - - binding.buttonClear.setVisible(setting.clearable) - binding.buttonClear.setOnClickListener { - adapter.onClearClick(setting, bindingAdapterPosition) - } - - setStyle(setting.isEditable, binding) - } - - override fun onClick(clicked: View) { - if (!setting.isEditable) { - return - } - - adapter.onGpuUnswizzleClick(setting, bindingAdapterPosition) - } - - override fun onLongClick(clicked: View): Boolean { - if (setting.isEditable) { - return adapter.onLongClick(setting, bindingAdapterPosition) - } - return false - } -} diff --git a/src/android/app/src/main/res/layout/dialog_gpu_unswizzle.xml b/src/android/app/src/main/res/layout/dialog_gpu_unswizzle.xml deleted file mode 100644 index 8299551618..0000000000 --- a/src/android/app/src/main/res/layout/dialog_gpu_unswizzle.xml +++ /dev/null @@ -1,96 +0,0 @@ - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - diff --git a/src/android/app/src/main/res/values-ar/strings.xml b/src/android/app/src/main/res/values-ar/strings.xml index 44f86d380f..f199e42efa 100644 --- a/src/android/app/src/main/res/values-ar/strings.xml +++ b/src/android/app/src/main/res/values-ar/strings.xml @@ -599,17 +599,6 @@ يُمكّن هذا الخيار من التعامل مع عملية إعادة تحجيم الألعاب بطريقة تقليدية باستخدام مسار إعادة التحجيم السريع استخدم تظليل غير متزامن يقوم بتجميع التظليل بشكل غير متزامن. قد يقلل ذلك من التقطعات ولكنه قد يؤدي أيضًا إلى حدوث أخطاء. - إعدادات إلغاء ترتيب بيانات وحدة معالجة الرسومات - قم بضبط معلمات فكّ تشابك النسيج المستندة إلى وحدة معالجة الرسومات أو تعطيلها تمامًا. اضبط هذه الإعدادات لتحقيق التوازن بين الأداء وجودة تحميل النسيج. - تفعيل إلغاء ترتيب بيانات وحدة معالجة الرسومات - تعطيل - الحد الأقصى لحجم النسيج في وحدة معالجة الرسومات بعد إعادة ترتيب البيانات - يُحدد هذا الخيار الحد الأقصى لحجم (ميغابايت) معالجة الصور باستخدام وحدة معالجة الرسومات. مع أن وحدة معالجة الرسومات أسرع في معالجة الصور المتوسطة والكبيرة، إلا أن وحدة المعالجة المركزية قد تكون أكثر كفاءة في معالجة الصور الصغيرة جدًا. اضبط هذا الخيار لتحقيق التوازن الأمثل بين سرعة معالجة الرسومات واستهلاك وحدة المعالجة المركزية. - حجم تدفق إلغاء ترتيب بيانات وحدة معالجة الرسومات - يحدد هذا الخيار حد البيانات لكل إطار لمعالجة الصور الكبيرة. القيم الأعلى تُسرّع تحميل الصور على حساب زيادة زمن استجابة الإطارات؛ أما القيم الأقل فتُقلل من الحمل الزائد على وحدة معالجة الرسومات، ولكنها قد تتسبب في ظهور الصور بشكل مفاجئ. - حجم كتلة إلغاء ترتيب بيانات وحدة معالجة الرسومات - يُحدد هذا الخيار عدد شرائح العمق التي تتم معالجتها لكل دفعة من الصور ثلاثية الأبعاد (3D). زيادة هذا العدد تُحسّن كفاءة الإنتاجية على وحدات معالجة الرسومات القوية، ولكنها قد تُسبب تقطعًا أو انقطاعًا في عمل برنامج التشغيل على الأجهزة ذات المواصفات الأقل قوة. - افتراضي إضافات @@ -1050,25 +1039,10 @@ كسر السرعة - صغير جدًا (16 ميغابايت) - صغير (32 ميغابايت) - قياسي (128 ميغابايت) - كبير (256 ميغابايت) - كبير جدًا (512 ميغابايت) - منخفض جدًا (4 ميغابايت) - منخفض (8 ميغابايت) - قياسي (16 ميغابايت) - متوسط (32 ميغابايت) - عالي (64 ميغابايت) - منخفض جدًا (32) - منخفض (64) - قياسي (128) - متوسط (256) - عالي (512) مئوية diff --git a/src/android/app/src/main/res/values-de/strings.xml b/src/android/app/src/main/res/values-de/strings.xml index 6ed158fc72..307f739870 100644 --- a/src/android/app/src/main/res/values-de/strings.xml +++ b/src/android/app/src/main/res/values-de/strings.xml @@ -508,7 +508,6 @@ Wird der Handheld-Modus verwendet, verringert es die Auflösung und erhöht die Überspringt bestimmte Cache-Invalidierungen auf CPU-Seite während Speicherupdates, reduziert die CPU-Auslastung und verbessert die Leistung. Kann in einigen Spielen zu Fehlern oder Abstürzen führen. Asynchrone Shader Kompiliert Shader asynchron. Dies kann Ruckler reduzieren, aber auch Grafikfehler verursachen. - Standard Erweiterungen @@ -918,25 +917,10 @@ Wirklich fortfahren? Übertaktung - Sehr klein (16 MB) - Klein (32 MB) - Normal (128 MB) - Groß (256 MB) - Sehr groß (512 MB) - Sehr niedrig (4 MB) - Niedrig (8 MB) - Normal (16 MB) - Mittel (32 MB) - Hoch (64 MB) - Sehr niedrig (32) - Niedrig (64) - Normal (128) - Mittel (256) - Hoch (512) Celsius diff --git a/src/android/app/src/main/res/values-es/strings.xml b/src/android/app/src/main/res/values-es/strings.xml index 448d55b92b..1f7faad22c 100644 --- a/src/android/app/src/main/res/values-es/strings.xml +++ b/src/android/app/src/main/res/values-es/strings.xml @@ -541,17 +541,6 @@ Permite el manejo de versiones anteriores para el paso de configuración de reescalado para juegos mediante el uso de una ruta de reescalado rápida. Usar sombreadores asíncronos Compila los sombreadores de forma asíncrona. Esto puede reducir los tirones, pero también puede introducir errores gráficos. - Ajustes de desentrelazado de la GPU - Configura los parámetros de desentrelazado de texturas basadas ​​en la GPU o desactívelos por completo. Modifique estos ajustes para equilibrar el rendimiento y la calidad de las texturas cargadas. - Activar desentrelazado de la GPU - Desactivado - Tamaño máximo de textura de desentrelazado de la GPU - Establece el tamaño máximo (en MB) para el desentrelazado de texturas basada en GPU. Aunque la GPU es más rápida para texturas medianas y grandes, la CPU puede ser más eficiente para texturas muy pequeñas. Ajuste este valor para encontrar el equilibrio entre la aceleración de la GPU y la sobrecarga de la CPU. - Tamaño del flujo de desentrelazado de la GPU - Establece el límite de datos por fotograma para desentrelazar texturas grandes. Los valores altos aceleran la carga de texturas, a coste de una mayor latencia por fotograma; los valores bajos reducen la carga de la GPU, pero pueden causar parpadeos visibles en las texturas. - Tamaño del trozo de desentrelazado de la GPU - Determina la cantidad de cortes de profundidad procesados ​​en un solo envío de texturas 3D. Aumentar este valor puede mejorar el rendimiento en una GPU de gama alta, pero puede causar tirones y problemas en los tiempos de respuesta en hardware más modesto. - Por defecto Extensiones @@ -988,25 +977,10 @@ Overclock - Muy pequeño (16 MB) - Pequeño (32 MB) - Normal (128 MB) - Grande (256 MB) - Muy grande (512 MB) - Muy bajo (4 MB) - Bajo (8 MB) - Normal (16 MB) - Medio (32 MB) - Alto (64 MB) - Muy bajo (32) - Bajo (64) - Normal (128) - Medio (256) - Alto (512) Celsius diff --git a/src/android/app/src/main/res/values-fr/strings.xml b/src/android/app/src/main/res/values-fr/strings.xml index 7507051dd7..a73d1ac9fb 100644 --- a/src/android/app/src/main/res/values-fr/strings.xml +++ b/src/android/app/src/main/res/values-fr/strings.xml @@ -482,8 +482,6 @@ Emuler BGR565 Utiliser les shaders asynchrones Compile les shaders de manière asynchrone. Cela peut réduire les saccades mais peut aussi provoquer des problèmes graphiques. - Désactivé - Par défaut Extensions @@ -884,25 +882,10 @@ 8 Go (Dangereux) - Très petit (16 Mo) - Petit (32 Mo) - Normal (128 Mo) - Large (256 Mo) - Très large (512 Mo) - Très faible (4 Mo) - Faible (8 Mo) - Normal (16 Mo) - Moyen (32 Mo) - Élevé (64 Mo) - Très faible (32) - Faible (64) - Normal (128) - Moyen (256) - Élevé (512) Celsius diff --git a/src/android/app/src/main/res/values-ru/strings.xml b/src/android/app/src/main/res/values-ru/strings.xml index 7a4b6cf91e..a839b3bfbc 100644 --- a/src/android/app/src/main/res/values-ru/strings.xml +++ b/src/android/app/src/main/res/values-ru/strings.xml @@ -579,17 +579,6 @@ Включает старый метод обработки этапа перенастройки масштабирования для игр за счёт использования быстрого алгоритма перемасштабирования. Использовать асинхронные шейдеры Компилирует шейдеры асинхронно. Это может уменьшить подтормаживания, но также может вызвать графические артефакты. - Настройки распаковки текстур (Unswizzle) - Настройте параметры распаковки текстур на стороне ГПУ либо полностью отключите эту функцию. Изменение этих параметров позволяет найти баланс между производительностью и качеством загрузки текстур. - Включить распаковку текстур (Unswizzle) - Отключено - Макс. размер текстуры Unswizzle - Задает максимальный размер (в МБ) текстур для преобразования формата (unswizzle) на ГПУ. Хотя ГПУ быстрее работает со средними и большими текстурами, ЦП может быть эффективнее для очень маленьких. Настройте это значение, чтобы найти баланс между ускорением на ГПУ и нагрузкой на ЦП. - Размер потока Unswizzle - Задает лимит данных на кадр для преобразования крупных текстур (unswizzle). Высокие значения ускоряют загрузку текстур ценой увеличения задержки кадра; низкие значения снижают нагрузку на ГПУ, но могут вызывать заметную постепенную подгрузку текстур. - Размер блока Unswizzle - Задает количество слоёв глубины, обрабатываемых за одну пачку для 3D-текстур. Увеличение этого значения улучшает пропускную способность на мощных ГПУ, но может вызывать подтормаживания или таймауты драйвера на слабом железе. - По умолчанию Расширения @@ -1030,25 +1019,10 @@ Разгон - Очень малый (16 МБ) - Малый (32 МБ) - Обычный (128 МБ) - Большой (256 МБ) - Очень большой (512 МБ) - Очень низкий (4 МБ) - Низкий (8 МБ) - Обычный (16 МБ) - Средний (32 МБ) - Высокий (64 МБ) - Очень малый (32) - Малый (64) - Обычный (128) - Средний (256) - Большой (512) Цельсий diff --git a/src/android/app/src/main/res/values-uk/strings.xml b/src/android/app/src/main/res/values-uk/strings.xml index 9923fc72c4..3f2cb94fb2 100644 --- a/src/android/app/src/main/res/values-uk/strings.xml +++ b/src/android/app/src/main/res/values-uk/strings.xml @@ -502,17 +502,6 @@ Вмикає застарілу обробку масштабування для ігор, використовуючи швидкий шлях масштабування Асинхронні шейдери Компілює шейдери асинхронно. Це може зменшити затримки, але також може спричинити графічні баги. - Налаштування розпакування за допомогою ГП - Налаштуйте розпакування текстур за допомогою ГП або повністю вимкнути його. Відкоригуйте ці налаштування, щоб урівноважити продуктивність і якість завантаження текстур. - Увімкнути розпакування за допомогою ГП - Вимкнено - Максимальний розмір текстур для розпакування ГП за допомогою ГП - Встановлює максимальний розмір (МБ) для розпакування текстур за допомогою ГП. ГП швидше справляється з текстурами середніх і великих розмірів, а ЦП ефективніший для дуже маленьких. Налаштуйте, щоб збалансувати ГП-прискоренням і навантаженням на ЦП. - Розмір потоку розпакування за допомогою ГП - Встановлює обмеження даних на кадр для розпакування великих текстур. Вищі значення пришвидшують завантаження текстур за рахунок більших кадрових затримок; менші значення зменшують перевантаження ГП але може спричинити помітні появи текстур. - Розмір блоків розпакування за допомогою ГП - Визначає кількість зрізів глибини, оброблених за партію 3D-текстур. Збільшення здатне покращити пропускну здатність на потужних ГП, але може призвести до затримок або затримок драйвера зі слабшим устаткуванням. - Стандартно Розширення @@ -929,25 +918,10 @@ 8 ГБ (Небезпечно) - Дуже малий (16 МБ) - Малий (32 МБ) - Нормальний (128 МБ) - Великий (256 МБ) - Дуже великий (512 МБ) - Дуже низький (4 МБ) - Низький (8 МБ) - Нормальний (16 МБ) - Середній (32 МБ) - Високий (64 МБ) - Дуже низький (32) - Низький (64) - Нормальний (128) - Середній (256) - Високий (512) Цельсій diff --git a/src/android/app/src/main/res/values-zh-rCN/strings.xml b/src/android/app/src/main/res/values-zh-rCN/strings.xml index 193d00c3d1..3cdb7278b3 100644 --- a/src/android/app/src/main/res/values-zh-rCN/strings.xml +++ b/src/android/app/src/main/res/values-zh-rCN/strings.xml @@ -560,17 +560,6 @@ 启用通过使用快速缩放路径，来为游戏提供缩放配置处理的传统处理方式 使用异步着色器 以异步方式编译着色器。采用此方式或可减少卡顿，但也可能引入故障点。 - GPU Unswizzle 设置 - 配置基于 GPU 的纹理 unswizzling 参数，或完全禁用该功能。通过调整这些设置，以尝试在性能与纹理加载质量之间取得平衡。 - 启用 GPU Unswizzle - 禁用 - GPU Unswizzle 最大纹理尺寸 - 设置基于 GPU 的纹理 unswizzling 的最大尺寸（MB）。虽然 GPU 处理中等和大型纹理的速度更快，但对于非常小的纹理，CPU 可能更为高效。通过调节此项设置，以平衡GPU 加速与 CPU 开销。 - GPU Unswizzle 流大小 - 设置用于 unswizzling 大型纹理时的每帧数据限制。较高的数值可以加速纹理的加载过程，但会带来更高的帧延迟。而较低的数值则可以降低 GPU 的开销，但也可能会导致可见的纹理闪现。 - GPU Unswizzle 块大小 - 定义了 3D 纹理每批次处理的深度切片数量。增加此数值可在高性能 GPU 上提升吞吐效率，但在性能较弱的硬件上可能会导致卡顿或驱动超时。 - 默认 扩展 @@ -1004,25 +993,10 @@ 超频 - 极小 (16 MB) - 较小 (32 MB) - 正常 (128 MB) - 较大 (256 MB) - 极大 (512 MB) - 极低 (4 MB) - 低 (8 MB) - 正常 (16 MB) - 中 (32 MB) - 高 (64 MB) - 极低 (32) - 低 (64) - 正常 (128) - 中 (256) - 高 (512) 摄氏度 diff --git a/src/android/app/src/main/res/values-zh-rTW/strings.xml b/src/android/app/src/main/res/values-zh-rTW/strings.xml index 0a1c3ba4c2..fa59e2a9b7 100644 --- a/src/android/app/src/main/res/values-zh-rTW/strings.xml +++ b/src/android/app/src/main/res/values-zh-rTW/strings.xml @@ -580,17 +580,6 @@ 透過快速重新縮放處理來啟用遊戲重新縮放設定階段的舊版處理方式 使用非同步著色器 使用非同步編譯著色器。這可能會減少卡頓，但也可能導致圖形錯誤。 - GPU Unswizzle 設定 - 設定基於 GPU 的 Unswizzling 參數或完全停用此功能。調整此設定以在性能與紋理載入品質中取得平衡 - 啟用 GPU Unswizzle - 停用 - GPU Unswizzle 最大紋理尺寸 - 設定 GPU 紋理 Unswizzling 的最大值 （MB）。雖然 GPU 處理中型和大型紋理的速度較快，但 CPU 處理非常小的紋理可能更有效率。整此設定以在 GPU 加速與 CPU 負載之間取得平衡 - GPU Unswizzle 流大小 - 設定每個影格處理大型紋理 Unswizzling 的資料上限。數值越高載入紋理的速度越快，但會增加畫面延遲；數值較低則能降低 GPU 負載，不過有可能導致至紋理載入不完全甚至突然出現 - GPU Unswizzle 塊大小 - 設定 3D 紋理每批次處理的深度切片數量。增加此數值可提升規格較佳 GPU 的處理效率，但在較低規格的硬體上可能導致卡頓或驅動程式逾時 - 預設 擴充功能 diff --git a/src/android/app/src/main/res/values/arrays.xml b/src/android/app/src/main/res/values/arrays.xml index b3aed5758c..63352386a4 100644 --- a/src/android/app/src/main/res/values/arrays.xml +++ b/src/android/app/src/main/res/values/arrays.xml @@ -591,54 +591,6 @@ 2 - - @string/gpu_texturesizeswizzle_verysmall - @string/gpu_texturesizeswizzle_small - @string/gpu_texturesizeswizzle_normal - @string/gpu_texturesizeswizzle_large - @string/gpu_texturesizeswizzle_verylarge - - - - 0 - 1 - 2 - 3 - 4 - - - - @string/gpu_swizzle_verylow - @string/gpu_swizzle_low - @string/gpu_swizzle_normal - @string/gpu_swizzle_medium - @string/gpu_swizzle_high - - - - 0 - 1 - 2 - 3 - 4 - - - - @string/gpu_swizzlechunk_verylow - @string/gpu_swizzlechunk_low - @string/gpu_swizzlechunk_normal - @string/gpu_swizzlechunk_medium - @string/gpu_swizzlechunk_high - - - - 0 - 1 - 2 - 3 - 4 - - @string/temperature_celsius @string/temperature_fahrenheit diff --git a/src/android/app/src/main/res/values/strings.xml b/src/android/app/src/main/res/values/strings.xml index 730f6bbade..9b941608a1 100644 --- a/src/android/app/src/main/res/values/strings.xml +++ b/src/android/app/src/main/res/values/strings.xml @@ -605,17 +605,6 @@ Enables a legacy handling for the rescale configuration pass for games by using a quick rescale path Use asynchronous shaders Compiles shaders asynchronously. This may reduce stutters but may also introduce glitches. - GPU Unswizzle Settings - Configure GPU-based texture unswizzling parameters or disable it entirely. Adjust these settings to balance performance and texture loading quality. - Enable GPU Unswizzle - Disabled - GPU Unswizzle Max Texture Size - Sets the maximum size (MB) for GPU-based texture unswizzling. While the GPU is faster for medium and large textures, the CPU may be more efficient for very small ones. Adjust this to find the balance between GPU acceleration and CPU overhead. - GPU Unswizzle Stream Size - Sets the data limit per frame for unswizzling large textures. Higher values speed up texture loading at the cost of higher frame latency; lower values reduce GPU overhead but may cause visible texture pop-in. - GPU Unswizzle Chunk Size - Defines the number of depth slices processed per batch for 3D textures. Increasing this improves throughput efficiency on powerful GPUs but may cause stuttering or driver timeouts on weaker hardware. - Default Extensions @@ -1065,25 +1054,10 @@ Overclock - Very Small (16 MB) - Small (32 MB) - Normal (128 MB) - Large (256 MB) - Very Large (512 MB) - Very Low (4 MB) - Low (8 MB) - Normal (16 MB) - Medium (32 MB) - High (64 MB) - Very Low (32) - Low (64) - Normal (128) - Medium (256) - High (512) Celsius diff --git a/src/common/settings.h b/src/common/settings.h index c1c2d4f0df..48d4ca427d 100644 --- a/src/common/settings.h +++ b/src/common/settings.h @@ -666,27 +666,6 @@ struct Values { SwitchableSetting use_asynchronous_shaders{linkage, false, "use_asynchronous_shaders", Category::RendererHacks}; - SwitchableSetting gpu_unswizzle_texture_size{linkage, - GpuUnswizzleSize::Large, - "gpu_unswizzle_texture_size", - Category::RendererHacks, - Specialization::Default}; - - SwitchableSetting gpu_unswizzle_stream_size{linkage, - GpuUnswizzle::Medium, - "gpu_unswizzle_stream_size", - Category::RendererHacks, - Specialization::Default}; - - SwitchableSetting gpu_unswizzle_chunk_size{linkage, - GpuUnswizzleChunk::Medium, - "gpu_unswizzle_chunk_size", - Category::RendererHacks, - Specialization::Default}; - - SwitchableSetting gpu_unswizzle_enabled{linkage, false, "gpu_unswizzle_enabled", - Category::RendererHacks}; - SwitchableSetting dyna_state{linkage, #if defined(__ANDROID__) ExtendedDynamicState::Disabled, diff --git a/src/common/settings_enums.h b/src/common/settings_enums.h index 4b406f4c35..1d7c25cf09 100644 --- a/src/common/settings_enums.h +++ b/src/common/settings_enums.h @@ -153,9 +153,6 @@ ENUM(ConsoleMode, Handheld, Docked); ENUM(AppletMode, HLE, LLE); ENUM(SpirvOptimizeMode, Never, OnLoad, Always); ENUM(GpuClock, Normal, Boost, Overclock) -ENUM(GpuUnswizzleSize, VerySmall, Small, Normal, Large, VeryLarge) -ENUM(GpuUnswizzle, VeryLow, Low, Normal, Medium, High) -ENUM(GpuUnswizzleChunk, VeryLow, Low, Normal, Medium, High) ENUM(TemperatureUnits, Celsius, Fahrenheit) ENUM(ExtendedDynamicState, Disabled, EDS1, EDS2, EDS3); ENUM(GpuLogLevel, Off, Errors, Standard, Verbose, All) diff --git a/src/qt_common/config/shared_translation.cpp b/src/qt_common/config/shared_translation.cpp index 4ef97c602a..96e569aa21 100644 --- a/src/qt_common/config/shared_translation.cpp +++ b/src/qt_common/config/shared_translation.cpp @@ -234,23 +234,6 @@ std::unique_ptr InitializeTranslations(QObject* parent) { INSERT(Settings, gpu_clock, tr("GPU Clocks"), tr("Makes the game believe GPU work finishes faster than it does, so it stops lowering " "resolution and render distance to fit the Switch's clocks.")); - INSERT(Settings, gpu_unswizzle_enabled, tr("GPU Unswizzle"), - tr("Accelerates BCn 3D texture decoding using GPU compute.\n" - "Disable if experiencing crashes or graphical glitches.")); - INSERT(Settings, gpu_unswizzle_texture_size, tr("GPU Unswizzle Max Texture Size"), - tr("Sets the maximum size (MiB) for GPU-based texture unswizzling.\n" - "While the GPU is faster for medium and large textures, the CPU may be more " - "efficient for very small ones.\n" - "Adjust this to find the balance between GPU acceleration and CPU overhead.")); - INSERT(Settings, gpu_unswizzle_stream_size, tr("GPU Unswizzle Stream Size"), - tr("Sets the maximum amount of texture data (in MiB) processed per frame.\n" - "Higher values can reduce stutter during texture loading but may impact frame " - "consistency.")); - INSERT(Settings, gpu_unswizzle_chunk_size, tr("GPU Unswizzle Chunk Size"), - tr("Determines the number of depth slices processed in a single dispatch.\n" - "Increasing this can improve throughput on high-end GPUs but may cause TDR or driver " - "timeouts on weaker hardware.")); - INSERT(Settings, use_vulkan_driver_pipeline_cache, tr("Use Vulkan pipeline cache"), tr("Enables GPU vendor-specific pipeline cache.\nThis option can improve shader loading " "time significantly in cases where the Vulkan driver does not store pipeline cache " @@ -654,30 +637,6 @@ std::unique_ptr ComboboxEnumeration(QObject* parent) { PAIR(GpuClock, Boost, tr("Boost")), PAIR(GpuClock, Overclock, tr("Overclock")), }}); - translations->insert({Settings::EnumMetadata::Index(), - { - PAIR(GpuUnswizzleSize, VerySmall, tr("Very Small (16 MB)")), - PAIR(GpuUnswizzleSize, Small, tr("Small (32 MB)")), - PAIR(GpuUnswizzleSize, Normal, tr("Normal (128 MB)")), - PAIR(GpuUnswizzleSize, Large, tr("Large (256 MB)")), - PAIR(GpuUnswizzleSize, VeryLarge, tr("Very Large (512 MB)")), - }}); - translations->insert({Settings::EnumMetadata::Index(), - { - PAIR(GpuUnswizzle, VeryLow, tr("Very Low (4 MB)")), - PAIR(GpuUnswizzle, Low, tr("Low (8 MB)")), - PAIR(GpuUnswizzle, Normal, tr("Normal (16 MB)")), - PAIR(GpuUnswizzle, Medium, tr("Medium (32 MB)")), - PAIR(GpuUnswizzle, High, tr("High (64 MB)")), - }}); - translations->insert({Settings::EnumMetadata::Index(), - { - PAIR(GpuUnswizzleChunk, VeryLow, tr("Very Low (32)")), - PAIR(GpuUnswizzleChunk, Low, tr("Low (64)")), - PAIR(GpuUnswizzleChunk, Normal, tr("Normal (128)")), - PAIR(GpuUnswizzleChunk, Medium, tr("Medium (256)")), - PAIR(GpuUnswizzleChunk, High, tr("High (512)")), - }}); translations->insert({Settings::EnumMetadata::Index(), { diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt index c4e7ac3f6f..a115bc7b64 100644 --- a/src/video_core/host_shaders/CMakeLists.txt +++ b/src/video_core/host_shaders/CMakeLists.txt @@ -22,7 +22,6 @@ set(SHADER_FILES ${CMAKE_CURRENT_SOURCE_DIR}/blit_depth_msaa.frag ${CMAKE_CURRENT_SOURCE_DIR}/blit_depth_stencil_msaa.frag ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_3d.comp - ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_3d_bcn.comp ${CMAKE_CURRENT_SOURCE_DIR}/convert_abgr8_to_d24s8.frag ${CMAKE_CURRENT_SOURCE_DIR}/convert_abgr8_to_d32f.frag ${CMAKE_CURRENT_SOURCE_DIR}/convert_d32f_to_abgr8.frag diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp index e059c683ff..b7797e3165 100644 --- a/src/video_core/host_shaders/astc_decoder.comp +++ b/src/video_core/host_shaders/astc_decoder.comp @@ -165,7 +165,26 @@ uint FastReplicateTo8(uint value, uint num_bits) { } uint FastReplicateTo6(uint value, uint num_bits) { - return ReplicateBits(value, num_bits, 6); + if (num_bits == 0) { + return 0; + } + if (num_bits >= 6) { + return value; + } + const uint v = value & ((1u << num_bits) - 1u); + if (num_bits == 1) { + return v * 63u; + } + if (num_bits == 2) { + return v * 21u; + } + if (num_bits == 3) { + return v * 9u; + } + if (num_bits == 4) { + return (v << 2u) | (v >> 2u); + } + return (v << 1u) | (v >> 4u); } uint Div3Floor(uint v) { @@ -958,35 +977,71 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, ui } uint UnquantizeTexelWeight(EncodingData val) { - uint encoding = Encoding(val), bitlen = NumBits(val), bitval = BitValue(val); - if (encoding == JUST_BITS) { - return (bitlen >= 1 && bitlen <= 5) - ? uint(floor(0.5f + float(bitval) * 64.0f / float((1 << bitlen) - 1))) - : FastReplicateTo6(bitval, bitlen); - } else if (encoding == TRIT || encoding == QUINT) { - uint B = 0, C = 0, D = 0; - uint b_mask = (0x3100 >> (bitlen * 4)) & 0xf; - uint b = (bitval >> 1) & b_mask; + const uint encoding = Encoding(val); + const uint bitlen = NumBits(val); + const uint bitval = BitValue(val); + const uint A = ReplicateBitTo7((bitval & 1)); + uint B = 0, C = 0, D = 0; + uint result = 0; + const uint bitlen_0_results[5] = {0, 16, 32, 48, 64}; + switch (encoding) { + case JUST_BITS: + result = FastReplicateTo6(bitval, bitlen); + break; + case TRIT: { D = QuintTritValue(val); - if (encoding == TRIT) { - switch (bitlen) { - case 0: return D * 32; //0,32,64 - case 1: C = 50; break; - case 2: C = 23; B = (b << 6) | (b << 2) | b; break; - case 3: C = 11; B = (b << 5) | b; break; - } - } else if (encoding == QUINT) { - switch (bitlen) { - case 0: return D * 16; //0, 16, 32, 48, 64 - case 1: C = 28; break; - case 2: C = 13; B = (b << 6) | (b << 1); break; - } + switch (bitlen) { + case 0: + return bitlen_0_results[D * 2]; + case 1: { + C = 50; + break; } - uint A = ReplicateBitTo7(bitval & 1); - uint res = (A & 0x20) | (((D * C + B) ^ A) >> 2); - return res + (res > 32 ? 1 : 0); + case 2: { + C = 23; + const uint b = (bitval >> 1) & 1; + B = (b << 6) | (b << 2) | b; + break; + } + case 3: { + C = 11; + const uint cb = (bitval >> 1) & 3; + B = (cb << 5) | cb; + break; + } + default: + break; + } + break; } - return 0; + case QUINT: { + D = QuintTritValue(val); + switch (bitlen) { + case 0: + return bitlen_0_results[D]; + case 1: { + C = 28; + break; + } + case 2: { + C = 13; + const uint b = (bitval >> 1) & 1; + B = (b << 6) | (b << 1); + break; + } + } + break; + } + } + if (encoding != JUST_BITS && bitlen > 0) { + result = D * C + B; + result ^= A; + result = (A & 0x20) | (result >> 2); + } + if (result > 32) { + result += 1; + } + return result; } void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) { @@ -1394,11 +1449,10 @@ void DecompressBlock(ivec3 coord) { } uint SwizzleOffset(uvec2 pos) { - return ((pos.x & 32u) << 3u) | - ((pos.y & 6u) << 5u) | - ((pos.x & 16u) << 1u) | - ((pos.y & 1u) << 4u) | - (pos.x & 15u); + const uint x = pos.x; + const uint y = pos.y; + return ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + + ((x % 32) / 16) * 32 + (y % 2) * 16 + (x % 16); } void main() { diff --git a/src/video_core/host_shaders/block_linear_unswizzle_2d.comp b/src/video_core/host_shaders/block_linear_unswizzle_2d.comp index 2a3446ca66..9cb0f2aca2 100644 --- a/src/video_core/host_shaders/block_linear_unswizzle_2d.comp +++ b/src/video_core/host_shaders/block_linear_unswizzle_2d.comp @@ -51,7 +51,7 @@ layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU128 { uvec4 u1 layout(binding = BINDING_OUTPUT_IMAGE) uniform writeonly uimage2DArray output_image; -layout(local_size_x = 32, local_size_y = 32, local_size_z = 1) in; +layout(local_size_x = 32, local_size_y = 8, local_size_z = 1) in; const uint GOB_SIZE_X = 64; const uint GOB_SIZE_Y = 8; @@ -65,19 +65,11 @@ const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_SHI const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1, GOB_SIZE_Y - 1); -uint SwizzleTable(uint pos) { - const uint t[8] = uint[]( - 0x12100200, 0x13110301, 0x16140604, 0x17150705, - 0x1a180a08, 0x1b190b09, 0x1e1c0e0c, 0x1f1d0f0d - ); - const uint i = pos >> 4; - const uint h = (t[i / 4] >> ((i % 4) * 8)) & 0xff; - return (h << 4) | (pos & 0xf); -} - uint SwizzleOffset(uvec2 pos) { pos = pos & SWIZZLE_MASK; - return SwizzleTable(pos.y * 64 + pos.x); + return ((pos.x & 32u) << 3u) | ((pos.y & 6u) << 5u) | + ((pos.x & 16u) << 1u) | ((pos.y & 1u) << 4u) | + (pos.x & 15u); } uvec4 ReadTexel(uint offset) { diff --git a/src/video_core/host_shaders/block_linear_unswizzle_3d.comp b/src/video_core/host_shaders/block_linear_unswizzle_3d.comp index 34e4f93307..dc23b32492 100644 --- a/src/video_core/host_shaders/block_linear_unswizzle_3d.comp +++ b/src/video_core/host_shaders/block_linear_unswizzle_3d.comp @@ -53,7 +53,7 @@ layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU128 { uvec4 u1 layout(binding = BINDING_OUTPUT_IMAGE) uniform writeonly uimage3D output_image; -layout(local_size_x = 16, local_size_y = 8, local_size_z = 8) in; +layout(local_size_x = 16, local_size_y = 8, local_size_z = 2) in; const uint GOB_SIZE_X = 64; const uint GOB_SIZE_Y = 8; @@ -67,19 +67,11 @@ const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_SHI const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1, GOB_SIZE_Y - 1); -uint SwizzleTable(uint pos) { - const uint t[8] = uint[]( - 0x12100200, 0x13110301, 0x16140604, 0x17150705, - 0x1a180a08, 0x1b190b09, 0x1e1c0e0c, 0x1f1d0f0d - ); - const uint i = pos >> 4; - const uint h = (t[i / 4] >> ((i % 4) * 8)) & 0xff; - return (h << 4) | (pos & 0xf); -} - uint SwizzleOffset(uvec2 pos) { pos = pos & SWIZZLE_MASK; - return SwizzleTable(pos.y * 64 + pos.x); + return ((pos.x & 32u) << 3u) | ((pos.y & 6u) << 5u) | + ((pos.x & 16u) << 1u) | ((pos.y & 1u) << 4u) | + (pos.x & 15u); } uvec4 ReadTexel(uint offset) { diff --git a/src/video_core/host_shaders/block_linear_unswizzle_3d_bcn.comp b/src/video_core/host_shaders/block_linear_unswizzle_3d_bcn.comp deleted file mode 100644 index e084837f25..0000000000 --- a/src/video_core/host_shaders/block_linear_unswizzle_3d_bcn.comp +++ /dev/null @@ -1,164 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project -// SPDX-License-Identifier: GPL-3.0-or-later - -#version 430 - -#ifdef VULKAN - #extension GL_EXT_shader_16bit_storage : require - #extension GL_EXT_shader_8bit_storage : require - #define HAS_EXTENDED_TYPES 1 - #define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants { - #define END_PUSH_CONSTANTS }; - #define UNIFORM(n) - #define BINDING_INPUT_BUFFER 0 - #define BINDING_OUTPUT_BUFFER 1 -#else - #extension GL_NV_gpu_shader5 : enable - #ifdef GL_NV_gpu_shader5 - #define HAS_EXTENDED_TYPES 1 - #else - #define HAS_EXTENDED_TYPES 0 - #endif - #define BEGIN_PUSH_CONSTANTS - #define END_PUSH_CONSTANTS - #define UNIFORM(n) layout(location = n) uniform - #define BINDING_INPUT_BUFFER 1 - #define BINDING_OUTPUT_BUFFER 0 -#endif - -// --- Push Constants / Uniforms --- -#ifdef VULKAN -layout(push_constant) uniform PushConstants { - uvec3 blocks_dim; // Offset 0 - uint bytes_per_block_log2; // Offset 12 - - uvec3 origin; // Offset 16 - uint slice_size; // Offset 28 - - uint block_size; // Offset 32 - uint x_shift; // Offset 36 - uint block_height; // Offset 40 - uint block_height_mask; // Offset 44 - - uint block_depth; // Offset 48 - uint block_depth_mask; // Offset 52 - int _pad; // Offset 56 - - ivec3 destination; // Offset 60 -} pc; -#else -BEGIN_PUSH_CONSTANTS - UNIFORM(0) uvec3 origin; - UNIFORM(1) ivec3 destination; - UNIFORM(2) uint bytes_per_block_log2; - UNIFORM(3) uint slice_size; - UNIFORM(4) uint block_size; - UNIFORM(5) uint x_shift; - UNIFORM(6) uint block_height; - UNIFORM(7) uint block_height_mask; - UNIFORM(8) uint block_depth; - UNIFORM(9) uint block_depth_mask; - UNIFORM(10) uvec3 blocks_dim; -END_PUSH_CONSTANTS -#define pc // Map pc prefix to nothing for OpenGL compatibility -#endif - -// --- Buffers --- -#if HAS_EXTENDED_TYPES -layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU8 { uint8_t u8data[]; }; -layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU16 { uint16_t u16data[]; }; -#endif -layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU32 { uint u32data[]; }; -layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU64 { uvec2 u64data[]; }; -layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU128 { uvec4 u128data[]; }; - -layout(binding = BINDING_OUTPUT_BUFFER, std430) writeonly buffer OutputBuffer { - uint out_u32[]; -}; - -// --- Constants --- -layout(local_size_x = 8, local_size_y = 8, local_size_z = 4) in; - -const uint GOB_SIZE_X = 64; -const uint GOB_SIZE_Y = 8; -const uint GOB_SIZE_Z = 1; -const uint GOB_SIZE = GOB_SIZE_X * GOB_SIZE_Y * GOB_SIZE_Z; - -const uint GOB_SIZE_X_SHIFT = 6; -const uint GOB_SIZE_Y_SHIFT = 3; -const uint GOB_SIZE_Z_SHIFT = 0; -const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_SHIFT; -const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1u, GOB_SIZE_Y - 1u); - -uint SwizzleTable(uint pos) { - const uint t[8] = uint[]( - 0x12100200, 0x13110301, 0x16140604, 0x17150705, - 0x1a180a08, 0x1b190b09, 0x1e1c0e0c, 0x1f1d0f0d - ); - const uint i = pos >> 4; - const uint h = (t[i / 4] >> ((i % 4) * 8)) & 0xff; - return (h << 4) | (pos & 0xf); -} - -// --- Helpers --- -uint SwizzleOffset(uvec2 pos) { - pos &= SWIZZLE_MASK; - return SwizzleTable(pos.y * 64u + pos.x); -} - -uvec4 ReadTexel(uint offset) { - uint bpl2 = pc.bytes_per_block_log2; - switch (bpl2) { -#if HAS_EXTENDED_TYPES - case 0u: return uvec4(u8data[offset], 0u, 0u, 0u); - case 1u: return uvec4(u16data[offset / 2u], 0u, 0u, 0u); -#else - case 0u: return uvec4(bitfieldExtract(u32data[offset / 4u], int((offset * 8u) & 24u), 8), 0u, 0u, 0u); - case 1u: return uvec4(bitfieldExtract(u32data[offset / 4u], int((offset * 8u) & 16u), 16), 0u, 0u, 0u); -#endif - case 2u: return uvec4(u32data[offset / 4u], 0u, 0u, 0u); - case 3u: return uvec4(u64data[offset / 8u], 0u, 0u); - case 4u: return u128data[offset / 16u]; - } - return uvec4(0u); -} - -void main() { - uvec3 block_coord = gl_GlobalInvocationID; - if (any(greaterThanEqual(block_coord, pc.blocks_dim))) { - return; - } - - uint bytes_per_block = 1u << pc.bytes_per_block_log2; - // Origin is in pixels, divide by 4 for block-space (e.g. BCn formats) - uvec3 pos; - pos.x = (block_coord.x + (pc.origin.x >> 2u)) * bytes_per_block; - pos.y = block_coord.y + (pc.origin.y >> 2u); - pos.z = block_coord.z + pc.origin.z; - - uint swizzle = SwizzleOffset(pos.xy); - uint block_y = pos.y >> GOB_SIZE_Y_SHIFT; - uint offset = 0u; - // Apply block-linear offsets - offset += (pos.z >> pc.block_depth) * pc.slice_size; - offset += (pos.z & pc.block_depth_mask) << (GOB_SIZE_SHIFT + pc.block_height); - offset += (block_y >> pc.block_height) * pc.block_size; - offset += (block_y & pc.block_height_mask) << GOB_SIZE_SHIFT; - offset += (pos.x >> GOB_SIZE_X_SHIFT) << pc.x_shift; - offset += swizzle; - - uvec4 texel = ReadTexel(offset); - - // Calculate linear output index - uint block_index = block_coord.x + - (block_coord.y * pc.blocks_dim.x) + - (block_coord.z * pc.blocks_dim.x * pc.blocks_dim.y); - uint out_idx = block_index * (bytes_per_block >> 2u); - - out_u32[out_idx] = texel.x; - out_u32[out_idx + 1u] = texel.y; - if (pc.bytes_per_block_log2 == 4u) { - out_u32[out_idx + 2u] = texel.z; - out_u32[out_idx + 3u] = texel.w; - } -} \ No newline at end of file diff --git a/src/video_core/host_shaders/pitch_unswizzle.comp b/src/video_core/host_shaders/pitch_unswizzle.comp index 7d23f594c7..dba874b414 100644 --- a/src/video_core/host_shaders/pitch_unswizzle.comp +++ b/src/video_core/host_shaders/pitch_unswizzle.comp @@ -33,7 +33,7 @@ BEGIN_PUSH_CONSTANTS UNIFORM(0) uvec2 origin; UNIFORM(1) ivec2 destination; -UNIFORM(2) uint bytes_per_block; +UNIFORM(2) uint bytes_per_block_log2; UNIFORM(3) uint pitch; END_PUSH_CONSTANTS @@ -47,26 +47,26 @@ layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU128 { layout(binding = BINDING_OUTPUT_IMAGE) writeonly uniform uimage2D output_image; -layout(local_size_x = 32, local_size_y = 32, local_size_z = 1) in; +layout(local_size_x = 32, local_size_y = 8, local_size_z = 1) in; uvec4 ReadTexel(uint offset) { - switch (bytes_per_block) { + switch (bytes_per_block_log2) { #if HAS_EXTENDED_TYPES - case 1: + case 0: return uvec4(u8data[offset], 0, 0, 0); - case 2: + case 1: return uvec4(u16data[offset / 2], 0, 0, 0); #else - case 1: + case 0: return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 24), 8), 0, 0, 0); - case 2: + case 1: return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 16), 16), 0, 0, 0); #endif - case 4: + case 2: return uvec4(u32data[offset / 4], 0, 0, 0); - case 8: + case 3: return uvec4(u64data[offset / 8], 0, 0); - case 16: + case 4: return u128data[offset / 16]; } return uvec4(0); @@ -76,7 +76,7 @@ void main() { uvec2 pos = gl_GlobalInvocationID.xy + origin; uint offset = 0; - offset += pos.x * bytes_per_block; + offset += pos.x << bytes_per_block_log2; offset += pos.y * pitch; const uvec4 texel = ReadTexel(offset); diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index 680cb2179f..48aa32bc90 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -556,7 +556,7 @@ void TextureCacheRuntime::Finish() { glFinish(); } -StagingBufferMap TextureCacheRuntime::UploadStagingBuffer(size_t size, bool deferred) { +StagingBufferMap TextureCacheRuntime::UploadStagingBuffer(size_t size) { return staging_buffer_pool.RequestUploadBuffer(size); } @@ -651,8 +651,7 @@ void TextureCacheRuntime::BlitFramebuffer(Framebuffer* dst, Framebuffer* src, } void TextureCacheRuntime::AccelerateImageUpload(Image& image, const StagingBufferMap& map, - std::span swizzles, - u32 z_start, u32 z_count) { + std::span swizzles) { switch (image.info.type) { case ImageType::e2D: if (IsPixelFormatASTC(image.info.format)) { diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index 8a830fa0c9..9f3b244a8d 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -77,7 +77,7 @@ public: void FlushDeferredClear() {} - StagingBufferMap UploadStagingBuffer(size_t size, bool deferred = false); + StagingBufferMap UploadStagingBuffer(size_t size); StagingBufferMap DownloadStagingBuffer(size_t size, bool deferred = false); @@ -121,8 +121,7 @@ public: Tegra::Engines::Fermi2D::Operation operation); void AccelerateImageUpload(Image& image, const StagingBufferMap& map, - std::span swizzles, - u32 z_start, u32 z_count); + std::span swizzles); void InsertUploadMemoryBarrier(); @@ -229,8 +228,6 @@ public: bool ScaleDown(bool ignore = false); - u64 allocation_tick; - private: void CopyBufferToImage(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset); diff --git a/src/video_core/renderer_opengl/util_shaders.cpp b/src/video_core/renderer_opengl/util_shaders.cpp index 7ced8c77b4..ef3fc05f2a 100644 --- a/src/video_core/renderer_opengl/util_shaders.cpp +++ b/src/video_core/renderer_opengl/util_shaders.cpp @@ -4,6 +4,7 @@ // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later +#include #include #include @@ -116,7 +117,7 @@ void UtilShaders::ASTCDecode(Image& image, const StagingBufferMap& map, void UtilShaders::BlockLinearUpload2D(Image& image, const StagingBufferMap& map, std::span swizzles) { - static constexpr Extent3D WORKGROUP_SIZE{32, 32, 1}; + static constexpr Extent3D WORKGROUP_SIZE{32, 8, 1}; static constexpr GLuint BINDING_INPUT_BUFFER = 0; static constexpr GLuint BINDING_OUTPUT_IMAGE = 0; @@ -151,7 +152,7 @@ void UtilShaders::BlockLinearUpload2D(Image& image, const StagingBufferMap& map, void UtilShaders::BlockLinearUpload3D(Image& image, const StagingBufferMap& map, std::span swizzles) { - static constexpr Extent3D WORKGROUP_SIZE{16, 8, 8}; + static constexpr Extent3D WORKGROUP_SIZE{16, 8, 2}; static constexpr GLuint BINDING_INPUT_BUFFER = 0; static constexpr GLuint BINDING_OUTPUT_IMAGE = 0; @@ -189,12 +190,12 @@ void UtilShaders::BlockLinearUpload3D(Image& image, const StagingBufferMap& map, void UtilShaders::PitchUpload(Image& image, const StagingBufferMap& map, std::span swizzles) { - static constexpr Extent3D WORKGROUP_SIZE{32, 32, 1}; + static constexpr Extent3D WORKGROUP_SIZE{32, 8, 1}; static constexpr GLuint BINDING_INPUT_BUFFER = 0; static constexpr GLuint BINDING_OUTPUT_IMAGE = 0; static constexpr GLuint LOC_ORIGIN = 0; static constexpr GLuint LOC_DESTINATION = 1; - static constexpr GLuint LOC_BYTES_PER_BLOCK = 2; + static constexpr GLuint LOC_BYTES_PER_BLOCK_LOG2 = 2; static constexpr GLuint LOC_PITCH = 3; const u32 bytes_per_block = BytesPerBlock(image.info.format); @@ -208,7 +209,7 @@ void UtilShaders::PitchUpload(Image& image, const StagingBufferMap& map, glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes); glUniform2ui(LOC_ORIGIN, 0, 0); glUniform2i(LOC_DESTINATION, 0, 0); - glUniform1ui(LOC_BYTES_PER_BLOCK, bytes_per_block); + glUniform1ui(LOC_BYTES_PER_BLOCK_LOG2, static_cast(std::countr_zero(bytes_per_block))); glUniform1ui(LOC_PITCH, pitch); glBindImageTexture(BINDING_OUTPUT_IMAGE, image.StorageHandle(), 0, GL_FALSE, 0, GL_WRITE_ONLY, format); diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index ab58560170..4685c6ae30 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -16,13 +16,16 @@ #include "common/common_types.h" #include "common/div_ceil.h" #include "common/vector_math.h" +#include #include "video_core/host_shaders/astc_decoder_comp_spv.h" #include "video_core/host_shaders/queries_prefix_scan_sum_comp_spv.h" #include "video_core/host_shaders/queries_prefix_scan_sum_nosubgroups_comp_spv.h" #include "video_core/host_shaders/resolve_conditional_render_comp_spv.h" #include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h" #include "video_core/host_shaders/vulkan_uint8_comp_spv.h" -#include "video_core/host_shaders/block_linear_unswizzle_3d_bcn_comp_spv.h" +#include "video_core/host_shaders/block_linear_unswizzle_2d_comp_spv.h" +#include "video_core/host_shaders/block_linear_unswizzle_3d_comp_spv.h" +#include "video_core/host_shaders/pitch_unswizzle_comp_spv.h" #include "video_core/renderer_vulkan/vk_compute_pass.h" #include "video_core/surface.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" @@ -185,6 +188,26 @@ struct AstcPushConstants { u32 block_height_mask; }; +struct BlockLinear3DImagePushConstants { + alignas(16) std::array origin; + alignas(16) std::array destination; + u32 bytes_per_block_log2; + u32 slice_size; + u32 block_size; + u32 x_shift; + u32 block_height; + u32 block_height_mask; + u32 block_depth; + u32 block_depth_mask; +}; + +struct PitchUnswizzlePushConstants { + std::array origin; + std::array destination; + u32 bytes_per_block_log2; + u32 pitch; +}; + struct QueriesPrefixScanPushConstants { u32 min_accumulation_base; u32 max_accumulation_base; @@ -616,260 +639,230 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, scheduler.Finish(); } -constexpr u32 BL3D_BINDING_INPUT_BUFFER = 0; -constexpr u32 BL3D_BINDING_OUTPUT_BUFFER = 1; +namespace { -struct alignas(16) BlockLinearUnswizzle3DPushConstants { - u32 blocks_dim[3]; // Offset 0 - u32 bytes_per_block_log2; // Offset 12 - - u32 origin[3]; // Offset 16 - u32 slice_size; // Offset 28 - - u32 block_size; // Offset 32 - u32 x_shift; // Offset 36 - u32 block_height; // Offset 40 - u32 block_height_mask; // Offset 44 - - u32 block_depth; // Offset 48 - u32 block_depth_mask; // Offset 52 - s32 _pad; // Offset 56 - - s32 destination[3]; // Offset 60 - s32 _pad_end; // Offset 72 -}; -static_assert(sizeof(BlockLinearUnswizzle3DPushConstants) <= 128); - -BlockLinearUnswizzle3DPass::BlockLinearUnswizzle3DPass( - const Device& device_, Scheduler& scheduler_, - DescriptorPool& descriptor_pool_, - StagingBufferPool& staging_buffer_pool_, - ComputePassDescriptorQueue& compute_pass_descriptor_queue_) - : ComputePass(device_, scheduler_, descriptor_pool_, - std::array{{ - { - .binding = BL3D_BINDING_INPUT_BUFFER, - .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, // block-linear input - .descriptorCount = 1, - .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, - .pImmutableSamplers = nullptr, - }, - { - .binding = BL3D_BINDING_OUTPUT_BUFFER, - .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, - .descriptorCount = 1, - .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, - .pImmutableSamplers = nullptr, - }, - }}, - std::array{{ - { - .dstBinding = BL3D_BINDING_INPUT_BUFFER, - .dstArrayElement = 0, - .descriptorCount = 1, - .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, - .offset = BL3D_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry), - .stride = sizeof(DescriptorUpdateEntry), - }, - { - .dstBinding = BL3D_BINDING_OUTPUT_BUFFER, - .dstArrayElement = 0, - .descriptorCount = 1, - .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, - .offset = BL3D_BINDING_OUTPUT_BUFFER * sizeof(DescriptorUpdateEntry), - .stride = sizeof(DescriptorUpdateEntry), - } - }}, - DescriptorBankInfo{ - .uniform_buffers = 0, - .storage_buffers = 2, - .texture_buffers = 0, - .image_buffers = 0, - .textures = 0, - .images = 0, - .score = 2, - }, - COMPUTE_PUSH_CONSTANT_RANGE, - BLOCK_LINEAR_UNSWIZZLE_3D_BCN_COMP_SPV), - scheduler{scheduler_}, - staging_buffer_pool{staging_buffer_pool_}, - compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} - -BlockLinearUnswizzle3DPass::~BlockLinearUnswizzle3DPass() = default; - -// God have mercy on my soul -void BlockLinearUnswizzle3DPass::Unswizzle( - Image& image, - const StagingBufferRef& swizzled, - std::span swizzles, - u32 z_start, u32 z_count) -{ - using namespace VideoCommon::Accelerated; - - const u32 MAX_BATCH_SLICES = (std::min)(z_count, image.info.size.depth); - - // Allocate or grow to cover this batch's slice count - image.AllocateComputeUnswizzleBuffer(MAX_BATCH_SLICES); - - ASSERT(swizzles.size() == 1); - const auto& sw = swizzles[0]; - const auto params = MakeBlockLinearSwizzle3DParams(sw, image.info); - - const u32 blocks_x = (image.info.size.width + 3) / 4; - const u32 blocks_y = (image.info.size.height + 3) / 4; - - scheduler.RequestOutsideRenderPassOperationContext(); - for (u32 z_offset = 0; z_offset < z_count; z_offset += MAX_BATCH_SLICES) { - const u32 current_chunk_slices = (std::min)(MAX_BATCH_SLICES, z_count - z_offset); - const u32 current_z_start = z_start + z_offset; - - UnswizzleChunk(image, swizzled, sw, params, blocks_x, blocks_y, - current_z_start, current_chunk_slices); +void RecordUnswizzleBeginBarrier(Scheduler& scheduler, VkPipeline vk_pipeline, VkImage vk_image, + VkImageAspectFlags aspect_mask, bool is_initialized) { + VkAccessFlags src_access = VK_ACCESS_NONE; + VkImageLayout old_layout = VK_IMAGE_LAYOUT_UNDEFINED; + VkPipelineStageFlags src_stage = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT; + if (is_initialized) { + src_access = VK_ACCESS_SHADER_WRITE_BIT; + old_layout = VK_IMAGE_LAYOUT_GENERAL; + src_stage = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER; } -} - -void BlockLinearUnswizzle3DPass::UnswizzleChunk( - Image& image, - const StagingBufferRef& swizzled, - const VideoCommon::SwizzleParameters& sw, - const BlockLinearSwizzle3DParams& params, - u32 blocks_x, u32 blocks_y, - u32 z_start, u32 z_count) -{ - BlockLinearUnswizzle3DPushConstants pc{}; - pc.origin[0] = params.origin[0]; - pc.origin[1] = params.origin[1]; - pc.origin[2] = z_start; // Current chunk's Z start - - pc.destination[0] = params.destination[0]; - pc.destination[1] = params.destination[1]; - pc.destination[2] = 0; // Shader writes to start of output buffer - - pc.bytes_per_block_log2 = params.bytes_per_block_log2; - pc.slice_size = params.slice_size; - pc.block_size = params.block_size; - pc.x_shift = params.x_shift; - pc.block_height = params.block_height; - pc.block_height_mask = params.block_height_mask; - pc.block_depth = params.block_depth; - pc.block_depth_mask = params.block_depth_mask; - - pc.blocks_dim[0] = blocks_x; - pc.blocks_dim[1] = blocks_y; - pc.blocks_dim[2] = z_count; // Only process the count - - compute_pass_descriptor_queue.Acquire(scheduler, 3); - compute_pass_descriptor_queue.AddBuffer(swizzled.buffer, - sw.buffer_offset + swizzled.offset, - image.guest_size_bytes - sw.buffer_offset); - compute_pass_descriptor_queue.AddBuffer(*image.compute_unswizzle_buffer, 0, - image.compute_unswizzle_buffer_size); - - const void* descriptor_data = compute_pass_descriptor_queue.UpdateData(); - const VkDescriptorSet set = descriptor_allocator.Commit(); - - const u32 gx = Common::DivCeil(blocks_x, 8u); - const u32 gy = Common::DivCeil(blocks_y, 8u); - const u32 gz = Common::DivCeil(z_count, 4u); - - const u32 bytes_per_block = 1u << pc.bytes_per_block_log2; - const VkDeviceSize output_slice_size = - static_cast(blocks_x) * blocks_y * bytes_per_block; - const VkDeviceSize barrier_size = output_slice_size * z_count; - - const bool is_first_chunk = (z_start == 0); - - const VkBuffer out_buffer = *image.compute_unswizzle_buffer; - const VkImage dst_image = image.Handle(); - const VkImageAspectFlags aspect = image.AspectMask(); - const u32 image_width = image.info.size.width; - const u32 image_height = image.info.size.height; - - scheduler.Record([this, set, descriptor_data, pc, gx, gy, gz, z_start, z_count, - barrier_size, is_first_chunk, out_buffer, dst_image, aspect, - image_width, image_height - ](vk::CommandBuffer cmdbuf) { - - if (dst_image == VK_NULL_HANDLE || out_buffer == VK_NULL_HANDLE) { - return; - } - - device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); - cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline); - cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); - cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(pc), &pc); - cmdbuf.Dispatch(gx, gy, gz); - - // Single barrier for compute -> transfer (buffer ready, image transition) - const VkBufferMemoryBarrier buffer_barrier{ - .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .buffer = out_buffer, - .offset = 0, - .size = barrier_size, - }; - - // Image layout transition - const VkImageMemoryBarrier pre_barrier{ + scheduler.Record([vk_pipeline, vk_image, aspect_mask, src_access, old_layout, + src_stage](vk::CommandBuffer cmdbuf) { + const VkImageMemoryBarrier image_barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, .pNext = nullptr, - .srcAccessMask = is_first_chunk ? VkAccessFlags{} : - static_cast(VK_ACCESS_TRANSFER_WRITE_BIT), - .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .oldLayout = is_first_chunk ? VK_IMAGE_LAYOUT_UNDEFINED : - VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, - .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .image = dst_image, - .subresourceRange = {aspect, 0, 1, 0, 1}, - }; - - // Single barrier handles both buffer and image - cmdbuf.PipelineBarrier( - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, - nullptr, buffer_barrier, pre_barrier - ); - - // Copy chunk to correct Z position in image - const VkBufferImageCopy copy{ - .bufferOffset = 0, // Read from start of staging buffer - .bufferRowLength = 0, - .bufferImageHeight = 0, - .imageSubresource = {aspect, 0, 0, 1}, - .imageOffset = {0, 0, static_cast(z_start)}, // Write to correct Z - .imageExtent = {image_width, image_height, z_count}, - }; - cmdbuf.CopyBufferToImage(out_buffer, dst_image, - VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copy); - - // Post-copy transition - const VkImageMemoryBarrier post_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .srcAccessMask = src_access, .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, - .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, + .oldLayout = old_layout, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .image = dst_image, - .subresourceRange = {aspect, 0, 1, 0, 1}, + .image = vk_image, + .subresourceRange{ + .aspectMask = aspect_mask, + .baseMipLevel = 0, + .levelCount = VK_REMAINING_MIP_LEVELS, + .baseArrayLayer = 0, + .layerCount = VK_REMAINING_ARRAY_LAYERS, + }, }; - - cmdbuf.PipelineBarrier( - VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - 0, - nullptr, nullptr, post_barrier - ); + cmdbuf.PipelineBarrier(src_stage, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier); + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline); }); } +void RecordUnswizzleEndBarrier(Scheduler& scheduler, VkImage vk_image, + VkImageAspectFlags aspect_mask) { + scheduler.Record([vk_image, aspect_mask](vk::CommandBuffer cmdbuf) { + const VkImageMemoryBarrier image_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | + VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = vk_image, + .subresourceRange{ + .aspectMask = aspect_mask, + .baseMipLevel = 0, + .levelCount = VK_REMAINING_MIP_LEVELS, + .baseArrayLayer = 0, + .layerCount = VK_REMAINING_ARRAY_LAYERS, + }, + }; + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, + vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, 0, image_barrier); + }); +} + +} // Anonymous namespace + +BlockLinearUnswizzleImage2DPass::BlockLinearUnswizzleImage2DPass( + const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_) + : ComputePass(device_, scheduler_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, + ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, + COMPUTE_PUSH_CONSTANT_RANGE, + BLOCK_LINEAR_UNSWIZZLE_2D_COMP_SPV), + scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, + compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} + +BlockLinearUnswizzleImage2DPass::~BlockLinearUnswizzleImage2DPass() = default; + +void BlockLinearUnswizzleImage2DPass::Unswizzle( + Image& image, const StagingBufferRef& map, + std::span swizzles) { + using namespace VideoCommon::Accelerated; + scheduler.RequestOutsideRenderPassOperationContext(); + const VkImageAspectFlags aspect_mask = image.AspectMask(); + const VkImage vk_image = image.Handle(); + RecordUnswizzleBeginBarrier(scheduler, *pipeline, vk_image, aspect_mask, + image.ExchangeInitialization()); + for (const VideoCommon::SwizzleParameters& swizzle : swizzles) { + const size_t input_offset = swizzle.buffer_offset + map.offset; + const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U); + const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U); + const u32 num_dispatches_z = image.info.resources.layers; + + compute_pass_descriptor_queue.Acquire(scheduler, 2); + compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset, + image.guest_size_bytes - swizzle.buffer_offset); + compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level)); + const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; + + const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info); + scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, params, + descriptor_data](vk::CommandBuffer cmdbuf) { + const VkDescriptorSet set = descriptor_allocator.Commit(); + device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); + cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params); + cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_dispatches_z); + }); + } + RecordUnswizzleEndBarrier(scheduler, vk_image, aspect_mask); + scheduler.Finish(); +} + +BlockLinearUnswizzleImage3DPass::BlockLinearUnswizzleImage3DPass( + const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_) + : ComputePass(device_, scheduler_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, + ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, + COMPUTE_PUSH_CONSTANT_RANGE, + BLOCK_LINEAR_UNSWIZZLE_3D_COMP_SPV), + scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, + compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} + +BlockLinearUnswizzleImage3DPass::~BlockLinearUnswizzleImage3DPass() = default; + +void BlockLinearUnswizzleImage3DPass::Unswizzle( + Image& image, const StagingBufferRef& map, + std::span swizzles) { + using namespace VideoCommon::Accelerated; + scheduler.RequestOutsideRenderPassOperationContext(); + const VkImageAspectFlags aspect_mask = image.AspectMask(); + const VkImage vk_image = image.Handle(); + RecordUnswizzleBeginBarrier(scheduler, *pipeline, vk_image, aspect_mask, + image.ExchangeInitialization()); + for (const VideoCommon::SwizzleParameters& swizzle : swizzles) { + const size_t input_offset = swizzle.buffer_offset + map.offset; + const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 16U); + const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U); + const u32 num_dispatches_z = Common::DivCeil(swizzle.num_tiles.depth, 2U); + + compute_pass_descriptor_queue.Acquire(scheduler, 2); + compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset, + image.guest_size_bytes - swizzle.buffer_offset); + compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level)); + const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; + + const auto p = MakeBlockLinearSwizzle3DParams(swizzle, image.info); + const BlockLinear3DImagePushConstants params{ + .origin = p.origin, + .destination = p.destination, + .bytes_per_block_log2 = p.bytes_per_block_log2, + .slice_size = p.slice_size, + .block_size = p.block_size, + .x_shift = p.x_shift, + .block_height = p.block_height, + .block_height_mask = p.block_height_mask, + .block_depth = p.block_depth, + .block_depth_mask = p.block_depth_mask, + }; + scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, params, + descriptor_data](vk::CommandBuffer cmdbuf) { + const VkDescriptorSet set = descriptor_allocator.Commit(); + device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); + cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params); + cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_dispatches_z); + }); + } + RecordUnswizzleEndBarrier(scheduler, vk_image, aspect_mask); + scheduler.Finish(); +} + +PitchUnswizzlePass::PitchUnswizzlePass(const Device& device_, Scheduler& scheduler_, + DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_) + : ComputePass(device_, scheduler_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, + ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, + COMPUTE_PUSH_CONSTANT_RANGE, + PITCH_UNSWIZZLE_COMP_SPV), + scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, + compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} + +PitchUnswizzlePass::~PitchUnswizzlePass() = default; + +void PitchUnswizzlePass::Unswizzle(Image& image, const StagingBufferRef& map, + std::span swizzles) { + const u32 bytes_per_block = VideoCore::Surface::BytesPerBlock(image.info.format); + ASSERT(std::has_single_bit(bytes_per_block)); + scheduler.RequestOutsideRenderPassOperationContext(); + const VkImageAspectFlags aspect_mask = image.AspectMask(); + const VkImage vk_image = image.Handle(); + RecordUnswizzleBeginBarrier(scheduler, *pipeline, vk_image, aspect_mask, + image.ExchangeInitialization()); + for (const VideoCommon::SwizzleParameters& swizzle : swizzles) { + const size_t input_offset = swizzle.buffer_offset + map.offset; + const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U); + const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U); + + compute_pass_descriptor_queue.Acquire(scheduler, 2); + compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset, + image.guest_size_bytes - swizzle.buffer_offset); + compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level)); + const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; + + const PitchUnswizzlePushConstants params{ + .origin = {0, 0}, + .destination = {0, 0}, + .bytes_per_block_log2 = static_cast(std::countr_zero(bytes_per_block)), + .pitch = image.info.pitch, + }; + scheduler.Record([this, num_dispatches_x, num_dispatches_y, params, + descriptor_data](vk::CommandBuffer cmdbuf) { + const VkDescriptorSet set = descriptor_allocator.Commit(); + device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); + cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params); + cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, 1); + }); + } + RecordUnswizzleEndBarrier(scheduler, vk_image, aspect_mask); + scheduler.Finish(); +} + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h index 3d30aa6bdc..811deee8a2 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.h +++ b/src/video_core/renderer_vulkan/vk_compute_pass.h @@ -17,7 +17,6 @@ #include "video_core/texture_cache/types.h" #include "video_core/vulkan_common/vulkan_memory_allocator.h" #include "video_core/vulkan_common/vulkan_wrapper.h" -#include "video_core/texture_cache/accelerated_swizzle.h" namespace VideoCommon { struct SwizzleParameters; @@ -25,8 +24,6 @@ struct SwizzleParameters; namespace Vulkan { -using VideoCommon::Accelerated::BlockLinearSwizzle3DParams; - class Device; class StagingBufferPool; class Scheduler; @@ -137,26 +134,50 @@ private: MemoryAllocator& memory_allocator; }; -class BlockLinearUnswizzle3DPass final : public ComputePass { +class BlockLinearUnswizzleImage2DPass final : public ComputePass { public: - explicit BlockLinearUnswizzle3DPass(const Device& device_, Scheduler& scheduler_, - DescriptorPool& descriptor_pool_, - StagingBufferPool& staging_buffer_pool_, - ComputePassDescriptorQueue& compute_pass_descriptor_queue_); - ~BlockLinearUnswizzle3DPass(); + explicit BlockLinearUnswizzleImage2DPass( + const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_); + ~BlockLinearUnswizzleImage2DPass(); - void Unswizzle(Image& image, - const StagingBufferRef& swizzled, - std::span swizzles, - u32 z_start, u32 z_count); + void Unswizzle(Image& image, const StagingBufferRef& map, + std::span swizzles); - void UnswizzleChunk( - Image& image, - const StagingBufferRef& swizzled, - const VideoCommon::SwizzleParameters& sw, - const BlockLinearSwizzle3DParams& params, - u32 blocks_x, u32 blocks_y, - u32 z_start, u32 z_count); +private: + Scheduler& scheduler; + StagingBufferPool& staging_buffer_pool; + ComputePassDescriptorQueue& compute_pass_descriptor_queue; +}; + +class BlockLinearUnswizzleImage3DPass final : public ComputePass { +public: + explicit BlockLinearUnswizzleImage3DPass( + const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_); + ~BlockLinearUnswizzleImage3DPass(); + + void Unswizzle(Image& image, const StagingBufferRef& map, + std::span swizzles); + +private: + Scheduler& scheduler; + StagingBufferPool& staging_buffer_pool; + ComputePassDescriptorQueue& compute_pass_descriptor_queue; +}; + +class PitchUnswizzlePass final : public ComputePass { +public: + explicit PitchUnswizzlePass(const Device& device_, Scheduler& scheduler_, + DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_); + ~PitchUnswizzlePass(); + + void Unswizzle(Image& image, const StagingBufferRef& map, + std::span swizzles); private: Scheduler& scheduler; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index bc2e82acae..4b37fee265 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -160,6 +160,73 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { info.size.depth == 1; } +[[nodiscard]] VkFormat UnswizzleStorageFormat(u32 bytes_per_block) { + switch (bytes_per_block) { + case 1: + return VK_FORMAT_R8_UINT; + case 2: + return VK_FORMAT_R16_UINT; + case 4: + return VK_FORMAT_R32_UINT; + case 8: + return VK_FORMAT_R32G32_UINT; + case 16: + return VK_FORMAT_R32G32B32A32_UINT; + default: + return VK_FORMAT_UNDEFINED; + } +} + +[[nodiscard]] bool IsUnswizzleStorageFormatSupported(const Device& device, u32 bytes_per_block) { + switch (bytes_per_block) { + case 1: + return device.IsStorageBuffer8BitAccessSupported(); + case 2: + return device.IsStorageBuffer16BitAccessSupported(); + case 4: + case 8: + case 16: + return true; + default: + return false; + } +} + +[[nodiscard]] bool IsUnswizzleAcceleratedFormat(const Device& device, PixelFormat format) { + if (IsPixelFormatASTC(format) || VideoCore::Surface::IsPixelFormatBCn(format)) { + return false; + } + if (VideoCore::Surface::GetFormatType(format) != SurfaceType::ColorTexture) { + return false; + } + return IsUnswizzleStorageFormatSupported(device, BytesPerBlock(format)); +} + +[[nodiscard]] bool WillUseAcceleratedUnswizzle(const Device& device, const ImageInfo& info) { + switch (info.type) { + case ImageType::e2D: + case ImageType::e3D: + case ImageType::Linear: + break; + default: + return false; + } + if (info.num_samples > 1 || !device.IsKhrImageFormatListSupported()) { + return false; + } + return IsUnswizzleAcceleratedFormat(device, info.format); +} + +[[nodiscard]] VkImageViewType UnswizzleStorageViewType(ImageType type) { + if (type == ImageType::e3D) { + return VK_IMAGE_VIEW_TYPE_3D; + } + if (type == ImageType::Linear) { + return VK_IMAGE_VIEW_TYPE_2D; + } + return VK_IMAGE_VIEW_TYPE_2D_ARRAY; +} + [[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info, std::optional format_override = {}) { auto format_info = @@ -249,7 +316,8 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { } [[nodiscard]] vk::ImageView MakeStorageView(const vk::Device& device, u32 level, VkImage image, - VkFormat format) { + VkFormat format, + VkImageViewType view_type = VK_IMAGE_VIEW_TYPE_2D_ARRAY) { static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO, .pNext = nullptr, @@ -260,7 +328,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { .pNext = &storage_image_view_usage_create_info, .flags = 0, .image = image, - .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY, + .viewType = view_type, .format = format, .components{ .r = VK_COMPONENT_SWIZZLE_IDENTITY, @@ -946,6 +1014,14 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched astc_decoder_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, compute_pass_descriptor_queue, memory_allocator); } + if (device.IsKhrImageFormatListSupported()) { + bl_unswizzle_2d_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, + compute_pass_descriptor_queue); + bl_unswizzle_3d_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, + compute_pass_descriptor_queue); + pitch_unswizzle_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, + compute_pass_descriptor_queue); + } if (!device.IsKhrImageFormatListSupported()) { return; } @@ -953,6 +1029,8 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched const auto image_format = static_cast(index_a); if (IsPixelFormatASTC(image_format) && !device.IsOptimalAstcSupported()) { view_formats[index_a].push_back(VK_FORMAT_A8B8G8R8_UNORM_PACK32); + } else if (IsUnswizzleAcceleratedFormat(device, image_format)) { + view_formats[index_a].push_back(UnswizzleStorageFormat(BytesPerBlock(image_format))); } for (size_t index_b = 0; index_b < VideoCore::Surface::MaxPixelFormat; index_b++) { const auto view_format = static_cast(index_b); @@ -964,18 +1042,14 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched } } - if (Settings::values.gpu_unswizzle_enabled.GetValue()) { - bl3d_unswizzle_pass.emplace(device, scheduler, descriptor_pool, - staging_buffer_pool, compute_pass_descriptor_queue); - } } void TextureCacheRuntime::Finish() { scheduler.Finish(); } -StagingBufferRef TextureCacheRuntime::UploadStagingBuffer(size_t size, bool deferred) { - return staging_buffer_pool.Request(size, MemoryUsage::Upload, deferred); +StagingBufferRef TextureCacheRuntime::UploadStagingBuffer(size_t size) { + return staging_buffer_pool.Request(size, MemoryUsage::Upload); } StagingBufferRef TextureCacheRuntime::DownloadStagingBuffer(size_t size, bool deferred) { @@ -1899,6 +1973,10 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu flags |= VideoCommon::ImageFlagBits::Converted; flags |= VideoCommon::ImageFlagBits::CostlyLoad; } + if (False(flags & VideoCommon::ImageFlagBits::Converted) && + WillUseAcceleratedUnswizzle(runtime->device, info)) { + flags |= VideoCommon::ImageFlagBits::AcceleratedUpload; + } if (runtime->device.HasDebuggingToolAttached()) { original_image.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); } @@ -1911,6 +1989,14 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu storage_image_views[level] = MakeStorageView(device, level, *original_image, storage_format); } + } else if (True(flags & VideoCommon::ImageFlagBits::AcceleratedUpload)) { + const auto& device = runtime->device.GetLogical(); + const VkFormat storage_format = UnswizzleStorageFormat(BytesPerBlock(info.format)); + const VkImageViewType view_type = UnswizzleStorageViewType(info.type); + for (s32 level = 0; level < info.resources.levels; ++level) { + storage_image_views[level] = + MakeStorageView(device, level, *original_image, storage_format, view_type); + } } } @@ -1927,48 +2013,6 @@ Image::~Image() { } } -void Image::AllocateComputeUnswizzleBuffer(u32 max_slices) { - using VideoCore::Surface::BytesPerBlock; - - const u32 block_bytes = BytesPerBlock(info.format); // 8 for BC1, 16 for BC6H - const u32 block_width = 4; - const u32 block_height = 4; - - // BCn is 4x4x1 blocks - const u32 blocks_x = (info.size.width + block_width - 1) / block_width; - const u32 blocks_y = (info.size.height + block_height - 1) / block_height; - const u32 blocks_z = (std::min)(max_slices, info.size.depth); - - const u64 block_count = - static_cast(blocks_x) * - static_cast(blocks_y) * - static_cast(blocks_z); - - const VkDeviceSize required_size = block_count * block_bytes; - if (has_compute_unswizzle_buffer && required_size <= compute_unswizzle_buffer_size) { - return; - } - - compute_unswizzle_buffer_size = required_size; - - VkBufferCreateInfo ci{ - .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .size = compute_unswizzle_buffer_size, - .usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | - VK_BUFFER_USAGE_TRANSFER_SRC_BIT, - .sharingMode = VK_SHARING_MODE_EXCLUSIVE, - .queueFamilyIndexCount = 0, - .pQueueFamilyIndices = nullptr, - }; - - compute_unswizzle_buffer = - runtime->memory_allocator.CreateBuffer(ci, MemoryUsage::DeviceLocal); - - has_compute_unswizzle_buffer = true; -} - void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset, std::span copies) { // TODO: Move this to another API @@ -2326,11 +2370,15 @@ VkImageView Image::StorageImageView(s32 level) noexcept { if (!view) { auto format_info = MaxwellToVK::SurfaceFormat(runtime->device, FormatType::Optimal, true, info.format); + VkImageViewType view_type = VK_IMAGE_VIEW_TYPE_2D_ARRAY; if (WillUseAcceleratedAstcDecode(runtime->device, info)) { format_info.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32; + } else if (True(flags & VideoCommon::ImageFlagBits::AcceleratedUpload)) { + format_info.format = UnswizzleStorageFormat(BytesPerBlock(info.format)); + view_type = UnswizzleStorageViewType(info.type); } view = MakeStorageView(runtime->device.GetLogical(), level, *(this->*current_image), - format_info.format); + format_info.format, view_type); } return *view; } @@ -3129,25 +3177,20 @@ VkRenderPass Framebuffer::RenderPassVariant(u32 color_clear_mask, bool depth_ste void TextureCacheRuntime::AccelerateImageUpload( Image& image, const StagingBufferRef& map, - std::span swizzles, - u32 z_start, u32 z_count) { - + std::span swizzles) { if (IsPixelFormatASTC(image.info.format)) { return astc_decoder_pass->Assemble(image, map, swizzles); } - - if (!Settings::values.gpu_unswizzle_enabled.GetValue() || !bl3d_unswizzle_pass) { - if (IsPixelFormatBCn(image.info.format) && image.info.type == ImageType::e3D) { - ASSERT(false && "GPU unswizzle is disabled for BCn 3D texture"); - } - ASSERT(false); - return; + switch (image.info.type) { + case ImageType::e2D: + return bl_unswizzle_2d_pass->Unswizzle(image, map, swizzles); + case ImageType::e3D: + return bl_unswizzle_3d_pass->Unswizzle(image, map, swizzles); + case ImageType::Linear: + return pitch_unswizzle_pass->Unswizzle(image, map, swizzles); + default: + break; } - - if (bl3d_unswizzle_pass && IsPixelFormatBCn(image.info.format) && image.info.type == ImageType::e3D && image.info.resources.levels == 1 && image.info.resources.layers == 1) { - return bl3d_unswizzle_pass->Unswizzle(image, map, swizzles, z_start, z_count); - } - ASSERT(false); } diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 2e9736f707..2e6b547e72 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -52,7 +52,7 @@ public: void Finish(); - StagingBufferRef UploadStagingBuffer(size_t size, bool deferred = false); + StagingBufferRef UploadStagingBuffer(size_t size); StagingBufferRef DownloadStagingBuffer(size_t size, bool deferred = false); @@ -98,8 +98,7 @@ public: } void AccelerateImageUpload(Image&, const StagingBufferRef&, - std::span, - u32 z_start, u32 z_count); + std::span); void InsertUploadMemoryBarrier() {} @@ -157,8 +156,9 @@ public: BlitImageHelper& blit_image_helper; RenderPassCache& render_pass_cache; std::optional astc_decoder_pass; - - std::optional bl3d_unswizzle_pass; + std::optional bl_unswizzle_2d_pass; + std::optional bl_unswizzle_3d_pass; + std::optional pitch_unswizzle_pass; const Settings::ResolutionScalingInfo& resolution; std::array, VideoCore::Surface::MaxPixelFormat> view_formats; @@ -334,8 +334,6 @@ public: void DownloadMemory(const StagingBufferRef& map, std::span copies); - void AllocateComputeUnswizzleImage(); - [[nodiscard]] VkImage Handle() const noexcept { return *(this->*current_image); } @@ -361,10 +359,6 @@ public: bool ScaleDown(bool ignore = false); - u64 allocation_tick; - - friend class BlockLinearUnswizzle3DPass; - private: bool BlitScaleHelper(bool scale_up); @@ -376,12 +370,6 @@ private: vk::Image original_image; vk::Image scaled_image; - vk::Buffer compute_unswizzle_buffer; - VkDeviceSize compute_unswizzle_buffer_size = 0; - bool has_compute_unswizzle_buffer = false; - - void AllocateComputeUnswizzleBuffer(u32 max_slices); - // Use a pointer to field because it is relative, so that the object can be // moved without breaking the reference. vk::Image Image::*current_image{}; diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 640bd38407..a248fe9c93 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -8,7 +8,6 @@ #include #include -#include #include "common/container/unordered_map.h" #include @@ -76,41 +75,6 @@ TextureCache

::TextureCache(Runtime& runtime_, Tegra::MaxwellDeviceMemoryManag critical_memory = DEFAULT_CRITICAL_MEMORY + 1_GiB; minimum_memory = 0; } - - const bool gpu_unswizzle_enabled = Settings::values.gpu_unswizzle_enabled.GetValue(); - - if (gpu_unswizzle_enabled) { - switch (Settings::values.gpu_unswizzle_texture_size.GetValue()) { - case Settings::GpuUnswizzleSize::VerySmall: gpu_unswizzle_maxsize = 16_MiB; break; - case Settings::GpuUnswizzleSize::Small: gpu_unswizzle_maxsize = 32_MiB; break; - case Settings::GpuUnswizzleSize::Normal: gpu_unswizzle_maxsize = 128_MiB; break; - case Settings::GpuUnswizzleSize::Large: gpu_unswizzle_maxsize = 256_MiB; break; - case Settings::GpuUnswizzleSize::VeryLarge: gpu_unswizzle_maxsize = 512_MiB; break; - default: gpu_unswizzle_maxsize = 128_MiB; break; - } - - switch (Settings::values.gpu_unswizzle_stream_size.GetValue()) { - case Settings::GpuUnswizzle::VeryLow: swizzle_chunk_size = 4_MiB; break; - case Settings::GpuUnswizzle::Low: swizzle_chunk_size = 8_MiB; break; - case Settings::GpuUnswizzle::Normal: swizzle_chunk_size = 16_MiB; break; - case Settings::GpuUnswizzle::Medium: swizzle_chunk_size = 32_MiB; break; - case Settings::GpuUnswizzle::High: swizzle_chunk_size = 64_MiB; break; - default: swizzle_chunk_size = 16_MiB; - } - - switch (Settings::values.gpu_unswizzle_chunk_size.GetValue()) { - case Settings::GpuUnswizzleChunk::VeryLow: swizzle_slices_per_batch = 32; break; - case Settings::GpuUnswizzleChunk::Low: swizzle_slices_per_batch = 64; break; - case Settings::GpuUnswizzleChunk::Normal: swizzle_slices_per_batch = 128; break; - case Settings::GpuUnswizzleChunk::Medium: swizzle_slices_per_batch = 256; break; - case Settings::GpuUnswizzleChunk::High: swizzle_slices_per_batch = 512; break; - default: swizzle_slices_per_batch = 128; - } - } else { - gpu_unswizzle_maxsize = 0; - swizzle_chunk_size = 0; - swizzle_slices_per_batch = 0; - } } template @@ -180,7 +144,6 @@ void TextureCache

::TickFrame() { sentenced_framebuffers.Tick(); sentenced_image_view.Tick(); TickAsyncDecode(); - TickAsyncUnswizzle(); runtime.TickFrame(); ++frame_tick; @@ -1132,19 +1095,6 @@ void TextureCache

::RefreshContents(Image& image, ImageId image_id) { return; } - const bool gpu_unswizzle_enabled = Settings::values.gpu_unswizzle_enabled.GetValue(); - - if (gpu_unswizzle_enabled && - IsPixelFormatBCn(image.info.format) && - image.info.type == ImageType::e3D && - image.info.resources.levels == 1 && - image.info.resources.layers == 1 && - MapSizeBytes(image) >= gpu_unswizzle_maxsize && - False(image.flags & ImageFlagBits::GpuModified)) { - - QueueAsyncUnswizzle(image, image_id); - return; - } auto staging = runtime.UploadStagingBuffer(MapSizeBytes(image)); UploadImageContents(image, staging); runtime.InsertUploadMemoryBarrier(); @@ -1160,7 +1110,7 @@ void TextureCache

::UploadImageContents(Image& image, StagingBuffer& staging) gpu_memory->ReadBlock(gpu_addr, mapped_span.data(), mapped_span.size_bytes(), VideoCommon::CacheType::NoTextureCache); const auto uploads = FullUploadSwizzles(image.info); - runtime.AccelerateImageUpload(image, staging, FixSmallVectorADL(uploads), 0, 0); + runtime.AccelerateImageUpload(image, staging, FixSmallVectorADL(uploads)); return; } @@ -1372,20 +1322,6 @@ void TextureCache

::QueueAsyncDecode(Image& image, ImageId image_id) { texture_decode_worker.QueueWork(std::move(func)); } -template -void TextureCache

::QueueAsyncUnswizzle(Image& image, ImageId image_id) { - if (True(image.flags & ImageFlagBits::IsDecoding)) { - return; - } - - image.flags |= ImageFlagBits::IsDecoding; - - unswizzle_queue.push_back({ - .image_id = image_id, - .info = image.info - }); -} - template void TextureCache

::TickAsyncDecode() { bool has_uploads{}; @@ -1411,83 +1347,6 @@ void TextureCache

::TickAsyncDecode() { } } -template -void TextureCache

::TickAsyncUnswizzle() { - if (unswizzle_queue.empty()) { - return; - } - - if(current_unswizzle_frame > 0) { - current_unswizzle_frame--; - return; - } - - PendingUnswizzle& task = unswizzle_queue.front(); - Image& image = slot_images[task.image_id]; - - if (!task.initialized) { - task.total_size = MapSizeBytes(image); - task.staging_buffer = runtime.UploadStagingBuffer(task.total_size, true); - - const auto& info = image.info; - const u32 bytes_per_block = BytesPerBlock(info.format); - const u32 width_blocks = Common::DivCeil(info.size.width, 4u); - const u32 height_blocks = Common::DivCeil(info.size.height, 4u); - - const u32 stride = width_blocks * bytes_per_block; - const u32 aligned_height = height_blocks; - task.bytes_per_slice = static_cast(stride) * aligned_height; - task.last_submitted_offset = 0; - task.initialized = true; - } - - // Read data - if (task.current_offset < task.total_size) { - const size_t remaining = task.total_size - task.current_offset; - - size_t copy_amount = (std::min)(swizzle_chunk_size, remaining); - - if (remaining > swizzle_chunk_size) { - copy_amount = (copy_amount / task.bytes_per_slice) * task.bytes_per_slice; - if (copy_amount == 0) copy_amount = task.bytes_per_slice; - } - - gpu_memory->ReadBlock(image.gpu_addr + task.current_offset, - task.staging_buffer.mapped_span.data() + task.current_offset, - copy_amount); - task.current_offset += copy_amount; - } - - const bool is_final_batch = task.current_offset >= task.total_size; - const size_t bytes_ready = task.current_offset - task.last_submitted_offset; - const u32 complete_slices = static_cast(bytes_ready / task.bytes_per_slice); - - if (complete_slices >= swizzle_slices_per_batch || (is_final_batch && complete_slices > 0)) { - const u32 z_start = static_cast(task.last_submitted_offset / task.bytes_per_slice); - const u32 slices_to_process = (std::min)(complete_slices, swizzle_slices_per_batch); - const u32 z_count = (std::min)(slices_to_process, image.info.size.depth - z_start); - - if (z_count > 0) { - const auto uploads = FullUploadSwizzles(task.info); - runtime.AccelerateImageUpload(image, task.staging_buffer, FixSmallVectorADL(uploads), z_start, z_count); - task.last_submitted_offset += (static_cast(z_count) * task.bytes_per_slice); - } - } - - // Check if complete - const u32 slices_submitted = static_cast(task.last_submitted_offset / task.bytes_per_slice); - const bool all_slices_submitted = slices_submitted >= image.info.size.depth; - - if (is_final_batch && all_slices_submitted) { - runtime.FreeDeferredStagingBuffer(task.staging_buffer); - image.flags &= ~ImageFlagBits::IsDecoding; - unswizzle_queue.pop_front(); - - // Wait 4 frames to process the next entry - current_unswizzle_frame = 4u; - } -} - template bool TextureCache

::ScaleUp(Image& image) { const bool has_copy = image.HasScaled(); @@ -1637,8 +1496,6 @@ ImageId TextureCache

::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, DA const ImageId new_image_id = slot_images.insert(runtime, new_info, gpu_addr, cpu_addr); Image& new_image = slot_images[new_image_id]; - new_image.allocation_tick = frame_tick; - if (!gpu_memory->IsContinuousRange(new_image.gpu_addr, new_image.guest_size_bytes) && new_info.is_sparse) { new_image.flags |= ImageFlagBits::Sparse; diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index 8a2318cc72..517afa4aa8 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -131,17 +131,6 @@ class TextureCache : public VideoCommon::ChannelSetupCaches> async_decodes; - std::deque unswizzle_queue; - u8 current_unswizzle_frame; - // Join caching boost::container::small_vector join_overlap_ids; ::Common::unordered_set join_overlaps_found; From 930b64858ed3107f6585ce212dea06d7a13a83c9 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Tue, 15 Sep 2026 23:38:47 -0400 Subject: [PATCH 02/69] Fix build --- src/video_core/renderer_vulkan/vk_compute_pass.cpp | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 4685c6ae30..b7bb4b00b4 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -711,7 +711,8 @@ BlockLinearUnswizzleImage2DPass::BlockLinearUnswizzleImage2DPass( ComputePassDescriptorQueue& compute_pass_descriptor_queue_) : ComputePass(device_, scheduler_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, - COMPUTE_PUSH_CONSTANT_RANGE, + COMPUTE_PUSH_CONSTANT_RANGE< + sizeof(VideoCommon::Accelerated::BlockLinearSwizzle2DParams)>, BLOCK_LINEAR_UNSWIZZLE_2D_COMP_SPV), scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} From 26031c0940795e45a66b4e5b43446c8a60bbb723 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Wed, 16 Sep 2026 00:54:01 -0400 Subject: [PATCH 03/69] Fixes on the astc decoder --- src/video_core/host_shaders/astc_decoder.comp | 96 +++++++------------ 1 file changed, 36 insertions(+), 60 deletions(-) diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp index b7797e3165..20f4af896e 100644 --- a/src/video_core/host_shaders/astc_decoder.comp +++ b/src/video_core/host_shaders/astc_decoder.comp @@ -980,67 +980,42 @@ uint UnquantizeTexelWeight(EncodingData val) { const uint encoding = Encoding(val); const uint bitlen = NumBits(val); const uint bitval = BitValue(val); - const uint A = ReplicateBitTo7((bitval & 1)); - uint B = 0, C = 0, D = 0; + const uint A = ReplicateBitTo7(bitval & 1u); uint result = 0; - const uint bitlen_0_results[5] = {0, 16, 32, 48, 64}; - switch (encoding) { - case JUST_BITS: + if (encoding == JUST_BITS) { result = FastReplicateTo6(bitval, bitlen); - break; - case TRIT: { - D = QuintTritValue(val); - switch (bitlen) { - case 0: - return bitlen_0_results[D * 2]; - case 1: { - C = 50; - break; + } else { + const uint D = QuintTritValue(val); + if (bitlen == 0u) { + if (encoding == TRIT) { + return D * 32u; + } + return D * 16u; } - case 2: { - C = 23; - const uint b = (bitval >> 1) & 1; - B = (b << 6) | (b << 2) | b; - break; + const uint b = (bitval >> 1u) & ((0x3100u >> (bitlen * 4u)) & 0xfu); + uint B = 0; + uint C = 0; + if (encoding == TRIT) { + if (bitlen == 1u) { + C = 50u; + } else if (bitlen == 2u) { + C = 23u; + B = (b << 6u) | (b << 2u) | b; + } else { + C = 11u; + B = (b << 5u) | b; + } + } else { + if (bitlen == 1u) { + C = 28u; + } else { + C = 13u; + B = (b << 6u) | (b << 1u); + } } - case 3: { - C = 11; - const uint cb = (bitval >> 1) & 3; - B = (cb << 5) | cb; - break; - } - default: - break; - } - break; - } - case QUINT: { - D = QuintTritValue(val); - switch (bitlen) { - case 0: - return bitlen_0_results[D]; - case 1: { - C = 28; - break; - } - case 2: { - C = 13; - const uint b = (bitval >> 1) & 1; - B = (b << 6) | (b << 1); - break; - } - } - break; - } - } - if (encoding != JUST_BITS && bitlen > 0) { - result = D * C + B; - result ^= A; - result = (A & 0x20) | (result >> 2); - } - if (result > 32) { - result += 1; + result = (A & 0x20u) | (((D * C + B) ^ A) >> 2u); } + result += 1u - uint(result <= 32u); return result; } @@ -1449,10 +1424,11 @@ void DecompressBlock(ivec3 coord) { } uint SwizzleOffset(uvec2 pos) { - const uint x = pos.x; - const uint y = pos.y; - return ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + - ((x % 32) / 16) * 32 + (y % 2) * 16 + (x % 16); + return ((pos.x & 32u) << 3u) | + ((pos.y & 6u) << 5u) | + ((pos.x & 16u) << 1u) | + ((pos.y & 1u) << 4u) | + (pos.x & 15u); } void main() { From b24c3d7226784f302e1740512af420d5977dadfa Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Wed, 16 Sep 2026 01:55:44 -0400 Subject: [PATCH 04/69] Keep refining astc decoding math --- src/video_core/host_shaders/astc_decoder.comp | 8 +--- .../renderer_vulkan/vk_texture_cache.cpp | 47 +++++++++++-------- src/video_core/texture_cache/texture_cache.h | 2 +- 3 files changed, 30 insertions(+), 27 deletions(-) diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp index 20f4af896e..2248a92a46 100644 --- a/src/video_core/host_shaders/astc_decoder.comp +++ b/src/video_core/host_shaders/astc_decoder.comp @@ -987,10 +987,7 @@ uint UnquantizeTexelWeight(EncodingData val) { } else { const uint D = QuintTritValue(val); if (bitlen == 0u) { - if (encoding == TRIT) { - return D * 32u; - } - return D * 16u; + return D << (4u | uint(encoding == TRIT)); } const uint b = (bitval >> 1u) & ((0x3100u >> (bitlen * 4u)) & 0xfu); uint B = 0; @@ -1015,8 +1012,7 @@ uint UnquantizeTexelWeight(EncodingData val) { } result = (A & 0x20u) | (((D * C + B) ^ A) >> 2u); } - result += 1u - uint(result <= 32u); - return result; + return result + (1u - uint(result <= 32u)); } void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) { diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 4b37fee265..bf8efaaf09 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -323,6 +323,10 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { .pNext = nullptr, .usage = VK_IMAGE_USAGE_STORAGE_BIT, }; + u32 layer_count = VK_REMAINING_ARRAY_LAYERS; + if (view_type == VK_IMAGE_VIEW_TYPE_2D || view_type == VK_IMAGE_VIEW_TYPE_3D) { + layer_count = 1; + } return device.CreateImageView(VkImageViewCreateInfo{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, .pNext = &storage_image_view_usage_create_info, @@ -341,7 +345,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { .baseMipLevel = level, .levelCount = 1, .baseArrayLayer = 0, - .layerCount = VK_REMAINING_ARRAY_LAYERS, + .layerCount = layer_count, }, }); } @@ -1989,14 +1993,6 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu storage_image_views[level] = MakeStorageView(device, level, *original_image, storage_format); } - } else if (True(flags & VideoCommon::ImageFlagBits::AcceleratedUpload)) { - const auto& device = runtime->device.GetLogical(); - const VkFormat storage_format = UnswizzleStorageFormat(BytesPerBlock(info.format)); - const VkImageViewType view_type = UnswizzleStorageViewType(info.type); - for (s32 level = 0; level < info.resources.levels; ++level) { - storage_image_views[level] = - MakeStorageView(device, level, *original_image, storage_format, view_type); - } } } @@ -3178,20 +3174,31 @@ VkRenderPass Framebuffer::RenderPassVariant(u32 color_clear_mask, bool depth_ste void TextureCacheRuntime::AccelerateImageUpload( Image& image, const StagingBufferRef& map, std::span swizzles) { + const bool is_rescaled = True(image.flags & ImageFlagBits::Rescaled); + if (is_rescaled) { + image.ScaleDown(true); + } if (IsPixelFormatASTC(image.info.format)) { - return astc_decoder_pass->Assemble(image, map, swizzles); + astc_decoder_pass->Assemble(image, map, swizzles); + } else { + switch (image.info.type) { + case ImageType::e2D: + bl_unswizzle_2d_pass->Unswizzle(image, map, swizzles); + break; + case ImageType::e3D: + bl_unswizzle_3d_pass->Unswizzle(image, map, swizzles); + break; + case ImageType::Linear: + pitch_unswizzle_pass->Unswizzle(image, map, swizzles); + break; + default: + ASSERT(false); + break; + } } - switch (image.info.type) { - case ImageType::e2D: - return bl_unswizzle_2d_pass->Unswizzle(image, map, swizzles); - case ImageType::e3D: - return bl_unswizzle_3d_pass->Unswizzle(image, map, swizzles); - case ImageType::Linear: - return pitch_unswizzle_pass->Unswizzle(image, map, swizzles); - default: - break; + if (is_rescaled) { + image.ScaleUp(); } - ASSERT(false); } void TextureCacheRuntime::TransitionImageLayout(Image& image) { diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index a248fe9c93..5b2cafc89d 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -1107,7 +1107,7 @@ void TextureCache

::UploadImageContents(Image& image, StagingBuffer& staging) const GPUVAddr gpu_addr = image.gpu_addr; if (True(image.flags & ImageFlagBits::AcceleratedUpload)) { - gpu_memory->ReadBlock(gpu_addr, mapped_span.data(), mapped_span.size_bytes(), + gpu_memory->ReadBlock(gpu_addr, mapped_span.data(), image.guest_size_bytes, VideoCommon::CacheType::NoTextureCache); const auto uploads = FullUploadSwizzles(image.info); runtime.AccelerateImageUpload(image, staging, FixSmallVectorADL(uploads)); From 66a25a3d79e6cf0dc095d07353b9767ee5c42575 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Wed, 16 Sep 2026 02:33:06 -0400 Subject: [PATCH 05/69] Some unvalid checks on astc decoder --- src/video_core/host_shaders/astc_decoder.comp | 11 ++++++++--- src/video_core/texture_cache/util.cpp | 4 ++-- 2 files changed, 10 insertions(+), 5 deletions(-) diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp index 2248a92a46..58dccb545f 100644 --- a/src/video_core/host_shaders/astc_decoder.comp +++ b/src/video_core/host_shaders/astc_decoder.comp @@ -296,14 +296,19 @@ uint ExtractBits(uvec4 payload, int offset, int bits) { if (bits > 32) { return 0; } - const int last_offset = offset + bits - 1; + if (offset >= 128) { + return 0; + } + const int avail_bits = min(bits, 128 - offset); + const int last_offset = offset + avail_bits - 1; const int shifted_offset = offset >> 5; if ((last_offset >> 5) == shifted_offset) { - return bitfieldExtract(payload[shifted_offset], offset & 31, bits); + return bitfieldExtract(payload[shifted_offset], offset & 31, avail_bits); } const int first_bits = 32 - (offset & 31); const int result_first = int(bitfieldExtract(payload[shifted_offset], offset & 31, first_bits)); - const int result_second = int(bitfieldExtract(payload[shifted_offset + 1], 0, bits - first_bits)); + const int result_second = + int(bitfieldExtract(payload[shifted_offset + 1], 0, avail_bits - first_bits)); return result_first | (result_second << first_bits); } diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp index e55d0752ec..25cf6e33b1 100644 --- a/src/video_core/texture_cache/util.cpp +++ b/src/video_core/texture_cache/util.cpp @@ -950,7 +950,7 @@ void ConvertImage(std::span input, const ImageInfo& info, std::span input, const ImageInfo& info, std::span Date: Wed, 16 Sep 2026 16:19:52 -0400 Subject: [PATCH 06/69] Keep working on ASTC --- .../renderer_vulkan/vk_texture_cache.cpp | 6 +- src/video_core/texture_cache/texture_cache.h | 25 +++- .../texture_cache/texture_cache_base.h | 1 + src/video_core/texture_cache/util.cpp | 74 +++++++++- src/video_core/texture_cache/util.h | 7 +- src/video_core/textures/astc.cpp | 126 +++++++++++++----- src/video_core/textures/astc.h | 19 +++ .../vulkan_common/vulkan_memory_allocator.cpp | 20 --- 8 files changed, 221 insertions(+), 57 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index bf8efaaf09..caff739187 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -57,6 +57,7 @@ using VideoCore::Surface::SurfaceType; namespace { constexpr bool ENABLE_MSAA_TILER_RESOLVE = true; constexpr bool ENABLE_MSAA_RESOLVE_CONSUME = true; +constexpr bool ENABLE_ACCELERATED_UNSWIZZLE = false; constexpr bool ENABLE_MSAA_COLOR_DISCARD = true; constexpr bool ENABLE_MSAA_DEPTH_STENCIL_DISCARD = true; @@ -193,6 +194,9 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { } [[nodiscard]] bool IsUnswizzleAcceleratedFormat(const Device& device, PixelFormat format) { + if (!ENABLE_ACCELERATED_UNSWIZZLE) { + return false; + } if (IsPixelFormatASTC(format) || VideoCore::Surface::IsPixelFormatBCn(format)) { return false; } @@ -1018,7 +1022,7 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched astc_decoder_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, compute_pass_descriptor_queue, memory_allocator); } - if (device.IsKhrImageFormatListSupported()) { + if (ENABLE_ACCELERATED_UNSWIZZLE && device.IsKhrImageFormatListSupported()) { bl_unswizzle_2d_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, compute_pass_descriptor_queue); bl_unswizzle_3d_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 5b2cafc89d..861d9f63d0 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -1117,6 +1117,12 @@ void TextureCache

::UploadImageContents(Image& image, StagingBuffer& staging) Tegra::Memory::GpuGuestMemory swizzle_data( *gpu_memory, gpu_addr, image.guest_size_bytes, &swizzle_data_buffer); if (True(image.flags & ImageFlagBits::Converted)) { + if (CanConvertFromGuest(image.info)) { + const auto copies = + FixSmallVectorADL(ConvertImageFromGuest(swizzle_data, image.info, mapped_span)); + image.UploadMemory(staging, copies); + return; + } unswizzle_data_buffer.resize_destructive(image.unswizzled_size_bytes); auto copies = FixSmallVectorADL(UnswizzleImage(*gpu_memory, gpu_addr, image.info, swizzle_data, unswizzle_data_buffer)); ConvertImage(unswizzle_data_buffer, image.info, mapped_span, copies); @@ -1302,10 +1308,27 @@ void TextureCache

::QueueAsyncDecode(Image& image, ImageId image_id) { decode->image_id = image_id; async_decodes.push_back(std::move(decode)); + const size_t out_size = MapSizeBytes(image); + if (CanConvertFromGuest(image.info)) { + decode_ptr->input_data.resize_destructive(image.guest_size_bytes); + gpu_memory->ReadBlockUnsafe(image.gpu_addr, decode_ptr->input_data.data(), + image.guest_size_bytes); + + texture_decode_worker.QueueWork([out_size, info = image.info, async_decode = decode_ptr] { + async_decode->decoded_data.resize_destructive(out_size); + auto copies = + ConvertImageFromGuest(async_decode->input_data, info, async_decode->decoded_data); + + std::unique_lock lock{async_decode->mutex}; + async_decode->copies = std::move(copies); + async_decode->complete = true; + }); + return; + } + std::vector local_unswizzle_data_buffer(image.unswizzled_size_bytes, 0); Tegra::Memory::GpuGuestMemory swizzle_data(*gpu_memory, image.gpu_addr, image.guest_size_bytes, &swizzle_data_buffer); auto copies = UnswizzleImage(*gpu_memory, image.gpu_addr, image.info, swizzle_data, local_unswizzle_data_buffer); - const size_t out_size = MapSizeBytes(image); auto func = [out_size, copies, info = image.info, input = std::move(local_unswizzle_data_buffer), diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index 517afa4aa8..8fbd7f61bc 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -63,6 +63,7 @@ struct ImageViewInOut { struct AsyncDecodeContext { ImageId image_id; + Common::ScratchBuffer input_data; Common::ScratchBuffer decoded_data; boost::container::small_vector copies; std::mutex mutex; diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp index 25cf6e33b1..8d275ff8eb 100644 --- a/src/video_core/texture_cache/util.cpp +++ b/src/video_core/texture_cache/util.cpp @@ -1,4 +1,4 @@ -// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project // SPDX-License-Identifier: GPL-3.0-or-later // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project @@ -994,6 +994,78 @@ void ConvertImage(std::span input, const ImageInfo& info, std::span ConvertImageFromGuest( + std::span input, const ImageInfo& info, std::span output) { + const u32 bpp_log2 = BytesPerBlockLog2(info.format); + const Extent2D tile_size = DefaultBlockSize(info.format); + const Extent3D size = info.size; + const LevelInfo level_info = MakeLevelInfo(info); + const s32 num_layers = info.resources.layers; + const s32 num_levels = info.resources.levels; + const std::array level_sizes = CalculateLevelSizes(level_info, num_levels); + const Extent2D gob = GobSize(bpp_log2, info.block.height, info.tile_width_spacing); + const u32 layer_size = CalculateLevelBytes(level_sizes, num_levels); + const u32 layer_stride = AlignLayerSize(layer_size, size, level_info.block, tile_size.height, + info.tile_width_spacing); + const u32 out_bytes_per_texel = BytesPerBlock(PixelFormat::A8B8G8R8_UNORM); + size_t guest_offset = 0; + u32 output_offset = 0; + boost::container::small_vector copies(num_levels); + + for (s32 level = 0; level < num_levels; ++level) { + const Extent3D level_size = AdjustMipSize(size, level); + const Extent3D num_tiles = AdjustTileSize(level_size, tile_size); + const Extent3D block = + AdjustMipBlockSize(num_tiles, level_info.block, level, level_info.num_levels); + const u32 stride_alignment = StrideAlignment(num_tiles, info.block, gob, bpp_log2); + const u32 stride = Common::AlignUpLog2(num_tiles.width, stride_alignment) << bpp_log2; + const u32 gobs_in_x = Common::DivCeilLog2(stride, GOB_SIZE_X_SHIFT); + const u32 gob_block_size = gobs_in_x << (GOB_SIZE_SHIFT + block.height + block.depth); + const u32 level_bytes = level_size.width * level_size.height * level_size.depth * + num_layers * out_bytes_per_texel; + copies[level] = BufferImageCopy{ + .buffer_offset = output_offset, + .buffer_size = level_bytes, + .buffer_row_length = level_size.width, + .buffer_image_height = level_size.height, + .image_subresource = + { + .base_level = level, + .base_layer = 0, + .num_layers = num_layers, + }, + .image_offset = {0, 0, 0}, + .image_extent = level_size, + }; + const Tegra::Texture::ASTC::BlockLinearLayout layout{ + .layer_stride = layer_stride, + .slice_size = + Common::DivCeilLog2(num_tiles.height, block.height + GOB_SIZE_Y_SHIFT) * + gob_block_size, + .block_size = gob_block_size, + .x_shift = GOB_SIZE_SHIFT + block.height + block.depth, + .gob_height = block.height, + .gob_height_mask = (1U << block.height) - 1, + .gob_depth = block.depth, + .gob_depth_mask = (1U << block.depth) - 1, + }; + Tegra::Texture::ASTC::DecompressBlockLinear( + input.subspan(guest_offset), level_size.width, level_size.height, level_size.depth, + num_layers, tile_size.width, tile_size.height, layout, output.subspan(output_offset)); + output_offset += level_bytes; + guest_offset += level_sizes[level]; + } + return copies; +} + boost::container::small_vector FullDownloadCopies(const ImageInfo& info) { const Extent3D size = info.size; const u32 bytes_per_block = BytesPerBlock(info.format); diff --git a/src/video_core/texture_cache/util.h b/src/video_core/texture_cache/util.h index 3e8bb00032..1bc4b53582 100644 --- a/src/video_core/texture_cache/util.h +++ b/src/video_core/texture_cache/util.h @@ -1,4 +1,4 @@ -// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project // SPDX-License-Identifier: GPL-3.0-or-later // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project @@ -72,6 +72,11 @@ struct OverlapResult { void ConvertImage(std::span input, const ImageInfo& info, std::span output, std::span copies); +[[nodiscard]] bool CanConvertFromGuest(const ImageInfo& info); + +[[nodiscard]] boost::container::small_vector ConvertImageFromGuest( + std::span input, const ImageInfo& info, std::span output); + [[nodiscard]] boost::container::small_vector FullDownloadCopies( const ImageInfo& info); diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index 0b6d06c49a..678e52a242 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -20,6 +20,7 @@ #include "common/common_types.h" #include #include "video_core/textures/astc.h" +#include "video_core/textures/decoders.h" #include "video_core/textures/workers.h" class InputBitStream { @@ -1671,7 +1672,7 @@ static void ComputeEndpoints(Pixel& ep1, Pixel& ep2, const u32*& colorValues, } static void FillVoidExtentLDR(InputBitStream& strm, std::span outBuf, u32 blockWidth, - u32 blockHeight) { + u32 blockHeight, u32 stride) { // Don't actually care about the void extent, just read the bits... for (s32 i = 0; i < 4; ++i) { strm.ReadBits<13>(); @@ -1688,7 +1689,7 @@ static void FillVoidExtentLDR(InputBitStream& strm, std::span outBuf, u32 b for (u32 j = 0; j < blockHeight; j++) { for (u32 i = 0; i < blockWidth; i++) { - outBuf[j * blockWidth + i] = rgba; + outBuf[j * stride + i] = rgba; } } } @@ -1726,52 +1727,52 @@ static u16 HalfToClampedByte(u16 half_bits) { return static_cast(clamped * 255.0f + 0.5f); } -static void FillError(std::span outBuf, u32 blockWidth, u32 blockHeight) { +static void FillError(std::span outBuf, u32 blockWidth, u32 blockHeight, u32 stride) { for (u32 j = 0; j < blockHeight; j++) { for (u32 i = 0; i < blockWidth; i++) { - outBuf[j * blockWidth + i] = 0x00000000; + outBuf[j * stride + i] = 0x00000000; } } } static void DecompressBlock(std::span inBuf, const u32 blockWidth, - const u32 blockHeight, std::span outBuf) { + const u32 blockHeight, std::span outBuf, u32 stride) { InputBitStream strm(inBuf); TexelWeightParams weightParams = DecodeBlockInfo(strm); // Was there an error? if (weightParams.m_bError) { assert(false && "Invalid block mode"); - FillError(outBuf, blockWidth, blockHeight); + FillError(outBuf, blockWidth, blockHeight, stride); return; } if (weightParams.m_bVoidExtentLDR) { - FillVoidExtentLDR(strm, outBuf, blockWidth, blockHeight); + FillVoidExtentLDR(strm, outBuf, blockWidth, blockHeight, stride); return; } if (weightParams.m_bVoidExtentHDR) { assert(false && "HDR void extent blocks are unsupported!"); - FillError(outBuf, blockWidth, blockHeight); + FillError(outBuf, blockWidth, blockHeight, stride); return; } if (weightParams.m_Width > blockWidth) { assert(false && "Texel weight grid width should be smaller than block width"); - FillError(outBuf, blockWidth, blockHeight); + FillError(outBuf, blockWidth, blockHeight, stride); return; } if (weightParams.m_Height > blockHeight) { assert(false && "Texel weight grid height should be smaller than block height"); - FillError(outBuf, blockWidth, blockHeight); + FillError(outBuf, blockWidth, blockHeight, stride); return; } if (weightParams.GetNumWeightValues() > 64) { assert(false && "Too many weights in the weight grid"); - FillError(outBuf, blockWidth, blockHeight); + FillError(outBuf, blockWidth, blockHeight, stride); return; } @@ -1781,7 +1782,7 @@ static void DecompressBlock(std::span inBuf, const u32 blockWidth, if (nPartitions == 4 && weightParams.m_bDualPlane) { assert(false && "Dual plane mode is incompatible with four partition blocks"); - FillError(outBuf, blockWidth, blockHeight); + FillError(outBuf, blockWidth, blockHeight, stride); return; } @@ -1813,7 +1814,7 @@ static void DecompressBlock(std::span inBuf, const u32 blockWidth, u32 nWeightBits = weightParams.GetPackedBitSize(); if (nWeightBits < 24 || nWeightBits > 96) { assert(false && "Invalid weight bit count"); - FillError(outBuf, blockWidth, blockHeight); + FillError(outBuf, blockWidth, blockHeight, stride); return; } s32 remainingBits = 128 - nWeightBits - static_cast(strm.GetBitsRead()); @@ -1995,41 +1996,54 @@ static void DecompressBlock(std::span inBuf, const u32 blockWidth, } } - outBuf[j * blockWidth + i] = p.Pack(); + outBuf[j * stride + i] = p.Pack(); } } +static void DecodeAndStoreBlock(std::span data, size_t src_offset, u32 x, u32 y, + u32 width, u32 height, u32 block_width, u32 block_height, + std::span out_slice) { + const u32 decomp_width = (std::min)(block_width, width - x); + const u32 decomp_height = (std::min)(block_height, height - y); + u32* const dst = out_slice.data() + size_t{y} * width + x; + if (src_offset + 16 > data.size()) { + for (u32 h = 0; h < decomp_height; ++h) { + std::memset(dst + size_t{h} * width, 0, decomp_width * 4); + } + return; + } + const std::span block_ptr{data.subspan(src_offset, 16)}; + if (decomp_width == block_width && decomp_height == block_height) { + const size_t touched = size_t{block_height - 1} * width + block_width; + DecompressBlock(block_ptr, block_width, block_height, std::span{dst, touched}, width); + return; + } + std::array staging; + DecompressBlock(block_ptr, block_width, block_height, staging, block_width); + for (u32 h = 0; h < decomp_height; ++h) { + std::memcpy(dst + size_t{h} * width, staging.data() + h * block_width, decomp_width * 4); + } +} + void Decompress(std::span data, uint32_t width, uint32_t height, uint32_t depth, uint32_t block_width, uint32_t block_height, std::span output) { const u32 rows = Common::DivideUp(height, block_height); const u32 cols = Common::DivideUp(width, block_width); Common::ThreadWorker& workers{GetThreadWorkers()}; + const std::span out{reinterpret_cast(output.data()), output.size() / 4}; + const size_t slice_texels = size_t{height} * width; for (u32 z = 0; z < depth; ++z) { - const u32 depth_offset = z * height * width * 4; + const std::span out_slice = out.subspan(z * slice_texels); for (u32 y_index = 0; y_index < rows; ++y_index) { - auto decompress_stride = [data, width, height, block_width, block_height, output, rows, - cols, z, depth_offset, y_index] { + auto decompress_stride = [data, width, height, block_width, block_height, out_slice, + rows, cols, z, y_index] { const u32 y = y_index * block_height; for (u32 x_index = 0; x_index < cols; ++x_index) { const u32 block_index = (z * rows * cols) + (y_index * cols) + x_index; - const u32 x = x_index * block_width; - - const std::span blockPtr{data.subspan(block_index * 16, 16)}; - - // Blocks can be at most 12x12 - std::array uncompData; - DecompressBlock(blockPtr, block_width, block_height, uncompData); - - u32 decompWidth = (std::min)(block_width, width - x); - u32 decompHeight = (std::min)(block_height, height - y); - - const std::span outRow = output.subspan(depth_offset + (y * width + x) * 4); - for (u32 h = 0; h < decompHeight; ++h) { - std::memcpy(outRow.data() + h * width * 4, - uncompData.data() + h * block_width, decompWidth * 4); - } + DecodeAndStoreBlock(data, size_t{block_index} * 16, x_index * block_width, y, + width, height, block_width, block_height, out_slice); } }; workers.QueueWork(std::move(decompress_stride)); @@ -2038,4 +2052,50 @@ void Decompress(std::span data, uint32_t width, uint32_t height, } } +void DecompressBlockLinear(std::span data, uint32_t width, uint32_t height, + uint32_t depth, uint32_t layers, uint32_t block_width, + uint32_t block_height, const BlockLinearLayout& layout, + std::span output) { + const u32 rows = Common::DivideUp(height, block_height); + const u32 cols = Common::DivideUp(width, block_width); + const size_t slice_texels = size_t{height} * width; + + Common::ThreadWorker& workers{GetThreadWorkers()}; + const std::span out{reinterpret_cast(output.data()), output.size() / 4}; + + for (u32 layer = 0; layer < layers; ++layer) { + const size_t layer_offset = size_t{layer} * layout.layer_stride; + for (u32 z = 0; z < depth; ++z) { + const size_t slice_offset = + layer_offset + (z >> layout.gob_depth) * size_t{layout.slice_size} + + ((z & layout.gob_depth_mask) << (GOB_SIZE_SHIFT + layout.gob_height)); + const std::span out_slice = + out.subspan((size_t{layer} * depth + z) * slice_texels); + for (u32 y_index = 0; y_index < rows; ++y_index) { + auto decompress_stride = [data, width, height, block_width, block_height, out_slice, + cols, layout, slice_offset, y_index] { + const u32 y = y_index * block_height; + const u32 gob_y = y_index >> GOB_SIZE_Y_SHIFT; + const size_t offset_y = + (gob_y >> layout.gob_height) * size_t{layout.block_size} + + ((gob_y & layout.gob_height_mask) << GOB_SIZE_SHIFT); + const u32 swizzled_y = ((y_index & 1) << 4) | ((y_index & 6) << 5); + for (u32 x_index = 0; x_index < cols; ++x_index) { + const u32 byte_x = x_index * 16; + const size_t offset_x = size_t{byte_x >> GOB_SIZE_X_SHIFT} + << layout.x_shift; + const u32 swizzled_x = ((x_index & 1) << 5) | ((x_index & 2) << 7); + const size_t src_offset = slice_offset + offset_y + offset_x + + (swizzled_x | swizzled_y); + DecodeAndStoreBlock(data, src_offset, x_index * block_width, y, width, + height, block_width, block_height, out_slice); + } + }; + workers.QueueWork(std::move(decompress_stride)); + } + } + } + workers.WaitForRequests(); +} + } // namespace Tegra::Texture::ASTC diff --git a/src/video_core/textures/astc.h b/src/video_core/textures/astc.h index afd3933c3e..5d59da7496 100644 --- a/src/video_core/textures/astc.h +++ b/src/video_core/textures/astc.h @@ -1,3 +1,6 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later @@ -5,7 +8,23 @@ namespace Tegra::Texture::ASTC { +struct BlockLinearLayout { + uint32_t layer_stride; + uint32_t slice_size; + uint32_t block_size; + uint32_t x_shift; + uint32_t gob_height; + uint32_t gob_height_mask; + uint32_t gob_depth; + uint32_t gob_depth_mask; +}; + void Decompress(std::span data, uint32_t width, uint32_t height, uint32_t depth, uint32_t block_width, uint32_t block_height, std::span output); +void DecompressBlockLinear(std::span data, uint32_t width, uint32_t height, + uint32_t depth, uint32_t layers, uint32_t block_width, + uint32_t block_height, const BlockLinearLayout& layout, + std::span output); + } // namespace Tegra::Texture::ASTC diff --git a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp index f4721543ae..4d2d5b1317 100644 --- a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp +++ b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp @@ -30,26 +30,6 @@ namespace { // Helpers translating MemoryUsage to flags/usage - [[maybe_unused]] VkMemoryPropertyFlags MemoryUsagePropertyFlags(MemoryUsage usage) { - switch (usage) { - case MemoryUsage::DeviceLocal: - return VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; - case MemoryUsage::Upload: - return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | - VK_MEMORY_PROPERTY_HOST_COHERENT_BIT; - case MemoryUsage::Download: - return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | - VK_MEMORY_PROPERTY_HOST_COHERENT_BIT | - VK_MEMORY_PROPERTY_HOST_CACHED_BIT; - case MemoryUsage::Stream: - return VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT | - VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | - VK_MEMORY_PROPERTY_HOST_COHERENT_BIT; - } - ASSERT_MSG(false, "Invalid memory usage={}", usage); - return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT; - } - [[nodiscard]] VkMemoryPropertyFlags MemoryUsagePreferredVmaFlags(MemoryUsage usage) { if (usage == MemoryUsage::Download) { return VK_MEMORY_PROPERTY_HOST_CACHED_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT; From 43622f463db2a750b3148884e35dcaa5635e691d Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Wed, 16 Sep 2026 16:59:20 -0400 Subject: [PATCH 07/69] A test for a corrupted unswizzled textures --- src/video_core/host_shaders/astc_decoder.comp | 26 +++++++------------ .../renderer_vulkan/vk_texture_cache.cpp | 6 ++++- 2 files changed, 14 insertions(+), 18 deletions(-) diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp index 58dccb545f..9a3954ca21 100644 --- a/src/video_core/host_shaders/astc_decoder.comp +++ b/src/video_core/host_shaders/astc_decoder.comp @@ -164,27 +164,19 @@ uint FastReplicateTo8(uint value, uint num_bits) { return ReplicateBits(value, num_bits, 8); } +const uint REPLICATE_TO_6_ODD_HALVES = ((63u >> 1u) << 5u) | ((21u >> 1u) << 10u) | + ((9u >> 1u) << 15u) | ((17u >> 1u) << 20u) | + ((33u >> 1u) << 25u); +const uint REPLICATE_TO_6_SHIFTS = (2u << 16u) | (4u << 20u); + uint FastReplicateTo6(uint value, uint num_bits) { - if (num_bits == 0) { - return 0; - } - if (num_bits >= 6) { + if (num_bits >= 6u) { return value; } const uint v = value & ((1u << num_bits) - 1u); - if (num_bits == 1) { - return v * 63u; - } - if (num_bits == 2) { - return v * 21u; - } - if (num_bits == 3) { - return v * 9u; - } - if (num_bits == 4) { - return (v << 2u) | (v >> 2u); - } - return (v << 1u) | (v >> 4u); + const uint c = ((REPLICATE_TO_6_ODD_HALVES >> (num_bits * 5u)) & 0x1fu) * 2u + 1u; + const uint s = (REPLICATE_TO_6_SHIFTS >> (num_bits * 4u)) & 0xfu; + return (v * c) >> s; } uint Div3Floor(uint v) { diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index caff739187..ff1ab0bf2c 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -57,7 +57,8 @@ using VideoCore::Surface::SurfaceType; namespace { constexpr bool ENABLE_MSAA_TILER_RESOLVE = true; constexpr bool ENABLE_MSAA_RESOLVE_CONSUME = true; -constexpr bool ENABLE_ACCELERATED_UNSWIZZLE = false; +constexpr bool ENABLE_ACCELERATED_UNSWIZZLE = true; +constexpr bool ENABLE_ACCELERATED_UNSWIZZLE_DISPATCH = false; constexpr bool ENABLE_MSAA_COLOR_DISCARD = true; constexpr bool ENABLE_MSAA_DEPTH_STENCIL_DISCARD = true; @@ -207,6 +208,9 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { } [[nodiscard]] bool WillUseAcceleratedUnswizzle(const Device& device, const ImageInfo& info) { + if (!ENABLE_ACCELERATED_UNSWIZZLE_DISPATCH) { + return false; + } switch (info.type) { case ImageType::e2D: case ImageType::e3D: From facee08b5ee9452a10ee0af0daf26544a7bcfdc9 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Wed, 16 Sep 2026 17:59:19 -0400 Subject: [PATCH 08/69] More tests to find out the texture corruption on previous unswizzle changes --- src/video_core/renderer_vulkan/vk_compute_pass.cpp | 5 ++--- src/video_core/renderer_vulkan/vk_texture_cache.cpp | 2 +- 2 files changed, 3 insertions(+), 4 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index b7bb4b00b4..76061d8cc9 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -647,7 +647,7 @@ void RecordUnswizzleBeginBarrier(Scheduler& scheduler, VkPipeline vk_pipeline, V VkImageLayout old_layout = VK_IMAGE_LAYOUT_UNDEFINED; VkPipelineStageFlags src_stage = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT; if (is_initialized) { - src_access = VK_ACCESS_SHADER_WRITE_BIT; + src_access = VK_ACCESS_MEMORY_WRITE_BIT; old_layout = VK_IMAGE_LAYOUT_GENERAL; src_stage = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER; } @@ -683,8 +683,7 @@ void RecordUnswizzleEndBarrier(Scheduler& scheduler, VkImage vk_image, .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, .pNext = nullptr, .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index ff1ab0bf2c..2b285ad3ea 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -58,7 +58,7 @@ namespace { constexpr bool ENABLE_MSAA_TILER_RESOLVE = true; constexpr bool ENABLE_MSAA_RESOLVE_CONSUME = true; constexpr bool ENABLE_ACCELERATED_UNSWIZZLE = true; -constexpr bool ENABLE_ACCELERATED_UNSWIZZLE_DISPATCH = false; +constexpr bool ENABLE_ACCELERATED_UNSWIZZLE_DISPATCH = true; constexpr bool ENABLE_MSAA_COLOR_DISCARD = true; constexpr bool ENABLE_MSAA_DEPTH_STENCIL_DISCARD = true; From 6221bd2c43e80bc0d9b1c303fd94481e23e8c3cb Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Wed, 16 Sep 2026 18:23:20 -0400 Subject: [PATCH 09/69] Reduce accelerated unswizzle path --- src/video_core/renderer_vulkan/vk_texture_cache.cpp | 3 +++ 1 file changed, 3 insertions(+) diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 2b285ad3ea..ed4105efa6 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -204,6 +204,9 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { if (VideoCore::Surface::GetFormatType(format) != SurfaceType::ColorTexture) { return false; } + if (!MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, false, format).storage) { + return false; + } return IsUnswizzleStorageFormatSupported(device, BytesPerBlock(format)); } From b016c8876878ecc737862fc607bf0ad1e9f88409 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Wed, 16 Sep 2026 21:12:40 -0400 Subject: [PATCH 10/69] Remove the unswizzle path on the extra shaders --- .../renderer_vulkan/vk_compute_pass.cpp | 250 ------------------ .../renderer_vulkan/vk_compute_pass.h | 51 ---- .../renderer_vulkan/vk_texture_cache.cpp | 130 +-------- .../renderer_vulkan/vk_texture_cache.h | 3 - 4 files changed, 6 insertions(+), 428 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 76061d8cc9..604a0d7cac 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -16,16 +16,12 @@ #include "common/common_types.h" #include "common/div_ceil.h" #include "common/vector_math.h" -#include #include "video_core/host_shaders/astc_decoder_comp_spv.h" #include "video_core/host_shaders/queries_prefix_scan_sum_comp_spv.h" #include "video_core/host_shaders/queries_prefix_scan_sum_nosubgroups_comp_spv.h" #include "video_core/host_shaders/resolve_conditional_render_comp_spv.h" #include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h" #include "video_core/host_shaders/vulkan_uint8_comp_spv.h" -#include "video_core/host_shaders/block_linear_unswizzle_2d_comp_spv.h" -#include "video_core/host_shaders/block_linear_unswizzle_3d_comp_spv.h" -#include "video_core/host_shaders/pitch_unswizzle_comp_spv.h" #include "video_core/renderer_vulkan/vk_compute_pass.h" #include "video_core/surface.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" @@ -188,26 +184,6 @@ struct AstcPushConstants { u32 block_height_mask; }; -struct BlockLinear3DImagePushConstants { - alignas(16) std::array origin; - alignas(16) std::array destination; - u32 bytes_per_block_log2; - u32 slice_size; - u32 block_size; - u32 x_shift; - u32 block_height; - u32 block_height_mask; - u32 block_depth; - u32 block_depth_mask; -}; - -struct PitchUnswizzlePushConstants { - std::array origin; - std::array destination; - u32 bytes_per_block_log2; - u32 pitch; -}; - struct QueriesPrefixScanPushConstants { u32 min_accumulation_base; u32 max_accumulation_base; @@ -639,230 +615,4 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, scheduler.Finish(); } -namespace { - -void RecordUnswizzleBeginBarrier(Scheduler& scheduler, VkPipeline vk_pipeline, VkImage vk_image, - VkImageAspectFlags aspect_mask, bool is_initialized) { - VkAccessFlags src_access = VK_ACCESS_NONE; - VkImageLayout old_layout = VK_IMAGE_LAYOUT_UNDEFINED; - VkPipelineStageFlags src_stage = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT; - if (is_initialized) { - src_access = VK_ACCESS_MEMORY_WRITE_BIT; - old_layout = VK_IMAGE_LAYOUT_GENERAL; - src_stage = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER; - } - scheduler.Record([vk_pipeline, vk_image, aspect_mask, src_access, old_layout, - src_stage](vk::CommandBuffer cmdbuf) { - const VkImageMemoryBarrier image_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = src_access, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, - .oldLayout = old_layout, - .newLayout = VK_IMAGE_LAYOUT_GENERAL, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .image = vk_image, - .subresourceRange{ - .aspectMask = aspect_mask, - .baseMipLevel = 0, - .levelCount = VK_REMAINING_MIP_LEVELS, - .baseArrayLayer = 0, - .layerCount = VK_REMAINING_ARRAY_LAYERS, - }, - }; - cmdbuf.PipelineBarrier(src_stage, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier); - cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline); - }); -} - -void RecordUnswizzleEndBarrier(Scheduler& scheduler, VkImage vk_image, - VkImageAspectFlags aspect_mask) { - scheduler.Record([vk_image, aspect_mask](vk::CommandBuffer cmdbuf) { - const VkImageMemoryBarrier image_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, - .oldLayout = VK_IMAGE_LAYOUT_GENERAL, - .newLayout = VK_IMAGE_LAYOUT_GENERAL, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .image = vk_image, - .subresourceRange{ - .aspectMask = aspect_mask, - .baseMipLevel = 0, - .levelCount = VK_REMAINING_MIP_LEVELS, - .baseArrayLayer = 0, - .layerCount = VK_REMAINING_ARRAY_LAYERS, - }, - }; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, 0, image_barrier); - }); -} - -} // Anonymous namespace - -BlockLinearUnswizzleImage2DPass::BlockLinearUnswizzleImage2DPass( - const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, - StagingBufferPool& staging_buffer_pool_, - ComputePassDescriptorQueue& compute_pass_descriptor_queue_) - : ComputePass(device_, scheduler_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, - ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, - COMPUTE_PUSH_CONSTANT_RANGE< - sizeof(VideoCommon::Accelerated::BlockLinearSwizzle2DParams)>, - BLOCK_LINEAR_UNSWIZZLE_2D_COMP_SPV), - scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, - compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} - -BlockLinearUnswizzleImage2DPass::~BlockLinearUnswizzleImage2DPass() = default; - -void BlockLinearUnswizzleImage2DPass::Unswizzle( - Image& image, const StagingBufferRef& map, - std::span swizzles) { - using namespace VideoCommon::Accelerated; - scheduler.RequestOutsideRenderPassOperationContext(); - const VkImageAspectFlags aspect_mask = image.AspectMask(); - const VkImage vk_image = image.Handle(); - RecordUnswizzleBeginBarrier(scheduler, *pipeline, vk_image, aspect_mask, - image.ExchangeInitialization()); - for (const VideoCommon::SwizzleParameters& swizzle : swizzles) { - const size_t input_offset = swizzle.buffer_offset + map.offset; - const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U); - const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U); - const u32 num_dispatches_z = image.info.resources.layers; - - compute_pass_descriptor_queue.Acquire(scheduler, 2); - compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset, - image.guest_size_bytes - swizzle.buffer_offset); - compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level)); - const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; - - const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info); - scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, params, - descriptor_data](vk::CommandBuffer cmdbuf) { - const VkDescriptorSet set = descriptor_allocator.Commit(); - device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); - cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); - cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params); - cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_dispatches_z); - }); - } - RecordUnswizzleEndBarrier(scheduler, vk_image, aspect_mask); - scheduler.Finish(); -} - -BlockLinearUnswizzleImage3DPass::BlockLinearUnswizzleImage3DPass( - const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, - StagingBufferPool& staging_buffer_pool_, - ComputePassDescriptorQueue& compute_pass_descriptor_queue_) - : ComputePass(device_, scheduler_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, - ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, - COMPUTE_PUSH_CONSTANT_RANGE, - BLOCK_LINEAR_UNSWIZZLE_3D_COMP_SPV), - scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, - compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} - -BlockLinearUnswizzleImage3DPass::~BlockLinearUnswizzleImage3DPass() = default; - -void BlockLinearUnswizzleImage3DPass::Unswizzle( - Image& image, const StagingBufferRef& map, - std::span swizzles) { - using namespace VideoCommon::Accelerated; - scheduler.RequestOutsideRenderPassOperationContext(); - const VkImageAspectFlags aspect_mask = image.AspectMask(); - const VkImage vk_image = image.Handle(); - RecordUnswizzleBeginBarrier(scheduler, *pipeline, vk_image, aspect_mask, - image.ExchangeInitialization()); - for (const VideoCommon::SwizzleParameters& swizzle : swizzles) { - const size_t input_offset = swizzle.buffer_offset + map.offset; - const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 16U); - const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U); - const u32 num_dispatches_z = Common::DivCeil(swizzle.num_tiles.depth, 2U); - - compute_pass_descriptor_queue.Acquire(scheduler, 2); - compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset, - image.guest_size_bytes - swizzle.buffer_offset); - compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level)); - const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; - - const auto p = MakeBlockLinearSwizzle3DParams(swizzle, image.info); - const BlockLinear3DImagePushConstants params{ - .origin = p.origin, - .destination = p.destination, - .bytes_per_block_log2 = p.bytes_per_block_log2, - .slice_size = p.slice_size, - .block_size = p.block_size, - .x_shift = p.x_shift, - .block_height = p.block_height, - .block_height_mask = p.block_height_mask, - .block_depth = p.block_depth, - .block_depth_mask = p.block_depth_mask, - }; - scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, params, - descriptor_data](vk::CommandBuffer cmdbuf) { - const VkDescriptorSet set = descriptor_allocator.Commit(); - device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); - cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); - cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params); - cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_dispatches_z); - }); - } - RecordUnswizzleEndBarrier(scheduler, vk_image, aspect_mask); - scheduler.Finish(); -} - -PitchUnswizzlePass::PitchUnswizzlePass(const Device& device_, Scheduler& scheduler_, - DescriptorPool& descriptor_pool_, - StagingBufferPool& staging_buffer_pool_, - ComputePassDescriptorQueue& compute_pass_descriptor_queue_) - : ComputePass(device_, scheduler_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, - ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, - COMPUTE_PUSH_CONSTANT_RANGE, - PITCH_UNSWIZZLE_COMP_SPV), - scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, - compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} - -PitchUnswizzlePass::~PitchUnswizzlePass() = default; - -void PitchUnswizzlePass::Unswizzle(Image& image, const StagingBufferRef& map, - std::span swizzles) { - const u32 bytes_per_block = VideoCore::Surface::BytesPerBlock(image.info.format); - ASSERT(std::has_single_bit(bytes_per_block)); - scheduler.RequestOutsideRenderPassOperationContext(); - const VkImageAspectFlags aspect_mask = image.AspectMask(); - const VkImage vk_image = image.Handle(); - RecordUnswizzleBeginBarrier(scheduler, *pipeline, vk_image, aspect_mask, - image.ExchangeInitialization()); - for (const VideoCommon::SwizzleParameters& swizzle : swizzles) { - const size_t input_offset = swizzle.buffer_offset + map.offset; - const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U); - const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U); - - compute_pass_descriptor_queue.Acquire(scheduler, 2); - compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset, - image.guest_size_bytes - swizzle.buffer_offset); - compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level)); - const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; - - const PitchUnswizzlePushConstants params{ - .origin = {0, 0}, - .destination = {0, 0}, - .bytes_per_block_log2 = static_cast(std::countr_zero(bytes_per_block)), - .pitch = image.info.pitch, - }; - scheduler.Record([this, num_dispatches_x, num_dispatches_y, params, - descriptor_data](vk::CommandBuffer cmdbuf) { - const VkDescriptorSet set = descriptor_allocator.Commit(); - device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); - cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); - cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params); - cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, 1); - }); - } - RecordUnswizzleEndBarrier(scheduler, vk_image, aspect_mask); - scheduler.Finish(); -} - } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h index 811deee8a2..12723099ca 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.h +++ b/src/video_core/renderer_vulkan/vk_compute_pass.h @@ -134,55 +134,4 @@ private: MemoryAllocator& memory_allocator; }; -class BlockLinearUnswizzleImage2DPass final : public ComputePass { -public: - explicit BlockLinearUnswizzleImage2DPass( - const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, - StagingBufferPool& staging_buffer_pool_, - ComputePassDescriptorQueue& compute_pass_descriptor_queue_); - ~BlockLinearUnswizzleImage2DPass(); - - void Unswizzle(Image& image, const StagingBufferRef& map, - std::span swizzles); - -private: - Scheduler& scheduler; - StagingBufferPool& staging_buffer_pool; - ComputePassDescriptorQueue& compute_pass_descriptor_queue; -}; - -class BlockLinearUnswizzleImage3DPass final : public ComputePass { -public: - explicit BlockLinearUnswizzleImage3DPass( - const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, - StagingBufferPool& staging_buffer_pool_, - ComputePassDescriptorQueue& compute_pass_descriptor_queue_); - ~BlockLinearUnswizzleImage3DPass(); - - void Unswizzle(Image& image, const StagingBufferRef& map, - std::span swizzles); - -private: - Scheduler& scheduler; - StagingBufferPool& staging_buffer_pool; - ComputePassDescriptorQueue& compute_pass_descriptor_queue; -}; - -class PitchUnswizzlePass final : public ComputePass { -public: - explicit PitchUnswizzlePass(const Device& device_, Scheduler& scheduler_, - DescriptorPool& descriptor_pool_, - StagingBufferPool& staging_buffer_pool_, - ComputePassDescriptorQueue& compute_pass_descriptor_queue_); - ~PitchUnswizzlePass(); - - void Unswizzle(Image& image, const StagingBufferRef& map, - std::span swizzles); - -private: - Scheduler& scheduler; - StagingBufferPool& staging_buffer_pool; - ComputePassDescriptorQueue& compute_pass_descriptor_queue; -}; - } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index ed4105efa6..5663220045 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -57,8 +57,6 @@ using VideoCore::Surface::SurfaceType; namespace { constexpr bool ENABLE_MSAA_TILER_RESOLVE = true; constexpr bool ENABLE_MSAA_RESOLVE_CONSUME = true; -constexpr bool ENABLE_ACCELERATED_UNSWIZZLE = true; -constexpr bool ENABLE_ACCELERATED_UNSWIZZLE_DISPATCH = true; constexpr bool ENABLE_MSAA_COLOR_DISCARD = true; constexpr bool ENABLE_MSAA_DEPTH_STENCIL_DISCARD = true; @@ -162,82 +160,6 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { info.size.depth == 1; } -[[nodiscard]] VkFormat UnswizzleStorageFormat(u32 bytes_per_block) { - switch (bytes_per_block) { - case 1: - return VK_FORMAT_R8_UINT; - case 2: - return VK_FORMAT_R16_UINT; - case 4: - return VK_FORMAT_R32_UINT; - case 8: - return VK_FORMAT_R32G32_UINT; - case 16: - return VK_FORMAT_R32G32B32A32_UINT; - default: - return VK_FORMAT_UNDEFINED; - } -} - -[[nodiscard]] bool IsUnswizzleStorageFormatSupported(const Device& device, u32 bytes_per_block) { - switch (bytes_per_block) { - case 1: - return device.IsStorageBuffer8BitAccessSupported(); - case 2: - return device.IsStorageBuffer16BitAccessSupported(); - case 4: - case 8: - case 16: - return true; - default: - return false; - } -} - -[[nodiscard]] bool IsUnswizzleAcceleratedFormat(const Device& device, PixelFormat format) { - if (!ENABLE_ACCELERATED_UNSWIZZLE) { - return false; - } - if (IsPixelFormatASTC(format) || VideoCore::Surface::IsPixelFormatBCn(format)) { - return false; - } - if (VideoCore::Surface::GetFormatType(format) != SurfaceType::ColorTexture) { - return false; - } - if (!MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, false, format).storage) { - return false; - } - return IsUnswizzleStorageFormatSupported(device, BytesPerBlock(format)); -} - -[[nodiscard]] bool WillUseAcceleratedUnswizzle(const Device& device, const ImageInfo& info) { - if (!ENABLE_ACCELERATED_UNSWIZZLE_DISPATCH) { - return false; - } - switch (info.type) { - case ImageType::e2D: - case ImageType::e3D: - case ImageType::Linear: - break; - default: - return false; - } - if (info.num_samples > 1 || !device.IsKhrImageFormatListSupported()) { - return false; - } - return IsUnswizzleAcceleratedFormat(device, info.format); -} - -[[nodiscard]] VkImageViewType UnswizzleStorageViewType(ImageType type) { - if (type == ImageType::e3D) { - return VK_IMAGE_VIEW_TYPE_3D; - } - if (type == ImageType::Linear) { - return VK_IMAGE_VIEW_TYPE_2D; - } - return VK_IMAGE_VIEW_TYPE_2D_ARRAY; -} - [[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info, std::optional format_override = {}) { auto format_info = @@ -327,23 +249,18 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { } [[nodiscard]] vk::ImageView MakeStorageView(const vk::Device& device, u32 level, VkImage image, - VkFormat format, - VkImageViewType view_type = VK_IMAGE_VIEW_TYPE_2D_ARRAY) { + VkFormat format) { static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO, .pNext = nullptr, .usage = VK_IMAGE_USAGE_STORAGE_BIT, }; - u32 layer_count = VK_REMAINING_ARRAY_LAYERS; - if (view_type == VK_IMAGE_VIEW_TYPE_2D || view_type == VK_IMAGE_VIEW_TYPE_3D) { - layer_count = 1; - } return device.CreateImageView(VkImageViewCreateInfo{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, .pNext = &storage_image_view_usage_create_info, .flags = 0, .image = image, - .viewType = view_type, + .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY, .format = format, .components{ .r = VK_COMPONENT_SWIZZLE_IDENTITY, @@ -356,7 +273,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { .baseMipLevel = level, .levelCount = 1, .baseArrayLayer = 0, - .layerCount = layer_count, + .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }); } @@ -1029,14 +946,6 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched astc_decoder_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, compute_pass_descriptor_queue, memory_allocator); } - if (ENABLE_ACCELERATED_UNSWIZZLE && device.IsKhrImageFormatListSupported()) { - bl_unswizzle_2d_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, - compute_pass_descriptor_queue); - bl_unswizzle_3d_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, - compute_pass_descriptor_queue); - pitch_unswizzle_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, - compute_pass_descriptor_queue); - } if (!device.IsKhrImageFormatListSupported()) { return; } @@ -1044,8 +953,6 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched const auto image_format = static_cast(index_a); if (IsPixelFormatASTC(image_format) && !device.IsOptimalAstcSupported()) { view_formats[index_a].push_back(VK_FORMAT_A8B8G8R8_UNORM_PACK32); - } else if (IsUnswizzleAcceleratedFormat(device, image_format)) { - view_formats[index_a].push_back(UnswizzleStorageFormat(BytesPerBlock(image_format))); } for (size_t index_b = 0; index_b < VideoCore::Surface::MaxPixelFormat; index_b++) { const auto view_format = static_cast(index_b); @@ -1988,10 +1895,6 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu flags |= VideoCommon::ImageFlagBits::Converted; flags |= VideoCommon::ImageFlagBits::CostlyLoad; } - if (False(flags & VideoCommon::ImageFlagBits::Converted) && - WillUseAcceleratedUnswizzle(runtime->device, info)) { - flags |= VideoCommon::ImageFlagBits::AcceleratedUpload; - } if (runtime->device.HasDebuggingToolAttached()) { original_image.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); } @@ -2377,15 +2280,11 @@ VkImageView Image::StorageImageView(s32 level) noexcept { if (!view) { auto format_info = MaxwellToVK::SurfaceFormat(runtime->device, FormatType::Optimal, true, info.format); - VkImageViewType view_type = VK_IMAGE_VIEW_TYPE_2D_ARRAY; if (WillUseAcceleratedAstcDecode(runtime->device, info)) { format_info.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32; - } else if (True(flags & VideoCommon::ImageFlagBits::AcceleratedUpload)) { - format_info.format = UnswizzleStorageFormat(BytesPerBlock(info.format)); - view_type = UnswizzleStorageViewType(info.type); } view = MakeStorageView(runtime->device.GetLogical(), level, *(this->*current_image), - format_info.format, view_type); + format_info.format); } return *view; } @@ -2778,7 +2677,7 @@ vk::ImageView ImageView::MakeView(VkFormat vk_format, VkImageAspectFlags aspect_ .pNext = nullptr, .flags = 0, .image = image_handle, - .viewType = view_type, + .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY, .format = vk_format, .components{ .r = VK_COMPONENT_SWIZZLE_IDENTITY, @@ -3189,24 +3088,7 @@ void TextureCacheRuntime::AccelerateImageUpload( if (is_rescaled) { image.ScaleDown(true); } - if (IsPixelFormatASTC(image.info.format)) { - astc_decoder_pass->Assemble(image, map, swizzles); - } else { - switch (image.info.type) { - case ImageType::e2D: - bl_unswizzle_2d_pass->Unswizzle(image, map, swizzles); - break; - case ImageType::e3D: - bl_unswizzle_3d_pass->Unswizzle(image, map, swizzles); - break; - case ImageType::Linear: - pitch_unswizzle_pass->Unswizzle(image, map, swizzles); - break; - default: - ASSERT(false); - break; - } - } + astc_decoder_pass->Assemble(image, map, swizzles); if (is_rescaled) { image.ScaleUp(); } diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 2e6b547e72..6510e10e50 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -156,9 +156,6 @@ public: BlitImageHelper& blit_image_helper; RenderPassCache& render_pass_cache; std::optional astc_decoder_pass; - std::optional bl_unswizzle_2d_pass; - std::optional bl_unswizzle_3d_pass; - std::optional pitch_unswizzle_pass; const Settings::ResolutionScalingInfo& resolution; std::array, VideoCore::Surface::MaxPixelFormat> view_formats; From 0cc76f1f22cc7ef376e45e26f3344b6a1b15fa5a Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Wed, 16 Sep 2026 21:26:19 -0400 Subject: [PATCH 11/69] First intent to relax the pressure on the gpu thread for unswizzling images --- src/video_core/textures/decoders.cpp | 63 ++++++++++++++++++---------- 1 file changed, 42 insertions(+), 21 deletions(-) diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index 25b0860b0f..d7eef06a4c 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -1,9 +1,10 @@ -// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project // SPDX-License-Identifier: GPL-3.0-or-later // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later +#include #include #include #include @@ -15,6 +16,7 @@ #include "common/div_ceil.h" #include "video_core/gpu.h" #include "video_core/textures/decoders.h" +#include "video_core/textures/workers.h" namespace Tegra::Texture { namespace { @@ -36,6 +38,8 @@ void incrpdep(u32& value) { value = ((value | ~mask) + swizzled_incr) & mask; } +constexpr u32 PARALLEL_SWIZZLE_BYTES = 256 * 1024; + template void SwizzleImpl(std::span output, std::span input, u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth, u32 stride) { @@ -58,37 +62,55 @@ void SwizzleImpl(std::span output, std::span input, u32 width, u32 const u32 block_depth_mask = (1U << block_depth) - 1; const u32 x_shift = GOB_SIZE_SHIFT + block_height + block_depth; - for (u32 slice = 0; slice < depth; ++slice) { + const auto process = [&](u32 slice, u32 line_begin, u32 line_end) { const u32 z = slice + origin_z; const u32 offset_z = (z >> block_depth) * slice_size + ((z & block_depth_mask) << (GOB_SIZE_SHIFT + block_height)); - for (u32 line = 0; line < height; ++line) { + const u32 slice_base = slice * pitch * height; + for (u32 line = line_begin; line < line_end; ++line) { const u32 y = line + origin_y; - const u32 swizzled_y = pdep(y); + const u32 swizzled_y = ((y & 1) << 4) | ((y & 6) << 5); const u32 block_y = y >> GOB_SIZE_Y_SHIFT; const u32 offset_y = (block_y >> block_height) * block_size + ((block_y & block_height_mask) << GOB_SIZE_SHIFT); + const u32 row_base = offset_z + offset_y + swizzled_y; + u32 linear_offset = slice_base + line * pitch; u32 swizzled_x = pdep(origin_x * BYTES_PER_PIXEL); for (u32 column = 0; column < width; - ++column, incrpdep(swizzled_x)) { + ++column, incrpdep(swizzled_x), + linear_offset += BYTES_PER_PIXEL) { const u32 x = (column + origin_x) * BYTES_PER_PIXEL; const u32 offset_x = (x >> GOB_SIZE_X_SHIFT) << x_shift; - const u32 base_swizzled_offset = offset_z + offset_y + offset_x; - const u32 swizzled_offset = base_swizzled_offset + (swizzled_x | swizzled_y); + const u32 swizzled_offset = row_base + offset_x + swizzled_x; - const u32 unswizzled_offset = - slice * pitch * height + line * pitch + column * BYTES_PER_PIXEL; - - u8* const dst = &output[TO_LINEAR ? swizzled_offset : unswizzled_offset]; - const u8* const src = &input[TO_LINEAR ? unswizzled_offset : swizzled_offset]; + u8* const dst = &output[TO_LINEAR ? swizzled_offset : linear_offset]; + const u8* const src = &input[TO_LINEAR ? linear_offset : swizzled_offset]; std::memcpy(dst, src, BYTES_PER_PIXEL); } } + }; + + const u32 rows_per_task = (std::max)(1U, PARALLEL_SWIZZLE_BYTES / (std::max)(1U, pitch)); + if (u64{depth} * height * pitch < PARALLEL_SWIZZLE_BYTES * 2) { + for (u32 slice = 0; slice < depth; ++slice) { + process(slice, 0, height); + } + return; } + Common::ThreadWorker& workers{GetThreadWorkers()}; + for (u32 slice = 0; slice < depth; ++slice) { + for (u32 line = 0; line < height; line += rows_per_task) { + const u32 line_end = (std::min)(line + rows_per_task, height); + workers.QueueWork([&process, slice, line, line_end] { + process(slice, line, line_end); + }); + } + } + workers.WaitForRequests(); } template @@ -123,26 +145,25 @@ void SwizzleSubrectImpl(std::span output, std::span input, u32 wid const u32 lines_in_y = (std::min)(unprocessed_lines, extent_y); for (u32 line = 0; line < lines_in_y; ++line) { const u32 y = line + origin_y; - const u32 swizzled_y = pdep(y); + const u32 swizzled_y = ((y & 1) << 4) | ((y & 6) << 5); const u32 block_y = y >> GOB_SIZE_Y_SHIFT; const u32 offset_y = (block_y >> block_height) * block_size + ((block_y & block_height_mask) << GOB_SIZE_SHIFT); + const u32 row_base = offset_z + offset_y + swizzled_y; + u32 linear_offset = slice * pitch * height + line * pitch; u32 swizzled_x = pdep(origin_x * BYTES_PER_PIXEL); for (u32 column = 0; column < extent_x; - ++column, incrpdep(swizzled_x)) { + ++column, incrpdep(swizzled_x), + linear_offset += BYTES_PER_PIXEL) { const u32 x = (column + origin_x) * BYTES_PER_PIXEL; const u32 offset_x = (x >> GOB_SIZE_X_SHIFT) << x_shift; - const u32 base_swizzled_offset = offset_z + offset_y + offset_x; - const u32 swizzled_offset = base_swizzled_offset + (swizzled_x | swizzled_y); + const u32 swizzled_offset = row_base + offset_x + swizzled_x; - const u32 unswizzled_offset = - slice * pitch * height + line * pitch + column * BYTES_PER_PIXEL; - - u8* const dst = &output[TO_LINEAR ? swizzled_offset : unswizzled_offset]; - const u8* const src = &input[TO_LINEAR ? unswizzled_offset : swizzled_offset]; + u8* const dst = &output[TO_LINEAR ? swizzled_offset : linear_offset]; + const u8* const src = &input[TO_LINEAR ? linear_offset : swizzled_offset]; std::memcpy(dst, src, BYTES_PER_PIXEL); } From 40036eb0f82358d9ba134c49d219d0ead8a6dc02 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Wed, 16 Sep 2026 22:40:54 -0400 Subject: [PATCH 12/69] Remove the thread opt for unswizzle images --- src/video_core/textures/decoders.cpp | 26 ++------------------------ 1 file changed, 2 insertions(+), 24 deletions(-) diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index d7eef06a4c..a914acace0 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -4,7 +4,6 @@ // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later -#include #include #include #include @@ -16,7 +15,6 @@ #include "common/div_ceil.h" #include "video_core/gpu.h" #include "video_core/textures/decoders.h" -#include "video_core/textures/workers.h" namespace Tegra::Texture { namespace { @@ -38,8 +36,6 @@ void incrpdep(u32& value) { value = ((value | ~mask) + swizzled_incr) & mask; } -constexpr u32 PARALLEL_SWIZZLE_BYTES = 256 * 1024; - template void SwizzleImpl(std::span output, std::span input, u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth, u32 stride) { @@ -62,12 +58,12 @@ void SwizzleImpl(std::span output, std::span input, u32 width, u32 const u32 block_depth_mask = (1U << block_depth) - 1; const u32 x_shift = GOB_SIZE_SHIFT + block_height + block_depth; - const auto process = [&](u32 slice, u32 line_begin, u32 line_end) { + for (u32 slice = 0; slice < depth; ++slice) { const u32 z = slice + origin_z; const u32 offset_z = (z >> block_depth) * slice_size + ((z & block_depth_mask) << (GOB_SIZE_SHIFT + block_height)); const u32 slice_base = slice * pitch * height; - for (u32 line = line_begin; line < line_end; ++line) { + for (u32 line = 0; line < height; ++line) { const u32 y = line + origin_y; const u32 swizzled_y = ((y & 1) << 4) | ((y & 6) << 5); @@ -92,25 +88,7 @@ void SwizzleImpl(std::span output, std::span input, u32 width, u32 std::memcpy(dst, src, BYTES_PER_PIXEL); } } - }; - - const u32 rows_per_task = (std::max)(1U, PARALLEL_SWIZZLE_BYTES / (std::max)(1U, pitch)); - if (u64{depth} * height * pitch < PARALLEL_SWIZZLE_BYTES * 2) { - for (u32 slice = 0; slice < depth; ++slice) { - process(slice, 0, height); - } - return; } - Common::ThreadWorker& workers{GetThreadWorkers()}; - for (u32 slice = 0; slice < depth; ++slice) { - for (u32 line = 0; line < height; line += rows_per_task) { - const u32 line_end = (std::min)(line + rows_per_task, height); - workers.QueueWork([&process, slice, line, line_end] { - process(slice, line, line_end); - }); - } - } - workers.WaitForRequests(); } template From a152596f29435f7cebcc0b3ce526282cbff3fde9 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Wed, 16 Sep 2026 23:57:44 -0400 Subject: [PATCH 13/69] Chinga su madre que perramente molesto es este puto juego --- src/video_core/renderer_vulkan/vk_texture_cache.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 5663220045..1e3520e21a 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -2677,7 +2677,7 @@ vk::ImageView ImageView::MakeView(VkFormat vk_format, VkImageAspectFlags aspect_ .pNext = nullptr, .flags = 0, .image = image_handle, - .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY, + .viewType = view_type, .format = vk_format, .components{ .r = VK_COMPONENT_SWIZZLE_IDENTITY, From bd82351d73b8c89c07868252bf09facba4d34473 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 00:56:50 -0400 Subject: [PATCH 14/69] First try to align GOB's --- src/video_core/textures/decoders.cpp | 47 +++++++++++++++------------- src/video_core/textures/decoders.h | 1 - 2 files changed, 26 insertions(+), 22 deletions(-) diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index a914acace0..29d44b77db 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -41,9 +41,9 @@ void SwizzleImpl(std::span output, std::span input, u32 width, u32 u32 block_height, u32 block_depth, u32 stride) { // The origin of the transformation can be configured here, leave it as zero as the current API // doesn't expose it. - static constexpr u32 origin_x = 0; static constexpr u32 origin_y = 0; static constexpr u32 origin_z = 0; + static constexpr u32 columns_per_gob = GOB_SIZE_X / BYTES_PER_PIXEL; // We can configure here a custom pitch // As it's not exposed 'width * BYTES_PER_PIXEL' will be the expected pitch. @@ -63,29 +63,37 @@ void SwizzleImpl(std::span output, std::span input, u32 width, u32 const u32 offset_z = (z >> block_depth) * slice_size + ((z & block_depth_mask) << (GOB_SIZE_SHIFT + block_height)); const u32 slice_base = slice * pitch * height; - for (u32 line = 0; line < height; ++line) { - const u32 y = line + origin_y; - const u32 swizzled_y = ((y & 1) << 4) | ((y & 6) << 5); - - const u32 block_y = y >> GOB_SIZE_Y_SHIFT; + for (u32 band = 0; band < height; band += GOB_SIZE_Y) { + const u32 band_end = (std::min)(band + GOB_SIZE_Y, height); + const u32 block_y = (band + origin_y) >> GOB_SIZE_Y_SHIFT; const u32 offset_y = (block_y >> block_height) * block_size + ((block_y & block_height_mask) << GOB_SIZE_SHIFT); - const u32 row_base = offset_z + offset_y + swizzled_y; - u32 linear_offset = slice_base + line * pitch; - u32 swizzled_x = pdep(origin_x * BYTES_PER_PIXEL); - for (u32 column = 0; column < width; - ++column, incrpdep(swizzled_x), - linear_offset += BYTES_PER_PIXEL) { - const u32 x = (column + origin_x) * BYTES_PER_PIXEL; - const u32 offset_x = (x >> GOB_SIZE_X_SHIFT) << x_shift; + const u32 band_base = offset_z + offset_y; + u32 band_linear = slice_base + band * pitch; + u32 offset_x = 0; + for (u32 gob = 0; gob < width; gob += columns_per_gob, + offset_x += 1U << x_shift, band_linear += GOB_SIZE_X) { + const u32 columns = (std::min)(columns_per_gob, width - gob); + const u32 gob_base = band_base + offset_x; + u32 linear_row = band_linear; + for (u32 line = band; line < band_end; ++line, linear_row += pitch) { + const u32 y = line + origin_y; + const u32 row_base = gob_base + ((y & 1) << 4) + ((y & 6) << 5); - const u32 swizzled_offset = row_base + offset_x + swizzled_x; + u32 linear_offset = linear_row; + u32 swizzled_x = 0; + for (u32 column = 0; column < columns; + ++column, incrpdep(swizzled_x), + linear_offset += BYTES_PER_PIXEL) { + const u32 swizzled_offset = row_base + swizzled_x; - u8* const dst = &output[TO_LINEAR ? swizzled_offset : linear_offset]; - const u8* const src = &input[TO_LINEAR ? linear_offset : swizzled_offset]; + u8* const dst = &output[TO_LINEAR ? swizzled_offset : linear_offset]; + const u8* const src = &input[TO_LINEAR ? linear_offset : swizzled_offset]; - std::memcpy(dst, src, BYTES_PER_PIXEL); + std::memcpy(dst, src, BYTES_PER_PIXEL); + } + } } } } @@ -163,11 +171,8 @@ void Swizzle(std::span output, std::span input, u32 bytes_per_pixe block_depth, stride_alignment); BPP_CASE(1) BPP_CASE(2) - BPP_CASE(3) BPP_CASE(4) - BPP_CASE(6) BPP_CASE(8) - BPP_CASE(12) BPP_CASE(16) #undef BPP_CASE default: diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h index 1e14a38753..8018f444be 100644 --- a/src/video_core/textures/decoders.h +++ b/src/video_core/textures/decoders.h @@ -24,7 +24,6 @@ constexpr u32 GOB_SIZE_Z_SHIFT = 0; constexpr u32 GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_SHIFT; constexpr u32 SWIZZLE_X_BITS = 0b100101111; -constexpr u32 SWIZZLE_Y_BITS = 0b011010000; /// Unswizzles a block linear texture into linear memory. void UnswizzleTexture(std::span output, std::span input, u32 bytes_per_pixel, From 052a04e610a38936b0e7d6100a8af8c8b90f7a4d Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 02:20:40 -0400 Subject: [PATCH 15/69] Correcting some bugs on the vulkan async presentation thread --- .../renderer_vulkan/renderer_vulkan.cpp | 6 +++-- .../renderer_vulkan/vk_present_manager.cpp | 26 ++++++++++++++++++- .../renderer_vulkan/vk_present_manager.h | 10 ++++++- 3 files changed, 38 insertions(+), 4 deletions(-) diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp index 4fd4c0233b..27ca531186 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp +++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp @@ -217,13 +217,15 @@ void RendererVulkan::Composite(std::span framebu scheduler.RequestOutsideRenderPassOperationContext(); blit_swapchain.DrawToFrame(device, rasterizer, frame, framebuffers, - render_window.GetFramebufferLayout(), swapchain.GetImageCount(), + render_window.GetFramebufferLayout(), + present_manager.SwapchainImageCount(), swapchain.GetImageViewFormat()); #ifdef HAS_LSFG void(frame_gen.WantedGenerations(present_manager.MaxExtraFrames())); - frame_gen.Process(device, frame, swapchain.GetImageFormat(), GuestExtent(framebuffers)); + frame_gen.Process(device, frame, present_manager.SwapchainImageFormat(), + GuestExtent(framebuffers)); const size_t generated_frames = frame_gen.GeneratedFrameCount(); for (size_t generation = 0; generation < generated_frames; ++generation) { diff --git a/src/video_core/renderer_vulkan/vk_present_manager.cpp b/src/video_core/renderer_vulkan/vk_present_manager.cpp index d173a3ba5f..9298764546 100644 --- a/src/video_core/renderer_vulkan/vk_present_manager.cpp +++ b/src/video_core/renderer_vulkan/vk_present_manager.cpp @@ -23,6 +23,7 @@ namespace Vulkan { namespace { constexpr size_t MAX_FRAMES_IN_FLIGHT = 7; +constexpr u32 MAX_PRESENT_ATTEMPTS = 3; #ifdef HAS_LSFG static_assert(MAX_FRAMES_IN_FLIGHT <= LSFG_MAX_TARGETS); #endif @@ -372,12 +373,33 @@ void PresentManager::SetImageCount() { #else image_count = std::min(swapchain.GetImageCount(), MAX_FRAMES_IN_FLIGHT); #endif + swapchain_image_count = swapchain.GetImageCount(); + swapchain_image_format = swapchain.GetImageFormat(); +} + +void PresentManager::DiscardFrame(Frame* frame) { + static constexpr VkPipelineStageFlags wait_stage = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT; + const VkSemaphore render_ready = *frame->render_ready; + const VkSubmitInfo submit_info{ + .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO, + .pNext = nullptr, + .waitSemaphoreCount = 1U, + .pWaitSemaphores = &render_ready, + .pWaitDstStageMask = &wait_stage, + .commandBufferCount = 0U, + .pCommandBuffers = nullptr, + .signalSemaphoreCount = 0U, + .pSignalSemaphores = nullptr, + }; + + std::scoped_lock submit_lock{scheduler.submit_mutex}; + void(device.GetGraphicsQueue().Submit(submit_info, *frame->present_done)); } void PresentManager::CopyToSwapchain(Frame* frame) { bool requires_recreation = false; - while (true) { + for (u32 attempt = 0; attempt < MAX_PRESENT_ATTEMPTS; ++attempt) { try { // Recreate surface and swapchain if needed. if (requires_recreation) { @@ -397,6 +419,8 @@ void PresentManager::CopyToSwapchain(Frame* frame) { requires_recreation = true; } } + + DiscardFrame(frame); } void PresentManager::CopyToSwapchainImpl(Frame* frame) { diff --git a/src/video_core/renderer_vulkan/vk_present_manager.h b/src/video_core/renderer_vulkan/vk_present_manager.h index 4741c55e61..ab2b76f10f 100644 --- a/src/video_core/renderer_vulkan/vk_present_manager.h +++ b/src/video_core/renderer_vulkan/vk_present_manager.h @@ -6,6 +6,7 @@ #pragma once +#include #include #include #include @@ -68,6 +69,9 @@ public: /// How many additional frames can be queued without stalling the render thread [[nodiscard]] size_t MaxExtraFrames() const; + [[nodiscard]] std::size_t SwapchainImageCount() const { return swapchain_image_count; } + [[nodiscard]] VkFormat SwapchainImageFormat() const { return swapchain_image_format; } + private: void PresentThread(std::stop_token token); @@ -77,6 +81,8 @@ private: void RecreateSwapchain(Frame* frame); + void DiscardFrame(Frame* frame); + void SetImageCount(); private: @@ -100,7 +106,9 @@ private: bool blit_supported; bool storage_supported; bool use_present_thread; - std::size_t image_count{}; + std::atomic image_count{}; + std::atomic swapchain_image_count{}; + std::atomic swapchain_image_format{}; }; } // namespace Vulkan From 86f826c313a097b89e2b846c725e834377661073 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 03:31:53 -0400 Subject: [PATCH 16/69] Remove astc recompression method --- src/common/settings.cpp | 1 - src/common/settings.h | 7 -- src/common/settings_enums.h | 1 - src/qt_common/config/shared_translation.cpp | 12 --- src/qt_common/config/uisettings.h | 1 - src/video_core/CMakeLists.txt | 4 +- .../renderer_opengl/gl_texture_cache.cpp | 30 +------ .../renderer_vulkan/maxwell_to_vk.cpp | 21 ++--- .../renderer_vulkan/vk_texture_cache.cpp | 7 +- src/video_core/surface.cpp | 10 +-- src/video_core/texture_cache/image_base.h | 2 + src/video_core/texture_cache/texture_cache.h | 28 ++++++ .../texture_cache/texture_cache_base.h | 2 + src/video_core/texture_cache/util.cpp | 59 +------------ src/video_core/textures/bcn.cpp | 86 ------------------- src/video_core/textures/bcn.h | 16 ---- 16 files changed, 45 insertions(+), 242 deletions(-) delete mode 100644 src/video_core/textures/bcn.cpp delete mode 100644 src/video_core/textures/bcn.h diff --git a/src/common/settings.cpp b/src/common/settings.cpp index bc46b61bb4..7605305a04 100644 --- a/src/common/settings.cpp +++ b/src/common/settings.cpp @@ -48,7 +48,6 @@ SWITCHABLE(AnisotropyMode, true); SWITCHABLE(AntiAliasing, false); SWITCHABLE(AspectRatio, true); SWITCHABLE(AstcDecodeMode, true); -SWITCHABLE(AstcRecompression, true); SWITCHABLE(AudioMode, true); SWITCHABLE(CpuBackend, true); SWITCHABLE(CpuAccuracy, true); diff --git a/src/common/settings.h b/src/common/settings.h index 48d4ca427d..d15427187c 100644 --- a/src/common/settings.h +++ b/src/common/settings.h @@ -65,7 +65,6 @@ SWITCHABLE(AnisotropyMode, true); SWITCHABLE(AntiAliasing, false); SWITCHABLE(AspectRatio, true); SWITCHABLE(AstcDecodeMode, true); -SWITCHABLE(AstcRecompression, true); SWITCHABLE(AudioMode, true); SWITCHABLE(CpuBackend, true); SWITCHABLE(CpuAccuracy, true); @@ -563,12 +562,6 @@ struct Values { true, true}; - SwitchableSetting astc_recompression{linkage, - AstcRecompression::Uncompressed, - "astc_recompression", - Category::RendererAdvanced}; - - SwitchableSetting sync_memory_operations{linkage, false, "sync_memory_operations", diff --git a/src/common/settings_enums.h b/src/common/settings_enums.h index 1d7c25cf09..73568e257c 100644 --- a/src/common/settings_enums.h +++ b/src/common/settings_enums.h @@ -130,7 +130,6 @@ ENUM(TimeZone, Auto, Default, Cet, Cst6Cdt, Cuba, Eet, Egypt, Eire, Est, Est5Edt Roc, Rok, Singapore, Turkey, Uct, Universal, Utc, WSu, Wet, Zulu); ENUM(AnisotropyMode, Automatic, Default, X2, X4, X8, X16); ENUM(AstcDecodeMode, Cpu, Gpu, CpuAsynchronous); -ENUM(AstcRecompression, Uncompressed, Bc1, Bc3); ENUM(FramePacingMode, Target_Auto, Target_30, Target_60, Target_90, Target_120); ENUM(VSyncMode, Immediate, Mailbox, Fifo, FifoRelaxed); ENUM(VramUsageMode, Conservative, Aggressive); diff --git a/src/qt_common/config/shared_translation.cpp b/src/qt_common/config/shared_translation.cpp index 96e569aa21..1f6792f86c 100644 --- a/src/qt_common/config/shared_translation.cpp +++ b/src/qt_common/config/shared_translation.cpp @@ -179,11 +179,6 @@ std::unique_ptr InitializeTranslations(QObject* parent) { "GPU: Use the GPU's compute shaders to decode ASTC textures (recommended).\n" "CPU Asynchronously: Use the CPU to decode ASTC textures on demand. Eliminates" "ASTC decoding\nstuttering but may present artifacts.")); - INSERT(Settings, astc_recompression, tr("ASTC Recompression Method:"), - tr("Most GPUs lack support for ASTC textures and must decompress to an" - "intermediate format: RGBA8.\n" - "BC1/BC3: The intermediate format will be recompressed to BC1 or BC3 format,\n" - " saving VRAM but degrading image quality.")); INSERT(Settings, frame_pacing_mode, tr("Frame Pacing Mode (Vulkan only)"), tr("Controls how the emulator manages frame pacing to reduce stuttering and make the " "frame rate smoother and more consistent.")); @@ -379,13 +374,6 @@ std::unique_ptr ComboboxEnumeration(QObject* parent) { PAIR(AstcDecodeMode, Gpu, tr("GPU")), PAIR(AstcDecodeMode, CpuAsynchronous, tr("CPU Asynchronous")), }}); - translations->insert( - {Settings::EnumMetadata::Index(), - { - PAIR(AstcRecompression, Uncompressed, tr("Uncompressed (Best quality)")), - PAIR(AstcRecompression, Bc1, tr("BC1 (Low quality)")), - PAIR(AstcRecompression, Bc3, tr("BC3 (Medium quality)")), - }}); translations->insert({Settings::EnumMetadata::Index(), { PAIR(FramePacingMode, Target_Auto, tr("Auto")), diff --git a/src/qt_common/config/uisettings.h b/src/qt_common/config/uisettings.h index 794d1a029b..58a538367a 100644 --- a/src/qt_common/config/uisettings.h +++ b/src/qt_common/config/uisettings.h @@ -262,6 +262,5 @@ Q_DECLARE_METATYPE(Settings::ResolutionSetup); Q_DECLARE_METATYPE(Settings::ScalingFilter); Q_DECLARE_METATYPE(Settings::AntiAliasing); Q_DECLARE_METATYPE(Settings::RendererBackend); -Q_DECLARE_METATYPE(Settings::AstcRecompression); Q_DECLARE_METATYPE(Settings::AstcDecodeMode); Q_DECLARE_METATYPE(Settings::SpirvOptimizeMode); diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index 64d14ab2a6..77883b6484 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -256,8 +256,6 @@ add_library(video_core STATIC texture_cache/util.h textures/astc.h textures/astc.cpp - textures/bcn.cpp - textures/bcn.h textures/decoders.cpp textures/decoders.h textures/texture.cpp @@ -406,7 +404,7 @@ if (ENABLE_OPENGL) endif() target_link_libraries(video_core PUBLIC common core) -target_link_libraries(video_core PUBLIC shader_recompiler stb bc_decoder gpu_logging) +target_link_libraries(video_core PUBLIC shader_recompiler bc_decoder gpu_logging) if (ENABLE_OPENGL) target_link_libraries(video_core PUBLIC glad) endif() diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index 48aa32bc90..2589e42109 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -235,9 +235,7 @@ void ApplySwizzle(GLuint handle, PixelFormat format, std::array& color) { if (Settings::values.accelerate_astc.GetValue() != Settings::AstcDecodeMode::Gpu) { return false; } - return Settings::values.astc_recompression.GetValue() == - Settings::AstcRecompression::Uncompressed && - info.size.depth == 1; + return info.size.depth == 1; } [[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info, @@ -1880,8 +1878,9 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu case Settings::AstcDecodeMode::Gpu: if (WillUseAcceleratedAstcDecode(runtime->device, info)) { flags |= VideoCommon::ImageFlagBits::AcceleratedUpload; + break; } - break; + [[fallthrough]]; case Settings::AstcDecodeMode::CpuAsynchronous: flags |= VideoCommon::ImageFlagBits::AsynchronousDecode; break; diff --git a/src/video_core/surface.cpp b/src/video_core/surface.cpp index 4603870bfa..fa9554db59 100644 --- a/src/video_core/surface.cpp +++ b/src/video_core/surface.cpp @@ -6,7 +6,6 @@ #include "common/common_types.h" #include "common/math_util.h" -#include "common/settings.h" #include "video_core/surface.h" namespace VideoCore::Surface { @@ -469,14 +468,7 @@ u64 TranscodedAstcSize(u64 base_size, PixelFormat format) { static_cast(DefaultBlockHeight(format)) * RGBA8_PIXEL_SIZE; const u64 uncompressed_size = (base_size * base_block_size) / BytesPerBlock(format); - switch (Settings::values.astc_recompression.GetValue()) { - case Settings::AstcRecompression::Bc1: - return uncompressed_size / 8; - case Settings::AstcRecompression::Bc3: - return uncompressed_size / 4; - default: - return uncompressed_size; - } + return uncompressed_size; } } // namespace VideoCore::Surface diff --git a/src/video_core/texture_cache/image_base.h b/src/video_core/texture_cache/image_base.h index 79fc8505d6..9e0a293962 100644 --- a/src/video_core/texture_cache/image_base.h +++ b/src/video_core/texture_cache/image_base.h @@ -95,6 +95,7 @@ struct ImageBase { u32 scale_rating = 0; u64 scale_tick = 0; bool has_scaled = false; + u32 guest_data_checks = 0; size_t channel = 0; @@ -105,6 +106,7 @@ struct ImageBase { VAddr cpu_addr_end = 0; u64 modification_tick = 0; + u64 guest_data_hash = 0; size_t lru_index = SIZE_MAX; std::array mip_level_offsets{}; diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 861d9f63d0..489ceda94e 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -12,6 +12,7 @@ #include #include "common/alignment.h" +#include "common/cityhash.h" #include "common/settings.h" #include "common/slot_vector.h" #include "video_core/control/channel_state.h" @@ -1074,6 +1075,29 @@ void TextureCache

::DownloadImageIntoBuffer(typename TextureCache

::Image* i } } +template +bool TextureCache

::IsAstcDataUnchanged(Image& image) { + static constexpr u32 CHECK_THRESHOLD = 4; + if (False(image.flags & ImageFlagBits::Converted) || + True(image.flags & ImageFlagBits::GpuModified) || + !IsPixelFormatASTC(image.info.format)) { + return false; + } + if (image.guest_data_checks < CHECK_THRESHOLD) { + ++image.guest_data_checks; + return false; + } + Tegra::Memory::GpuGuestMemory guest_data( + *gpu_memory, image.gpu_addr, image.guest_size_bytes, &swizzle_data_buffer); + const u64 hash = Common::CityHash64(reinterpret_cast(guest_data.data()), + image.guest_size_bytes); + if (image.guest_data_hash == hash) { + return true; + } + image.guest_data_hash = hash; + return false; +} + template void TextureCache

::RefreshContents(Image& image, ImageId image_id) { if (False(image.flags & ImageFlagBits::CpuModified)) { @@ -1085,6 +1109,10 @@ void TextureCache

::RefreshContents(Image& image, ImageId image_id) { TrackImage(image, image_id); + if (IsAstcDataUnchanged(image)) { + return; + } + if (image.info.num_samples > 1 && !runtime.CanUploadMSAA()) { LOG_WARNING(HW_GPU, "MSAA image uploads are not implemented"); runtime.TransitionImageLayout(image); diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index 8fbd7f61bc..29f5bdde83 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -298,6 +298,8 @@ private: void RefreshContents(Image& image, ImageId image_id); + [[nodiscard]] bool IsAstcDataUnchanged(Image& image); + /// Upload data from guest to an image template void UploadImageContents(Image& image, StagingBuffer& staging_buffer); diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp index 8d275ff8eb..d8cae4f3ee 100644 --- a/src/video_core/texture_cache/util.cpp +++ b/src/video_core/texture_cache/util.cpp @@ -22,7 +22,6 @@ #include "common/common_types.h" #include "common/div_ceil.h" #include "common/scratch_buffer.h" -#include "common/settings.h" #include "video_core/compatible_formats.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/guest_memory.h" @@ -34,7 +33,6 @@ #include "video_core/texture_cache/samples_helper.h" #include "video_core/texture_cache/util.h" #include "video_core/textures/astc.h" -#include "video_core/textures/bcn.h" #include "video_core/textures/decoders.h" namespace VideoCommon { @@ -608,21 +606,6 @@ u32 CalculateConvertedSizeBytes(const ImageInfo& info) noexcept { return info.size.width * BytesPerBlock(info.format); } static constexpr Extent2D TILE_SIZE{1, 1}; - if (IsPixelFormatASTC(info.format) && Settings::values.astc_recompression.GetValue() != - Settings::AstcRecompression::Uncompressed) { - const u32 bpp_div = - Settings::values.astc_recompression.GetValue() == Settings::AstcRecompression::Bc1 ? 2 - : 1; - // NumBlocksPerLayer doesn't account for this correctly, so we have to do it manually. - u32 output_size = 0; - for (s32 i = 0; i < info.resources.levels; i++) { - const auto mip_size = AdjustMipSize(info.size, i); - const u32 plane_dim = - Common::AlignUp(mip_size.width, 4U) * Common::AlignUp(mip_size.height, 4U); - output_size += (plane_dim * info.size.depth * info.resources.layers) / bpp_div; - } - return output_size; - } return NumBlocksPerLayer(info, TILE_SIZE) * info.resources.layers * ConvertedBytesPerBlock(info.format); } @@ -925,7 +908,6 @@ boost::container::small_vector UnswizzleImage(Tegra::Memory void ConvertImage(std::span input, const ImageInfo& info, std::span output, std::span copies) { u32 output_offset = 0; - Common::ScratchBuffer decode_scratch; const Extent2D tile_size = DefaultBlockSize(info.format); for (BufferImageCopy& copy : copies) { @@ -940,10 +922,7 @@ void ConvertImage(std::span input, const ImageInfo& info, std::span input, const ImageInfo& info, std::span(copy.buffer_size); } else { DecompressBCn(input_offset, output.subspan(output_offset), copy, info.format); output_offset += copy.image_extent.width * copy.image_extent.height * @@ -995,11 +944,7 @@ void ConvertImage(std::span input, const ImageInfo& info, std::span ConvertImageFromGuest( diff --git a/src/video_core/textures/bcn.cpp b/src/video_core/textures/bcn.cpp deleted file mode 100644 index 16ddbe3202..0000000000 --- a/src/video_core/textures/bcn.cpp +++ /dev/null @@ -1,86 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later - -#include -#include -#include "common/alignment.h" -#include "video_core/textures/bcn.h" -#include "video_core/textures/workers.h" - -namespace Tegra::Texture::BCN { - -using BCNCompressor = void(u8* block_output, const u8* block_input, bool any_alpha); - -template -void CompressBCN(std::span data, uint32_t width, uint32_t height, uint32_t depth, - std::span output, BCNCompressor f) { - constexpr u8 alpha_threshold = 128; - constexpr u32 bytes_per_px = 4; - const u32 plane_dim = width * height; - - Common::ThreadWorker& workers{GetThreadWorkers()}; - - for (u32 z = 0; z < depth; z++) { - for (u32 y = 0; y < height; y += 4) { - auto compress_row = [z, y, width, height, plane_dim, f, data, output]() { - for (u32 x = 0; x < width; x += 4) { - // Gather 4x4 block of RGBA texels - u8 input_colors[4][4][4]; - bool any_alpha = false; - - for (u32 j = 0; j < 4; j++) { - for (u32 i = 0; i < 4; i++) { - const size_t coord = - (z * plane_dim + (y + j) * width + (x + i)) * bytes_per_px; - - if ((x + i < width) && (y + j < height)) { - if constexpr (ThresholdAlpha) { - if (data[coord + 3] >= alpha_threshold) { - input_colors[j][i][0] = data[coord + 0]; - input_colors[j][i][1] = data[coord + 1]; - input_colors[j][i][2] = data[coord + 2]; - input_colors[j][i][3] = 255; - } else { - any_alpha = true; - memset(input_colors[j][i], 0, bytes_per_px); - } - } else { - memcpy(input_colors[j][i], &data[coord], bytes_per_px); - } - } else { - memset(input_colors[j][i], 0, bytes_per_px); - } - } - } - - const u32 bytes_per_row = BytesPerBlock * Common::DivideUp(width, 4U); - const u32 bytes_per_plane = bytes_per_row * Common::DivideUp(height, 4U); - f(output.data() + z * bytes_per_plane + (y / 4) * bytes_per_row + - (x / 4) * BytesPerBlock, - reinterpret_cast(input_colors), any_alpha); - } - }; - workers.QueueWork(std::move(compress_row)); - } - workers.WaitForRequests(); - } -} - -void CompressBC1(std::span data, uint32_t width, uint32_t height, uint32_t depth, - std::span output) { - CompressBCN<8, true>(data, width, height, depth, output, - [](u8* block_output, const u8* block_input, bool any_alpha) { - stb_compress_bc1_block(block_output, block_input, any_alpha, - STB_DXT_NORMAL); - }); -} - -void CompressBC3(std::span data, uint32_t width, uint32_t height, uint32_t depth, - std::span output) { - CompressBCN<16, false>(data, width, height, depth, output, - [](u8* block_output, const u8* block_input, bool any_alpha) { - stb_compress_bc3_block(block_output, block_input, STB_DXT_NORMAL); - }); -} - -} // namespace Tegra::Texture::BCN diff --git a/src/video_core/textures/bcn.h b/src/video_core/textures/bcn.h deleted file mode 100644 index d5d2a16c91..0000000000 --- a/src/video_core/textures/bcn.h +++ /dev/null @@ -1,16 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later - -#pragma once - -#include - -#include "common/common_types.h" - -namespace Tegra::Texture::BCN { - -void CompressBC1(std::span data, u32 width, u32 height, u32 depth, std::span output); - -void CompressBC3(std::span data, u32 width, u32 height, u32 depth, std::span output); - -} // namespace Tegra::Texture::BCN From f955dfedcbbcef9ce98801f0aec1df600a3b7a10 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 04:59:25 -0400 Subject: [PATCH 17/69] remove async vulkan presentation toggle android --- .../yuzu_emu/features/settings/model/BooleanSetting.kt | 1 - .../yuzu_emu/features/settings/model/view/SettingsItem.kt | 7 ------- .../features/settings/ui/SettingsFragmentPresenter.kt | 1 - src/android/app/src/main/res/values-ar/strings.xml | 2 -- src/android/app/src/main/res/values-es/strings.xml | 2 -- src/android/app/src/main/res/values-ru/strings.xml | 2 -- src/android/app/src/main/res/values-uk/strings.xml | 2 -- src/android/app/src/main/res/values-zh-rCN/strings.xml | 2 -- src/android/app/src/main/res/values-zh-rTW/strings.xml | 2 -- src/android/app/src/main/res/values/strings.xml | 2 -- src/common/settings.h | 2 +- 11 files changed, 1 insertion(+), 24 deletions(-) diff --git a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/BooleanSetting.kt b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/BooleanSetting.kt index c0f85d0f60..63747cd241 100644 --- a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/BooleanSetting.kt +++ b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/BooleanSetting.kt @@ -25,7 +25,6 @@ enum class BooleanSetting(override val key: String) : AbstractBooleanSetting { RENDERER_USE_DISK_SHADER_CACHE("use_disk_shader_cache"), RENDERER_FORCE_MAX_CLOCK("force_max_clock"), RENDERER_ASYNCHRONOUS_GPU_EMULATION("use_asynchronous_gpu_emulation"), - RENDERER_ASYNC_PRESENTATION("async_presentation"), RENDERER_ASYNCHRONOUS_SHADERS("use_asynchronous_shaders"), RENDERER_REACTIVE_FLUSHING("use_reactive_flushing"), ENABLE_BUFFER_HISTORY("enable_buffer_history"), diff --git a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt index 8c07d394b0..b6b42c366c 100644 --- a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt +++ b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt @@ -751,13 +751,6 @@ abstract class SettingsItem( descriptionId = R.string.renderer_asynchronous_gpu_emulation_description ) ) - put( - SwitchSetting( - BooleanSetting.RENDERER_ASYNC_PRESENTATION, - titleId = R.string.renderer_async_presentation, - descriptionId = R.string.renderer_async_presentation_description - ) - ) put( SingleChoiceSetting( IntSetting.DMA_ACCURACY, diff --git a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsFragmentPresenter.kt b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsFragmentPresenter.kt index 53e0e3a3d7..4cfa885fb4 100644 --- a/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsFragmentPresenter.kt +++ b/src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsFragmentPresenter.kt @@ -560,7 +560,6 @@ class SettingsFragmentPresenter( add(BooleanSetting.RENDERER_ASYNCHRONOUS_SHADERS.key) add(IntSetting.ANDROID_PIPELINE_WORKERS.key) add(BooleanSetting.RENDERER_ASYNCHRONOUS_GPU_EMULATION.key) - add(BooleanSetting.RENDERER_ASYNC_PRESENTATION.key) add(HeaderSetting(R.string.extensions)) diff --git a/src/android/app/src/main/res/values-ar/strings.xml b/src/android/app/src/main/res/values-ar/strings.xml index f199e42efa..cf2c968f99 100644 --- a/src/android/app/src/main/res/values-ar/strings.xml +++ b/src/android/app/src/main/res/values-ar/strings.xml @@ -574,8 +574,6 @@ يجبر وحدة معالجة الرسومات على العمل بأقصى سرعة ممكنة (سيظل يتم تطبيق القيود الحرارية). محاكاة غير متزامنة لوحدة معالجة الرسومات يمكن لهذه الحيلة أن تزيد الأداء عن طريق تشغيل محاكاة وحدة معالجة الرسومات بشكل غير متزامن على حساب مشاكل الرسومات وزيادة معدلات الأعطال بسبب العمليات المتعلقة بالتوقيت. - عرض غير متزامن - يمكن لهذه الحيلة أن تزيد من الأداء عن طريق نقل عملية العرض إلى خيط معالجة منفصل على حساب مشاكل الرسوميات. استخدم التنظيف التفاعلي يحسن دقة العرض في بعض الألعاب على حساب الأداء. تمكين سجل التخزين المؤقت diff --git a/src/android/app/src/main/res/values-es/strings.xml b/src/android/app/src/main/res/values-es/strings.xml index 1f7faad22c..11c7fe296f 100644 --- a/src/android/app/src/main/res/values-es/strings.xml +++ b/src/android/app/src/main/res/values-es/strings.xml @@ -517,8 +517,6 @@ Fuerza a la GPU a ejecutarse a la velocidad máxima de reloj posible (se seguirán aplicando restricciones térmicas). Emulación de GPU asíncrona Este hack puede aumentar el rendimiento ejecutando la emulación de la GPU de forma asíncrona, a costa de problemas gráficos y un aumento en la tasa de fallos debido a operaciones relacionadas con la sincronización. - Presentación asíncrona - Este hack puede aumentar el rendimiento al mover la presentación a un hilo independiente de la CPU a costa de problemas gráficos. Usar limpieza reactiva Mejora la precisión de renderizado en algunos juegos, pero reduce el rendimiento. Activar el historial del búfer diff --git a/src/android/app/src/main/res/values-ru/strings.xml b/src/android/app/src/main/res/values-ru/strings.xml index a839b3bfbc..6dd50bbf29 100644 --- a/src/android/app/src/main/res/values-ru/strings.xml +++ b/src/android/app/src/main/res/values-ru/strings.xml @@ -554,8 +554,6 @@ Заставляет ГПУ работать на максимально возможных тактовых частотах (тепловые ограничения все равно будут применяться). Асинхронная эмуляция ГПУ Может повысить производительность за счёт асинхронного запуска эмуляции ГПУ, но ценой появления графических ошибок и увеличения частоты вылетов из-за операций, зависящих от синхронизации. - Асинхронная презентация - Может повысить производительность за счёт перемещения вывода кадров в отдельный поток ЦП, но ценой возникновения графических проблем. Реактивная очистка Повышение точности рендеринга в некоторых играх за счет снижения производительности. Включить историю буфера diff --git a/src/android/app/src/main/res/values-uk/strings.xml b/src/android/app/src/main/res/values-uk/strings.xml index 3f2cb94fb2..62c9c090e1 100644 --- a/src/android/app/src/main/res/values-uk/strings.xml +++ b/src/android/app/src/main/res/values-uk/strings.xml @@ -481,8 +481,6 @@ Змушує GPU працювати на максимальній тактовій частоті. Асинхронна емуляція ГП Це обхідне рішення може покращити продуктивність завдяки асинхронному виконанню емуляції ГП, але спричинить проблеми з графікою та збільшить частоту збоїв чутливих до таймінгів операцій. - Асинхронне подання - Це обхідне рішення може покращити продуктивність завдяки переміщенню подання на окремий потік ЦП, але спричинить проблеми з графікою. Реактивне очищення Покращує точність рендерингу в деяких іграх. Увімкнути історію буфера diff --git a/src/android/app/src/main/res/values-zh-rCN/strings.xml b/src/android/app/src/main/res/values-zh-rCN/strings.xml index 3cdb7278b3..568d1d384e 100644 --- a/src/android/app/src/main/res/values-zh-rCN/strings.xml +++ b/src/android/app/src/main/res/values-zh-rCN/strings.xml @@ -535,8 +535,6 @@ 强制 GPU 以最大时钟运行 (温控依然生效)。 GPU 异步模拟 此技巧可通过异步运行 GPU 模拟来提升性能，但在执行与时序相关的操作时，可能带来图形显示问题以及增加崩溃概率。 - 异步呈现 - 此技巧通过将图形呈现移至独立的 CPU 线程来提升性能，但可能会带来图形显示问题。 启用反应性刷新 通过牺牲性能来提升某些游戏的渲染精度。 启用缓冲区历史 diff --git a/src/android/app/src/main/res/values-zh-rTW/strings.xml b/src/android/app/src/main/res/values-zh-rTW/strings.xml index fa59e2a9b7..15a3b9645d 100644 --- a/src/android/app/src/main/res/values-zh-rTW/strings.xml +++ b/src/android/app/src/main/res/values-zh-rTW/strings.xml @@ -555,8 +555,6 @@ 強制 GPU 以可能的最大時脈執行 (熱溫限制仍會被套用) GPU 非同步模擬 此修改可透過使用 GPU 非同步模擬來提升性能，不過可能會導致圖形問題且會提高在進行時序相關操作時當機的機率 - 非同步呈現 - 此修改可以透過將渲染移至單獨的 CPU 執行緒來提升性能，但可能會導致圖形問題 使用重新啟用排清 犧牲效能，以改善部分遊戲的轉譯準確度 啟用緩衝區歷史 diff --git a/src/android/app/src/main/res/values/strings.xml b/src/android/app/src/main/res/values/strings.xml index 9b941608a1..025922df8d 100644 --- a/src/android/app/src/main/res/values/strings.xml +++ b/src/android/app/src/main/res/values/strings.xml @@ -580,8 +580,6 @@ Forces the GPU to run at the maximum possible clocks (thermal constraints will still be applied). GPU async emulation This hack can increase performance by running GPU emulation asynchronously at the cost of graphical issues and increased crash rates by timing-related operations. - Asynchronous presentation - This hack can increase performance by moving presentation to a separate CPU thread at the cost of graphical issues. Use reactive flushing Improves rendering accuracy in some games at the cost of performance. Enable buffer history diff --git a/src/common/settings.h b/src/common/settings.h index d15427187c..10c5042539 100644 --- a/src/common/settings.h +++ b/src/common/settings.h @@ -634,7 +634,7 @@ struct Values { true}; SwitchableSetting async_presentation{linkage, #ifdef __ANDROID__ - false, + true, #else false, #endif From 691b63e79b24e9b44d8d98ef0af6649d4059eda3 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 05:12:09 -0400 Subject: [PATCH 18/69] Quick adjustments on overlapped buffers --- src/video_core/buffer_cache/buffer_cache.h | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 4c345dd213..3165483388 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -1680,6 +1680,11 @@ BufferId BufferCache

::CreateBuffer(DAddr device_addr, u32 wanted_size, wanted_size = static_cast(device_addr_end - device_addr); const OverlapResult overlap = ResolveOverlaps(device_addr, wanted_size); const u32 size = static_cast(overlap.end - overlap.begin); + if constexpr (requires(Buffer& buffer) { buffer.IsSparseCompatible(); }) { + for (const BufferId overlap_id : overlap.ids) { + sparse_compatible |= slot_buffers[overlap_id].IsSparseCompatible(); + } + } const BufferId new_buffer_id = slot_buffers.insert(runtime, overlap.begin, size, sparse_compatible); auto& new_buffer = slot_buffers[new_buffer_id]; From 64a307c6dc6fd25978c7821c4ace708dca601d2a Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 05:42:58 -0400 Subject: [PATCH 19/69] Port some changes from Lizzie's optimization with some exceptions --- src/video_core/host_shaders/astc_decoder.comp | 1257 ++++++----------- 1 file changed, 398 insertions(+), 859 deletions(-) diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp index 9a3954ca21..af6fac0e84 100644 --- a/src/video_core/host_shaders/astc_decoder.comp +++ b/src/video_core/host_shaders/astc_decoder.comp @@ -55,6 +55,7 @@ const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT; const uint BYTES_PER_BLOCK_LOG2 = 4; +// DO NOT CHANGE - code depends on the value of these! const uint JUST_BITS = 0u; const uint QUINT = 1u; const uint TRIT = 2u; @@ -62,27 +63,45 @@ const uint TRIT = 2u; // ASTC Encodings data, sorted in ascending order based on their BitLength value // (see GetBitLength() function) const uint encoding_values[22] = uint[]( - (JUST_BITS), (JUST_BITS | (1u << 8u)), (TRIT), (JUST_BITS | (2u << 8u)), - (QUINT), (TRIT | (1u << 8u)), (JUST_BITS | (3u << 8u)), (QUINT | (1u << 8u)), - (TRIT | (2u << 8u)), (JUST_BITS | (4u << 8u)), (QUINT | (2u << 8u)), (TRIT | (3u << 8u)), - (JUST_BITS | (5u << 8u)), (QUINT | (3u << 8u)), (TRIT | (4u << 8u)), (JUST_BITS | (6u << 8u)), - (QUINT | (4u << 8u)), (TRIT | (5u << 8u)), (JUST_BITS | (7u << 8u)), (QUINT | (5u << 8u)), - (TRIT | (6u << 8u)), (JUST_BITS | (8u << 8u))); + (JUST_BITS), + (JUST_BITS | (1u << 8u)), + (TRIT), + (JUST_BITS | (2u << 8u)), + (QUINT), + (TRIT | (1u << 8u)), + (JUST_BITS | (3u << 8u)), + (QUINT | (1u << 8u)), + (TRIT | (2u << 8u)), + (JUST_BITS | (4u << 8u)), + (QUINT | (2u << 8u)), + (TRIT | (3u << 8u)), + (JUST_BITS | (5u << 8u)), + (QUINT | (3u << 8u)), + (TRIT | (4u << 8u)), + (JUST_BITS | (6u << 8u)), + (QUINT | (4u << 8u)), + (TRIT | (5u << 8u)), + (JUST_BITS | (7u << 8u)), + (QUINT | (5u << 8u)), + (TRIT | (6u << 8u)), + (JUST_BITS | (8u << 8u)) +); // Input ASTC texture globals -int total_bitsread = 0; -uvec4 local_buff; - +uint total_bitsread = 0; // Color data globals uvec4 color_endpoint_data; -int color_bitsread = 0; +uint color_bitsread = 0; +// Global "vector" to be pushed into when decoding +// At most will require BLOCK_WIDTH x BLOCK_HEIGHT in single plane mode +// At most will require BLOCK_WIDTH x BLOCK_HEIGHT x 2 in dual plane mode +// So the maximum would be 144 (12 x 12) elements, x 2 for two planes #define MAX_WEIGHT_VALUES 64 -uint result_vector[MAX_WEIGHT_VALUES]; +uint result_vector[MAX_WEIGHT_VALUES + 1]; int result_index = 0; uint result_vector_max_index; -bool result_limit_reached = false; // EncodingData helpers uint Encoding(EncodingData val) { @@ -118,12 +137,10 @@ EncodingData CreateEncodingData(uint encoding, uint num_bits, uint bit_val, uint void ResultEmplaceBack(EncodingData val) { - if (result_index >= result_vector_max_index) { - // Alert callers to avoid decoding more than needed by this phase - result_limit_reached = true; - return; - } - result_vector[result_index] = val.data; + // A = result_index, B = result_vector_max_index + // A >= B -> A - B >= 0 + // A < B -> A - B < 0 + result_vector[min(uint(MAX_WEIGHT_VALUES), result_index)] = val.data; ++result_index; } @@ -139,60 +156,31 @@ uint ReplicateBitTo9(uint value) { return value * 511; } -uint ReplicateBits(uint value, uint num_bits, uint to_bit) { - if (value == 0 || num_bits == 0) { - return 0; - } - if (num_bits >= to_bit) { - return value; - } - const uint v = value & uint((1 << num_bits) - 1); - uint res = v; - uint reslen = num_bits; - while (reslen < to_bit) { - const uint num_dst_bits_to_shift_up = min(num_bits, to_bit - reslen); - const uint num_src_bits_to_shift_down = num_bits - num_dst_bits_to_shift_up; - - res <<= num_dst_bits_to_shift_up; - res |= (v >> num_src_bits_to_shift_down); - reslen += num_bits; - } - return res; +const uint mod6_table = 0 + | ((6 % 1) << (2 * 1)) | ((6 % 2) << (2 * 2)) + | ((6 % 3) << (2 * 3)) | ((6 % 4) << (2 * 4)) + | ((6 % 5) << (2 * 5)) | ((6 % 6) << (2 * 6)); +const uint mod8_table = 0 + | ((8 % 1) << (2 * 1)) | ((8 % 2) << (2 * 2)) + | ((8 % 3) << (2 * 3)) | ((8 % 4) << (2 * 4)) + | ((8 % 5) << (2 * 5)) | ((8 % 6) << (2 * 6)) + | ((8 % 7) << (2 * 7)) | ((8 % 8) << (2 * 8)); +// Assumes num_bits < to_bit, num_bits and to_bit != 0 +uint ReplicateBits(uint value, uint num_bits, uint to_bit, uint table) { + const uint repl = value & ((1 << num_bits) - 1); + const uint shift = (table >> (num_bits << 1)) & 3; + uint v = repl; + v |= v << (num_bits << 0); // [ xxxx xxrr ] + v |= v << (num_bits << 1); // [ xxxx rrrr ] + v |= v << (num_bits << 2); // [ rrrr rrrr ] + v = (v << shift) | repl >> (num_bits - shift); + return v & ((1 << to_bit) - 1); } - uint FastReplicateTo8(uint value, uint num_bits) { - return ReplicateBits(value, num_bits, 8); + return ReplicateBits(value, num_bits, 8, mod8_table); } - -const uint REPLICATE_TO_6_ODD_HALVES = ((63u >> 1u) << 5u) | ((21u >> 1u) << 10u) | - ((9u >> 1u) << 15u) | ((17u >> 1u) << 20u) | - ((33u >> 1u) << 25u); -const uint REPLICATE_TO_6_SHIFTS = (2u << 16u) | (4u << 20u); - uint FastReplicateTo6(uint value, uint num_bits) { - if (num_bits >= 6u) { - return value; - } - const uint v = value & ((1u << num_bits) - 1u); - const uint c = ((REPLICATE_TO_6_ODD_HALVES >> (num_bits * 5u)) & 0x1fu) * 2u + 1u; - const uint s = (REPLICATE_TO_6_SHIFTS >> (num_bits * 4u)) & 0xfu; - return (v * c) >> s; -} - -uint Div3Floor(uint v) { - return (v * 0x5556) >> 16; -} - -uint Div3Ceil(uint v) { - return Div3Floor(v + 2); -} - -uint Div5Floor(uint v) { - return (v * 0x3334) >> 16; -} - -uint Div5Ceil(uint v) { - return Div5Floor(v + 4); + return ReplicateBits(value, num_bits, 6, mod6_table); } uint Hash52(uint p) { @@ -209,144 +197,85 @@ uint Hash52(uint p) { return p; } -uint Select2DPartition(uint seed, uint x, uint y, uint partition_count) { - if ((block_dims.y * block_dims.x) < 32) { - x <<= 1; - y <<= 1; - } - +uint Select2DPartition(uint seed, uvec2 pos, uint partition_count) { + pos <<= uint(block_dims.y * block_dims.x < 32); seed += (partition_count - 1) * 1024; - const uint rnum = Hash52(uint(seed)); - uint seed1 = uint(rnum & 0xF); - uint seed2 = uint((rnum >> 4) & 0xF); - uint seed3 = uint((rnum >> 8) & 0xF); - uint seed4 = uint((rnum >> 12) & 0xF); - uint seed5 = uint((rnum >> 16) & 0xF); - uint seed6 = uint((rnum >> 20) & 0xF); - uint seed7 = uint((rnum >> 24) & 0xF); - uint seed8 = uint((rnum >> 28) & 0xF); - - seed1 = (seed1 * seed1); - seed2 = (seed2 * seed2); - seed3 = (seed3 * seed3); - seed4 = (seed4 * seed4); - seed5 = (seed5 * seed5); - seed6 = (seed6 * seed6); - seed7 = (seed7 * seed7); - seed8 = (seed8 * seed8); - - uint sh1, sh2; - if ((seed & 1) > 0) { - sh1 = (seed & 2) > 0 ? 4 : 5; - sh2 = (partition_count == 3) ? 6 : 5; - } else { - sh1 = (partition_count == 3) ? 6 : 5; - sh2 = (seed & 2) > 0 ? 4 : 5; - } - seed1 >>= sh1; - seed2 >>= sh2; - seed3 >>= sh1; - seed4 >>= sh2; - seed5 >>= sh1; - seed6 >>= sh2; - seed7 >>= sh1; - seed8 >>= sh2; - - uint a = seed1 * x + seed2 * y + (rnum >> 14); - uint b = seed3 * x + seed4 * y + (rnum >> 10); - uint c = seed5 * x + seed6 * y + (rnum >> 6); - uint d = seed7 * x + seed8 * y + (rnum >> 2); - - a &= 0x3F; - b &= 0x3F; - c &= 0x3F; - d &= 0x3F; - - if (partition_count < 4) { - d = 0; - } - if (partition_count < 3) { - c = 0; - } - - if (a >= b && a >= c && a >= d) { + uvec2 shift = uvec2( + (seed & 2) > 0 ? 4 : 5, + (partition_count == 3) ? 6 : 5 + ); + shift.xy = (seed & 1) > 0 ? shift.xy : shift.yx; + uvec4 rseed[2] = uvec4[]( + (uvec4(rnum) >> uvec4(0, 4, 8, 12)) & 0xf, + (uvec4(rnum) >> uvec4(16, 20, 24, 28)) & 0xf + ); + rseed[0] = (rseed[0] * rseed[0]) >> shift.xyxy; + rseed[1] = (rseed[1] * rseed[1]) >> shift.xyxy; + const uvec4 rnum_vec = uvec4(rnum) >> uvec4(14, 10, 6, 2); + const uvec4 result_mask = ((uvec4(0, 1, 2, 3) - partition_count) >> 8) & 0x3f; + uvec4 result = uvec4( + rseed[0].xz * pos.xx + rseed[0].yw * pos.yy + rnum_vec.xy, + rseed[1].xz * pos.xx + rseed[1].yw * pos.yy + rnum_vec.zw + ) & result_mask; + if (result.x >= result.y && result.x >= result.z && result.x >= result.w) { return 0; - } else if (b >= c && b >= d) { + } else if (result.y >= result.z && result.y >= result.w) { return 1; - } else if (c >= d) { + } else if (result.z >= result.w) { return 2; } else { return 3; } } -uint ExtractBits(uvec4 payload, int offset, int bits) { - if (bits <= 0) { +uint ExtractBits(uvec4 payload, uint offset, uint bits) { + if (bits == 0 || bits > 32 || offset >= 128) return 0; - } - if (bits > 32) { - return 0; - } - if (offset >= 128) { - return 0; - } - const int avail_bits = min(bits, 128 - offset); - const int last_offset = offset + avail_bits - 1; - const int shifted_offset = offset >> 5; - if ((last_offset >> 5) == shifted_offset) { - return bitfieldExtract(payload[shifted_offset], offset & 31, avail_bits); - } - const int first_bits = 32 - (offset & 31); - const int result_first = int(bitfieldExtract(payload[shifted_offset], offset & 31, first_bits)); - const int result_second = - int(bitfieldExtract(payload[shifted_offset + 1], 0, avail_bits - first_bits)); - return result_first | (result_second << first_bits); + const uint avail_bits = min(bits, 128 - offset); + const uint last_offset = offset + avail_bits - 1; + const uint shifted_offset = offset >> 5; + if ((last_offset >> 5) == shifted_offset) + return bitfieldExtract(payload[shifted_offset], int(offset & 31), int(avail_bits)); + const uint first_bits = 32 - (offset & 31); + const uvec4 next = uvec4(payload.yzw, 0); + return bitfieldExtract(payload[shifted_offset], int(offset & 31), int(first_bits)) + | (bitfieldExtract(next[shifted_offset], 0, int(avail_bits - first_bits)) << first_bits); } -uint StreamBits(uint num_bits) { - const int int_bits = int(num_bits); - const uint ret = ExtractBits(local_buff, total_bitsread, int_bits); - total_bitsread += int_bits; +uint StreamBits(uvec4 local_buff, uint num_bits) { + const uint ret = ExtractBits(local_buff, total_bitsread, num_bits); + total_bitsread += num_bits; return ret; } void SkipBits(uint num_bits) { - const int int_bits = int(num_bits); - total_bitsread += int_bits; + total_bitsread += num_bits; } uint StreamColorBits(uint num_bits) { - const int int_bits = int(num_bits); - const uint ret = ExtractBits(color_endpoint_data, color_bitsread, int_bits); - color_bitsread += int_bits; + const uint ret = ExtractBits(color_endpoint_data, color_bitsread, num_bits); + color_bitsread += num_bits; return ret; } -EncodingData GetEncodingFromVector(uint index) { - const uint data = result_vector[index]; - return EncodingData(data); -} - // Returns the number of bits required to encode n_vals values. uint GetBitLength(uint n_vals, uint encoding_index) { + // uint Div3Floor(uint v) { return (v * 0x5556) >> 16; } + // uint Div3Ceil(uint v) { return Div3Floor(v + 2); } + // uint Div5Floor(uint v) { return (v * 0x3334) >> 16; } + // uint Div5Ceil(uint v) { return Div5Floor(v + 4); } const EncodingData encoding_value = EncodingData(encoding_values[encoding_index]); const uint encoding = Encoding(encoding_value); - uint total_bits = NumBits(encoding_value) * n_vals; - if (encoding == TRIT) { - total_bits += Div5Ceil(n_vals * 8); - } else if (encoding == QUINT) { - total_bits += Div3Ceil(n_vals * 7); - } - return total_bits; + const uint num_bits = NumBits(encoding_value); + const uvec3 div_constant = uvec3(0, 0x5556, 0x3334); + return num_bits * n_vals + + ((((n_vals * ((0x870 >> (encoding << 2)) & 0xf)) + ((0x420 >> (encoding << 2)) & 0xf)) + * div_constant[encoding]) >> 16); } uint GetNumWeightValues(uvec2 size, bool dual_plane) { - uint n_vals = size.x * size.y; - if (dual_plane) { - n_vals *= 2; - } - return n_vals; + return (size.x * size.y) << uint(dual_plane); } uint GetPackedBitSize(uvec2 size, bool dual_plane, uint max_weight) { @@ -375,13 +304,16 @@ void DecodeQuintBlock(uint num_bits) { if (BitsOp(qQ.w, 1, 2) == 3 && BitsOp(qQ.w, 5, 6) == 0) { qQ.x = 4; qQ.y = 4; - qQ.z = (BitsBracket(qQ.w, 0) << 2) | ((BitsBracket(qQ.w, 4) & ~BitsBracket(qQ.w, 0)) << 1) | - (BitsBracket(qQ.w, 3) & ~BitsBracket(qQ.w, 0)); + qQ.z = (BitsBracket(qQ.w, 0) << 2) + | ((BitsBracket(qQ.w, 4) & ~BitsBracket(qQ.w, 0)) << 1) + | (BitsBracket(qQ.w, 3) & ~BitsBracket(qQ.w, 0)); } else { uint C = 0; if (BitsOp(qQ.w, 1, 2) == 3) { qQ.z = 4; - C = (BitsOp(qQ.w, 3, 4) << 3) | ((~BitsOp(qQ.w, 5, 6) & 3) << 1) | BitsBracket(qQ.w, 0); + C = (BitsOp(qQ.w, 3, 4) << 3) + | ((~BitsOp(qQ.w, 5, 6) & 3) << 1) + | BitsBracket(qQ.w, 0); } else { qQ.z = BitsOp(qQ.w, 5, 6); C = BitsOp(qQ.w, 0, 4); @@ -394,39 +326,37 @@ void DecodeQuintBlock(uint num_bits) { qQ.x = BitsOp(C, 0, 2); } } - for (uint i = 0; i < 3; i++) { - const EncodingData val = CreateEncodingData(QUINT, num_bits, m[i], qQ[i]); - ResultEmplaceBack(val); - } + for (uint i = 0; i < 3; i++) + ResultEmplaceBack(CreateEncodingData(QUINT, num_bits, m[i], qQ[i])); } void DecodeTritBlock(uint num_bits) { uvec4 m; uvec4 t; - uvec3 Tm5t5; + uvec3 tm5t5; m[0] = StreamColorBits(num_bits); - Tm5t5.x = StreamColorBits(2); + tm5t5.x = StreamColorBits(2); m[1] = StreamColorBits(num_bits); - Tm5t5.x |= StreamColorBits(2) << 2; + tm5t5.x |= StreamColorBits(2) << 2; m[2] = StreamColorBits(num_bits); - Tm5t5.x |= StreamColorBits(1) << 4; + tm5t5.x |= StreamColorBits(1) << 4; m[3] = StreamColorBits(num_bits); - Tm5t5.x |= StreamColorBits(2) << 5; - Tm5t5.y = StreamColorBits(num_bits); - Tm5t5.x |= StreamColorBits(1) << 7; + tm5t5.x |= StreamColorBits(2) << 5; + tm5t5.y = StreamColorBits(num_bits); + tm5t5.x |= StreamColorBits(1) << 7; uint C = 0; - if (BitsOp(Tm5t5.x, 2, 4) == 7) { - C = (BitsOp(Tm5t5.x, 5, 7) << 2) | BitsOp(Tm5t5.x, 0, 1); - Tm5t5.z = 2; + if (BitsOp(tm5t5.x, 2, 4) == 7) { + C = (BitsOp(tm5t5.x, 5, 7) << 2) | BitsOp(tm5t5.x, 0, 1); + tm5t5.z = 2; t[3] = 2; } else { - C = BitsOp(Tm5t5.x, 0, 4); - if (BitsOp(Tm5t5.x, 5, 6) == 3) { - Tm5t5.z = 2; - t[3] = BitsBracket(Tm5t5.x, 7); + C = BitsOp(tm5t5.x, 0, 4); + if (BitsOp(tm5t5.x, 5, 6) == 3) { + tm5t5.z = 2; + t[3] = BitsBracket(tm5t5.x, 7); } else { - Tm5t5.z = BitsBracket(Tm5t5.x, 7); - t[3] = BitsOp(Tm5t5.x, 5, 6); + tm5t5.z = BitsBracket(tm5t5.x, 7); + t[3] = BitsOp(tm5t5.x, 5, 6); } } if (BitsOp(C, 0, 1) == 3) { @@ -442,156 +372,147 @@ void DecodeTritBlock(uint num_bits) { t[1] = BitsOp(C, 2, 3); t[0] = (BitsBracket(C, 1) << 1) | (BitsBracket(C, 0) & ~BitsBracket(C, 1)); } - for (uint i = 0; i < 4; i++) { - const EncodingData val = CreateEncodingData(TRIT, num_bits, m[i], t[i]); - ResultEmplaceBack(val); - } - const EncodingData val = CreateEncodingData(TRIT, num_bits, Tm5t5.y, Tm5t5.z); - ResultEmplaceBack(val); + for (uint i = 0; i < 4; i++) + ResultEmplaceBack(CreateEncodingData(TRIT, num_bits, m[i], t[i])); + ResultEmplaceBack(CreateEncodingData(TRIT, num_bits, tm5t5.y, tm5t5.z)); } void DecodeIntegerSequence(uint max_range, uint num_values) { EncodingData val = EncodingData(encoding_values[max_range]); const uint encoding = Encoding(val); const uint num_bits = NumBits(val); - uint vals_decoded = 0; - while (vals_decoded < num_values && !result_limit_reached) { + for (uint i = 0; i < num_values && result_index < result_vector_max_index; ) { switch (encoding) { case QUINT: DecodeQuintBlock(num_bits); - vals_decoded += 3; + i += 3; break; case TRIT: DecodeTritBlock(num_bits); - vals_decoded += 5; + i += 5; break; case JUST_BITS: BitValue(val, StreamColorBits(num_bits)); ResultEmplaceBack(val); - vals_decoded++; + i++; break; } } } -void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits, out uint color_values[32]) { - uint num_values = 0; - for (uint i = 0; i < num_partitions; i++) { - num_values += ((modes[i] >> 2) + 1) << 1; +uint DecodeSingleColorValue(EncodingData val) { + const uint encoding = Encoding(val); + const uint bitlen = NumBits(val); + const uint bitval = BitValue(val); + uint B = 0, C = 0, D = 0; + uint A = ReplicateBitTo9((bitval & 1)); + switch (encoding) { + case JUST_BITS: + break; + case TRIT: { + D = QuintTritValue(val); + switch (bitlen) { + case 1: + C = 204; + break; + case 2: { + C = 93; + const uint b = (bitval >> 1) & 1; + B = (b << 8) | (b << 4) | (b << 2) | (b << 1); + break; + } + case 3: { + C = 44; + const uint cb = (bitval >> 1) & 3; + B = (cb << 7) | (cb << 2) | cb; + break; + } + case 4: { + C = 22; + const uint dcb = (bitval >> 1) & 7; + B = (dcb << 6) | dcb; + break; + } + case 5: { + C = 11; + const uint edcb = (bitval >> 1) & 0xF; + B = (edcb << 5) | (edcb >> 2); + break; + } + case 6: { + C = 5; + const uint fedcb = (bitval >> 1) & 0x1F; + B = (fedcb << 4) | (fedcb >> 4); + break; + } + } + break; } + case QUINT: { + D = QuintTritValue(val); + switch (bitlen) { + case 1: + C = 113; + break; + case 2: { + C = 54; + const uint b = (bitval >> 1) & 1; + B = (b << 8) | (b << 3) | (b << 2); + break; + } + case 3: { + C = 26; + const uint cb = (bitval >> 1) & 3; + B = (cb << 7) | (cb << 1) | (cb >> 1); + break; + } + case 4: { + C = 13; + const uint dcb = (bitval >> 1) & 7; + B = (dcb << 6) | (dcb >> 1); + break; + } + case 5: { + C = 6; + const uint edcb = (bitval >> 1) & 0xF; + B = (edcb << 5) | (edcb >> 3); + break; + } + } + break; + } + } + uint unq = D * C + B; + unq = unq ^ A; + unq = (A & 0x80) | (unq >> 2); + return encoding == JUST_BITS ? FastReplicateTo8(bitval, bitlen) : unq; +} + +void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits, out uint color_values[32]) { + // TODO: modes[] zero on invalid, so less ops + const uvec4 num_values_tmp = (((modes >> 2) + 1) << 1) & ((uvec4(0, 1, 2, 3) - num_partitions) >> 8); + uint num_values = num_values_tmp.x + num_values_tmp.y + num_values_tmp.z + num_values_tmp.w; // Find the largest encoding that's within color_data_bits // TODO(ameerj): profile with binary search - int range = 0; + uint range = 0; while (++range < encoding_values.length()) { - const uint bit_length = GetBitLength(num_values, range); + const uint bit_length = GetBitLength(num_values, int(range)); if (bit_length > color_data_bits) { break; } } + const uint upper_bound = num_values; DecodeIntegerSequence(range - 1, num_values); - uint out_index = 0; - for (int itr = 0; itr < result_index; ++itr) { - if (out_index >= num_values) { - break; - } - const EncodingData val = GetEncodingFromVector(itr); - const uint encoding = Encoding(val); - const uint bitlen = NumBits(val); - const uint bitval = BitValue(val); - uint A = 0, B = 0, C = 0, D = 0; - A = ReplicateBitTo9((bitval & 1)); - switch (encoding) { - case JUST_BITS: - color_values[out_index++] = FastReplicateTo8(bitval, bitlen); - break; - case TRIT: { - D = QuintTritValue(val); - switch (bitlen) { - case 1: - C = 204; - break; - case 2: { - C = 93; - const uint b = (bitval >> 1) & 1; - B = (b << 8) | (b << 4) | (b << 2) | (b << 1); - break; - } - case 3: { - C = 44; - const uint cb = (bitval >> 1) & 3; - B = (cb << 7) | (cb << 2) | cb; - break; - } - case 4: { - C = 22; - const uint dcb = (bitval >> 1) & 7; - B = (dcb << 6) | dcb; - break; - } - case 5: { - C = 11; - const uint edcb = (bitval >> 1) & 0xF; - B = (edcb << 5) | (edcb >> 2); - break; - } - case 6: { - C = 5; - const uint fedcb = (bitval >> 1) & 0x1F; - B = (fedcb << 4) | (fedcb >> 4); - break; - } - } - break; - } - case QUINT: { - D = QuintTritValue(val); - switch (bitlen) { - case 1: - C = 113; - break; - case 2: { - C = 54; - const uint b = (bitval >> 1) & 1; - B = (b << 8) | (b << 3) | (b << 2); - break; - } - case 3: { - C = 26; - const uint cb = (bitval >> 1) & 3; - B = (cb << 7) | (cb << 1) | (cb >> 1); - break; - } - case 4: { - C = 13; - const uint dcb = (bitval >> 1) & 7; - B = (dcb << 6) | (dcb >> 1); - break; - } - case 5: { - C = 6; - const uint edcb = (bitval >> 1) & 0xF; - B = (edcb << 5) | (edcb >> 3); - break; - } - } - break; - } - } - if (encoding != JUST_BITS) { - uint T = (D * C) + B; - T ^= A; - T = (A & 0x80) | (T >> 2); - color_values[out_index++] = T; - } + for (int i = 0; i < upper_bound; ++i) { + color_values[i + 1] = DecodeSingleColorValue(EncodingData(result_vector[i])); } } ivec2 BitTransferSigned(int a, int b) { - ivec2 transferred; - transferred.y = b >> 1; - transferred.y |= a & 0x80; - transferred.x = a >> 1; - transferred.x &= 0x3F; + ivec2 transferred = ivec2( + (a >> 1) & 0x3F, + (b >> 1) | (a & 0x80) + ); if ((transferred.x & 0x20) > 0) { transferred.x -= 0x40; } @@ -606,169 +527,17 @@ ivec4 BlueContract(int a, int r, int g, int b) { return ivec4(a, (r + b) >> 1, (g + b) >> 1, b); } -bool IsHDRColorEndpointMode(uint cem) { - return cem == 2u || cem == 3u || cem == 7u || cem == 11u || cem == 14u || cem == 15u; -} - -int SignExtend(int value, uint nbits) { - int sign_bit = 1 << (nbits - 1u); - return (value ^ sign_bit) - sign_bit; -} - -void DecodeHDREndpointMode7(uint v0, uint v1, uint v2, uint v3, out ivec3 e0, out ivec3 e1) { - uint modeval = ((v0 & 0xC0u) >> 6u) | ((v1 & 0x80u) >> 5u) | ((v2 & 0x80u) >> 4u); - - uint majcomp; - uint mode; - if ((modeval & 0xCu) != 0xCu) { - majcomp = modeval >> 2u; - mode = modeval & 3u; - } else if (modeval != 0xFu) { - majcomp = modeval & 3u; - mode = 4u; - } else { - majcomp = 0u; - mode = 5u; - } - - int red = int(v0 & 0x3Fu); - int green = int(v1 & 0x1Fu); - int blue = int(v2 & 0x1Fu); - int scale = int(v3 & 0x1Fu); - - uint x0 = (v1 >> 6u) & 1u; - uint x1 = (v1 >> 5u) & 1u; - uint x2 = (v2 >> 6u) & 1u; - uint x3 = (v2 >> 5u) & 1u; - uint x4 = (v3 >> 7u) & 1u; - uint x5 = (v3 >> 6u) & 1u; - uint x6 = (v3 >> 5u) & 1u; - - uint ohm = 1u << mode; - if ((ohm & 0x30u) != 0u) green |= int(x0 << 6u); - if ((ohm & 0x3Au) != 0u) green |= int(x1 << 5u); - if ((ohm & 0x30u) != 0u) blue |= int(x2 << 6u); - if ((ohm & 0x3Au) != 0u) blue |= int(x3 << 5u); - if ((ohm & 0x3Du) != 0u) scale |= int(x6 << 5u); - if ((ohm & 0x2Du) != 0u) scale |= int(x5 << 6u); - if ((ohm & 0x04u) != 0u) scale |= int(x4 << 7u); - if ((ohm & 0x3Bu) != 0u) red |= int(x4 << 6u); - if ((ohm & 0x04u) != 0u) red |= int(x3 << 6u); - if ((ohm & 0x10u) != 0u) red |= int(x5 << 7u); - if ((ohm & 0x0Fu) != 0u) red |= int(x2 << 7u); - if ((ohm & 0x05u) != 0u) red |= int(x1 << 8u); - if ((ohm & 0x0Au) != 0u) red |= int(x0 << 8u); - if ((ohm & 0x05u) != 0u) red |= int(x0 << 9u); - if ((ohm & 0x02u) != 0u) red |= int(x6 << 9u); - if ((ohm & 0x01u) != 0u) red |= int(x3 << 10u); - if ((ohm & 0x02u) != 0u) red |= int(x5 << 10u); - - int shamts[6] = int[](1, 1, 2, 3, 4, 5); - int shamt = shamts[mode]; - red <<= shamt; - green <<= shamt; - blue <<= shamt; - scale <<= shamt; - - if (mode != 5u) { - green = red - green; - blue = red - blue; - } - - if (majcomp == 1u) { - int t = red; red = green; green = t; - } - if (majcomp == 2u) { - int t = red; red = blue; blue = t; - } - - e1 = ivec3(clamp(red, 0, 0xFFF), clamp(green, 0, 0xFFF), clamp(blue, 0, 0xFFF)); - e0 = ivec3(clamp(red - scale, 0, 0xFFF), clamp(green - scale, 0, 0xFFF), - clamp(blue - scale, 0, 0xFFF)); -} - -void DecodeHDREndpointMode11(uint v0, uint v1, uint v2, uint v3, uint v4, uint v5, out ivec3 e0, - out ivec3 e1) { - uint majcomp = ((v4 & 0x80u) >> 7u) | ((v5 & 0x80u) >> 6u); - if (majcomp == 3u) { - e0 = ivec3(int(v0 << 4u), int(v2 << 4u), int((v4 & 0x7Fu) << 5u)); - e1 = ivec3(int(v1 << 4u), int(v3 << 4u), int((v5 & 0x7Fu) << 5u)); - return; - } - - uint mode = ((v1 & 0x80u) >> 7u) | ((v2 & 0x80u) >> 6u) | ((v3 & 0x80u) >> 5u); - int va = int(v0 | ((v1 & 0x40u) << 2u)); - int vb0 = int(v2 & 0x3Fu); - int vb1 = int(v3 & 0x3Fu); - int vc = int(v1 & 0x3Fu); - int vd0 = int(v4 & 0x7Fu); - int vd1 = int(v5 & 0x7Fu); - - int dbitstab[8] = int[](7, 6, 7, 6, 5, 6, 5, 6); - vd0 = SignExtend(vd0, uint(dbitstab[mode])); - vd1 = SignExtend(vd1, uint(dbitstab[mode])); - - uint x0 = (v2 >> 6u) & 1u; - uint x1 = (v3 >> 6u) & 1u; - uint x2 = (v4 >> 6u) & 1u; - uint x3 = (v5 >> 6u) & 1u; - uint x4 = (v4 >> 5u) & 1u; - uint x5 = (v5 >> 5u) & 1u; - - uint ohm = 1u << mode; - if ((ohm & 0xA4u) != 0u) va |= int(x0 << 9u); - if ((ohm & 0x08u) != 0u) va |= int(x2 << 9u); - if ((ohm & 0x50u) != 0u) va |= int(x4 << 9u); - if ((ohm & 0x50u) != 0u) va |= int(x5 << 10u); - if ((ohm & 0xA0u) != 0u) va |= int(x1 << 10u); - if ((ohm & 0xC0u) != 0u) va |= int(x2 << 11u); - if ((ohm & 0x04u) != 0u) vc |= int(x1 << 6u); - if ((ohm & 0xE8u) != 0u) vc |= int(x3 << 6u); - if ((ohm & 0x20u) != 0u) vc |= int(x2 << 7u); - if ((ohm & 0x5Bu) != 0u) vb0 |= int(x0 << 6u); - if ((ohm & 0x5Bu) != 0u) vb1 |= int(x1 << 6u); - if ((ohm & 0x12u) != 0u) vb0 |= int(x2 << 7u); - if ((ohm & 0x12u) != 0u) vb1 |= int(x3 << 7u); - - int shamt = (int(mode) >> 1) ^ 3; - va <<= shamt; - vb0 <<= shamt; - vb1 <<= shamt; - vc <<= shamt; - vd0 <<= shamt; - vd1 <<= shamt; - - int r1 = clamp(va, 0, 0xFFF); - int g1 = clamp(va - vb0, 0, 0xFFF); - int b1 = clamp(va - vb1, 0, 0xFFF); - int r0 = clamp(va - vc, 0, 0xFFF); - int g0 = clamp(va - vb0 - vc - vd0, 0, 0xFFF); - int b0 = clamp(va - vb1 - vc - vd1, 0, 0xFFF); - - if (majcomp == 1u) { - int t; - t = r0; r0 = g0; g0 = t; - t = r1; r1 = g1; g1 = t; - } else if (majcomp == 2u) { - int t; - t = r0; r0 = b0; b0 = t; - t = r1; r1 = b1; b1 = t; - } - e0 = ivec3(r0, g0, b0); - e1 = ivec3(r1, g1, b1); -} - void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, uint color_values[32], inout uint colvals_index) { #define READ_UINT_VALUES(N) \ uvec4 V[2]; \ for (uint i = 0; i < N; i++) { \ - V[i / 4][i % 4] = color_values[colvals_index++]; \ + V[i >> 2][i & 3] = color_values[++colvals_index]; \ } #define READ_INT_VALUES(N) \ ivec4 V[2]; \ for (uint i = 0; i < N; i++) { \ - V[i / 4][i % 4] = int(color_values[colvals_index++]); \ + V[i >> 2][i & 3] = int(color_values[++colvals_index]); \ } switch (color_endpoint_mode) { @@ -794,12 +563,8 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, ui } case 5: { READ_INT_VALUES(4) - ivec2 transferred = BitTransferSigned(V[0].y, V[0].x); - V[0].y = transferred.x; - V[0].x = transferred.y; - transferred = BitTransferSigned(V[0].w, V[0].z); - V[0].w = transferred.x; - V[0].z = transferred.y; + V[0].yx = BitTransferSigned(V[0].y, V[0].x); + V[0].wz = BitTransferSigned(V[0].w, V[0].z); ep1 = ClampByte(ivec4(V[0].z, V[0].x, V[0].x, V[0].x)); ep2 = ClampByte(ivec4(V[0].z + V[0].w, V[0].x + V[0].y, V[0].x + V[0].y, V[0].x + V[0].y)); break; @@ -823,15 +588,9 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, ui } case 9: { READ_INT_VALUES(6) - ivec2 transferred = BitTransferSigned(V[0].y, V[0].x); - V[0].y = transferred.x; - V[0].x = transferred.y; - transferred = BitTransferSigned(V[0].w, V[0].z); - V[0].w = transferred.x; - V[0].z = transferred.y; - transferred = BitTransferSigned(V[1].y, V[1].x); - V[1].y = transferred.x; - V[1].x = transferred.y; + V[0].yx = BitTransferSigned(V[0].y, V[0].x); + V[0].wz = BitTransferSigned(V[0].w, V[0].z); + V[1].yx = BitTransferSigned(V[1].y, V[1].x); if ((V[0].y + V[0].w + V[1].y) >= 0) { ep1 = ClampByte(ivec4(0xFF, V[0].x, V[0].z, V[1].x)); ep2 = ClampByte(ivec4(0xFF, V[0].x + V[0].y, V[0].z + V[0].w, V[1].x + V[1].y)); @@ -860,21 +619,10 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, ui } case 13: { READ_INT_VALUES(8) - ivec2 transferred = BitTransferSigned(V[0].y, V[0].x); - V[0].y = transferred.x; - V[0].x = transferred.y; - transferred = BitTransferSigned(V[0].w, V[0].z); - V[0].w = transferred.x; - V[0].z = transferred.y; - - transferred = BitTransferSigned(V[1].y, V[1].x); - V[1].y = transferred.x; - V[1].x = transferred.y; - - transferred = BitTransferSigned(V[1].w, V[1].z); - V[1].w = transferred.x; - V[1].z = transferred.y; - + V[0].yx = BitTransferSigned(V[0].y, V[0].x); + V[0].wz = BitTransferSigned(V[0].w, V[0].z); + V[1].yx = BitTransferSigned(V[1].y, V[1].x); + V[1].wz = BitTransferSigned(V[1].w, V[1].z); if ((V[0].y + V[0].w + V[1].y) >= 0) { ep1 = ClampByte(ivec4(V[1].z, V[0].x, V[0].z, V[1].x)); ep2 = ClampByte(ivec4(V[1].w + V[1].z, V[0].x + V[0].y, V[0].z + V[0].w, V[1].x + V[1].y)); @@ -884,86 +632,8 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, ui } break; } - case 2: { - READ_UINT_VALUES(2) - uint y0, y1; - if (V[0].y >= V[0].x) { - y0 = V[0].x << 4u; - y1 = V[0].y << 4u; - } else { - y0 = (V[0].y << 4u) + 8u; - y1 = (V[0].x << 4u) - 8u; - } - ep1 = uvec4(0x780u, y0, y0, y0); - ep2 = uvec4(0x780u, y1, y1, y1); - break; - } - case 3: { - READ_UINT_VALUES(2) - uint y0, d; - if ((V[0].x & 0x80u) != 0u) { - y0 = ((V[0].y & 0xE0u) << 4u) | ((V[0].x & 0x7Fu) << 2u); - d = (V[0].y & 0x1Fu) << 2u; - } else { - y0 = ((V[0].y & 0xF0u) << 4u) | ((V[0].x & 0x7Fu) << 1u); - d = (V[0].y & 0x0Fu) << 1u; - } - const uint y1 = min(y0 + d, 0xFFFu); - ep1 = uvec4(0x780u, y0, y0, y0); - ep2 = uvec4(0x780u, y1, y1, y1); - break; - } - case 7: { - READ_UINT_VALUES(4) - ivec3 e0, e1; - DecodeHDREndpointMode7(V[0].x, V[0].y, V[0].z, V[0].w, e0, e1); - ep1 = uvec4(0x780u, uint(e0.x), uint(e0.y), uint(e0.z)); - ep2 = uvec4(0x780u, uint(e1.x), uint(e1.y), uint(e1.z)); - break; - } - case 11: { - READ_UINT_VALUES(6) - ivec3 e0, e1; - DecodeHDREndpointMode11(V[0].x, V[0].y, V[0].z, V[0].w, V[1].x, V[1].y, e0, e1); - ep1 = uvec4(0x780u, uint(e0.x), uint(e0.y), uint(e0.z)); - ep2 = uvec4(0x780u, uint(e1.x), uint(e1.y), uint(e1.z)); - break; - } - case 14: { - READ_UINT_VALUES(8) - ivec3 e0, e1; - DecodeHDREndpointMode11(V[0].x, V[0].y, V[0].z, V[0].w, V[1].x, V[1].y, e0, e1); - ep1 = uvec4(V[1].z, uint(e0.x), uint(e0.y), uint(e0.z)); - ep2 = uvec4(V[1].w, uint(e1.x), uint(e1.y), uint(e1.z)); - break; - } - case 15: { - READ_UINT_VALUES(8) - ivec3 e0, e1; - DecodeHDREndpointMode11(V[0].x, V[0].y, V[0].z, V[0].w, V[1].x, V[1].y, e0, e1); - const uint mode = ((V[1].z >> 7u) & 1u) | ((V[1].w >> 6u) & 2u); - int a6 = int(V[1].z & 0x7Fu); - int a7 = int(V[1].w & 0x7Fu); - int alpha0, alpha1; - if (mode == 3u) { - alpha0 = a6 << 5; - alpha1 = a7 << 5; - } else { - a6 |= (a7 << int(mode + 1u)) & 0x780; - a7 &= int(0x3Fu >> mode); - a7 ^= int(0x20u >> mode); - a7 -= int(0x20u >> mode); - a6 <<= int(4u - mode); - a7 <<= int(4u - mode); - a7 += a6; - alpha0 = a6; - alpha1 = clamp(a7, 0, 0xFFF); - } - ep1 = uvec4(uint(alpha0), uint(e0.x), uint(e0.y), uint(e0.z)); - ep2 = uvec4(uint(alpha1), uint(e1.x), uint(e1.y), uint(e1.z)); - break; - } default: { + // HDR mode, or more likely a bug computing the color_endpoint_mode ep1 = uvec4(0xFF, 0xFF, 0, 0); ep2 = uvec4(0xFF, 0xFF, 0, 0); break; @@ -974,149 +644,84 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, ui } uint UnquantizeTexelWeight(EncodingData val) { - const uint encoding = Encoding(val); - const uint bitlen = NumBits(val); - const uint bitval = BitValue(val); - const uint A = ReplicateBitTo7(bitval & 1u); - uint result = 0; + uint encoding = Encoding(val), bitlen = NumBits(val), bitval = BitValue(val); if (encoding == JUST_BITS) { - result = FastReplicateTo6(bitval, bitlen); - } else { - const uint D = QuintTritValue(val); - if (bitlen == 0u) { - return D << (4u | uint(encoding == TRIT)); - } - const uint b = (bitval >> 1u) & ((0x3100u >> (bitlen * 4u)) & 0xfu); - uint B = 0; - uint C = 0; + return (bitlen >= 1 && bitlen <= 5) + ? uint(floor(0.5f + float(bitval) * 64.0f / float((1 << bitlen) - 1))) + : FastReplicateTo6(bitval, bitlen); + } else if (encoding == TRIT || encoding == QUINT) { + uint B = 0, C = 0, D = 0; + uint b_mask = (0x3100 >> (bitlen << 2)) & 0xf; + uint b = (bitval >> 1) & b_mask; + D = QuintTritValue(val); if (encoding == TRIT) { - if (bitlen == 1u) { - C = 50u; - } else if (bitlen == 2u) { - C = 23u; - B = (b << 6u) | (b << 2u) | b; - } else { - C = 11u; - B = (b << 5u) | b; + switch (bitlen) { + case 0: return D << 5; //0,32,64 + case 1: C = 50; break; + case 2: C = 23; B = (b << 6) | (b << 2) | b; break; + case 3: C = 11; B = (b << 5) | b; break; } - } else { - if (bitlen == 1u) { - C = 28u; - } else { - C = 13u; - B = (b << 6u) | (b << 1u); + } else if (encoding == QUINT) { + switch (bitlen) { + case 0: return D << 4; //0, 16, 32, 48, 64 + case 1: C = 28; break; + case 2: C = 13; B = (b << 6) | (b << 1); break; } } - result = (A & 0x20u) | (((D * C + B) ^ A) >> 2u); + uint A = ReplicateBitTo7(bitval & 1); + uint res = (A & 0x20) | (((D * C + B) ^ A) >> 2); + return res + (res > 32 ? 1 : 0); } - return result + (1u - uint(result <= 32u)); + return 0; } void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) { const uint num_planes = is_dual_plane ? 2 : 1; const uint area = size.x * size.y; const uint loop_count = min(result_index, area * num_planes); - for (uint itr = 0; itr < loop_count; ++itr) { - result_vector[itr] = - UnquantizeTexelWeight(GetEncodingFromVector(itr)); + for (uint i = 0; i < loop_count; ++i) { + result_vector[i] = UnquantizeTexelWeight(EncodingData(result_vector[i])); } } uint GetUnquantizedTexelWeight(uint offset_base, uint plane, bool is_dual_plane) { - const uint offset = is_dual_plane ? 2 * offset_base + plane : offset_base; + const uint offset = is_dual_plane ? (offset_base << 1) + plane : offset_base; return result_vector[offset]; } -uvec4 GetUnquantizedWeightVector(uint t, uint s, uvec2 size, uint plane_index, bool is_dual_plane) { - const uint Ds = uint((block_dims.x * 0.5f + 1024) / (block_dims.x - 1)); - const uint Dt = uint((block_dims.y * 0.5f + 1024) / (block_dims.y - 1)); +uvec4 GetUnquantizedWeightVector(uvec2 pos, uvec2 size, uint plane_index, bool is_dual_plane) { const uint area = size.x * size.y; - - const uint cs = Ds * s; - const uint ct = Dt * t; - const uint gs = (cs * (size.x - 1) + 32) >> 6; - const uint gt = (ct * (size.y - 1) + 32) >> 6; - const uint js = gs >> 4; - const uint fs = gs & 0xF; - const uint jt = gt >> 4; - const uint ft = gt & 0x0F; - const uint w11 = (fs * ft + 8) >> 4; - const uint w10 = ft - w11; - const uint w01 = fs - w11; - const uint w00 = 16 - fs - ft + w11; + const uvec2 D = uvec2((block_dims * 0.5f + 1024.0f) / (block_dims - 1.0f)); + const uvec2 c = D * pos; + const uvec2 g = (c * (size - 1) + 32) >> 6; + const uvec2 j = g >> 4; + const uvec2 f = g & 0xF; + const uint w11 = (f.x * f.y + 8) >> 4; + const uint w10 = f.y - w11; + const uint w01 = f.x - w11; + const uint w00 = 16 - f.x - f.y + w11; const uvec4 w = uvec4(w00, w01, w10, w11); - const uint v0 = jt * size.x + js; - - uvec4 p0 = uvec4(0); - uvec4 p1 = uvec4(0); - - if (v0 < area) { - const uint offset_base = v0; - p0.x = GetUnquantizedTexelWeight(offset_base, 0, is_dual_plane); - p1.x = GetUnquantizedTexelWeight(offset_base, 1, is_dual_plane); - } - if ((v0 + 1) < (area)) { - const uint offset_base = v0 + 1; - p0.y = GetUnquantizedTexelWeight(offset_base, 0, is_dual_plane); - p1.y = GetUnquantizedTexelWeight(offset_base, 1, is_dual_plane); - } - if ((v0 + size.x) < (area)) { - const uint offset_base = v0 + size.x; - p0.z = GetUnquantizedTexelWeight(offset_base, 0, is_dual_plane); - p1.z = GetUnquantizedTexelWeight(offset_base, 1, is_dual_plane); - } - if ((v0 + size.x + 1) < (area)) { - const uint offset_base = v0 + size.x + 1; - p0.w = GetUnquantizedTexelWeight(offset_base, 0, is_dual_plane); - p1.w = GetUnquantizedTexelWeight(offset_base, 1, is_dual_plane); - } - - const uint primary_weight = (uint(dot(p0, w)) + 8) >> 4; - - uvec4 weight_vec = uvec4(primary_weight); - - if (is_dual_plane) { - const uint secondary_weight = (uint(dot(p1, w)) + 8) >> 4; - for (uint c = 0; c < 4; c++) { - const bool is_secondary = ((plane_index + 1u) & 3u) == c; - weight_vec[c] = is_secondary ? secondary_weight : primary_weight; - } - } - return weight_vec; + const uint v0 = j.y * size.x + j.x; + // + const uvec4 offset_base = uvec4(v0, v0 + 1, v0 + size.x, v0 + size.x + 1); + const uvec4 cmp_mask = uvec4(0) - uvec4(lessThan(offset_base, uvec4(area))); + uvec4 p0 = uvec4( + GetUnquantizedTexelWeight(offset_base.x, 0, is_dual_plane) & cmp_mask.x, + GetUnquantizedTexelWeight(offset_base.y, 0, is_dual_plane) & cmp_mask.y, + GetUnquantizedTexelWeight(offset_base.z, 0, is_dual_plane) & cmp_mask.z, + GetUnquantizedTexelWeight(offset_base.w, 0, is_dual_plane) & cmp_mask.w + ); + uvec4 p1 = uvec4( + GetUnquantizedTexelWeight(offset_base.x, 1, is_dual_plane) & cmp_mask.x, + GetUnquantizedTexelWeight(offset_base.y, 1, is_dual_plane) & cmp_mask.y, + GetUnquantizedTexelWeight(offset_base.z, 1, is_dual_plane) & cmp_mask.z, + GetUnquantizedTexelWeight(offset_base.w, 1, is_dual_plane) & cmp_mask.w + ); + const uvec2 final_weight = (uvec2(uint(dot(p0, w)), uint(dot(p1, w))) + 8) >> 4; + const uvec4 plane_mask = uvec4(0) - (uvec4(equal(uvec4(((plane_index + 1u) & 3u)), uvec4(0, 1, 2, 3))) & uvec4(0 - uint(is_dual_plane))); + return (final_weight.yyyy & plane_mask) | (final_weight.xxxx & ~plane_mask); } -int FindLayout(uint mode) { - if ((mode & 3) != 0) { - if ((mode & 8) != 0) { - if ((mode & 4) != 0) { - if ((mode & 0x100) != 0) { - return 4; - } - return 3; - } - return 2; - } - if ((mode & 4) != 0) { - return 1; - } - return 0; - } - if ((mode & 0x100) != 0) { - if ((mode & 0x80) != 0) { - if ((mode & 0x20) != 0) { - return 8; - } - return 7; - } - return 9; - } - if ((mode & 0x80) != 0) { - return 6; - } - return 5; -} - - void FillError(ivec3 coord) { for (uint j = 0; j < block_dims.y; j++) { for (uint i = 0; i < block_dims.x; i++) { @@ -1125,12 +730,13 @@ void FillError(ivec3 coord) { } } -void FillVoidExtentLDR(ivec3 coord) { +void FillVoidExtentLDR(uvec4 local_buff, ivec3 coord) { + // TODO: If you do extract bits, remember that it may be 11, or OTHER SkipBits(52); - const uint r_u = StreamBits(16); - const uint g_u = StreamBits(16); - const uint b_u = StreamBits(16); - const uint a_u = StreamBits(16); + const uint r_u = StreamBits(local_buff, 16); + const uint g_u = StreamBits(local_buff, 16); + const uint b_u = StreamBits(local_buff, 16); + const uint a_u = StreamBits(local_buff, 16); const float a = float(a_u) / 65535.0f; const float r = float(r_u) / 65535.0f; const float g = float(g_u) / 65535.0f; @@ -1142,13 +748,13 @@ void FillVoidExtentLDR(ivec3 coord) { } } -bool IsError(uint mode) { +bool IsError(uvec4 local_buff, uint mode) { if ((mode & 0x1ff) == 0x1fc) { if ((mode & 0x200) != 0) { // params.void_extent_hdr = true; return true; } - if ((mode & 0x400) == 0 || StreamBits(1) == 0) { + if ((mode & 0x400) == 0 || StreamBits(local_buff, 1) == 0) { return true; } return false; @@ -1162,9 +768,36 @@ bool IsError(uint mode) { return false; } -uvec2 DecodeBlockSize(uint mode) { +int FindLayout(uint mode) { + // Possible (Relevant), x = dont care, (in hex) + // Before shift: 02x, 08x, 10x, 00x, 0Ax, 12x, 18x, 1Ax + // After shift: 01, 04, 08, 00, 05, 09, 0c, 0d + // Reassemble and remove middle 0x040 (moronic fuzzy table) + // 0000 -> 0 0020 -> 1 0040 -> 0 0060 -> 1 + // 0080 -> 2 00a0 -> 3 00c0 -> 2 00e0 -> 3 + // 0100 -> 4 0120 -> 5 0140 -> 4 0160 -> 5 + // 0180 -> 6 01a0 -> 7 01c0 -> 6 01e0 -> 7 + // Key ranges: 01a0 -> 7, 0180 -> 6, 0100 -> 4, 0080 -> 2 + // 8>>2 = 2, 4>>2 = 1 :: () = 0, (4) = 1, (8) = 2, (8,4) = 3 + const uint mask = 0 - uint((mode & 3) != 0); + const uint sh3_mode = (mode >> 2) & 3; + const uint sh0_mode = ((mode >> 6) & 6) | ((mode >> 5) & 1); + const uint fl_const_table = 0 + | ((1) << (2 * 4)) //0080 -> 2, 1 + 5 = 6 + | ((1) << (3 * 4)) + | ((4) << (4 * 4)) //0100 -> 4, 4 + 5 = 9 + | ((4) << (5 * 4)) + | ((2) << (6 * 4)) //0180 -> 6, 2 + 5 = 7 + | ((3) << (7 * 4)) //01a0 -> 7, 3 + 5 = 8 + ; + const uint if_mode3_t = sh3_mode + uint((mode & 0x10c) == 0x10c); + const uint if_mode3_f = 5 + ((fl_const_table >> (sh0_mode << 2)) & 7); + return int((if_mode3_t & mask) | (if_mode3_f & ~mask)); +} + +uvec2 DecodeBlockSize(uint mode_layout, uint mode) { uint A, B; - switch (FindLayout(mode)) { + switch (mode_layout) { case 0: A = (mode >> 5) & 0x3; B = (mode >> 7) & 0x3; @@ -1204,130 +837,81 @@ uvec2 DecodeBlockSize(uint mode) { } } -uint DecodeMaxWeight(uint mode) { - const uint mode_layout = FindLayout(mode); - uint weight_index = (mode & 0x10) != 0 ? 1 : 0; - if (mode_layout < 5) { - weight_index |= (mode & 0x3) << 1; - } else { - weight_index |= (mode & 0xc) >> 1; - } - weight_index -= 2; - if ((mode_layout != 9) && ((mode & 0x200) != 0)) { - weight_index += 6; - } - return weight_index + 1; +uint DecodeMaxWeight(uint mode_layout, uint mode) { + const uint mode0_or = (mode & 0x3) << 1; + const uint mode5_or = (mode & 0xc) >> 1; + const uint cmp_moden = 0 - uint(mode_layout < 5); + const uint cmp_add6 = 0 - (uint(mode_layout != 9) & (mode >> 9) & 1); + return (((mode >> 4) & 1) + | (mode0_or & cmp_moden) | (mode5_or & ~cmp_moden)) + + (6 & cmp_add6) - 1; } -void DecompressBlock(ivec3 coord) { - uint mode = StreamBits(11); - if (IsError(mode)) { +void DecompressBlock(uvec4 local_buff, ivec3 coord) { + uint mode = StreamBits(local_buff, 11); + if (IsError(local_buff, mode)) { FillError(coord); return; } if ((mode & 0x1ff) == 0x1fc) { // params.void_extent_ldr = true; - FillVoidExtentLDR(coord); + FillVoidExtentLDR(local_buff, coord); return; } - const uvec2 size_params = DecodeBlockSize(mode); - if ((size_params.x > block_dims.x) || (size_params.y > block_dims.y)) { - FillError(coord); - return; - } - const uint num_partitions = StreamBits(2) + 1; + const uint num_partitions = StreamBits(local_buff, 2) + 1; const uint mode_layout = FindLayout(mode); const bool dual_plane = (mode_layout != 9) && ((mode & 0x400) != 0); - if (num_partitions > 4 || (num_partitions == 4 && dual_plane)) { - FillError(coord); - return; - } - if (GetNumWeightValues(size_params, dual_plane) > MAX_WEIGHT_VALUES) { + const uvec2 size_params = DecodeBlockSize(mode_layout, mode); + if (size_params.x > block_dims.x || size_params.y > block_dims.y + || num_partitions > 4 || (num_partitions == 4 && dual_plane) + || GetNumWeightValues(size_params, dual_plane) > MAX_WEIGHT_VALUES) { FillError(coord); return; } uint partition_index = 1; uvec4 color_endpoint_mode = uvec4(0); - uint ced_pointer = 0; uint base_cem = 0; if (num_partitions == 1) { - color_endpoint_mode.x = StreamBits(4); + color_endpoint_mode.x = StreamBits(local_buff, 4); partition_index = 0; } else { - partition_index = StreamBits(10); - base_cem = StreamBits(6); + partition_index = StreamBits(local_buff, 10); + base_cem = StreamBits(local_buff, 6); } const uint base_mode = base_cem & 3; - const uint max_weight = DecodeMaxWeight(mode); + const uint max_weight = DecodeMaxWeight(mode_layout, mode); const uint weight_bits = GetPackedBitSize(size_params, dual_plane, max_weight); - if (weight_bits < 24 || weight_bits > 96) { - FillError(coord); - return; - } - uint remaining_bits = 128 - weight_bits - total_bitsread; - uint extra_cem_bits = 0; - if (base_mode > 0) { - switch (num_partitions) { - case 2: - extra_cem_bits += 2; - break; - case 3: - extra_cem_bits += 5; - break; - case 4: - extra_cem_bits += 8; - break; - default: - FillError(coord); - return; - } - } - remaining_bits -= extra_cem_bits; + const uint extra_cem_bits = base_mode > 0 ? ((0x85200 >> (num_partitions << 2)) & 0x0f) : 0; const uint plane_selector_bits = dual_plane ? 2 : 0; + uint remaining_bits = 128 - weight_bits - total_bitsread; + remaining_bits -= extra_cem_bits; remaining_bits -= plane_selector_bits; if (remaining_bits > 128) { // Bad data, more remaining bits than 4 bytes // return early - FillError(coord); return; } // Read color data... const uint color_data_bits = remaining_bits; - while (remaining_bits > 0) { + for (uint i = 0; remaining_bits > 0 && i < 4; ++i) { const int nb = int(min(remaining_bits, 32U)); - const uint b = StreamBits(nb); - color_endpoint_data[ced_pointer] = uint(bitfieldExtract(b, 0, nb)); - ++ced_pointer; + color_endpoint_data[i] = StreamBits(local_buff, nb); remaining_bits -= nb; } - const uint plane_index = uint(StreamBits(plane_selector_bits)); + + // color_endpoint_mode assumed to be 0 on invalids/out of "range" + const uint plane_index = uint(StreamBits(local_buff, plane_selector_bits)); + const uvec4 cem_mask = (uvec4(0, 1, 2, 3) - num_partitions) >> 8; if (base_mode > 0) { - const uint extra_cem = StreamBits(extra_cem_bits); - uint cem = (extra_cem << 6) | base_cem; - cem >>= 2; - uvec4 C = uvec4(0); - for (uint i = 0; i < num_partitions; i++) { - C[i] = (cem & 1); - cem >>= 1; - } - uvec4 M = uvec4(0); - for (uint i = 0; i < num_partitions; i++) { - M[i] = cem & 3; - cem >>= 2; - } - for (uint i = 0; i < num_partitions; i++) { - color_endpoint_mode[i] = base_mode; - if (C[i] == 0) { - --color_endpoint_mode[i]; - } - color_endpoint_mode[i] <<= 2; - color_endpoint_mode[i] |= M[i]; - } + const uint extra_cem = StreamBits(local_buff, extra_cem_bits); + const uint cem = ((extra_cem << 6) | base_cem) >> 2; + const uint c0 = cem & ((1 << num_partitions) - 1); + const uint c1 = (cem >> num_partitions) & ((1 << (num_partitions << 1)) - 1); + const uvec4 c = (uvec4(c0) >> uvec4(0, 1, 2, 3)) & 1; + const uvec4 m = (uvec4(c1) >> uvec4(0, 2, 4, 6)) & 3; + color_endpoint_mode = (((uvec4(base_mode) - (1 - c)) << 2) | m) & cem_mask; } else if (num_partitions > 1) { - const uint cem = base_cem >> 2; - for (uint i = 0; i < num_partitions; i++) { - color_endpoint_mode[i] = cem; - } + color_endpoint_mode = uvec4(base_cem >> 2) & cem_mask; } uvec4 endpoints0[4]; @@ -1338,90 +922,46 @@ void DecompressBlock(ivec3 coord) { uint color_values[32]; uint colvals_index = 0; DecodeColorValues(color_endpoint_mode, num_partitions, color_data_bits, color_values); - for (uint i = 0; i < num_partitions; i++) { - ComputeEndpoints(endpoints0[i], endpoints1[i], color_endpoint_mode[i], color_values, - colvals_index); - } + for (uint i = 0; i < num_partitions; i++) + ComputeEndpoints(endpoints0[i], endpoints1[i], color_endpoint_mode[i], color_values, colvals_index); } color_endpoint_data = local_buff; color_endpoint_data = bitfieldReverse(color_endpoint_data).wzyx; const uint clear_byte_start = (weight_bits >> 3) + 1; - const uint byte_insert = ExtractBits(color_endpoint_data, int(clear_byte_start - 1) * 8, 8) & - uint(((1 << (weight_bits % 8)) - 1)); + const uint byte_insert = ExtractBits(color_endpoint_data, (clear_byte_start - 1) << 3, 8) & uint(((1 << (weight_bits & 7)) - 1)); const uint vec_index = (clear_byte_start - 1) >> 2; - color_endpoint_data[vec_index] = bitfieldInsert(color_endpoint_data[vec_index], byte_insert, - int((clear_byte_start - 1) % 4) * 8, 8); - for (uint i = clear_byte_start; i < 16; ++i) { - const uint idx = i >> 2; - color_endpoint_data[idx] = bitfieldInsert(color_endpoint_data[idx], 0, int(i % 4) * 8, 8); - } + color_endpoint_data[vec_index] = bitfieldInsert(color_endpoint_data[vec_index], byte_insert, int((clear_byte_start - 1) & 3) << 3, 8); + for (uint i = clear_byte_start; i < 16; ++i) + color_endpoint_data[i >> 2] = bitfieldInsert(color_endpoint_data[i >> 2], 0, int(i & 3) << 3, 8); // Re-init vector variables for next decode phase result_index = 0; color_bitsread = 0; - result_limit_reached = false; // The limit for the Unquantize phase, avoids decoding more data than needed. - result_vector_max_index = size_params.x * size_params.y; - if (dual_plane) { - result_vector_max_index *= 2; - } + result_vector_max_index = (size_params.x * size_params.y) << uint(dual_plane); DecodeIntegerSequence(max_weight, GetNumWeightValues(size_params, dual_plane)); - UnquantizeTexelWeights(size_params, dual_plane); for (uint j = 0; j < block_dims.y; j++) { for (uint i = 0; i < block_dims.x; i++) { - uint local_partition = 0; - if (num_partitions > 1) { - local_partition = Select2DPartition(partition_index, i, j, num_partitions); - } - const uint local_cem = color_endpoint_mode[local_partition]; - const uvec4 weight_vec = GetUnquantizedWeightVector(j, i, size_params, plane_index, dual_plane); - - vec4 p; - if (IsHDRColorEndpointMode(local_cem)) { - const uvec4 C0 = endpoints0[local_partition] << 4u; - const uvec4 C1 = endpoints1[local_partition] << 4u; - const uvec4 C = (C0 * (uvec4(64) - weight_vec) + C1 * weight_vec + uvec4(32)) / 64u; - const uvec4 E = (C & uvec4(0xF800u)) >> 11u; - const uvec4 M = C & uvec4(0x7FFu); - const uvec4 Mt_lo = 3u * M; - const uvec4 Mt_mid = 4u * M - 512u; - const uvec4 Mt_hi = 5u * M - 2048u; - const uvec4 Mt = - mix(Mt_lo, mix(Mt_mid, Mt_hi, greaterThanEqual(M, uvec4(1536u))), - greaterThanEqual(M, uvec4(512u))); - const uvec4 Cf = (E << 10u) + (Mt >> 3u); - const uvec4 half_bits = mix(Cf, uvec4(0x7BFFu), greaterThanEqual(Cf, uvec4(0x7C00u))); - p = vec4(unpackHalf2x16(half_bits.x).x, unpackHalf2x16(half_bits.y).x, - unpackHalf2x16(half_bits.z).x, unpackHalf2x16(half_bits.w).x); - - if (local_cem == 14u) { - const uint a0 = ReplicateByteTo16(uvec4(endpoints0[local_partition].x)).x; - const uint a1 = ReplicateByteTo16(uvec4(endpoints1[local_partition].x)).x; - const uint Ca = (a0 * (64u - weight_vec.x) + a1 * weight_vec.x + 32u) / 64u; - p.x = float(Ca) / 65535.0f; - } - } else { - const uvec4 C0 = ReplicateByteTo16(endpoints0[local_partition]); - const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]); - const vec4 Cf = - vec4((C0 * (uvec4(64) - weight_vec) + C1 * weight_vec + uvec4(32)) / 64); - p = Cf / 65535.0f; - } - - imageStore(dest_image, coord + ivec3(i, j, 0), clamp(p, 0.0f, 1.0f).gbar); + const uint local_partition = Select2DPartition(partition_index, uvec2(i, j), num_partitions) & (0 - uint(num_partitions > 1)); + const uvec4 C0 = ReplicateByteTo16(endpoints0[local_partition]); + const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]); + const uvec4 weight_vec = GetUnquantizedWeightVector(uvec2(i, j), size_params, plane_index, dual_plane); + const vec4 Cf = vec4((C0 * (uvec4(64) - weight_vec) + C1 * weight_vec + 32) >> 6); + const vec4 p = (Cf / 65535.0f); + imageStore(dest_image, coord + ivec3(i, j, 0), p.gbar); } } } uint SwizzleOffset(uvec2 pos) { - return ((pos.x & 32u) << 3u) | - ((pos.y & 6u) << 5u) | - ((pos.x & 16u) << 1u) | - ((pos.y & 1u) << 4u) | - (pos.x & 15u); + return ((pos.x & 32u) << 3u) + | ((pos.y & 6u) << 5u) + | ((pos.x & 16u) << 1u) + | ((pos.y & 1u) << 4u) + | (pos.x & 15u); } void main() { @@ -1441,6 +981,5 @@ void main() { if (any(greaterThanEqual(coord, imageSize(dest_image)))) { return; } - local_buff = astc_data[offset / 16]; - DecompressBlock(coord); + DecompressBlock(astc_data[offset >> 4], coord); } From 27bcaba75cd9e98f63bb470f97915776a64d00e8 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 06:12:25 -0400 Subject: [PATCH 20/69] first intent to pinpoint performance decrease on virtual buffer table + multi-range --- .../buffer_cache/virtual_range_cache.h | 91 ++----------------- 1 file changed, 10 insertions(+), 81 deletions(-) diff --git a/src/video_core/buffer_cache/virtual_range_cache.h b/src/video_core/buffer_cache/virtual_range_cache.h index f4284d6abb..378df18656 100644 --- a/src/video_core/buffer_cache/virtual_range_cache.h +++ b/src/video_core/buffer_cache/virtual_range_cache.h @@ -5,9 +5,7 @@ #include #include -#include #include -#include #include @@ -28,23 +26,22 @@ using VirtualSegments = boost::container::small_vector; class VirtualRangeCache { public: static constexpr size_t MAX_ENTRIES = 8192; - static constexpr size_t MAX_DEFERRED = 4096; const VirtualSegments* Query(Tegra::MemoryManager& memory, GPUVAddr gpu_addr, u32 size) { - if (has_deferred.load(std::memory_order_acquire)) { - ApplyDeferred(); - } + const u64 current = generation.load(std::memory_order_acquire); if (entries.size() > MAX_ENTRIES) { entries.clear(); } const size_t as_id = memory.GetID(); const u64 key = MakeKey(as_id, gpu_addr); const auto it = entries.find(key); - if (it != entries.end() && it->second.as_id == as_id && - it->second.gpu_addr == gpu_addr && it->second.size == size) { + if (it != entries.end() && it->second.generation == current && + it->second.as_id == as_id && it->second.gpu_addr == gpu_addr && + it->second.size == size) { return &it->second.segments; } Entry entry{}; + entry.generation = current; entry.as_id = as_id; entry.gpu_addr = gpu_addr; entry.size = size; @@ -79,95 +76,27 @@ public: return &result.first->second.segments; } - void Unmap(size_t as_id, GPUVAddr gpu_addr, u64 size) { - if (size == 0) { - return; + void Unmap(size_t, GPUVAddr, u64 size) { + if (size != 0) { + generation.fetch_add(1, std::memory_order_release); } - { - std::scoped_lock lock{deferred_mutex}; - if (!deferred.empty()) { - DeferredUnmap& last = deferred.back(); - if (last.as_id == as_id && last.gpu_addr + last.size == gpu_addr) { - last.size += size; - has_deferred.store(true, std::memory_order_release); - return; - } - } - if (deferred.size() >= MAX_DEFERRED) { - deferred.clear(); - deferred_overflow = true; - } else { - deferred.push_back(DeferredUnmap{ - .as_id = as_id, - .gpu_addr = gpu_addr, - .size = size, - }); - } - } - has_deferred.store(true, std::memory_order_release); } private: struct Entry { VirtualSegments segments; + u64 generation{}; size_t as_id{}; GPUVAddr gpu_addr{}; u32 size{}; }; - struct DeferredUnmap { - size_t as_id; - GPUVAddr gpu_addr; - u64 size; - }; - static u64 MakeKey(size_t as_id, GPUVAddr gpu_addr) { return (static_cast(as_id) << 48) ^ gpu_addr; } - void ApplyDeferred() { - std::vector pending; - bool overflow = false; - { - std::scoped_lock lock{deferred_mutex}; - has_deferred.store(false, std::memory_order_release); - pending.swap(deferred); - overflow = deferred_overflow; - deferred_overflow = false; - } - if (overflow) { - entries.clear(); - return; - } - if (pending.empty() || entries.empty()) { - return; - } - for (auto it = entries.begin(); it != entries.end();) { - const Entry& entry = it->second; - const GPUVAddr entry_end = entry.gpu_addr + entry.size; - bool overlaps = false; - for (const DeferredUnmap& unmap : pending) { - if (unmap.as_id != entry.as_id) { - continue; - } - if (entry.gpu_addr < unmap.gpu_addr + unmap.size && unmap.gpu_addr < entry_end) { - overlaps = true; - break; - } - } - if (overlaps) { - it = entries.erase(it); - } else { - ++it; - } - } - } - ::Common::unordered_map entries; - std::vector deferred; - std::mutex deferred_mutex; - std::atomic has_deferred{false}; - bool deferred_overflow{}; + std::atomic generation{}; }; } // namespace VideoCommon From d633c43aae5f1e74ac3ee3d54b5349e469a89ca6 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 06:35:08 -0400 Subject: [PATCH 21/69] Remove static pipelines --- src/video_core/renderer_vulkan/blit_image.cpp | 18 ++-- .../renderer_vulkan/present/util.cpp | 2 +- .../renderer_vulkan/vk_compute_pass.cpp | 2 +- .../vulkan_common/vulkan_device.cpp | 100 ------------------ src/video_core/vulkan_common/vulkan_device.h | 9 -- 5 files changed, 11 insertions(+), 120 deletions(-) diff --git a/src/video_core/renderer_vulkan/blit_image.cpp b/src/video_core/renderer_vulkan/blit_image.cpp index ddd52e9426..f9d9460c37 100644 --- a/src/video_core/renderer_vulkan/blit_image.cpp +++ b/src/video_core/renderer_vulkan/blit_image.cpp @@ -1281,7 +1281,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceColorPipeline(const BlitImagePipelineKe .subpass = 0, .basePipelineHandle = VK_NULL_HANDLE, .basePipelineIndex = 0, - }, device.StaticPipelineCache())); + })); return *blit_color_pipelines.back(); } @@ -1313,7 +1313,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceDepthStencilPipeline(const BlitImagePip .subpass = 0, .basePipelineHandle = VK_NULL_HANDLE, .basePipelineIndex = 0, - }, device.StaticPipelineCache())); + })); return *blit_depth_stencil_pipelines.back(); } @@ -1366,7 +1366,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearColorPipeline(const BlitImagePipel .subpass = 0, .basePipelineHandle = VK_NULL_HANDLE, .basePipelineIndex = 0, - }, device.StaticPipelineCache())); + })); return *clear_color_pipelines.back(); } @@ -1422,7 +1422,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearStencilPipeline( .subpass = 0, .basePipelineHandle = VK_NULL_HANDLE, .basePipelineIndex = 0, - }, device.StaticPipelineCache())); + })); return *clear_stencil_pipelines.back(); } @@ -1465,7 +1465,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceBlitColorMSAAPipeline(const BlitMSAAPip .subpass = 0, .basePipelineHandle = VK_NULL_HANDLE, .basePipelineIndex = 0, - }, device.StaticPipelineCache())); + })); return *blit_msaa_color_pipelines.back(); } @@ -1584,7 +1584,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceResolveDepthStencilPipeline(VkRenderPas .subpass = 0, .basePipelineHandle = VK_NULL_HANDLE, .basePipelineIndex = 0, - }, device.StaticPipelineCache())); + })); return *pipelines.back(); } @@ -1697,7 +1697,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceMSAACopyPipeline(const MSAACopyPipeline .subpass = 0, .basePipelineHandle = VK_NULL_HANDLE, .basePipelineIndex = 0, - }, device.StaticPipelineCache())); + })); return *msaa_copy_pipelines.back(); } @@ -1817,7 +1817,7 @@ void BlitImageHelper::ConvertPipelineEx(vk::Pipeline& pipeline, VkRenderPass ren .subpass = 0, .basePipelineHandle = VK_NULL_HANDLE, .basePipelineIndex = 0, - }, device.StaticPipelineCache()); + }); } void BlitImageHelper::ConvertPipelineColorTargetEx(vk::Pipeline& pipeline, VkRenderPass renderpass, @@ -1860,7 +1860,7 @@ void BlitImageHelper::ConvertPipeline(vk::Pipeline& pipeline, VkRenderPass rende .subpass = 0, .basePipelineHandle = VK_NULL_HANDLE, .basePipelineIndex = 0, - }, device.StaticPipelineCache()); + }); } } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/present/util.cpp b/src/video_core/renderer_vulkan/present/util.cpp index 2ea7870f01..11250687d6 100644 --- a/src/video_core/renderer_vulkan/present/util.cpp +++ b/src/video_core/renderer_vulkan/present/util.cpp @@ -521,7 +521,7 @@ static vk::Pipeline CreateWrappedPipelineImpl( .subpass = 0, .basePipelineHandle = 0, .basePipelineIndex = 0, - }, device.StaticPipelineCache()); + }); } vk::Pipeline CreateWrappedPipeline(const Device& device, vk::RenderPass& renderpass, diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 604a0d7cac..5aeef5ad3c 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -267,7 +267,7 @@ ComputePass::ComputePass(const Device& device_, Scheduler& scheduler, Descriptor .layout = *layout, .basePipelineHandle = {}, .basePipelineIndex = 0, - }, device.StaticPipelineCache()); + }); } ComputePass::~ComputePass() = default; diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index be09bca8d3..d79e8d37fe 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -7,8 +7,6 @@ #include #include #include -#include -#include #include #include #include "common/container/unordered_map.h" @@ -19,8 +17,6 @@ #include #include "common/assert.h" -#include "common/fs/fs.h" -#include "common/fs/path_util.h" #include "common/literals.h" #include #include "common/settings.h" @@ -398,17 +394,6 @@ std::vector ExtensionListForVulkan( return output; } -constexpr std::array STATIC_CACHE_MAGIC_NUMBER{'e', 'd', 'e', 'n', 's', 't', 'p', 'c'}; -constexpr u32 STATIC_CACHE_VERSION = 1; - -std::filesystem::path StaticPipelineCacheFilename() { - const auto shader_dir = Common::FS::GetEdenPath(Common::FS::EdenPath::ShaderDir); - if (!Common::FS::CreateDir(shader_dir)) { - return {}; - } - return shader_dir / "vulkan_static_pipelines.bin"; -} - } // Anonymous namespace void Device::RemoveExtension(bool& extension, const std::string& extension_name) { @@ -766,100 +751,15 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR vk::Check(vmaCreateAllocator(&allocator_info, &allocator)); - owns_static_pipeline_cache = surface != VkSurfaceKHR{}; - LoadStaticPipelineCache(); - // Initialize GPU logging if enabled InitializeGPULogging(); } Device::~Device() { - SaveStaticPipelineCache(); ShutdownGPULogging(); vmaDestroyAllocator(allocator); } -void Device::LoadStaticPipelineCache() { - const auto create = [this](size_t size, const void* data) { - static_pipeline_cache = logical.CreatePipelineCache({ - .sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .initialDataSize = size, - .pInitialData = data, - }); - }; - if (!owns_static_pipeline_cache) { - create(0, nullptr); - return; - } - const auto filename = StaticPipelineCacheFilename(); - if (filename.empty()) { - create(0, nullptr); - return; - } - std::vector data; - try { - std::ifstream file(filename, std::ios::binary | std::ios::ate); - if (!file.is_open()) { - create(0, nullptr); - return; - } - file.exceptions(std::ifstream::failbit | std::ifstream::badbit); - const size_t total = static_cast(file.tellg()); - file.seekg(0, std::ios::beg); - std::array magic{}; - u32 version{}; - if (total < magic.size() + sizeof(version)) { - create(0, nullptr); - return; - } - file.read(magic.data(), magic.size()) - .read(reinterpret_cast(&version), sizeof(version)); - if (magic != STATIC_CACHE_MAGIC_NUMBER || version != STATIC_CACHE_VERSION) { - create(0, nullptr); - return; - } - data.resize(total - magic.size() - sizeof(version)); - file.read(data.data(), static_cast(data.size())); - } catch (const std::ios_base::failure& e) { - create(0, nullptr); - return; - } - create(data.size(), data.empty() ? nullptr : data.data()); -} - -void Device::SaveStaticPipelineCache() const { - if (!owns_static_pipeline_cache || !static_pipeline_cache) { - return; - } - const auto filename = StaticPipelineCacheFilename(); - if (filename.empty()) { - return; - } - size_t size = 0; - std::vector data; - static_pipeline_cache.Read(&size, nullptr); - if (size == 0) { - return; - } - data.resize(size); - static_pipeline_cache.Read(&size, data.data()); - try { - std::ofstream file(filename, std::ios::binary | std::ios::trunc); - file.exceptions(std::ofstream::failbit); - if (!file.is_open()) { - return; - } - file.write(STATIC_CACHE_MAGIC_NUMBER.data(), STATIC_CACHE_MAGIC_NUMBER.size()) - .write(reinterpret_cast(&STATIC_CACHE_VERSION), - sizeof(STATIC_CACHE_VERSION)) - .write(data.data(), static_cast(size)); - } catch (const std::ios_base::failure& e) { - Common::FS::RemoveFile(filename); - } -} - VkFormat Device::GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage, FormatType format_type) const { if (IsFormatSupported(wanted_format, wanted_usage, format_type)) { diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 077b1c3e23..45b0ad7b17 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -273,10 +273,6 @@ public: return physical; } - VkPipelineCache StaticPipelineCache() const noexcept { - return *static_pipeline_cache; - } - /// Returns the main graphics queue. vk::Queue GetGraphicsQueue() const { return graphics_queue; @@ -1135,9 +1131,6 @@ private: /// Returns true if the device natively supports blitting depth stencil images. bool TestDepthStencilBlits(VkFormat format) const; - void LoadStaticPipelineCache(); - void SaveStaticPipelineCache() const; - private: VkInstance instance; ///< Vulkan instance. VmaAllocator allocator; ///< VMA allocator. @@ -1146,8 +1139,6 @@ private: vk::Device logical; ///< Logical device. vk::Queue graphics_queue; ///< Main graphics queue. vk::Queue present_queue; ///< Main present queue. - vk::PipelineCache static_pipeline_cache; - bool owns_static_pipeline_cache{}; u32 instance_version{}; ///< Vulkan instance version. u32 graphics_family{}; ///< Main graphics queue family index. u32 present_family{}; ///< Main present queue family index. From 8cdf552c28056ca204f449655f675041817a003a Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 06:55:48 -0400 Subject: [PATCH 22/69] Some adjustments to vertex attributes --- src/video_core/engines/draw_manager.cpp | 36 +++++++++++++++---- src/video_core/engines/maxwell_3d.h | 3 +- .../renderer_vulkan/vk_graphics_pipeline.cpp | 3 +- .../renderer_vulkan/vk_rasterizer.cpp | 6 +++- .../vulkan_common/vulkan_device.cpp | 8 +++++ src/video_core/vulkan_common/vulkan_device.h | 6 ++++ 6 files changed, 52 insertions(+), 10 deletions(-) diff --git a/src/video_core/engines/draw_manager.cpp b/src/video_core/engines/draw_manager.cpp index 5891c25622..cbac98c8c6 100644 --- a/src/video_core/engines/draw_manager.cpp +++ b/src/video_core/engines/draw_manager.cpp @@ -14,6 +14,10 @@ namespace Tegra::Engines { void Maxwell3D::DrawManager::ProcessMethodCall(Maxwell3D& maxwell3d, u32 method, u32 argument) { + if (draw_state.draw_mode == DrawMode::InstanceArray && + method != MAXWELL3D_REG_INDEX(vertex_array_instance_subsequent)) { + DrawDeferred(maxwell3d); + } switch (method) { case MAXWELL3D_REG_INDEX(clear_surface): return Clear(maxwell3d, 1); @@ -73,6 +77,15 @@ void Maxwell3D::DrawManager::Clear(Maxwell3D& maxwell3d, u32 layer_count) { } void Maxwell3D::DrawManager::DrawDeferred(Maxwell3D& maxwell3d) { + if (draw_state.draw_mode == DrawMode::InstanceArray) { + const u32 instance_count = draw_state.instance_count + 1; + draw_state.draw_mode = DrawMode::General; + draw_state.instance_count = 0; + if (maxwell3d.ShouldExecute()) { + maxwell3d.rasterizer->Draw(false, instance_count); + } + return; + } if (draw_state.draw_mode != DrawMode::Instance || draw_state.instance_count == 0) { return; } @@ -89,16 +102,25 @@ void Maxwell3D::DrawManager::DrawArray(Maxwell3D& maxwell3d, Maxwell3D::Regs::Pr } void Maxwell3D::DrawManager::DrawArrayInstanced(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 vertex_first, u32 vertex_count, bool subsequent) { + if (subsequent && draw_state.draw_mode == DrawMode::InstanceArray && + instance_topology == topology && draw_state.vertex_buffer.first == vertex_first && + draw_state.vertex_buffer.count == vertex_count) { + ++draw_state.instance_count; + return; + } + u32 base_instance = 0; + if (subsequent) { + base_instance = draw_state.base_instance + draw_state.instance_count + 1; + } + DrawDeferred(maxwell3d); + instance_topology = topology; draw_state.topology = topology; draw_state.vertex_buffer.first = vertex_first; draw_state.vertex_buffer.count = vertex_count; - if (!subsequent) { - draw_state.instance_count = 1; - } - draw_state.base_instance = draw_state.instance_count - 1; - draw_state.draw_mode = DrawMode::Instance; - draw_state.instance_count++; - ProcessDraw(maxwell3d, false, 1); + draw_state.base_instance = base_instance; + draw_state.instance_count = 0; + draw_state.draw_mode = DrawMode::InstanceArray; + UpdateTopology(maxwell3d); } void Maxwell3D::DrawManager::DrawIndex(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 index_first, u32 index_count, u32 base_index, u32 base_instance, u32 num_instances) { diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 0cab2ae986..277304ed76 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -3034,7 +3034,7 @@ public: // clang-format on struct DrawManager { - enum class DrawMode : u32 { General = 0, Instance, InlineIndex }; + enum class DrawMode : u32 { General = 0, Instance, InlineIndex, InstanceArray }; struct State { Maxwell3D::Regs::PrimitiveTopology topology{}; DrawMode draw_mode{}; @@ -3086,6 +3086,7 @@ public: void ProcessDraw(Maxwell3D& maxwell3d, bool draw_indexed, u32 instance_count); void ProcessDrawIndirect(Maxwell3D& maxwell3d); State draw_state{}; + Maxwell3D::Regs::PrimitiveTopology instance_topology{}; DrawTextureState draw_texture_state{}; IndirectParams indirect_state{}; }; diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index 8b5b0bc9c5..7ee1c608c0 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -705,7 +705,8 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) { if (instanced) { vertex_binding_divisors.push_back({ .binding = static_cast(index), - .divisor = key.state.binding_divisors[index], + .divisor = (std::min)(key.state.binding_divisors[index], + device.GetMaxVertexAttribDivisor()), }); } } diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index f43bea1220..92c5574691 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -1918,13 +1918,17 @@ void RasterizerVulkan::UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs) for (u32 binding = 0; binding < max_bindings; ++binding) { const auto& input_binding{regs.vertex_streams[binding]}; const bool is_instanced{regs.vertex_stream_instances.IsInstancingEnabled(binding)}; + u32 divisor = 1; + if (is_instanced) { + divisor = (std::min)(input_binding.frequency, device.GetMaxVertexAttribDivisor()); + } bindings.push_back({ .sType = VK_STRUCTURE_TYPE_VERTEX_INPUT_BINDING_DESCRIPTION_2_EXT, .pNext = nullptr, .binding = binding, .stride = input_binding.stride, .inputRate = is_instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX, - .divisor = is_instanced ? input_binding.frequency : 1, + .divisor = divisor, }); } diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index d79e8d37fe..9a9bfdd43b 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -1130,11 +1130,19 @@ bool Device::GetSuitability(bool requires_swapchain) { SetNext(next, properties.custom_border_color); } + if (extensions.vertex_attribute_divisor) { + properties.vertex_attribute_divisor.sType = + VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT; + SetNext(next, properties.vertex_attribute_divisor); + } + // Perform the property fetch. physical.GetProperties2(properties2); // Store base properties properties.properties = properties2.properties; + max_vertex_attrib_divisor = + (std::max)(1U, properties.vertex_attribute_divisor.maxVertexAttribDivisor); // Unload extensions if feature support is insufficient. RemoveUnsuitableExtensions(); diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 45b0ad7b17..c6d6df16b1 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -984,6 +984,10 @@ FN_MAX_LIMIT_LIST return properties.properties.limits.maxVertexInputBindings; } + u32 GetMaxVertexAttribDivisor() const { + return max_vertex_attrib_divisor; + } + u32 GetMaxViewports() const { return properties.properties.limits.maxViewports; } @@ -1188,6 +1192,7 @@ private: VkPhysicalDeviceMaintenance5PropertiesKHR maintenance5{}; VkPhysicalDeviceDepthStencilResolveProperties depth_stencil_resolve{}; VkPhysicalDeviceCustomBorderColorPropertiesEXT custom_border_color{}; + VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT vertex_attribute_divisor{}; VkPhysicalDeviceProperties properties{}; }; @@ -1195,6 +1200,7 @@ private: Extensions extensions{}; Features features{}; Properties properties{}; + u32 max_vertex_attrib_divisor{1}; VkPhysicalDeviceFeatures2 features2{}; VkPhysicalDeviceProperties2 properties2{}; From 521d3922f5ccaf96250a70a9676aa2cb57eb38ae Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 10:44:42 -0400 Subject: [PATCH 23/69] Fix build --- src/video_core/engines/draw_manager.cpp | 3 +++ 1 file changed, 3 insertions(+) diff --git a/src/video_core/engines/draw_manager.cpp b/src/video_core/engines/draw_manager.cpp index cbac98c8c6..682b5ae048 100644 --- a/src/video_core/engines/draw_manager.cpp +++ b/src/video_core/engines/draw_manager.cpp @@ -208,6 +208,9 @@ void Maxwell3D::DrawManager::DrawBegin(Maxwell3D& maxwell3d) { void Maxwell3D::DrawManager::DrawEnd(Maxwell3D& maxwell3d, u32 instance_count, bool force_draw) { switch (draw_state.draw_mode) { + case DrawMode::InstanceArray: + DrawDeferred(maxwell3d); + break; case DrawMode::Instance: if (!force_draw) { break; From b10212fe9473aeb745b73b61bf4848438a5b58f3 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 12:58:32 -0400 Subject: [PATCH 24/69] Let's try to change a bit the redundancy --- src/video_core/buffer_cache/buffer_cache.h | 17 +++++----- .../buffer_cache/buffer_cache_base.h | 3 +- src/video_core/engines/draw_manager.cpp | 33 ++++++++++++++----- src/video_core/engines/maxwell_3d.h | 1 + .../renderer_vulkan/vk_rasterizer.cpp | 23 ++++++++----- .../renderer_vulkan/vk_rasterizer.h | 2 +- 6 files changed, 50 insertions(+), 29 deletions(-) diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 3165483388..5705954ae6 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -1006,7 +1006,7 @@ void BufferCache

::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 } template -void BufferCache

::ResolveMultiRangeStorage(Binding& binding, bool is_written, +void BufferCache

::ResolveMultiRangeStorage(Binding& binding, std::vector& pool) { binding.segment_first = 0; binding.segment_count = 0; @@ -1014,15 +1014,17 @@ void BufferCache

::ResolveMultiRangeStorage(Binding& binding, bool is_written, if (binding.gpu_addr == 0 || binding.size == 0) { return; } - if (is_written && !runtime.PrefersSparseSources()) { - return; - } const VirtualSegments* found = virtual_ranges.Query(*gpu_memory, binding.gpu_addr, binding.size); if (!found || found->size() < 2) { return; } const VirtualSegments segments = *found; + for (const VirtualSegment& segment : segments) { + if (memory_tracker.IsRegionGpuModified(segment.device_addr, segment.size)) { + return; + } + } const u32 first = static_cast(pool.size()); const bool prefer_sparse = runtime.PrefersSparseSources(); for (const VirtualSegment& segment : segments) { @@ -1439,8 +1441,7 @@ void BufferCache

::UpdateStorageBuffers(size_t stage) { Binding& binding = channel_state->storage_buffers[stage][index]; const BufferId buffer_id = FindBuffer(binding.device_addr, binding.size, false); binding.buffer_id = buffer_id; - const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0; - ResolveMultiRangeStorage(binding, is_written, graphics_segments); + ResolveMultiRangeStorage(binding, graphics_segments); }); } @@ -1504,9 +1505,7 @@ void BufferCache

::UpdateComputeStorageBuffers() { // Resolve buffer Binding& binding = channel_state->compute_storage_buffers[index]; binding.buffer_id = FindBuffer(binding.device_addr, binding.size, false); - const bool is_written = - ((channel_state->written_compute_storage_buffers >> index) & 1) != 0; - ResolveMultiRangeStorage(binding, is_written, compute_segments); + ResolveMultiRangeStorage(binding, compute_segments); }); } diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h index 0dbb368c47..61b3e9bbab 100644 --- a/src/video_core/buffer_cache/buffer_cache_base.h +++ b/src/video_core/buffer_cache/buffer_cache_base.h @@ -228,8 +228,7 @@ public: bool BindMultiRangeStorage(const Binding& binding, bool is_written, std::span pool); - void ResolveMultiRangeStorage(Binding& binding, bool is_written, - std::vector& pool); + void ResolveMultiRangeStorage(Binding& binding, std::vector& pool); void UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t size); diff --git a/src/video_core/engines/draw_manager.cpp b/src/video_core/engines/draw_manager.cpp index 682b5ae048..b38847db6b 100644 --- a/src/video_core/engines/draw_manager.cpp +++ b/src/video_core/engines/draw_manager.cpp @@ -76,16 +76,20 @@ void Maxwell3D::DrawManager::Clear(Maxwell3D& maxwell3d, u32 layer_count) { } } -void Maxwell3D::DrawManager::DrawDeferred(Maxwell3D& maxwell3d) { - if (draw_state.draw_mode == DrawMode::InstanceArray) { - const u32 instance_count = draw_state.instance_count + 1; - draw_state.draw_mode = DrawMode::General; - draw_state.instance_count = 0; - if (maxwell3d.ShouldExecute()) { - maxwell3d.rasterizer->Draw(false, instance_count); - } +void Maxwell3D::DrawManager::FlushInstanceArray(Maxwell3D& maxwell3d) { + if (draw_state.draw_mode != DrawMode::InstanceArray) { return; } + const u32 instance_count = draw_state.instance_count + 1; + draw_state.draw_mode = DrawMode::General; + draw_state.instance_count = 0; + if (draw_state.vertex_buffer.count != 0 && maxwell3d.ShouldExecute()) { + maxwell3d.rasterizer->Draw(false, instance_count); + } +} + +void Maxwell3D::DrawManager::DrawDeferred(Maxwell3D& maxwell3d) { + FlushInstanceArray(maxwell3d); if (draw_state.draw_mode != DrawMode::Instance || draw_state.instance_count == 0) { return; } @@ -94,6 +98,7 @@ void Maxwell3D::DrawManager::DrawDeferred(Maxwell3D& maxwell3d) { } void Maxwell3D::DrawManager::DrawArray(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 vertex_first, u32 vertex_count, u32 base_instance, u32 num_instances) { + FlushInstanceArray(maxwell3d); draw_state.topology = topology; draw_state.vertex_buffer.first = vertex_first; draw_state.vertex_buffer.count = vertex_count; @@ -124,6 +129,7 @@ void Maxwell3D::DrawManager::DrawArrayInstanced(Maxwell3D& maxwell3d, Maxwell3D: } void Maxwell3D::DrawManager::DrawIndex(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 index_first, u32 index_count, u32 base_index, u32 base_instance, u32 num_instances) { + FlushInstanceArray(maxwell3d); draw_state.topology = topology; draw_state.index_buffer = maxwell3d.regs.index_buffer; draw_state.index_buffer.first = index_first; @@ -134,11 +140,13 @@ void Maxwell3D::DrawManager::DrawIndex(Maxwell3D& maxwell3d, Maxwell3D::Regs::Pr } void Maxwell3D::DrawManager::DrawArrayIndirect(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology) { + FlushInstanceArray(maxwell3d); draw_state.topology = topology; ProcessDrawIndirect(maxwell3d); } void Maxwell3D::DrawManager::DrawIndexedIndirect(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 index_first, u32 index_count) { + FlushInstanceArray(maxwell3d); draw_state.topology = topology; draw_state.index_buffer = maxwell3d.regs.index_buffer; draw_state.index_buffer.first = index_first; @@ -209,7 +217,7 @@ void Maxwell3D::DrawManager::DrawBegin(Maxwell3D& maxwell3d) { void Maxwell3D::DrawManager::DrawEnd(Maxwell3D& maxwell3d, u32 instance_count, bool force_draw) { switch (draw_state.draw_mode) { case DrawMode::InstanceArray: - DrawDeferred(maxwell3d); + FlushInstanceArray(maxwell3d); break; case DrawMode::Instance: if (!force_draw) { @@ -301,6 +309,13 @@ void Maxwell3D::DrawManager::UpdateTopology(Maxwell3D& maxwell3d) { void Maxwell3D::DrawManager::ProcessDraw(Maxwell3D& maxwell3d, bool draw_indexed, u32 instance_count) { LOG_TRACE(HW_GPU, "called, topology={}, count={}", draw_state.topology, draw_indexed ? draw_state.index_buffer.count : draw_state.vertex_buffer.count); UpdateTopology(maxwell3d); + u32 count = draw_state.vertex_buffer.count; + if (draw_indexed) { + count = draw_state.index_buffer.count; + } + if (count == 0) { + return; + } if (maxwell3d.ShouldExecute()) { maxwell3d.rasterizer->Draw(draw_indexed, instance_count); } diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 277304ed76..ae4b761ced 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -3071,6 +3071,7 @@ public: void ProcessMethodCall(Maxwell3D& maxwell3d, u32 method, u32 argument); void Clear(Maxwell3D& maxwell3d, u32 layer_count); void DrawDeferred(Maxwell3D& maxwell3d); + void FlushInstanceArray(Maxwell3D& maxwell3d); void DrawArray(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 vertex_first, u32 vertex_count, u32 base_instance, u32 num_instances); void DrawArrayInstanced(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 vertex_first, u32 vertex_count, bool subsequent); void DrawIndex(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 index_first, u32 index_count, u32 base_index, u32 base_instance, u32 num_instances); diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 92c5574691..b5eb0c5763 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -166,7 +166,11 @@ DrawParams MakeDrawParams(const Tegra::Engines::Maxwell3D::DrawManager::State& d params.base_vertex = 0; params.is_indexed = true; } else if (draw_state.topology == Maxwell::PrimitiveTopology::QuadStrip) { - params.num_vertices = (params.num_vertices - 2) / 2 * 6; + u32 strips = 0; + if (params.num_vertices >= 2) { + strips = (params.num_vertices - 2) / 2; + } + params.num_vertices = strips * 6; params.base_vertex = 0; params.is_indexed = true; } @@ -251,7 +255,7 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) { if (!pipeline->Configure(is_indexed)) return; - UpdateDynamicStates(); + UpdateDynamicStates(pipeline->HasDynamicVertexInput()); query_cache.NotifySegment(true); HandleTransformFeedback(); @@ -361,7 +365,12 @@ void RasterizerVulkan::DrawTexture() { texture_cache.SynchronizeDescriptors(false); texture_cache.UpdateRenderTargets(false); - UpdateDynamicStates(); + bool dynamic_vertex_input = false; + if (device.IsExtVertexInputDynamicStateSupported()) { + GraphicsPipeline* const gp = pipeline_cache.CurrentGraphicsPipeline(); + dynamic_vertex_input = gp && gp->HasDynamicVertexInput(); + } + UpdateDynamicStates(dynamic_vertex_input); query_cache.NotifySegment(true); query_cache.CounterEnable(VideoCommon::QueryType::ZPassPixelCount64, maxwell3d->regs.zpass_pixel_count_enable); @@ -1060,7 +1069,7 @@ bool AccelerateDMA::BufferToImage(const Tegra::DMA::ImageCopy& copy_info, return DmaBufferImageCopy(copy_info, buffer_operand, image_operand); } -void RasterizerVulkan::UpdateDynamicStates() { +void RasterizerVulkan::UpdateDynamicStates(bool dynamic_vertex_input) { auto& regs = maxwell3d->regs; auto& flags = maxwell3d->dirty.flags; const auto topology = maxwell3d->draw_manager.draw_state.topology; @@ -1137,10 +1146,8 @@ void RasterizerVulkan::UpdateDynamicStates() { UpdateColorWriteEnable(regs); } - if (device.IsExtVertexInputDynamicStateSupported()) { - if (auto* gp = pipeline_cache.CurrentGraphicsPipeline(); gp && gp->HasDynamicVertexInput()) { - UpdateVertexInput(regs); - } + if (dynamic_vertex_input) { + UpdateVertexInput(regs); } } diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index 7470aa4f14..59bcf65ec4 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -159,7 +159,7 @@ private: void FlushWork(); - void UpdateDynamicStates(); + void UpdateDynamicStates(bool dynamic_vertex_input); void HandleTransformFeedback(); From f92e326e174666aeb30b8441a3dcc32415330022 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 15:51:32 -0400 Subject: [PATCH 25/69] Adjustments from the previous change for pipelines --- src/video_core/buffer_cache/buffer_cache_base.h | 6 +++++- src/video_core/engines/draw_manager.cpp | 9 +-------- .../renderer_vulkan/vk_rasterizer.cpp | 17 ++++++++++++++--- src/video_core/renderer_vulkan/vk_rasterizer.h | 2 +- 4 files changed, 21 insertions(+), 13 deletions(-) diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h index 61b3e9bbab..18997088d3 100644 --- a/src/video_core/buffer_cache/buffer_cache_base.h +++ b/src/video_core/buffer_cache/buffer_cache_base.h @@ -228,7 +228,11 @@ public: bool BindMultiRangeStorage(const Binding& binding, bool is_written, std::span pool); - void ResolveMultiRangeStorage(Binding& binding, std::vector& pool); + + + + + void ResolveMultiRangeStorage(Binding& binding, std::vector& pool); void UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t size); diff --git a/src/video_core/engines/draw_manager.cpp b/src/video_core/engines/draw_manager.cpp index b38847db6b..177d83700b 100644 --- a/src/video_core/engines/draw_manager.cpp +++ b/src/video_core/engines/draw_manager.cpp @@ -83,7 +83,7 @@ void Maxwell3D::DrawManager::FlushInstanceArray(Maxwell3D& maxwell3d) { const u32 instance_count = draw_state.instance_count + 1; draw_state.draw_mode = DrawMode::General; draw_state.instance_count = 0; - if (draw_state.vertex_buffer.count != 0 && maxwell3d.ShouldExecute()) { + if (maxwell3d.ShouldExecute()) { maxwell3d.rasterizer->Draw(false, instance_count); } } @@ -309,13 +309,6 @@ void Maxwell3D::DrawManager::UpdateTopology(Maxwell3D& maxwell3d) { void Maxwell3D::DrawManager::ProcessDraw(Maxwell3D& maxwell3d, bool draw_indexed, u32 instance_count) { LOG_TRACE(HW_GPU, "called, topology={}, count={}", draw_state.topology, draw_indexed ? draw_state.index_buffer.count : draw_state.vertex_buffer.count); UpdateTopology(maxwell3d); - u32 count = draw_state.vertex_buffer.count; - if (draw_indexed) { - count = draw_state.index_buffer.count; - } - if (count == 0) { - return; - } if (maxwell3d.ShouldExecute()) { maxwell3d.rasterizer->Draw(draw_indexed, instance_count); } diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index b5eb0c5763..7346c5c269 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -237,7 +237,7 @@ RasterizerVulkan::~RasterizerVulkan() { } template -void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) { +void RasterizerVulkan::PrepareDraw(bool is_indexed, bool skip_empty, Func&& draw_func) { SCOPE_EXIT { gpu.TickWork(); @@ -245,6 +245,17 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) { FlushWork(); gpu_memory->FlushCaching(); + if (skip_empty) { + const auto& draw_state = maxwell3d->draw_manager.draw_state; + u32 count = draw_state.vertex_buffer.count; + if (is_indexed) { + count = draw_state.index_buffer.count; + } + if (count == 0) { + return; + } + } + GraphicsPipeline* const pipeline{pipeline_cache.CurrentGraphicsPipeline()}; if (!pipeline) { return; @@ -264,7 +275,7 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) { } void RasterizerVulkan::Draw(bool is_indexed, u32 instance_count) { - PrepareDraw(is_indexed, [this, is_indexed, instance_count] { + PrepareDraw(is_indexed, true, [this, is_indexed, instance_count] { const auto& draw_state = maxwell3d->draw_manager.draw_state; const u32 num_instances{instance_count}; const DrawParams draw_params{MakeDrawParams(draw_state, num_instances, is_indexed)}; @@ -299,7 +310,7 @@ void RasterizerVulkan::Draw(bool is_indexed, u32 instance_count) { void RasterizerVulkan::DrawIndirect() { const auto& params = maxwell3d->draw_manager.indirect_state; buffer_cache.SetDrawIndirect(¶ms); - PrepareDraw(params.is_indexed, [this, ¶ms] { + PrepareDraw(params.is_indexed, false, [this, ¶ms] { const auto indirect_buffer = buffer_cache.GetDrawIndirectBuffer(); const auto& buffer = indirect_buffer.first; const auto& offset = indirect_buffer.second; diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index 59bcf65ec4..bb7c5ff09e 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -155,7 +155,7 @@ private: static constexpr VkDeviceSize DEFAULT_BUFFER_SIZE = 4 * sizeof(float); template - void PrepareDraw(bool is_indexed, Func&&); + void PrepareDraw(bool is_indexed, bool skip_empty, Func&&); void FlushWork(); From 377478cd46db5091356e025369d5b230697859c8 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 16:25:34 -0400 Subject: [PATCH 26/69] A quick topology test --- src/video_core/engines/draw_manager.cpp | 58 ++++++++++++++++--- src/video_core/host_shaders/CMakeLists.txt | 1 + .../host_shaders/vulkan_indirect_quads.comp | 33 +++++++++++ src/video_core/macro.cpp | 8 ++- .../renderer_vulkan/vk_compute_pass.cpp | 51 ++++++++++++++++ .../renderer_vulkan/vk_compute_pass.h | 19 ++++++ .../renderer_vulkan/vk_rasterizer.cpp | 41 +++++++++---- .../renderer_vulkan/vk_rasterizer.h | 2 + .../vk_staging_buffer_pool.cpp | 3 +- 9 files changed, 192 insertions(+), 24 deletions(-) create mode 100644 src/video_core/host_shaders/vulkan_indirect_quads.comp diff --git a/src/video_core/engines/draw_manager.cpp b/src/video_core/engines/draw_manager.cpp index 177d83700b..0d3558a8f3 100644 --- a/src/video_core/engines/draw_manager.cpp +++ b/src/video_core/engines/draw_manager.cpp @@ -282,28 +282,68 @@ void Maxwell3D::DrawManager::DrawTexture(Maxwell3D& maxwell3d) { } void Maxwell3D::DrawManager::UpdateTopology(Maxwell3D& maxwell3d) { + using Topology = Maxwell3D::Regs::PrimitiveTopology; + using Override = Maxwell3D::Regs::PrimitiveTopologyOverride; switch (maxwell3d.regs.primitive_topology_control) { case Maxwell3D::Regs::PrimitiveTopologyControl::UseInBeginMethods: break; case Maxwell3D::Regs::PrimitiveTopologyControl::UseSeparateState: switch (maxwell3d.regs.topology_override) { - case Maxwell3D::Regs::PrimitiveTopologyOverride::None: + case Override::None: break; - case Maxwell3D::Regs::PrimitiveTopologyOverride::Points: - draw_state.topology = Maxwell3D::Regs::PrimitiveTopology::Points; + case Override::Points: + case Override::LegacyPoints: + draw_state.topology = Topology::Points; break; - case Maxwell3D::Regs::PrimitiveTopologyOverride::Lines: - draw_state.topology = Maxwell3D::Regs::PrimitiveTopology::Lines; + case Override::Lines: + case Override::LegacyLines: + case Override::LegacyLinesImm: + case Override::LegacyIndexedLines: + case Override::LegacyIndexedLines2: + draw_state.topology = Topology::Lines; break; - case Maxwell3D::Regs::PrimitiveTopologyOverride::LineStrip: - draw_state.topology = Maxwell3D::Regs::PrimitiveTopology::LineStrip; + case Override::LineStrip: + case Override::LegacyLineStrip: + case Override::LegacyIndexedLineStrip: + draw_state.topology = Topology::LineStrip; break; - default: - draw_state.topology = Maxwell3D::Regs::PrimitiveTopology(maxwell3d.regs.topology_override); + case Override::Triangles: + case Override::LegacyTriangles: + case Override::LegacyIndexedTriangles: + case Override::LegacyIndexedTriangles2: + draw_state.topology = Topology::Triangles; + break; + case Override::TriangleStrip: + case Override::LegacyTriangleStrip: + case Override::LegacyIndexedTriangleStrip: + draw_state.topology = Topology::TriangleStrip; + break; + case Override::LegacyTriangleFan: + case Override::LegacyTriangleFanImm: + case Override::LegacyIndexedTriangleFan: + draw_state.topology = Topology::TriangleFan; + break; + case Override::LinesAdjacency: + draw_state.topology = Topology::LinesAdjacency; + break; + case Override::LineStripAdjacency: + draw_state.topology = Topology::LineStripAdjacency; + break; + case Override::TrianglesAdjacency: + draw_state.topology = Topology::TrianglesAdjacency; + break; + case Override::TriangleStripAdjacency: + draw_state.topology = Topology::TriangleStripAdjacency; + break; + case Override::Patches: + draw_state.topology = Topology::Patches; break; } break; } + if (u32(draw_state.topology) > u32(Topology::Patches)) { + draw_state.topology = Topology::Triangles; + } } void Maxwell3D::DrawManager::ProcessDraw(Maxwell3D& maxwell3d, bool draw_indexed, u32 instance_count) { diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt index a115bc7b64..468bdc19ec 100644 --- a/src/video_core/host_shaders/CMakeLists.txt +++ b/src/video_core/host_shaders/CMakeLists.txt @@ -78,6 +78,7 @@ set(SHADER_FILES ${CMAKE_CURRENT_SOURCE_DIR}/vulkan_fidelityfx_fsr_easu_fp32.frag ${CMAKE_CURRENT_SOURCE_DIR}/vulkan_fidelityfx_fsr_rcas_fp16.frag ${CMAKE_CURRENT_SOURCE_DIR}/vulkan_fidelityfx_fsr_rcas_fp32.frag + ${CMAKE_CURRENT_SOURCE_DIR}/vulkan_indirect_quads.comp ${CMAKE_CURRENT_SOURCE_DIR}/vulkan_present.frag ${CMAKE_CURRENT_SOURCE_DIR}/vulkan_present.vert ${CMAKE_CURRENT_SOURCE_DIR}/vulkan_present_scaleforce_fp16.frag diff --git a/src/video_core/host_shaders/vulkan_indirect_quads.comp b/src/video_core/host_shaders/vulkan_indirect_quads.comp new file mode 100644 index 0000000000..01bc2f961f --- /dev/null +++ b/src/video_core/host_shaders/vulkan_indirect_quads.comp @@ -0,0 +1,33 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#version 460 core + +layout (local_size_x = 32) in; + +layout (std430, set = 0, binding = 0) readonly buffer InputBuffer { + uint input_commands[]; +}; + +layout (std430, set = 0, binding = 1) writeonly buffer OutputBuffer { + uint output_commands[]; +}; + +layout (push_constant) uniform PushConstants { + uint num_draws; + uint src_stride; +}; + +void main() { + uint draw = gl_GlobalInvocationID.x; + if (draw >= num_draws) { + return; + } + uint src = draw * src_stride; + uint dst = draw * 5; + output_commands[dst + 0] = input_commands[src + 0] / 4 * 6; + output_commands[dst + 1] = input_commands[src + 1]; + output_commands[dst + 2] = input_commands[src + 2] / 4 * 6; + output_commands[dst + 3] = input_commands[src + 3]; + output_commands[dst + 4] = input_commands[src + 4]; +} diff --git a/src/video_core/macro.cpp b/src/video_core/macro.cpp index 86f528af15..ac658f7534 100644 --- a/src/video_core/macro.cpp +++ b/src/video_core/macro.cpp @@ -62,6 +62,10 @@ bool IsTopologySafe(Maxwell3D::Regs::PrimitiveTopology topology) { } } +bool IsTopologySafeIndexedIndirect(Maxwell3D::Regs::PrimitiveTopology topology) { + return IsTopologySafe(topology) || topology == Maxwell3D::Regs::PrimitiveTopology::Quads; +} + } // Anonymous namespace void HLE_DrawArraysIndirect::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, std::span parameters, [[maybe_unused]] u32 method) { @@ -125,7 +129,7 @@ void HLE_DrawArraysIndirect::Fallback(Core::System& system, Engines::Maxwell3D& void HLE_DrawIndexedIndirect::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, std::span parameters, [[maybe_unused]] u32 method) { auto topology = static_cast(parameters[0]); - if (!maxwell3d.AnyParametersDirty() || !IsTopologySafe(topology)) { + if (!maxwell3d.AnyParametersDirty() || !IsTopologySafeIndexedIndirect(topology)) { Fallback(system, maxwell3d, parameters); return; } @@ -198,7 +202,7 @@ void HLE_MultiLayerClear::Execute(Core::System& system, Engines::Maxwell3D& maxw } void HLE_MultiDrawIndexedIndirectCount::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, std::span parameters, [[maybe_unused]] u32 method) { const auto topology = Maxwell3D::Regs::PrimitiveTopology(parameters[2]); - if (IsTopologySafe(topology)) { + if (IsTopologySafeIndexedIndirect(topology)) { const u32 start_indirect = parameters[0]; const u32 end_indirect = parameters[1]; if (start_indirect >= end_indirect) { diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 5aeef5ad3c..a92155881d 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -20,6 +20,7 @@ #include "video_core/host_shaders/queries_prefix_scan_sum_comp_spv.h" #include "video_core/host_shaders/queries_prefix_scan_sum_nosubgroups_comp_spv.h" #include "video_core/host_shaders/resolve_conditional_render_comp_spv.h" +#include "video_core/host_shaders/vulkan_indirect_quads_comp_spv.h" #include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h" #include "video_core/host_shaders/vulkan_uint8_comp_spv.h" #include "video_core/renderer_vulkan/vk_compute_pass.h" @@ -375,6 +376,56 @@ std::pair QuadIndexedPass::Assemble( return {staging.buffer, staging.offset}; } +IndirectQuadsPass::IndirectQuadsPass(const Device& device_, Scheduler& scheduler_, + DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_) + : ComputePass(device_, scheduler_, descriptor_pool_, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS, + INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO, + COMPUTE_PUSH_CONSTANT_RANGE, VULKAN_INDIRECT_QUADS_COMP_SPV), + scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, + compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} + +IndirectQuadsPass::~IndirectQuadsPass() = default; + +std::pair IndirectQuadsPass::Assemble(u32 num_draws, u32 stride, + VkBuffer src_buffer, u32 src_offset) { + u32 src_stride = stride / static_cast(sizeof(u32)); + if (src_stride < COMMAND_WORDS) { + src_stride = COMMAND_WORDS; + } + const u32 input_size = num_draws * src_stride * static_cast(sizeof(u32)); + const std::size_t staging_size = std::size_t(num_draws) * COMMAND_WORDS * sizeof(u32); + const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal); + + compute_pass_descriptor_queue.Acquire(scheduler, 2); + compute_pass_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size); + compute_pass_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size); + const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; + + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([this, descriptor_data, num_draws, src_stride](vk::CommandBuffer cmdbuf) { + static constexpr u32 DISPATCH_SIZE = 32; + static constexpr VkMemoryBarrier WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_INDIRECT_COMMAND_READ_BIT, + }; + const std::array push_constants{num_draws, src_stride}; + const VkDescriptorSet set = descriptor_allocator.Commit(); + device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); + cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(push_constants), + &push_constants); + cmdbuf.Dispatch(Common::DivCeil(num_draws, DISPATCH_SIZE), 1, 1); + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, + VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT, 0, WRITE_BARRIER); + }); + return {staging.buffer, staging.offset}; +} + ConditionalRenderingResolvePass::ConditionalRenderingResolvePass( const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, ComputePassDescriptorQueue& compute_pass_descriptor_queue_) diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h index 12723099ca..3fdd2c6f01 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.h +++ b/src/video_core/renderer_vulkan/vk_compute_pass.h @@ -88,6 +88,25 @@ private: ComputePassDescriptorQueue& compute_pass_descriptor_queue; }; +class IndirectQuadsPass final : public ComputePass { +public: + explicit IndirectQuadsPass(const Device& device_, Scheduler& scheduler_, + DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_); + ~IndirectQuadsPass(); + + static constexpr u32 COMMAND_WORDS = 5; + + std::pair Assemble(u32 num_draws, u32 stride, VkBuffer src_buffer, + u32 src_offset); + +private: + Scheduler& scheduler; + StagingBufferPool& staging_buffer_pool; + ComputePassDescriptorQueue& compute_pass_descriptor_queue; +}; + class ConditionalRenderingResolvePass final : public ComputePass { public: explicit ConditionalRenderingResolvePass( diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 7346c5c269..7fa889e22f 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -212,6 +212,8 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra compute_pass_descriptor_queue(device, UpdateDescriptorQueue::COMPUTE_FRAME_PAYLOAD_SIZE), descriptor_buffer_ring(device, memory_allocator), blit_image(device, scheduler, state_tracker, descriptor_pool), render_pass_cache(device), + indirect_quads_pass(device, scheduler, descriptor_pool, staging_pool, + compute_pass_descriptor_queue), texture_cache_runtime{ device, scheduler, memory_allocator, staging_pool, blit_image, render_pass_cache, descriptor_pool, compute_pass_descriptor_queue}, @@ -314,6 +316,18 @@ void RasterizerVulkan::DrawIndirect() { const auto indirect_buffer = buffer_cache.GetDrawIndirectBuffer(); const auto& buffer = indirect_buffer.first; const auto& offset = indirect_buffer.second; + VkBuffer command_buffer = buffer->Handle(); + VkDeviceSize command_offset = offset; + u32 command_stride = static_cast(params.stride); + if (params.is_indexed && + maxwell3d->draw_manager.draw_state.topology == Maxwell::PrimitiveTopology::Quads) { + const auto patched = indirect_quads_pass.Assemble( + static_cast(params.max_draw_counts), command_stride, command_buffer, + static_cast(offset)); + command_buffer = patched.first; + command_offset = patched.second; + command_stride = IndirectQuadsPass::COMMAND_WORDS * static_cast(sizeof(u32)); + } if (params.is_byte_count) { scheduler.Record([buffer_obj = buffer->Handle(), offset, stride = params.stride](vk::CommandBuffer cmdbuf) { @@ -326,29 +340,32 @@ void RasterizerVulkan::DrawIndirect() { const auto count = buffer_cache.GetDrawIndirectCount(); const auto& draw_buffer = count.first; const auto& offset_base = count.second; - scheduler.Record([draw_buffer_obj = draw_buffer->Handle(), - buffer_obj = buffer->Handle(), offset_base, offset, + scheduler.Record([draw_buffer_obj = draw_buffer->Handle(), command_buffer, + offset_base, command_offset, command_stride, params](vk::CommandBuffer cmdbuf) { if (params.is_indexed) { - cmdbuf.DrawIndexedIndirectCount( - buffer_obj, offset, draw_buffer_obj, offset_base, - static_cast(params.max_draw_counts), static_cast(params.stride)); + cmdbuf.DrawIndexedIndirectCount(command_buffer, command_offset, draw_buffer_obj, + offset_base, + static_cast(params.max_draw_counts), + command_stride); } else { - cmdbuf.DrawIndirectCount(buffer_obj, offset, draw_buffer_obj, offset_base, + cmdbuf.DrawIndirectCount(command_buffer, command_offset, draw_buffer_obj, + offset_base, static_cast(params.max_draw_counts), - static_cast(params.stride)); + command_stride); } }); return; } - scheduler.Record([buffer_obj = buffer->Handle(), offset, params](vk::CommandBuffer cmdbuf) { + scheduler.Record([command_buffer, command_offset, command_stride, + params](vk::CommandBuffer cmdbuf) { if (params.is_indexed) { - cmdbuf.DrawIndexedIndirect(buffer_obj, offset, + cmdbuf.DrawIndexedIndirect(command_buffer, command_offset, static_cast(params.max_draw_counts), - static_cast(params.stride)); + command_stride); } else { - cmdbuf.DrawIndirect(buffer_obj, offset, static_cast(params.max_draw_counts), - static_cast(params.stride)); + cmdbuf.DrawIndirect(command_buffer, command_offset, + static_cast(params.max_draw_counts), command_stride); } }); diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index bb7c5ff09e..52ab0742a7 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -17,6 +17,7 @@ #include "video_core/rasterizer_interface.h" #include "video_core/renderer_vulkan/blit_image.h" #include "video_core/renderer_vulkan/vk_buffer_cache.h" +#include "video_core/renderer_vulkan/vk_compute_pass.h" #include "video_core/renderer_vulkan/vk_descriptor_buffer.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" #include "video_core/renderer_vulkan/vk_fence_manager.h" @@ -211,6 +212,7 @@ private: DescriptorBufferRing descriptor_buffer_ring; BlitImageHelper blit_image; RenderPassCache render_pass_cache; + IndirectQuadsPass indirect_quads_pass; TextureCacheRuntime texture_cache_runtime; TextureCache texture_cache; diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp index 7b6571d33d..b847b66e7c 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp @@ -211,7 +211,8 @@ StagingBufferRef StagingBufferPool::CreateStagingBuffer(size_t size, MemoryUsage .size = 1ULL << log2_size, .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | - VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_VERTEX_BUFFER_BIT, + VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | + VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT, .sharingMode = VK_SHARING_MODE_EXCLUSIVE, .queueFamilyIndexCount = 0, .pQueueFamilyIndices = nullptr, From 10528f1ceb6d98ebef77ed985b41bbbec4107132 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 17 Sep 2026 17:19:27 -0400 Subject: [PATCH 27/69] Revert some previous changes partially + maintenance5 changes --- .../buffer_cache/virtual_range_cache.h | 92 +++++++++++++++++-- src/video_core/macro.cpp | 8 +- .../renderer_vulkan/vk_buffer_cache.cpp | 13 ++- .../renderer_vulkan/vk_compute_pass.cpp | 12 ++- .../renderer_vulkan/vk_rasterizer.cpp | 11 --- .../renderer_vulkan/vk_rasterizer.h | 2 - .../vulkan_common/vulkan_wrapper.cpp | 1 + src/video_core/vulkan_common/vulkan_wrapper.h | 6 ++ 8 files changed, 114 insertions(+), 31 deletions(-) diff --git a/src/video_core/buffer_cache/virtual_range_cache.h b/src/video_core/buffer_cache/virtual_range_cache.h index 378df18656..542fb52b23 100644 --- a/src/video_core/buffer_cache/virtual_range_cache.h +++ b/src/video_core/buffer_cache/virtual_range_cache.h @@ -5,7 +5,9 @@ #include #include +#include #include +#include #include @@ -26,22 +28,23 @@ using VirtualSegments = boost::container::small_vector; class VirtualRangeCache { public: static constexpr size_t MAX_ENTRIES = 8192; + static constexpr size_t MAX_DEFERRED = 4096; const VirtualSegments* Query(Tegra::MemoryManager& memory, GPUVAddr gpu_addr, u32 size) { - const u64 current = generation.load(std::memory_order_acquire); + if (has_deferred.load(std::memory_order_acquire)) { + ApplyDeferred(); + } if (entries.size() > MAX_ENTRIES) { entries.clear(); } const size_t as_id = memory.GetID(); const u64 key = MakeKey(as_id, gpu_addr); const auto it = entries.find(key); - if (it != entries.end() && it->second.generation == current && - it->second.as_id == as_id && it->second.gpu_addr == gpu_addr && - it->second.size == size) { + if (it != entries.end() && it->second.as_id == as_id && + it->second.gpu_addr == gpu_addr && it->second.size == size) { return &it->second.segments; } Entry entry{}; - entry.generation = current; entry.as_id = as_id; entry.gpu_addr = gpu_addr; entry.size = size; @@ -76,27 +79,96 @@ public: return &result.first->second.segments; } - void Unmap(size_t, GPUVAddr, u64 size) { - if (size != 0) { - generation.fetch_add(1, std::memory_order_release); + void Unmap(size_t as_id, GPUVAddr gpu_addr, u64 size) { + if (size == 0) { + return; } + { + std::scoped_lock lock{deferred_mutex}; + if (!deferred.empty()) { + DeferredUnmap& last = deferred.back(); + if (last.as_id == as_id && last.gpu_addr + last.size == gpu_addr) { + last.size += size; + has_deferred.store(true, std::memory_order_release); + return; + } + } + if (deferred.size() >= MAX_DEFERRED) { + deferred.clear(); + deferred_overflow = true; + } else { + deferred.push_back(DeferredUnmap{ + .as_id = as_id, + .gpu_addr = gpu_addr, + .size = size, + }); + } + } + has_deferred.store(true, std::memory_order_release); } private: struct Entry { VirtualSegments segments; - u64 generation{}; size_t as_id{}; GPUVAddr gpu_addr{}; u32 size{}; }; + struct DeferredUnmap { + size_t as_id; + GPUVAddr gpu_addr; + u64 size; + }; + static u64 MakeKey(size_t as_id, GPUVAddr gpu_addr) { return (static_cast(as_id) << 48) ^ gpu_addr; } + void ApplyDeferred() { + bool overflow = false; + { + std::scoped_lock lock{deferred_mutex}; + has_deferred.store(false, std::memory_order_release); + pending.clear(); + pending.swap(deferred); + overflow = deferred_overflow; + deferred_overflow = false; + } + if (overflow) { + entries.clear(); + return; + } + if (pending.empty() || entries.empty()) { + return; + } + for (auto it = entries.begin(); it != entries.end();) { + const Entry& entry = it->second; + const GPUVAddr entry_end = entry.gpu_addr + entry.size; + bool overlaps = false; + for (const DeferredUnmap& unmap : pending) { + if (unmap.as_id != entry.as_id) { + continue; + } + if (entry.gpu_addr < unmap.gpu_addr + unmap.size && unmap.gpu_addr < entry_end) { + overlaps = true; + break; + } + } + if (overlaps) { + it = entries.erase(it); + } else { + ++it; + } + } + } + ::Common::unordered_map entries; - std::atomic generation{}; + std::vector deferred; + std::vector pending; + std::mutex deferred_mutex; + std::atomic has_deferred{false}; + bool deferred_overflow{}; }; } // namespace VideoCommon diff --git a/src/video_core/macro.cpp b/src/video_core/macro.cpp index ac658f7534..86f528af15 100644 --- a/src/video_core/macro.cpp +++ b/src/video_core/macro.cpp @@ -62,10 +62,6 @@ bool IsTopologySafe(Maxwell3D::Regs::PrimitiveTopology topology) { } } -bool IsTopologySafeIndexedIndirect(Maxwell3D::Regs::PrimitiveTopology topology) { - return IsTopologySafe(topology) || topology == Maxwell3D::Regs::PrimitiveTopology::Quads; -} - } // Anonymous namespace void HLE_DrawArraysIndirect::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, std::span parameters, [[maybe_unused]] u32 method) { @@ -129,7 +125,7 @@ void HLE_DrawArraysIndirect::Fallback(Core::System& system, Engines::Maxwell3D& void HLE_DrawIndexedIndirect::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, std::span parameters, [[maybe_unused]] u32 method) { auto topology = static_cast(parameters[0]); - if (!maxwell3d.AnyParametersDirty() || !IsTopologySafeIndexedIndirect(topology)) { + if (!maxwell3d.AnyParametersDirty() || !IsTopologySafe(topology)) { Fallback(system, maxwell3d, parameters); return; } @@ -202,7 +198,7 @@ void HLE_MultiLayerClear::Execute(Core::System& system, Engines::Maxwell3D& maxw } void HLE_MultiDrawIndexedIndirectCount::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, std::span parameters, [[maybe_unused]] u32 method) { const auto topology = Maxwell3D::Regs::PrimitiveTopology(parameters[2]); - if (IsTopologySafeIndexedIndirect(topology)) { + if (IsTopologySafe(topology)) { const u32 start_indirect = parameters[0]; const u32 end_indirect = parameters[1]; if (start_indirect >= end_indirect) { diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 958f3b2a2e..f397f9faf4 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -578,29 +578,40 @@ bool BufferCacheRuntime::BindMultiRangeStorageBuffer(u64 key, bool is_written) { void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format, u32 base_vertex, u32 num_indices, VkBuffer buffer, - u32 offset, [[maybe_unused]] u32 size) { + u32 offset, u32 size) { VkIndexType vk_index_type = MaxwellToVK::IndexFormat(index_format); VkDeviceSize vk_offset = offset; + VkDeviceSize vk_size = size; VkBuffer vk_buffer = buffer; if (topology == PrimitiveTopology::Quads || topology == PrimitiveTopology::QuadStrip) { vk_index_type = VK_INDEX_TYPE_UINT32; + vk_size = VK_WHOLE_SIZE; std::tie(vk_buffer, vk_offset) = quad_index_pass.Assemble(index_format, num_indices, base_vertex, buffer, offset, topology == PrimitiveTopology::QuadStrip); } else if (vk_index_type == VK_INDEX_TYPE_UINT8_EXT && !device.IsExtIndexTypeUint8Supported()) { vk_index_type = VK_INDEX_TYPE_UINT16; if (uint8_pass) { + vk_size = VK_WHOLE_SIZE; std::tie(vk_buffer, vk_offset) = uint8_pass->Assemble(num_indices, buffer, offset); } else if (device.GetDriverID() == VK_DRIVER_ID_QUALCOMM_PROPRIETARY) { ReserveNullBuffer(); vk_buffer = *null_buffer; vk_offset = 0; + vk_size = VK_WHOLE_SIZE; } } if (vk_buffer == VK_NULL_HANDLE) { // Vulkan doesn't support null index buffers. Replace it with our own null buffer. ReserveNullBuffer(); vk_buffer = *null_buffer; + vk_size = VK_WHOLE_SIZE; + } + if (device.IsKhrMaintenance5Supported()) { + scheduler.Record([vk_buffer, vk_offset, vk_size, vk_index_type](vk::CommandBuffer cmdbuf) { + cmdbuf.BindIndexBuffer2KHR(vk_buffer, vk_offset, vk_size, vk_index_type); + }); + return; } scheduler.Record([vk_buffer, vk_offset, vk_index_type](vk::CommandBuffer cmdbuf) { cmdbuf.BindIndexBuffer(vk_buffer, vk_offset, vk_index_type); diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index a92155881d..2fe69a8dc4 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -342,7 +342,17 @@ std::pair QuadIndexedPass::Assemble( return 2; }(); const u32 input_size = num_vertices << index_shift; - const u32 num_tri_vertices = (is_strip ? (num_vertices - 2) / 2 : num_vertices / 4) * 6; + u32 quads = num_vertices / 4; + if (is_strip) { + quads = 0; + if (num_vertices >= 2) { + quads = (num_vertices - 2) / 2; + } + } + if (quads == 0) { + quads = 1; + } + const u32 num_tri_vertices = quads * 6; const std::size_t staging_size = num_tri_vertices * sizeof(u32); const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal); diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 7fa889e22f..c7978ea9fc 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -212,8 +212,6 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra compute_pass_descriptor_queue(device, UpdateDescriptorQueue::COMPUTE_FRAME_PAYLOAD_SIZE), descriptor_buffer_ring(device, memory_allocator), blit_image(device, scheduler, state_tracker, descriptor_pool), render_pass_cache(device), - indirect_quads_pass(device, scheduler, descriptor_pool, staging_pool, - compute_pass_descriptor_queue), texture_cache_runtime{ device, scheduler, memory_allocator, staging_pool, blit_image, render_pass_cache, descriptor_pool, compute_pass_descriptor_queue}, @@ -319,15 +317,6 @@ void RasterizerVulkan::DrawIndirect() { VkBuffer command_buffer = buffer->Handle(); VkDeviceSize command_offset = offset; u32 command_stride = static_cast(params.stride); - if (params.is_indexed && - maxwell3d->draw_manager.draw_state.topology == Maxwell::PrimitiveTopology::Quads) { - const auto patched = indirect_quads_pass.Assemble( - static_cast(params.max_draw_counts), command_stride, command_buffer, - static_cast(offset)); - command_buffer = patched.first; - command_offset = patched.second; - command_stride = IndirectQuadsPass::COMMAND_WORDS * static_cast(sizeof(u32)); - } if (params.is_byte_count) { scheduler.Record([buffer_obj = buffer->Handle(), offset, stride = params.stride](vk::CommandBuffer cmdbuf) { diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index 52ab0742a7..bb7c5ff09e 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -17,7 +17,6 @@ #include "video_core/rasterizer_interface.h" #include "video_core/renderer_vulkan/blit_image.h" #include "video_core/renderer_vulkan/vk_buffer_cache.h" -#include "video_core/renderer_vulkan/vk_compute_pass.h" #include "video_core/renderer_vulkan/vk_descriptor_buffer.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" #include "video_core/renderer_vulkan/vk_fence_manager.h" @@ -212,7 +211,6 @@ private: DescriptorBufferRing descriptor_buffer_ring; BlitImageHelper blit_image; RenderPassCache render_pass_cache; - IndirectQuadsPass indirect_quads_pass; TextureCacheRuntime texture_cache_runtime; TextureCache texture_cache; diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp index 5044881598..652f864890 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.cpp +++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp @@ -96,6 +96,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCmdBeginDebugUtilsLabelEXT); X(vkCmdBindDescriptorSets); X(vkCmdBindIndexBuffer); + X(vkCmdBindIndexBuffer2KHR); X(vkCmdBindPipeline); X(vkCmdBindTransformFeedbackBuffersEXT); X(vkCmdBindVertexBuffers); diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index 97d2ba3700..2a2c5fedc7 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -211,6 +211,7 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdBeginTransformFeedbackEXT vkCmdBeginTransformFeedbackEXT{}; PFN_vkCmdBindDescriptorSets vkCmdBindDescriptorSets{}; PFN_vkCmdBindIndexBuffer vkCmdBindIndexBuffer{}; + PFN_vkCmdBindIndexBuffer2KHR vkCmdBindIndexBuffer2KHR{}; PFN_vkCmdBindPipeline vkCmdBindPipeline{}; PFN_vkCmdBindTransformFeedbackBuffersEXT vkCmdBindTransformFeedbackBuffersEXT{}; PFN_vkCmdBindVertexBuffers vkCmdBindVertexBuffers{}; @@ -1280,6 +1281,11 @@ public: dld->vkCmdBindIndexBuffer(handle, buffer, offset, index_type); } + void BindIndexBuffer2KHR(VkBuffer buffer, VkDeviceSize offset, VkDeviceSize size, + VkIndexType index_type) const noexcept { + dld->vkCmdBindIndexBuffer2KHR(handle, buffer, offset, size, index_type); + } + void BindVertexBuffers(u32 first, u32 count, const VkBuffer* buffers, const VkDeviceSize* offsets) const noexcept { dld->vkCmdBindVertexBuffers(handle, first, count, buffers, offsets); From 17adc28b2c8a5f0dfe2abcbf306570c5b84d3e9f Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Fri, 18 Sep 2026 01:18:03 -0400 Subject: [PATCH 28/69] Remove some dead accesors --- .../renderer_vulkan/maxwell_to_vk.cpp | 8 ++- .../renderer_vulkan/vk_graphics_pipeline.cpp | 2 +- .../renderer_vulkan/vk_multi_range_buffer.cpp | 10 +++- .../renderer_vulkan/vk_rasterizer.cpp | 3 + .../vulkan_common/vulkan_device.cpp | 26 +------- src/video_core/vulkan_common/vulkan_device.h | 59 +++++-------------- .../vulkan_common/vulkan_wrapper.cpp | 23 ++++++++ src/video_core/vulkan_common/vulkan_wrapper.h | 3 + 8 files changed, 61 insertions(+), 73 deletions(-) diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp index b730ae4d58..dae9fe0269 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp @@ -70,12 +70,14 @@ VkSamplerAddressMode WrapMode(const Device& device, } ASSERT(false); return VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE; - case Tegra::Texture::WrapMode::MirrorOnceClampToEdge: - return VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE; case Tegra::Texture::WrapMode::MirrorOnceBorder: UNIMPLEMENTED(); - return VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE; + [[fallthrough]]; + case Tegra::Texture::WrapMode::MirrorOnceClampToEdge: case Tegra::Texture::WrapMode::MirrorOnceClampOGL: + if (!device.IsKhrSamplerMirrorClampToEdgeSupported()) { + return VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE; + } return VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE; default: UNIMPLEMENTED_MSG("Unimplemented wrap mode={}", wrap_mode); diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index 7ee1c608c0..101fab301e 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -740,7 +740,7 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) { .vertexBindingDivisorCount = static_cast(vertex_binding_divisors.size()), .pVertexBindingDivisors = vertex_binding_divisors.data(), }; - if (!vertex_binding_divisors.empty()) { + if (!vertex_binding_divisors.empty() && device.IsExtVertexAttributeDivisorSupported()) { vertex_input_ci.pNext = &input_divisor_ci; } const bool has_tess_stages = spv_modules[1] || spv_modules[2]; diff --git a/src/video_core/renderer_vulkan/vk_multi_range_buffer.cpp b/src/video_core/renderer_vulkan/vk_multi_range_buffer.cpp index 4d84390002..6d1ae5fc53 100644 --- a/src/video_core/renderer_vulkan/vk_multi_range_buffer.cpp +++ b/src/video_core/renderer_vulkan/vk_multi_range_buffer.cpp @@ -32,8 +32,6 @@ MultiRangeBufferCache::MultiRangeBufferCache(const Device& device) { VkDeviceSize MultiRangeBufferCache::QueryBlockSize(const Device& device, u32& memory_type_bits) const { - const VkDevice logical = *device.GetLogical(); - const auto& dld = device.GetDispatchLoader(); const VkBufferCreateInfo probe_ci{ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, .pNext = nullptr, @@ -44,6 +42,14 @@ VkDeviceSize MultiRangeBufferCache::QueryBlockSize(const Device& device, .queueFamilyIndexCount = 0, .pQueueFamilyIndices = nullptr, }; + if (device.IsKhrMaintenance4Supported()) { + const VkMemoryRequirements reqs = + device.GetLogical().GetDeviceBufferMemoryRequirements(probe_ci); + memory_type_bits = reqs.memoryTypeBits; + return reqs.alignment; + } + const VkDevice logical = *device.GetLogical(); + const auto& dld = device.GetDispatchLoader(); VkBuffer probe{}; if (dld.vkCreateBuffer(logical, &probe_ci, nullptr, &probe) != VK_SUCCESS) { return 0; diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index c7978ea9fc..fd95ea288b 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -326,6 +326,9 @@ void RasterizerVulkan::DrawIndirect() { return; } if (params.include_count) { + if (!device.IsDrawIndirectCountSupported()) { + return; + } const auto count = buffer_cache.GetDrawIndirectCount(); const auto& draw_buffer = count.first; const auto& offset_base = count.second; diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 9a9bfdd43b..27379be506 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -949,6 +949,8 @@ bool Device::GetSuitability(bool requires_swapchain) { extensions.depth_stencil_resolve = extensions.depth_stencil_resolve && (instance_version >= VK_API_VERSION_1_2 || extensions.create_renderpass2); + extensions.draw_indirect_count = + extensions.draw_indirect_count || instance_version >= VK_API_VERSION_1_2; RemoveExtensionIfUnsuitable(extensions.depth_stencil_resolve, VK_KHR_DEPTH_STENCIL_RESOLVE_EXTENSION_NAME); @@ -1408,18 +1410,6 @@ void Device::RemoveUnsuitableExtensions() { features.workgroup_memory_explicit_layout, VK_KHR_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_EXTENSION_NAME); - // VK_KHR_maintenance1 - extensions.maintenance1 = loaded_extensions.contains(VK_KHR_MAINTENANCE_1_EXTENSION_NAME); - RemoveExtensionIfUnsuitable(extensions.maintenance1, VK_KHR_MAINTENANCE_1_EXTENSION_NAME); - - // VK_KHR_maintenance2 - extensions.maintenance2 = loaded_extensions.contains(VK_KHR_MAINTENANCE_2_EXTENSION_NAME); - RemoveExtensionIfUnsuitable(extensions.maintenance2, VK_KHR_MAINTENANCE_2_EXTENSION_NAME); - - // VK_KHR_maintenance3 - extensions.maintenance3 = loaded_extensions.contains(VK_KHR_MAINTENANCE_3_EXTENSION_NAME); - RemoveExtensionIfUnsuitable(extensions.maintenance3, VK_KHR_MAINTENANCE_3_EXTENSION_NAME); - // VK_KHR_maintenance4 extensions.maintenance4 = features.maintenance4.maintenance4; RemoveExtensionFeatureIfUnsuitable(extensions.maintenance4, features.maintenance4, @@ -1430,18 +1420,6 @@ void Device::RemoveUnsuitableExtensions() { RemoveExtensionFeatureIfUnsuitable(extensions.maintenance5, features.maintenance5, VK_KHR_MAINTENANCE_5_EXTENSION_NAME); - // VK_KHR_maintenance6 - extensions.maintenance6 = features.maintenance6.maintenance6; - RemoveExtensionFeatureIfUnsuitable(extensions.maintenance6, features.maintenance6, - VK_KHR_MAINTENANCE_6_EXTENSION_NAME); - - // VK_KHR_maintenance7 - extensions.maintenance7 = loaded_extensions.contains(VK_KHR_MAINTENANCE_7_EXTENSION_NAME); - RemoveExtensionIfUnsuitable(extensions.maintenance7, VK_KHR_MAINTENANCE_7_EXTENSION_NAME); - - // VK_KHR_maintenance8 - extensions.maintenance8 = loaded_extensions.contains(VK_KHR_MAINTENANCE_8_EXTENSION_NAME); - RemoveExtensionIfUnsuitable(extensions.maintenance8, VK_KHR_MAINTENANCE_8_EXTENSION_NAME); // VK_KHR_synchronization2 extensions.synchronization2 = features.synchronization2.synchronization2; diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index c6d6df16b1..8b93aad373 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -72,7 +72,6 @@ VK_DEFINE_HANDLE(VmaAllocator) FEATURE(EXT, TransformFeedback, TRANSFORM_FEEDBACK, transform_feedback) \ FEATURE(EXT, VertexInputDynamicState, VERTEX_INPUT_DYNAMIC_STATE, vertex_input_dynamic_state) \ FEATURE(KHR, Maintenance5, MAINTENANCE_5, maintenance5) \ - FEATURE(KHR, Maintenance6, MAINTENANCE_6, maintenance6) \ FEATURE(KHR, PipelineExecutableProperties, PIPELINE_EXECUTABLE_PROPERTIES, \ pipeline_executable_properties) \ FEATURE(KHR, ShaderQuadControl, SHADER_QUAD_CONTROL, shader_quad_control) \ @@ -104,17 +103,11 @@ VK_DEFINE_HANDLE(VmaAllocator) EXTENSION(KHR, SWAPCHAIN, swapchain) \ EXTENSION(KHR, SWAPCHAIN_MUTABLE_FORMAT, swapchain_mutable_format) \ EXTENSION(KHR, IMAGE_FORMAT_LIST, image_format_list) \ - EXTENSION(KHR, MAINTENANCE_1, maintenance1) \ - EXTENSION(KHR, MAINTENANCE_2, maintenance2) \ - EXTENSION(KHR, MAINTENANCE_3, maintenance3) \ - EXTENSION(KHR, MAINTENANCE_7, maintenance7) \ - EXTENSION(KHR, MAINTENANCE_8, maintenance8) \ EXTENSION(NV, DEVICE_DIAGNOSTICS_CONFIG, device_diagnostics_config) \ EXTENSION(NV, GEOMETRY_SHADER_PASSTHROUGH, geometry_shader_passthrough) \ EXTENSION(NV, VIEWPORT_ARRAY2, viewport_array2) \ EXTENSION(NV, VIEWPORT_SWIZZLE, viewport_swizzle) \ EXTENSION(EXT, FILTER_CUBIC, filter_cubic) \ - EXTENSION(IMG, FILTER_CUBIC, filter_cubic_img) \ EXTENSION(QCOM, FILTER_CUBIC_WEIGHTS, filter_cubic_weights) // Define extensions which must be supported. @@ -549,11 +542,6 @@ FN_MAX_LIMIT_LIST return extensions.geometry_shader_passthrough; } - /// Returns true if the device supports VK_KHR_uniform_buffer_standard_layout. - bool IsKhrUniformBufferStandardLayoutSupported() const { - return extensions.uniform_buffer_standard_layout; - } - /// Returns true if the device supports VK_KHR_push_descriptor. bool IsKhrPushDescriptorSupported() const { return extensions.push_descriptor; @@ -1004,27 +992,27 @@ FN_MAX_LIMIT_LIST return features2.features.multiViewport; } - /// Returns true if the device supports VK_KHR_maintenance1. - bool IsKhrMaintenance1Supported() const { - return extensions.maintenance1; - } - - /// Returns true if the device supports VK_KHR_maintenance2. - bool IsKhrMaintenance2Supported() const { - return extensions.maintenance2; - } - - /// Returns true if the device supports VK_KHR_maintenance3. - bool IsKhrMaintenance3Supported() const { - return extensions.maintenance3; - } - + /// Returns true if the device supports VK_KHR_maintenance5. /// Returns true if the device supports VK_KHR_maintenance4. bool IsKhrMaintenance4Supported() const { return extensions.maintenance4; } - /// Returns true if the device supports VK_KHR_maintenance5. + /// Returns true if the device can read the draw count from a buffer. + bool IsDrawIndirectCountSupported() const { + return extensions.draw_indirect_count; + } + + /// Returns true if the device supports VK_EXT_vertex_attribute_divisor. + bool IsExtVertexAttributeDivisorSupported() const { + return extensions.vertex_attribute_divisor; + } + + /// Returns true if the device supports VK_KHR_sampler_mirror_clamp_to_edge. + bool IsKhrSamplerMirrorClampToEdgeSupported() const { + return extensions.sampler_mirror_clamp_to_edge; + } + bool IsKhrMaintenance5Supported() const { return extensions.maintenance5; } @@ -1046,21 +1034,6 @@ FN_MAX_LIMIT_LIST properties.maintenance5.earlyFragmentSampleMaskTestBeforeSampleCounting; } - /// Returns true if the device supports VK_KHR_maintenance6. - bool IsKhrMaintenance6Supported() const { - return extensions.maintenance6; - } - - /// Returns true if the device supports VK_KHR_maintenance7. - bool IsKhrMaintenance7Supported() const { - return extensions.maintenance7; - } - - /// Returns true if the device supports VK_KHR_maintenance8. - bool IsKhrMaintenance8Supported() const { - return extensions.maintenance8; - } - /// Returns true if the device supports UINT8 index buffer conversion via compute shader. bool SupportsUint8Indices() const { return features.bit8_storage.storageBuffer8BitAccess && diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp index 652f864890..d5012ddd83 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.cpp +++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp @@ -216,6 +216,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkFreeDescriptorSets); X(vkFreeMemory); X(vkGetBufferMemoryRequirements2); + X(vkGetDeviceBufferMemoryRequirements); X(vkGetDeviceQueue); X(vkGetEventStatus); X(vkGetFenceStatus); @@ -266,6 +267,12 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { Proc(dld.vkCmdDrawIndexedIndirectCount, dld, "vkCmdDrawIndexedIndirectCountKHR", device); } + // Maintenance4 queries are core in Vulkan 1.3, otherwise require VK_KHR_maintenance4 + if (!dld.vkGetDeviceBufferMemoryRequirements) { + Proc(dld.vkGetDeviceBufferMemoryRequirements, dld, + "vkGetDeviceBufferMemoryRequirementsKHR", device); + } + // Synchronization2 is core in Vulkan 1.3, otherwise requires VK_KHR_synchronization2 if (!dld.vkCmdPipelineBarrier2) { Proc(dld.vkCmdPipelineBarrier2, dld, "vkCmdPipelineBarrier2KHR", device); @@ -874,6 +881,22 @@ VkMemoryRequirements Device::GetBufferMemoryRequirements(VkBuffer buffer, return requirements.memoryRequirements; } +VkMemoryRequirements Device::GetDeviceBufferMemoryRequirements( + const VkBufferCreateInfo& ci) const noexcept { + const VkDeviceBufferMemoryRequirements info{ + .sType = VK_STRUCTURE_TYPE_DEVICE_BUFFER_MEMORY_REQUIREMENTS, + .pNext = nullptr, + .pCreateInfo = &ci, + }; + VkMemoryRequirements2 requirements{ + .sType = VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2, + .pNext = nullptr, + .memoryRequirements{}, + }; + dld->vkGetDeviceBufferMemoryRequirements(handle, &info, &requirements); + return requirements.memoryRequirements; +} + VkMemoryRequirements Device::GetImageMemoryRequirements(VkImage image) const noexcept { VkMemoryRequirements requirements; dld->vkGetImageMemoryRequirements(handle, image, &requirements); diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index 2a2c5fedc7..5c2d78bc2c 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -332,6 +332,7 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkFreeDescriptorSets vkFreeDescriptorSets{}; PFN_vkFreeMemory vkFreeMemory{}; PFN_vkGetBufferMemoryRequirements2 vkGetBufferMemoryRequirements2{}; + PFN_vkGetDeviceBufferMemoryRequirements vkGetDeviceBufferMemoryRequirements{}; PFN_vkGetDeviceQueue vkGetDeviceQueue{}; PFN_vkGetEventStatus vkGetEventStatus{}; PFN_vkGetFenceStatus vkGetFenceStatus{}; @@ -1105,6 +1106,8 @@ public: VkMemoryRequirements GetBufferMemoryRequirements(VkBuffer buffer, void* pnext = nullptr) const noexcept; + VkMemoryRequirements GetDeviceBufferMemoryRequirements(const VkBufferCreateInfo& ci) const noexcept; + VkMemoryRequirements GetImageMemoryRequirements(VkImage image) const noexcept; std::vector GetPipelineExecutablePropertiesKHR( From 08741c919a9b4affa44b178b2f6d367bedb6347b Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Fri, 18 Sep 2026 01:56:42 -0400 Subject: [PATCH 29/69] Some changes to GC --- src/video_core/CMakeLists.txt | 1 + src/video_core/buffer_cache/buffer_cache.h | 70 ++++++++++++------- .../buffer_cache/buffer_cache_base.h | 8 +++ src/video_core/cache_reclaim.h | 35 ++++++++++ src/video_core/texture_cache/texture_cache.h | 42 +++++------ .../texture_cache/texture_cache_base.h | 10 ++- 6 files changed, 113 insertions(+), 53 deletions(-) create mode 100644 src/video_core/cache_reclaim.h diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index 77883b6484..82f0b68863 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -22,6 +22,7 @@ add_library(video_core STATIC buffer_cache/usage_tracker.h buffer_cache/virtual_range_cache.h buffer_cache/word_manager.h + cache_reclaim.h cache_types.h capture.h cdma_pusher.cpp diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 5705954ae6..061611bdb7 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -30,34 +30,52 @@ BufferCache

::BufferCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, R #ifdef YUZU_LEGACY immediately_free = (Settings::values.vram_usage_mode.GetValue() == Settings::VramUsageMode::Aggressive); #endif - if (!runtime.CanReportMemoryUsage()) { - minimum_memory = DEFAULT_EXPECTED_MEMORY; - critical_memory = DEFAULT_CRITICAL_MEMORY; - return; - } - - const s64 device_local_memory = static_cast(runtime.GetDeviceLocalMemory()); - const s64 min_spacing_expected = device_local_memory - 1_GiB; - const s64 min_spacing_critical = device_local_memory - 512_MiB; - const s64 mem_threshold = (std::min)(device_local_memory, TARGET_THRESHOLD); - const s64 min_vacancy_expected = (6 * mem_threshold) / 10; - const s64 min_vacancy_critical = (2 * mem_threshold) / 10; - minimum_memory = static_cast( - (std::max)((std::min)(device_local_memory - min_vacancy_expected, min_spacing_expected), - DEFAULT_EXPECTED_MEMORY)); - critical_memory = static_cast( - (std::max)((std::min)(device_local_memory - min_vacancy_critical, min_spacing_critical), - DEFAULT_CRITICAL_MEMORY)); + device_local_memory = runtime.GetDeviceLocalMemory(); + const auto thresholds = VideoCommon::MakeReclaimThresholds( + device_local_memory, static_cast(TARGET_THRESHOLD), + static_cast(DEFAULT_EXPECTED_MEMORY), static_cast(DEFAULT_CRITICAL_MEMORY)); + minimum_memory = thresholds.minimum; + expected_memory = thresholds.expected; + critical_memory = thresholds.critical; } template BufferCache

::~BufferCache() = default; +template +void BufferCache

::ReclaimInline() { + if (total_used_memory < minimum_memory) { + return; + } + int num_iterations = 8; + const auto clean_up = [this, &num_iterations](BufferId buffer_id) { + if (num_iterations == 0) { + return true; + } + Buffer& buffer = slot_buffers[buffer_id]; + if (memory_tracker.IsRegionGpuModified(buffer.CpuAddr(), buffer.SizeBytes())) { + return false; + } + --num_iterations; + DeleteBuffer(buffer_id); + return false; + }; + lru_cache.ForEachItemBelow(frame_tick - INLINE_TICKS_TO_DESTROY, clean_up); +} + template void BufferCache

::RunGarbageCollector() { - const bool aggressive_gc = total_used_memory >= critical_memory; - const u64 ticks_to_destroy = aggressive_gc ? 60 : 120; - int num_iterations = aggressive_gc ? 64 : 32; + const bool aggressive_gc = heap_pressure || total_used_memory >= critical_memory; + const bool priority_gc = aggressive_gc || total_used_memory >= expected_memory; + u64 ticks_to_destroy = 120; + int num_iterations = 32; + if (aggressive_gc) { + ticks_to_destroy = 30; + num_iterations = 64; + } else if (priority_gc) { + ticks_to_destroy = 60; + num_iterations = 48; + } const auto clean_up = [this, &num_iterations](BufferId buffer_id) { if (num_iterations == 0) { return true; @@ -96,11 +114,12 @@ void BufferCache

::TickFrame() { const bool skip_preferred = hits * 256 < shots * 251; channel_state->uniform_buffer_skip_cache_size = skip_preferred ? DEFAULT_SKIP_CACHE_SIZE : 0; - // If we can obtain the memory info, use it instead of the estimate. - if (runtime.CanReportMemoryUsage()) { - total_used_memory = runtime.GetDeviceMemoryUsage(); + heap_pressure = false; + if (device_local_memory != 0 && runtime.CanReportMemoryUsage()) { + heap_pressure = runtime.GetDeviceMemoryUsage() + HEAP_PRESSURE_HEADROOM >= + device_local_memory; } - if (total_used_memory >= minimum_memory) { + if (total_used_memory >= minimum_memory || heap_pressure) { RunGarbageCollector(); } ++frame_tick; @@ -1674,6 +1693,7 @@ void BufferCache

::JoinOverlap(BufferId new_buffer_id, BufferId overlap_id, template BufferId BufferCache

::CreateBuffer(DAddr device_addr, u32 wanted_size, bool sparse_compatible) { + ReclaimInline(); DAddr device_addr_end = Common::AlignUp(device_addr + wanted_size, CACHING_PAGESIZE); device_addr = Common::AlignDown(device_addr, CACHING_PAGESIZE); wanted_size = static_cast(device_addr_end - device_addr); diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h index 18997088d3..8afa93b433 100644 --- a/src/video_core/buffer_cache/buffer_cache_base.h +++ b/src/video_core/buffer_cache/buffer_cache_base.h @@ -31,6 +31,7 @@ #include "video_core/buffer_cache/buffer_base.h" #include "video_core/buffer_cache/virtual_range_cache.h" #include "video_core/control/channel_state_cache.h" +#include "video_core/cache_reclaim.h" #include "video_core/delayed_destruction_ring.h" #include "video_core/dirty_flags.h" #include "video_core/engines/maxwell_3d.h" @@ -199,6 +200,8 @@ class BufferCache : public VideoCommon::ChannelSetupCaches compute_segments; u64 frame_tick = 0; u64 total_used_memory = 0; + u64 device_local_memory = 0; u64 minimum_memory = 0; + u64 expected_memory = 0; u64 critical_memory = 0; + bool heap_pressure = false; BufferId inline_buffer_id; #ifdef YUZU_LEGACY bool immediately_free = false; diff --git a/src/video_core/cache_reclaim.h b/src/video_core/cache_reclaim.h new file mode 100644 index 0000000000..aa5c43552f --- /dev/null +++ b/src/video_core/cache_reclaim.h @@ -0,0 +1,35 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include + +#include "common/common_types.h" + +namespace VideoCommon { + +struct ReclaimThresholds { + u64 minimum{}; + u64 expected{}; + u64 critical{}; +}; + +[[nodiscard]] constexpr ReclaimThresholds MakeReclaimThresholds(u64 device_local_memory, + u64 target_threshold, + u64 default_expected, + u64 default_critical) { + u64 critical = default_critical; + if (device_local_memory != 0) { + const u64 budget = (std::min)(device_local_memory, target_threshold); + critical = (std::min)(critical, budget / 2); + } + const u64 expected = (std::min)(default_expected, (critical * 3) / 4); + return ReclaimThresholds{ + .minimum = (expected * 3) / 4, + .expected = expected, + .critical = critical, + }; +} + +} // namespace VideoCommon diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 489ceda94e..2c3a19eb20 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -58,24 +58,14 @@ TextureCache

::TextureCache(Runtime& runtime_, Tegra::MaxwellDeviceMemoryManag void(slot_samplers.insert(runtime, sampler_descriptor)); if constexpr (HAS_DEVICE_MEMORY_INFO) { - const s64 device_local_memory = static_cast(runtime.GetDeviceLocalMemory()); - const s64 min_spacing_expected = device_local_memory - 1_GiB; - const s64 min_spacing_critical = device_local_memory - 512_MiB; - const s64 mem_threshold = (std::min)(device_local_memory, TARGET_THRESHOLD); - const s64 min_vacancy_expected = (6 * mem_threshold) / 10; - const s64 min_vacancy_critical = (2 * mem_threshold) / 10; - expected_memory = static_cast( - (std::max)((std::min)(device_local_memory - min_vacancy_expected, min_spacing_expected), - DEFAULT_EXPECTED_MEMORY)); - critical_memory = static_cast( - (std::max)((std::min)(device_local_memory - min_vacancy_critical, min_spacing_critical), - DEFAULT_CRITICAL_MEMORY)); - minimum_memory = static_cast((device_local_memory - mem_threshold) / 2); - } else { - expected_memory = DEFAULT_EXPECTED_MEMORY + 512_MiB; - critical_memory = DEFAULT_CRITICAL_MEMORY + 1_GiB; - minimum_memory = 0; + device_local_memory = runtime.GetDeviceLocalMemory(); } + const auto thresholds = VideoCommon::MakeReclaimThresholds( + device_local_memory, static_cast(TARGET_THRESHOLD), + static_cast(DEFAULT_EXPECTED_MEMORY), static_cast(DEFAULT_CRITICAL_MEMORY)); + minimum_memory = thresholds.minimum; + expected_memory = thresholds.expected; + critical_memory = thresholds.critical; } template @@ -85,8 +75,9 @@ void TextureCache

::RunGarbageCollector() { u64 ticks_to_destroy = 0; size_t num_iterations = 0; const auto Configure = [&](bool allow_aggressive) { - high_priority_mode = total_used_memory >= expected_memory; - aggressive_mode = allow_aggressive && total_used_memory >= critical_memory; + high_priority_mode = heap_pressure || total_used_memory >= expected_memory; + aggressive_mode = + allow_aggressive && (heap_pressure || total_used_memory >= critical_memory); ticks_to_destroy = aggressive_mode ? 10ULL : high_priority_mode ? 25ULL : 50ULL; num_iterations = aggressive_mode ? 40 : (high_priority_mode ? 20 : 10); }; @@ -100,7 +91,7 @@ void TextureCache

::RunGarbageCollector() { return false; } const bool must_download = IsDownloadable(image) && False(image.flags & ImageFlagBits::BadOverlap); - if ((!aggressive_mode && True(image.flags & ImageFlagBits::CostlyLoad)) || (!high_priority_mode && must_download)) { + if (!aggressive_mode && (must_download || True(image.flags & ImageFlagBits::CostlyLoad))) { return false; } if (must_download) { @@ -126,7 +117,7 @@ void TextureCache

::RunGarbageCollector() { }; Configure(false); lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, Cleanup); - if (total_used_memory >= critical_memory) { + if (heap_pressure || total_used_memory >= critical_memory) { Configure(true); lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, Cleanup); } @@ -134,11 +125,12 @@ void TextureCache

::RunGarbageCollector() { template void TextureCache

::TickFrame() { - // If we can obtain the memory info, use it instead of the estimate. - if (runtime.CanReportMemoryUsage()) { - total_used_memory = runtime.GetDeviceMemoryUsage(); + heap_pressure = false; + if (device_local_memory != 0 && runtime.CanReportMemoryUsage()) { + heap_pressure = runtime.GetDeviceMemoryUsage() + HEAP_PRESSURE_HEADROOM >= + device_local_memory; } - if (total_used_memory > minimum_memory) { + if (total_used_memory > minimum_memory || heap_pressure) { RunGarbageCollector(); } sentenced_images.Tick(); diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index 29f5bdde83..63b83050e6 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -31,6 +31,7 @@ #include "common/thread_worker.h" #include "video_core/compatible_formats.h" #include "video_core/control/channel_state_cache.h" +#include "video_core/cache_reclaim.h" #include "video_core/delayed_destruction_ring.h" #include "video_core/engines/fermi_2d.h" #include "video_core/surface.h" @@ -121,6 +122,7 @@ class TextureCache : public VideoCommon::ChannelSetupCaches Date: Fri, 18 Sep 2026 03:30:06 -0400 Subject: [PATCH 30/69] Quick change to previous GC adjustment --- src/video_core/buffer_cache/buffer_cache.h | 18 ++++++++++++------ .../buffer_cache/buffer_cache_base.h | 1 + src/video_core/cache_reclaim.h | 5 ++++- src/video_core/texture_cache/texture_cache.h | 9 ++++----- .../texture_cache/texture_cache_base.h | 4 ++-- 5 files changed, 23 insertions(+), 14 deletions(-) diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 061611bdb7..6066609e45 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -33,10 +33,12 @@ BufferCache

::BufferCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, R device_local_memory = runtime.GetDeviceLocalMemory(); const auto thresholds = VideoCommon::MakeReclaimThresholds( device_local_memory, static_cast(TARGET_THRESHOLD), - static_cast(DEFAULT_EXPECTED_MEMORY), static_cast(DEFAULT_CRITICAL_MEMORY)); + static_cast(DEFAULT_EXPECTED_MEMORY), static_cast(DEFAULT_CRITICAL_MEMORY), + HEAP_PRESSURE_HEADROOM); minimum_memory = thresholds.minimum; expected_memory = thresholds.expected; critical_memory = thresholds.critical; + heap_headroom = thresholds.headroom; } template @@ -52,11 +54,11 @@ void BufferCache

::ReclaimInline() { if (num_iterations == 0) { return true; } + --num_iterations; Buffer& buffer = slot_buffers[buffer_id]; if (memory_tracker.IsRegionGpuModified(buffer.CpuAddr(), buffer.SizeBytes())) { return false; } - --num_iterations; DeleteBuffer(buffer_id); return false; }; @@ -116,8 +118,7 @@ void BufferCache

::TickFrame() { heap_pressure = false; if (device_local_memory != 0 && runtime.CanReportMemoryUsage()) { - heap_pressure = runtime.GetDeviceMemoryUsage() + HEAP_PRESSURE_HEADROOM >= - device_local_memory; + heap_pressure = runtime.GetDeviceMemoryUsage() + heap_headroom >= device_local_memory; } if (total_used_memory >= minimum_memory || heap_pressure) { RunGarbageCollector(); @@ -363,6 +364,7 @@ void BufferCache

::DisableGraphicsUniformBuffer(size_t stage, u32 index) { template void BufferCache

::UpdateGraphicsBuffers(bool is_indexed) { + ReclaimInline(); do { channel_state->has_deleted_buffers = false; DoUpdateGraphicsBuffers(is_indexed); @@ -371,6 +373,7 @@ void BufferCache

::UpdateGraphicsBuffers(bool is_indexed) { template void BufferCache

::UpdateComputeBuffers() { + ReclaimInline(); do { channel_state->has_deleted_buffers = false; DoUpdateComputeBuffers(); @@ -1343,7 +1346,7 @@ void BufferCache

::UpdateIndexBuffer() { inline_buffer_id = CreateBuffer(0, buffer_size, false); } if (slot_buffers[inline_buffer_id].SizeBytes() < buffer_size) [[unlikely]] { - slot_buffers.erase(inline_buffer_id); + DeleteBuffer(inline_buffer_id, true); inline_buffer_id = CreateBuffer(0, buffer_size, false); } channel_state->index_buffer = Binding{ @@ -1693,7 +1696,6 @@ void BufferCache

::JoinOverlap(BufferId new_buffer_id, BufferId overlap_id, template BufferId BufferCache

::CreateBuffer(DAddr device_addr, u32 wanted_size, bool sparse_compatible) { - ReclaimInline(); DAddr device_addr_end = Common::AlignUp(device_addr + wanted_size, CACHING_PAGESIZE); device_addr = Common::AlignDown(device_addr, CACHING_PAGESIZE); wanted_size = static_cast(device_addr_end - device_addr); @@ -1994,6 +1996,10 @@ void BufferCache

::DeleteBuffer(BufferId buffer_id, bool do_not_mark) { memory_tracker.MarkRegionAsCpuModified(buffer.CpuAddr(), buffer.SizeBytes()); } + if (inline_buffer_id == buffer_id) { + inline_buffer_id = NULL_BUFFER_ID; + } + Unregister(buffer_id); #ifdef YUZU_LEGACY diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h index 8afa93b433..b4af2820d5 100644 --- a/src/video_core/buffer_cache/buffer_cache_base.h +++ b/src/video_core/buffer_cache/buffer_cache_base.h @@ -549,6 +549,7 @@ private: u64 minimum_memory = 0; u64 expected_memory = 0; u64 critical_memory = 0; + u64 heap_headroom = 0; bool heap_pressure = false; BufferId inline_buffer_id; #ifdef YUZU_LEGACY diff --git a/src/video_core/cache_reclaim.h b/src/video_core/cache_reclaim.h index aa5c43552f..a8870ea96b 100644 --- a/src/video_core/cache_reclaim.h +++ b/src/video_core/cache_reclaim.h @@ -13,12 +13,14 @@ struct ReclaimThresholds { u64 minimum{}; u64 expected{}; u64 critical{}; + u64 headroom{}; }; [[nodiscard]] constexpr ReclaimThresholds MakeReclaimThresholds(u64 device_local_memory, u64 target_threshold, u64 default_expected, - u64 default_critical) { + u64 default_critical, + u64 default_headroom) { u64 critical = default_critical; if (device_local_memory != 0) { const u64 budget = (std::min)(device_local_memory, target_threshold); @@ -29,6 +31,7 @@ struct ReclaimThresholds { .minimum = (expected * 3) / 4, .expected = expected, .critical = critical, + .headroom = (std::clamp)(device_local_memory / 4, default_headroom, default_headroom * 2), }; } diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 2c3a19eb20..4092036b9e 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -62,10 +62,12 @@ TextureCache

::TextureCache(Runtime& runtime_, Tegra::MaxwellDeviceMemoryManag } const auto thresholds = VideoCommon::MakeReclaimThresholds( device_local_memory, static_cast(TARGET_THRESHOLD), - static_cast(DEFAULT_EXPECTED_MEMORY), static_cast(DEFAULT_CRITICAL_MEMORY)); + static_cast(DEFAULT_EXPECTED_MEMORY), static_cast(DEFAULT_CRITICAL_MEMORY), + HEAP_PRESSURE_HEADROOM); minimum_memory = thresholds.minimum; expected_memory = thresholds.expected; critical_memory = thresholds.critical; + heap_headroom = thresholds.headroom; } template @@ -127,8 +129,7 @@ template void TextureCache

::TickFrame() { heap_pressure = false; if (device_local_memory != 0 && runtime.CanReportMemoryUsage()) { - heap_pressure = runtime.GetDeviceMemoryUsage() + HEAP_PRESSURE_HEADROOM >= - device_local_memory; + heap_pressure = runtime.GetDeviceMemoryUsage() + heap_headroom >= device_local_memory; } if (total_used_memory > minimum_memory || heap_pressure) { RunGarbageCollector(); @@ -1320,8 +1321,6 @@ u64 TextureCache

::GetScaledImageSizeBytes(const ImageBase& image) { template void TextureCache

::QueueAsyncDecode(Image& image, ImageId image_id) { UNIMPLEMENTED_IF(False(image.flags & ImageFlagBits::Converted)); - LOG_INFO(HW_GPU, "Queuing async texture decode"); - image.flags |= ImageFlagBits::IsDecoding; auto decode = std::make_unique(); auto* decode_ptr = decode.get(); diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index 63b83050e6..7a46134490 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -446,6 +446,7 @@ private: u64 minimum_memory = 0; u64 expected_memory = 0; u64 critical_memory = 0; + u64 heap_headroom = 0; bool heap_pressure = false; struct BufferDownload { @@ -498,8 +499,7 @@ private: u64 modification_tick = 0; u64 frame_tick = 0; - Common::ThreadWorker texture_decode_worker{1, "TextureDecoder", {}, - Common::ThreadPlacement::Efficiency}; + Common::ThreadWorker texture_decode_worker{1, "TextureDecoder"}; std::vector> async_decodes; // Join caching From e9436f8094e3a2ab05b2abc15a75706832e6a18a Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Fri, 18 Sep 2026 04:28:12 -0400 Subject: [PATCH 31/69] A test to remove uneeded Finish() --- src/video_core/renderer_vulkan/vk_compute_pass.cpp | 11 ++++++----- src/video_core/renderer_vulkan/vk_texture_cache.cpp | 10 +--------- 2 files changed, 7 insertions(+), 14 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 2fe69a8dc4..914174b92c 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -596,8 +596,9 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, const VkImageMemoryBarrier image_barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, .pNext = nullptr, - .srcAccessMask = static_cast(is_initialized ? VK_ACCESS_SHADER_WRITE_BIT - : VK_ACCESS_NONE), + .srcAccessMask = static_cast( + is_initialized ? VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT + : VK_ACCESS_NONE), .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED, .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -656,7 +657,8 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, .pNext = nullptr, .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | + VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -671,9 +673,8 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, }, }; cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, 0, image_barrier); + vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, 0, image_barrier); }); - scheduler.Finish(); } } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index b5ce435d83..1eaaf1eba4 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -1899,14 +1899,6 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu } current_image = &Image::original_image; storage_image_views.resize(info.resources.levels); - if (WillUseAcceleratedAstcDecode(runtime->device, info)) { - const auto& device = runtime->device.GetLogical(); - const VkFormat storage_format = VK_FORMAT_A8B8G8R8_UNORM_PACK32; - for (s32 level = 0; level < info.resources.levels; ++level) { - storage_image_views[level] = - MakeStorageView(device, level, *original_image, storage_format); - } - } } Image::Image(const VideoCommon::NullImageParams& params) : VideoCommon::ImageBase{params} {} @@ -2282,7 +2274,7 @@ VkImageView Image::StorageImageView(s32 level) noexcept { if (WillUseAcceleratedAstcDecode(runtime->device, info)) { format_info.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32; } - view = MakeStorageView(runtime->device.GetLogical(), level, *(this->*current_image), + view = MakeStorageView(runtime->device.GetLogical(), level, *original_image, format_info.format); } return *view; From 3cabafc06a8ac255e4a466fbb4711e41b386c644 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sat, 19 Sep 2026 03:30:18 -0400 Subject: [PATCH 32/69] Re-add unswizzle linear 2D to the pass --- .../renderer_vulkan/vk_compute_pass.cpp | 170 ++++++++++++------ .../renderer_vulkan/vk_compute_pass.h | 17 ++ .../renderer_vulkan/vk_texture_cache.cpp | 65 ++++++- .../renderer_vulkan/vk_texture_cache.h | 1 + 4 files changed, 200 insertions(+), 53 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 914174b92c..b542ccceb5 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -17,6 +17,7 @@ #include "common/div_ceil.h" #include "common/vector_math.h" #include "video_core/host_shaders/astc_decoder_comp_spv.h" +#include "video_core/host_shaders/block_linear_unswizzle_2d_comp_spv.h" #include "video_core/host_shaders/queries_prefix_scan_sum_comp_spv.h" #include "video_core/host_shaders/queries_prefix_scan_sum_nosubgroups_comp_spv.h" #include "video_core/host_shaders/resolve_conditional_render_comp_spv.h" @@ -565,6 +566,72 @@ void QueriesPrefixScanPass::Run(VkBuffer accumulation_buffer, VkBuffer dst_buffe } } +namespace { + +void RecordUnswizzleBeginBarrier(Scheduler& scheduler, VkPipeline vk_pipeline, VkImage vk_image, + VkImageAspectFlags aspect_mask, bool is_initialized) { + VkAccessFlags src_access = VK_ACCESS_NONE; + VkImageLayout old_layout = VK_IMAGE_LAYOUT_UNDEFINED; + VkPipelineStageFlags src_stage = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT; + if (is_initialized) { + src_access = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT; + old_layout = VK_IMAGE_LAYOUT_GENERAL; + src_stage = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER; + } + scheduler.Record([vk_pipeline, vk_image, aspect_mask, src_access, old_layout, + src_stage](vk::CommandBuffer cmdbuf) { + const VkImageMemoryBarrier image_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = src_access, + .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, + .oldLayout = old_layout, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = vk_image, + .subresourceRange{ + .aspectMask = aspect_mask, + .baseMipLevel = 0, + .levelCount = VK_REMAINING_MIP_LEVELS, + .baseArrayLayer = 0, + .layerCount = VK_REMAINING_ARRAY_LAYERS, + }, + }; + cmdbuf.PipelineBarrier(src_stage, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier); + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline); + }); +} + +void RecordUnswizzleEndBarrier(Scheduler& scheduler, VkImage vk_image, + VkImageAspectFlags aspect_mask) { + scheduler.Record([vk_image, aspect_mask](vk::CommandBuffer cmdbuf) { + const VkImageMemoryBarrier image_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | + VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = vk_image, + .subresourceRange{ + .aspectMask = aspect_mask, + .baseMipLevel = 0, + .levelCount = VK_REMAINING_MIP_LEVELS, + .baseArrayLayer = 0, + .layerCount = VK_REMAINING_ARRAY_LAYERS, + }, + }; + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, + vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, 0, image_barrier); + }); +} + +} // Anonymous namespace + ASTCDecoderPass::ASTCDecoderPass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, @@ -587,37 +654,10 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, VideoCore::Surface::DefaultBlockHeight(image.info.format), }; scheduler.RequestOutsideRenderPassOperationContext(); - const VkPipeline vk_pipeline = *pipeline; const VkImageAspectFlags aspect_mask = image.AspectMask(); const VkImage vk_image = image.Handle(); - const bool is_initialized = image.ExchangeInitialization(); - scheduler.Record([vk_pipeline, vk_image, aspect_mask, - is_initialized](vk::CommandBuffer cmdbuf) { - const VkImageMemoryBarrier image_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = static_cast( - is_initialized ? VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT - : VK_ACCESS_NONE), - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, - .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED, - .newLayout = VK_IMAGE_LAYOUT_GENERAL, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .image = vk_image, - .subresourceRange{ - .aspectMask = aspect_mask, - .baseMipLevel = 0, - .levelCount = VK_REMAINING_MIP_LEVELS, - .baseArrayLayer = 0, - .layerCount = VK_REMAINING_ARRAY_LAYERS, - }, - }; - cmdbuf.PipelineBarrier(is_initialized ? vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER - : VkPipelineStageFlags(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT), - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier); - cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline); - }); + RecordUnswizzleBeginBarrier(scheduler, *pipeline, vk_image, aspect_mask, + image.ExchangeInitialization()); for (const VideoCommon::SwizzleParameters& swizzle : swizzles) { const size_t input_offset = swizzle.buffer_offset + map.offset; const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 8U); @@ -652,29 +692,55 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_dispatches_z); }); } - scheduler.Record([vk_image, aspect_mask](vk::CommandBuffer cmdbuf) { - const VkImageMemoryBarrier image_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, - .oldLayout = VK_IMAGE_LAYOUT_GENERAL, - .newLayout = VK_IMAGE_LAYOUT_GENERAL, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .image = vk_image, - .subresourceRange{ - .aspectMask = aspect_mask, - .baseMipLevel = 0, - .levelCount = VK_REMAINING_MIP_LEVELS, - .baseArrayLayer = 0, - .layerCount = VK_REMAINING_ARRAY_LAYERS, - }, - }; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, 0, image_barrier); - }); + RecordUnswizzleEndBarrier(scheduler, vk_image, aspect_mask); +} + +BlockLinearUnswizzleImage2DPass::BlockLinearUnswizzleImage2DPass( + const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_) + : ComputePass(device_, scheduler_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, + ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, + COMPUTE_PUSH_CONSTANT_RANGE< + sizeof(VideoCommon::Accelerated::BlockLinearSwizzle2DParams)>, + BLOCK_LINEAR_UNSWIZZLE_2D_COMP_SPV), + scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, + compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} + +BlockLinearUnswizzleImage2DPass::~BlockLinearUnswizzleImage2DPass() = default; + +void BlockLinearUnswizzleImage2DPass::Unswizzle( + Image& image, const StagingBufferRef& map, + std::span swizzles) { + using namespace VideoCommon::Accelerated; + scheduler.RequestOutsideRenderPassOperationContext(); + const VkImageAspectFlags aspect_mask = image.AspectMask(); + const VkImage vk_image = image.Handle(); + RecordUnswizzleBeginBarrier(scheduler, *pipeline, vk_image, aspect_mask, + image.ExchangeInitialization()); + for (const VideoCommon::SwizzleParameters& swizzle : swizzles) { + const size_t input_offset = swizzle.buffer_offset + map.offset; + const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U); + const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U); + const u32 num_dispatches_z = image.info.resources.layers; + + compute_pass_descriptor_queue.Acquire(scheduler, 2); + compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset, + image.guest_size_bytes - swizzle.buffer_offset); + compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level)); + const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; + + const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info); + scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, params, + descriptor_data](vk::CommandBuffer cmdbuf) { + const VkDescriptorSet set = descriptor_allocator.Commit(); + device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); + cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params); + cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_dispatches_z); + }); + } + RecordUnswizzleEndBarrier(scheduler, vk_image, aspect_mask); } } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h index 3fdd2c6f01..c0bc1e6e88 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.h +++ b/src/video_core/renderer_vulkan/vk_compute_pass.h @@ -153,4 +153,21 @@ private: MemoryAllocator& memory_allocator; }; +class BlockLinearUnswizzleImage2DPass final : public ComputePass { +public: + explicit BlockLinearUnswizzleImage2DPass( + const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_); + ~BlockLinearUnswizzleImage2DPass(); + + void Unswizzle(Image& image, const StagingBufferRef& map, + std::span swizzles); + +private: + Scheduler& scheduler; + StagingBufferPool& staging_buffer_pool; + ComputePassDescriptorQueue& compute_pass_descriptor_queue; +}; + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 1eaaf1eba4..c44546c254 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -158,6 +158,52 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { return info.size.depth == 1; } +[[nodiscard]] VkFormat UnswizzleStorageFormat(u32 bytes_per_block) { + switch (bytes_per_block) { + case 1: + return VK_FORMAT_R8_UINT; + case 2: + return VK_FORMAT_R16_UINT; + case 4: + return VK_FORMAT_R32_UINT; + case 8: + return VK_FORMAT_R32G32_UINT; + case 16: + return VK_FORMAT_R32G32B32A32_UINT; + default: + return VK_FORMAT_UNDEFINED; + } +} + +[[nodiscard]] bool IsUnswizzleAcceleratedFormat(const Device& device, PixelFormat format) { + if (!device.IsStorageBuffer8BitAccessSupported() || + !device.IsStorageBuffer16BitAccessSupported()) { + return false; + } + if (IsPixelFormatASTC(format) || VideoCore::Surface::IsPixelFormatBCn(format)) { + return false; + } + if (VideoCore::Surface::GetFormatType(format) != SurfaceType::ColorTexture) { + return false; + } + const VkFormat storage_format = UnswizzleStorageFormat(BytesPerBlock(format)); + if (storage_format == VK_FORMAT_UNDEFINED) { + return false; + } + return device.IsFormatSupported(storage_format, VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT, + FormatType::Optimal); +} + +[[nodiscard]] bool WillUseAcceleratedUnswizzle(const Device& device, const ImageInfo& info) { + if (info.type != ImageType::e2D || info.num_samples > 1) { + return false; + } + if (!device.IsKhrImageFormatListSupported()) { + return false; + } + return IsUnswizzleAcceleratedFormat(device, info.format); +} + [[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info, std::optional format_override = {}) { auto format_info = @@ -944,6 +990,11 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched astc_decoder_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, compute_pass_descriptor_queue, memory_allocator); } + if (device.IsKhrImageFormatListSupported() && device.IsStorageBuffer8BitAccessSupported() && + device.IsStorageBuffer16BitAccessSupported()) { + bl_unswizzle_2d_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, + compute_pass_descriptor_queue); + } if (!device.IsKhrImageFormatListSupported()) { return; } @@ -951,6 +1002,8 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched const auto image_format = static_cast(index_a); if (IsPixelFormatASTC(image_format) && !device.IsOptimalAstcSupported()) { view_formats[index_a].push_back(VK_FORMAT_A8B8G8R8_UNORM_PACK32); + } else if (IsUnswizzleAcceleratedFormat(device, image_format)) { + view_formats[index_a].push_back(UnswizzleStorageFormat(BytesPerBlock(image_format))); } for (size_t index_b = 0; index_b < VideoCore::Surface::MaxPixelFormat; index_b++) { const auto view_format = static_cast(index_b); @@ -1894,6 +1947,10 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu flags |= VideoCommon::ImageFlagBits::Converted; flags |= VideoCommon::ImageFlagBits::CostlyLoad; } + if (False(flags & VideoCommon::ImageFlagBits::Converted) && + WillUseAcceleratedUnswizzle(runtime->device, info)) { + flags |= VideoCommon::ImageFlagBits::AcceleratedUpload; + } if (runtime->device.HasDebuggingToolAttached()) { original_image.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); } @@ -2273,6 +2330,8 @@ VkImageView Image::StorageImageView(s32 level) noexcept { MaxwellToVK::SurfaceFormat(runtime->device, FormatType::Optimal, true, info.format); if (WillUseAcceleratedAstcDecode(runtime->device, info)) { format_info.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32; + } else if (True(flags & VideoCommon::ImageFlagBits::AcceleratedUpload)) { + format_info.format = UnswizzleStorageFormat(BytesPerBlock(info.format)); } view = MakeStorageView(runtime->device.GetLogical(), level, *original_image, format_info.format); @@ -3079,7 +3138,11 @@ void TextureCacheRuntime::AccelerateImageUpload( if (is_rescaled) { image.ScaleDown(true); } - astc_decoder_pass->Assemble(image, map, swizzles); + if (IsPixelFormatASTC(image.info.format)) { + astc_decoder_pass->Assemble(image, map, swizzles); + } else { + bl_unswizzle_2d_pass->Unswizzle(image, map, swizzles); + } if (is_rescaled) { image.ScaleUp(); } diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 6510e10e50..8c86bdd5d2 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -156,6 +156,7 @@ public: BlitImageHelper& blit_image_helper; RenderPassCache& render_pass_cache; std::optional astc_decoder_pass; + std::optional bl_unswizzle_2d_pass; const Settings::ResolutionScalingInfo& resolution; std::array, VideoCore::Surface::MaxPixelFormat> view_formats; From 51d64c298d7127dd7a7e616887196ef6b8f184f3 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sat, 19 Sep 2026 04:40:22 -0400 Subject: [PATCH 33/69] Some reviews on storage views --- .../renderer_vulkan/vk_texture_cache.cpp | 35 +++++++++++++++++-- 1 file changed, 32 insertions(+), 3 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index c44546c254..f3142215ff 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -180,7 +180,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { !device.IsStorageBuffer16BitAccessSupported()) { return false; } - if (IsPixelFormatASTC(format) || VideoCore::Surface::IsPixelFormatBCn(format)) { + if (DefaultBlockWidth(format) != 1 || DefaultBlockHeight(format) != 1) { return false; } if (VideoCore::Surface::GetFormatType(format) != SurfaceType::ColorTexture) { @@ -204,6 +204,29 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { return IsUnswizzleAcceleratedFormat(device, info.format); } +[[nodiscard]] std::array ShaderStorageViewFormats(PixelFormat format) { + if (VideoCore::Surface::GetFormatType(format) != SurfaceType::ColorTexture) { + return {}; + } + if (DefaultBlockWidth(format) != 1 || DefaultBlockHeight(format) != 1) { + return {}; + } + switch (BytesPerBlock(format)) { + case 1: + return {VK_FORMAT_R8_UINT, VK_FORMAT_R8_SINT}; + case 2: + return {VK_FORMAT_R16_UINT, VK_FORMAT_R16_SINT}; + case 4: + return {VK_FORMAT_R32_UINT, VK_FORMAT_UNDEFINED}; + case 8: + return {VK_FORMAT_R32G32_UINT, VK_FORMAT_UNDEFINED}; + case 16: + return {VK_FORMAT_R32G32B32A32_UINT, VK_FORMAT_UNDEFINED}; + default: + return {}; + } +} + [[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info, std::optional format_override = {}) { auto format_info = @@ -1002,8 +1025,6 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched const auto image_format = static_cast(index_a); if (IsPixelFormatASTC(image_format) && !device.IsOptimalAstcSupported()) { view_formats[index_a].push_back(VK_FORMAT_A8B8G8R8_UNORM_PACK32); - } else if (IsUnswizzleAcceleratedFormat(device, image_format)) { - view_formats[index_a].push_back(UnswizzleStorageFormat(BytesPerBlock(image_format))); } for (size_t index_b = 0; index_b < VideoCore::Surface::MaxPixelFormat; index_b++) { const auto view_format = static_cast(index_b); @@ -1013,6 +1034,14 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched view_formats[index_a].push_back(view_info.format); } } + auto& formats = view_formats[index_a]; + for (const VkFormat storage_format : ShaderStorageViewFormats(image_format)) { + if (storage_format == VK_FORMAT_UNDEFINED || + std::ranges::find(formats, storage_format) != formats.end()) { + continue; + } + formats.push_back(storage_format); + } } } From 1529e30435165f6df8f10222da55cbff8294a2e9 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sat, 19 Sep 2026 05:18:56 -0400 Subject: [PATCH 34/69] quick tests with dependency flags per attachment stage --- .../renderer_vulkan/vk_render_pass_cache.cpp | 86 +++++++++++++------ 1 file changed, 61 insertions(+), 25 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp b/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp index 4d8fa15d16..fc65596eb8 100644 --- a/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp @@ -4,6 +4,8 @@ // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later +#include + #include "common/container/unordered_map.h" #include @@ -229,17 +231,48 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { .preserveAttachmentCount = 0, .pPreserveAttachments = nullptr, }; - const VkSubpassDependency dependency{ - .srcSubpass = 0, // Current subpass - .dstSubpass = 0, // Same subpass (self-dependency) - .srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT, + static constexpr VkPipelineStageFlags ATTACHMENT_STAGES = + VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | + VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT; + static constexpr VkAccessFlags ATTACHMENT_WRITE_ACCESS = + VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; + static constexpr VkAccessFlags ATTACHMENT_ACCESS = + ATTACHMENT_WRITE_ACCESS | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | + VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT; + static constexpr VkPipelineStageFlags OUTSIDE_STAGES = + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | + VK_PIPELINE_STAGE_TRANSFER_BIT; + + const std::array dependencies{ + VkSubpassDependency{ + .srcSubpass = VK_SUBPASS_EXTERNAL, + .dstSubpass = 0, + .srcStageMask = OUTSIDE_STAGES, + .dstStageMask = ATTACHMENT_STAGES, + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = ATTACHMENT_ACCESS, + .dependencyFlags = 0, + }, + VkSubpassDependency{ + .srcSubpass = 0, + .dstSubpass = 0, + .srcStageMask = ATTACHMENT_STAGES, .dstStageMask = VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, - .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, + .srcAccessMask = ATTACHMENT_WRITE_ACCESS, .dstAccessMask = VK_ACCESS_SHADER_READ_BIT, - .dependencyFlags = VK_DEPENDENCY_BY_REGION_BIT + .dependencyFlags = VK_DEPENDENCY_BY_REGION_BIT, + }, + VkSubpassDependency{ + .srcSubpass = 0, + .dstSubpass = VK_SUBPASS_EXTERNAL, + .srcStageMask = ATTACHMENT_STAGES, + .dstStageMask = OUTSIDE_STAGES, + .srcAccessMask = ATTACHMENT_WRITE_ACCESS, + .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT | + VK_ACCESS_TRANSFER_WRITE_BIT, + .dependencyFlags = 0, + }, }; if (device->IsKhrCreateRenderPass2Supported()) { @@ -299,18 +332,21 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { .preserveAttachmentCount = 0, .pPreserveAttachments = nullptr, }; - const VkSubpassDependency2 dependency2{ - .sType = VK_STRUCTURE_TYPE_SUBPASS_DEPENDENCY_2, - .pNext = nullptr, - .srcSubpass = dependency.srcSubpass, - .dstSubpass = dependency.dstSubpass, - .srcStageMask = dependency.srcStageMask, - .dstStageMask = dependency.dstStageMask, - .srcAccessMask = dependency.srcAccessMask, - .dstAccessMask = dependency.dstAccessMask, - .dependencyFlags = dependency.dependencyFlags, - .viewOffset = 0, - }; + boost::container::static_vector dependencies2; + for (const VkSubpassDependency& dependency : dependencies) { + dependencies2.push_back(VkSubpassDependency2{ + .sType = VK_STRUCTURE_TYPE_SUBPASS_DEPENDENCY_2, + .pNext = nullptr, + .srcSubpass = dependency.srcSubpass, + .dstSubpass = dependency.dstSubpass, + .srcStageMask = dependency.srcStageMask, + .dstStageMask = dependency.dstStageMask, + .srcAccessMask = dependency.srcAccessMask, + .dstAccessMask = dependency.dstAccessMask, + .dependencyFlags = dependency.dependencyFlags, + .viewOffset = 0, + }); + } pair->second = device->GetLogical().CreateRenderPass2({ .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2, .pNext = nullptr, @@ -319,8 +355,8 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { .pAttachments = descriptions2.empty() ? nullptr : descriptions2.data(), .subpassCount = 1, .pSubpasses = &subpass2, - .dependencyCount = 1, - .pDependencies = &dependency2, + .dependencyCount = static_cast(dependencies2.size()), + .pDependencies = dependencies2.data(), .correlatedViewMaskCount = 0, .pCorrelatedViewMasks = nullptr, }); @@ -335,8 +371,8 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { .pAttachments = descriptions.empty() ? nullptr : descriptions.data(), .subpassCount = 1, .pSubpasses = &subpass, - .dependencyCount = 1, - .pDependencies = &dependency, + .dependencyCount = static_cast(dependencies.size()), + .pDependencies = dependencies.data(), }); return *pair->second; } From 20d3a4370ab5ada66df337de57930024b7174792 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sat, 19 Sep 2026 05:42:55 -0400 Subject: [PATCH 35/69] Remove the dependency test + remove color texture filter per shasder storage view --- .../renderer_vulkan/vk_render_pass_cache.cpp | 86 ++++++------------- .../renderer_vulkan/vk_texture_cache.cpp | 3 - 2 files changed, 25 insertions(+), 64 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp b/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp index fc65596eb8..4d8fa15d16 100644 --- a/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp @@ -4,8 +4,6 @@ // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later -#include - #include "common/container/unordered_map.h" #include @@ -231,48 +229,17 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { .preserveAttachmentCount = 0, .pPreserveAttachments = nullptr, }; - static constexpr VkPipelineStageFlags ATTACHMENT_STAGES = - VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT; - static constexpr VkAccessFlags ATTACHMENT_WRITE_ACCESS = - VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; - static constexpr VkAccessFlags ATTACHMENT_ACCESS = - ATTACHMENT_WRITE_ACCESS | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT; - static constexpr VkPipelineStageFlags OUTSIDE_STAGES = - VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | - VK_PIPELINE_STAGE_TRANSFER_BIT; - - const std::array dependencies{ - VkSubpassDependency{ - .srcSubpass = VK_SUBPASS_EXTERNAL, - .dstSubpass = 0, - .srcStageMask = OUTSIDE_STAGES, - .dstStageMask = ATTACHMENT_STAGES, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = ATTACHMENT_ACCESS, - .dependencyFlags = 0, - }, - VkSubpassDependency{ - .srcSubpass = 0, - .dstSubpass = 0, - .srcStageMask = ATTACHMENT_STAGES, + const VkSubpassDependency dependency{ + .srcSubpass = 0, // Current subpass + .dstSubpass = 0, // Same subpass (self-dependency) + .srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | + VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT, .dstStageMask = VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, - .srcAccessMask = ATTACHMENT_WRITE_ACCESS, + .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, .dstAccessMask = VK_ACCESS_SHADER_READ_BIT, - .dependencyFlags = VK_DEPENDENCY_BY_REGION_BIT, - }, - VkSubpassDependency{ - .srcSubpass = 0, - .dstSubpass = VK_SUBPASS_EXTERNAL, - .srcStageMask = ATTACHMENT_STAGES, - .dstStageMask = OUTSIDE_STAGES, - .srcAccessMask = ATTACHMENT_WRITE_ACCESS, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT | - VK_ACCESS_TRANSFER_WRITE_BIT, - .dependencyFlags = 0, - }, + .dependencyFlags = VK_DEPENDENCY_BY_REGION_BIT }; if (device->IsKhrCreateRenderPass2Supported()) { @@ -332,21 +299,18 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { .preserveAttachmentCount = 0, .pPreserveAttachments = nullptr, }; - boost::container::static_vector dependencies2; - for (const VkSubpassDependency& dependency : dependencies) { - dependencies2.push_back(VkSubpassDependency2{ - .sType = VK_STRUCTURE_TYPE_SUBPASS_DEPENDENCY_2, - .pNext = nullptr, - .srcSubpass = dependency.srcSubpass, - .dstSubpass = dependency.dstSubpass, - .srcStageMask = dependency.srcStageMask, - .dstStageMask = dependency.dstStageMask, - .srcAccessMask = dependency.srcAccessMask, - .dstAccessMask = dependency.dstAccessMask, - .dependencyFlags = dependency.dependencyFlags, - .viewOffset = 0, - }); - } + const VkSubpassDependency2 dependency2{ + .sType = VK_STRUCTURE_TYPE_SUBPASS_DEPENDENCY_2, + .pNext = nullptr, + .srcSubpass = dependency.srcSubpass, + .dstSubpass = dependency.dstSubpass, + .srcStageMask = dependency.srcStageMask, + .dstStageMask = dependency.dstStageMask, + .srcAccessMask = dependency.srcAccessMask, + .dstAccessMask = dependency.dstAccessMask, + .dependencyFlags = dependency.dependencyFlags, + .viewOffset = 0, + }; pair->second = device->GetLogical().CreateRenderPass2({ .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2, .pNext = nullptr, @@ -355,8 +319,8 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { .pAttachments = descriptions2.empty() ? nullptr : descriptions2.data(), .subpassCount = 1, .pSubpasses = &subpass2, - .dependencyCount = static_cast(dependencies2.size()), - .pDependencies = dependencies2.data(), + .dependencyCount = 1, + .pDependencies = &dependency2, .correlatedViewMaskCount = 0, .pCorrelatedViewMasks = nullptr, }); @@ -371,8 +335,8 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { .pAttachments = descriptions.empty() ? nullptr : descriptions.data(), .subpassCount = 1, .pSubpasses = &subpass, - .dependencyCount = static_cast(dependencies.size()), - .pDependencies = dependencies.data(), + .dependencyCount = 1, + .pDependencies = &dependency, }); return *pair->second; } diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index f3142215ff..4c2617144e 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -205,9 +205,6 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { } [[nodiscard]] std::array ShaderStorageViewFormats(PixelFormat format) { - if (VideoCore::Surface::GetFormatType(format) != SurfaceType::ColorTexture) { - return {}; - } if (DefaultBlockWidth(format) != 1 || DefaultBlockHeight(format) != 1) { return {}; } From 0dd78bbd07828f2d1c67b58eca3f3204352a0e9d Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sat, 19 Sep 2026 06:12:59 -0400 Subject: [PATCH 36/69] Re-add unswizzle 3D pass --- .../renderer_vulkan/maxwell_to_vk.cpp | 16 +--- .../renderer_vulkan/vk_compute_pass.cpp | 73 +++++++++++++++++++ .../renderer_vulkan/vk_compute_pass.h | 17 +++++ .../renderer_vulkan/vk_texture_cache.cpp | 24 ++++-- .../renderer_vulkan/vk_texture_cache.h | 1 + src/video_core/surface.cpp | 4 - src/video_core/surface.h | 4 - .../texture_cache/format_lookup_table.cpp | 9 --- .../vulkan_common/vulkan_device.cpp | 4 - 9 files changed, 111 insertions(+), 41 deletions(-) diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp index dae9fe0269..12c311ad35 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp @@ -224,10 +224,6 @@ FormatInfo SurfaceFormat(const Device& device, FormatType format_type, bool with SURFACE_FORMAT_ELEM(VK_FORMAT_ETC2_R8G8B8_SRGB_BLOCK, 0, ETC2_RGB_SRGB) \ SURFACE_FORMAT_ELEM(VK_FORMAT_ETC2_R8G8B8A8_SRGB_BLOCK, 0, ETC2_RGBA_SRGB) \ SURFACE_FORMAT_ELEM(VK_FORMAT_ETC2_R8G8B8A1_SRGB_BLOCK, 0, ETC2_RGB_PTA_SRGB) \ - SURFACE_FORMAT_ELEM(VK_FORMAT_EAC_R11_UNORM_BLOCK, 0, EAC_R11_UNORM) \ - SURFACE_FORMAT_ELEM(VK_FORMAT_EAC_R11_SNORM_BLOCK, 0, EAC_R11_SNORM) \ - SURFACE_FORMAT_ELEM(VK_FORMAT_EAC_R11G11_UNORM_BLOCK, 0, EAC_R11G11_UNORM) \ - SURFACE_FORMAT_ELEM(VK_FORMAT_EAC_R11G11_SNORM_BLOCK, 0, EAC_R11G11_SNORM) \ /* Depth formats */ \ SURFACE_FORMAT_ELEM(VK_FORMAT_D32_SFLOAT, usage_attachable, D32_FLOAT) \ SURFACE_FORMAT_ELEM(VK_FORMAT_D16_UNORM, usage_attachable, D16_UNORM) \ @@ -273,17 +269,7 @@ FormatInfo SurfaceFormat(const Device& device, FormatType format_type, bool with } } else if (!device.IsOptimalEtc2Supported() && VideoCore::Surface::IsPixelFormatETC2(pixel_format)) { // Transcode on hardware that doesn't support ETC2 natively - if (pixel_format == PixelFormat::EAC_R11_SNORM) { - tuple.format = VK_FORMAT_R8_SNORM; - } else if (pixel_format == PixelFormat::EAC_R11_UNORM) { - tuple.format = VK_FORMAT_R8_UNORM; - } else if (pixel_format == PixelFormat::EAC_R11G11_SNORM) { - tuple.format = VK_FORMAT_R8G8_SNORM; - } else if (pixel_format == PixelFormat::EAC_R11G11_UNORM) { - tuple.format = VK_FORMAT_R8G8_UNORM; - } else { - tuple.format = is_srgb ? VK_FORMAT_A8B8G8R8_SRGB_PACK32 : VK_FORMAT_A8B8G8R8_UNORM_PACK32; - } + tuple.format = is_srgb ? VK_FORMAT_A8B8G8R8_SRGB_PACK32 : VK_FORMAT_A8B8G8R8_UNORM_PACK32; } bool const attachable = (tuple.usage & usage_attachable) != 0; bool const storage = (tuple.usage & usage_storage) != 0; diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index b542ccceb5..d8137d6fab 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -18,6 +18,7 @@ #include "common/vector_math.h" #include "video_core/host_shaders/astc_decoder_comp_spv.h" #include "video_core/host_shaders/block_linear_unswizzle_2d_comp_spv.h" +#include "video_core/host_shaders/block_linear_unswizzle_3d_comp_spv.h" #include "video_core/host_shaders/queries_prefix_scan_sum_comp_spv.h" #include "video_core/host_shaders/queries_prefix_scan_sum_nosubgroups_comp_spv.h" #include "video_core/host_shaders/resolve_conditional_render_comp_spv.h" @@ -186,6 +187,19 @@ struct AstcPushConstants { u32 block_height_mask; }; +struct BlockLinear3DImagePushConstants { + alignas(16) std::array origin; + alignas(16) std::array destination; + u32 bytes_per_block_log2; + u32 slice_size; + u32 block_size; + u32 x_shift; + u32 block_height; + u32 block_height_mask; + u32 block_depth; + u32 block_depth_mask; +}; + struct QueriesPrefixScanPushConstants { u32 min_accumulation_base; u32 max_accumulation_base; @@ -743,4 +757,63 @@ void BlockLinearUnswizzleImage2DPass::Unswizzle( RecordUnswizzleEndBarrier(scheduler, vk_image, aspect_mask); } +BlockLinearUnswizzleImage3DPass::BlockLinearUnswizzleImage3DPass( + const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_) + : ComputePass(device_, scheduler_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, + ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, + COMPUTE_PUSH_CONSTANT_RANGE, + BLOCK_LINEAR_UNSWIZZLE_3D_COMP_SPV), + scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, + compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} + +BlockLinearUnswizzleImage3DPass::~BlockLinearUnswizzleImage3DPass() = default; + +void BlockLinearUnswizzleImage3DPass::Unswizzle( + Image& image, const StagingBufferRef& map, + std::span swizzles) { + using namespace VideoCommon::Accelerated; + scheduler.RequestOutsideRenderPassOperationContext(); + const VkImageAspectFlags aspect_mask = image.AspectMask(); + const VkImage vk_image = image.Handle(); + RecordUnswizzleBeginBarrier(scheduler, *pipeline, vk_image, aspect_mask, + image.ExchangeInitialization()); + for (const VideoCommon::SwizzleParameters& swizzle : swizzles) { + const size_t input_offset = swizzle.buffer_offset + map.offset; + const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 16U); + const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U); + const u32 num_dispatches_z = Common::DivCeil(swizzle.num_tiles.depth, 2U); + + compute_pass_descriptor_queue.Acquire(scheduler, 2); + compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset, + image.guest_size_bytes - swizzle.buffer_offset); + compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level)); + const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; + + const auto p = MakeBlockLinearSwizzle3DParams(swizzle, image.info); + const BlockLinear3DImagePushConstants params{ + .origin = p.origin, + .destination = p.destination, + .bytes_per_block_log2 = p.bytes_per_block_log2, + .slice_size = p.slice_size, + .block_size = p.block_size, + .x_shift = p.x_shift, + .block_height = p.block_height, + .block_height_mask = p.block_height_mask, + .block_depth = p.block_depth, + .block_depth_mask = p.block_depth_mask, + }; + scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, params, + descriptor_data](vk::CommandBuffer cmdbuf) { + const VkDescriptorSet set = descriptor_allocator.Commit(); + device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); + cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params); + cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_dispatches_z); + }); + } + RecordUnswizzleEndBarrier(scheduler, vk_image, aspect_mask); +} + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h index c0bc1e6e88..d791554d32 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.h +++ b/src/video_core/renderer_vulkan/vk_compute_pass.h @@ -153,6 +153,23 @@ private: MemoryAllocator& memory_allocator; }; +class BlockLinearUnswizzleImage3DPass final : public ComputePass { +public: + explicit BlockLinearUnswizzleImage3DPass( + const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_); + ~BlockLinearUnswizzleImage3DPass(); + + void Unswizzle(Image& image, const StagingBufferRef& map, + std::span swizzles); + +private: + Scheduler& scheduler; + StagingBufferPool& staging_buffer_pool; + ComputePassDescriptorQueue& compute_pass_descriptor_queue; +}; + class BlockLinearUnswizzleImage2DPass final : public ComputePass { public: explicit BlockLinearUnswizzleImage2DPass( diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 4c2617144e..fb64e89574 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -195,7 +195,8 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { } [[nodiscard]] bool WillUseAcceleratedUnswizzle(const Device& device, const ImageInfo& info) { - if (info.type != ImageType::e2D || info.num_samples > 1) { + const bool supported_type = info.type == ImageType::e2D || info.type == ImageType::e3D; + if (!supported_type || info.num_samples > 1) { return false; } if (!device.IsKhrImageFormatListSupported()) { @@ -313,7 +314,12 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { } [[nodiscard]] vk::ImageView MakeStorageView(const vk::Device& device, u32 level, VkImage image, - VkFormat format) { + VkFormat format, + VkImageViewType view_type) { + u32 layer_count = VK_REMAINING_ARRAY_LAYERS; + if (view_type == VK_IMAGE_VIEW_TYPE_3D) { + layer_count = 1; + } static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO, .pNext = nullptr, @@ -324,7 +330,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { .pNext = &storage_image_view_usage_create_info, .flags = 0, .image = image, - .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY, + .viewType = view_type, .format = format, .components{ .r = VK_COMPONENT_SWIZZLE_IDENTITY, @@ -337,7 +343,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { .baseMipLevel = level, .levelCount = 1, .baseArrayLayer = 0, - .layerCount = VK_REMAINING_ARRAY_LAYERS, + .layerCount = layer_count, }, }); } @@ -1014,6 +1020,8 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched device.IsStorageBuffer16BitAccessSupported()) { bl_unswizzle_2d_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, compute_pass_descriptor_queue); + bl_unswizzle_3d_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, + compute_pass_descriptor_queue); } if (!device.IsKhrImageFormatListSupported()) { return; @@ -2354,13 +2362,17 @@ VkImageView Image::StorageImageView(s32 level) noexcept { if (!view) { auto format_info = MaxwellToVK::SurfaceFormat(runtime->device, FormatType::Optimal, true, info.format); + VkImageViewType view_type = VK_IMAGE_VIEW_TYPE_2D_ARRAY; if (WillUseAcceleratedAstcDecode(runtime->device, info)) { format_info.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32; } else if (True(flags & VideoCommon::ImageFlagBits::AcceleratedUpload)) { format_info.format = UnswizzleStorageFormat(BytesPerBlock(info.format)); + if (info.type == ImageType::e3D) { + view_type = VK_IMAGE_VIEW_TYPE_3D; + } } view = MakeStorageView(runtime->device.GetLogical(), level, *original_image, - format_info.format); + format_info.format, view_type); } return *view; } @@ -3166,6 +3178,8 @@ void TextureCacheRuntime::AccelerateImageUpload( } if (IsPixelFormatASTC(image.info.format)) { astc_decoder_pass->Assemble(image, map, swizzles); + } else if (image.info.type == ImageType::e3D) { + bl_unswizzle_3d_pass->Unswizzle(image, map, swizzles); } else { bl_unswizzle_2d_pass->Unswizzle(image, map, swizzles); } diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 8c86bdd5d2..5023a94b8b 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -157,6 +157,7 @@ public: RenderPassCache& render_pass_cache; std::optional astc_decoder_pass; std::optional bl_unswizzle_2d_pass; + std::optional bl_unswizzle_3d_pass; const Settings::ResolutionScalingInfo& resolution; std::array, VideoCore::Surface::MaxPixelFormat> view_formats; diff --git a/src/video_core/surface.cpp b/src/video_core/surface.cpp index fa9554db59..d1b9bf4151 100644 --- a/src/video_core/surface.cpp +++ b/src/video_core/surface.cpp @@ -343,10 +343,6 @@ bool IsPixelFormatETC2(PixelFormat format) { case PixelFormat::ETC2_RGB_SRGB: case PixelFormat::ETC2_RGBA_SRGB: case PixelFormat::ETC2_RGB_PTA_SRGB: - case PixelFormat::EAC_R11_UNORM: - case PixelFormat::EAC_R11_SNORM: - case PixelFormat::EAC_R11G11_UNORM: - case PixelFormat::EAC_R11G11_SNORM: return true; default: return false; diff --git a/src/video_core/surface.h b/src/video_core/surface.h index 8e1973d3b9..5611dbdffa 100644 --- a/src/video_core/surface.h +++ b/src/video_core/surface.h @@ -119,10 +119,6 @@ namespace VideoCore::Surface { PIXEL_FORMAT_ELEM(ETC2_RGB_SRGB, 4, 4, 64) \ PIXEL_FORMAT_ELEM(ETC2_RGBA_SRGB, 4, 4, 128) \ PIXEL_FORMAT_ELEM(ETC2_RGB_PTA_SRGB, 4, 4, 64) \ - PIXEL_FORMAT_ELEM(EAC_R11_UNORM, 4, 4, 64) \ - PIXEL_FORMAT_ELEM(EAC_R11_SNORM, 4, 4, 64) \ - PIXEL_FORMAT_ELEM(EAC_R11G11_UNORM, 4, 4, 128) \ - PIXEL_FORMAT_ELEM(EAC_R11G11_SNORM, 4, 4, 128) \ /* Depth formats */ \ PIXEL_FORMAT_ELEM(D32_FLOAT, 1, 1, 32) \ PIXEL_FORMAT_ELEM(D16_UNORM, 1, 1, 16) \ diff --git a/src/video_core/texture_cache/format_lookup_table.cpp b/src/video_core/texture_cache/format_lookup_table.cpp index d6e019cddf..ea68a1c063 100644 --- a/src/video_core/texture_cache/format_lookup_table.cpp +++ b/src/video_core/texture_cache/format_lookup_table.cpp @@ -204,15 +204,6 @@ PixelFormat PixelFormatFromTextureInfo(TextureFormat format, ComponentType red, return PixelFormat::ETC2_RGB_PTA_SRGB; case Hash(TextureFormat::ETC2_RGBA, UNORM, SRGB): return PixelFormat::ETC2_RGBA_SRGB; - /* EAC */ - case Hash(TextureFormat::EAC, UNORM): - return PixelFormat::EAC_R11_UNORM; - case Hash(TextureFormat::EAC, SNORM): - return PixelFormat::EAC_R11_SNORM; - case Hash(TextureFormat::EACX2, UNORM): - return PixelFormat::EAC_R11G11_UNORM; - case Hash(TextureFormat::EACX2, SNORM): - return PixelFormat::EAC_R11G11_SNORM; /* ASTC */ case Hash(TextureFormat::ASTC_2D_4X4, UNORM, LINEAR): return PixelFormat::ASTC_2D_4X4_UNORM; diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 27379be506..1e15f05156 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -301,10 +301,6 @@ VkFormatFeatureFlags GetFormatFeatures(VkFormatProperties properties, FormatType VK_FORMAT_ETC2_R8G8B8_SRGB_BLOCK, VK_FORMAT_ETC2_R8G8B8A8_SRGB_BLOCK, VK_FORMAT_ETC2_R8G8B8A1_SRGB_BLOCK, - VK_FORMAT_EAC_R11_UNORM_BLOCK, - VK_FORMAT_EAC_R11_SNORM_BLOCK, - VK_FORMAT_EAC_R11G11_UNORM_BLOCK, - VK_FORMAT_EAC_R11G11_SNORM_BLOCK, }; ::Common::unordered_map format_properties; for (const auto format : formats) { From 3a22f908a3275039113bfb3ffe78f0d40688c142 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sat, 19 Sep 2026 06:44:30 -0400 Subject: [PATCH 37/69] Re-add pitch 2D pass --- .../renderer_vulkan/vk_compute_pass.cpp | 61 +++++++++++++++++++ .../renderer_vulkan/vk_compute_pass.h | 17 ++++++ .../renderer_vulkan/vk_texture_cache.cpp | 11 +++- .../renderer_vulkan/vk_texture_cache.h | 1 + 4 files changed, 88 insertions(+), 2 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index d8137d6fab..4dd65f0269 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -12,6 +12,8 @@ #include "video_core/renderer_vulkan/vk_texture_cache.h" +#include + #include "common/assert.h" #include "common/common_types.h" #include "common/div_ceil.h" @@ -19,6 +21,7 @@ #include "video_core/host_shaders/astc_decoder_comp_spv.h" #include "video_core/host_shaders/block_linear_unswizzle_2d_comp_spv.h" #include "video_core/host_shaders/block_linear_unswizzle_3d_comp_spv.h" +#include "video_core/host_shaders/pitch_unswizzle_comp_spv.h" #include "video_core/host_shaders/queries_prefix_scan_sum_comp_spv.h" #include "video_core/host_shaders/queries_prefix_scan_sum_nosubgroups_comp_spv.h" #include "video_core/host_shaders/resolve_conditional_render_comp_spv.h" @@ -200,6 +203,13 @@ struct BlockLinear3DImagePushConstants { u32 block_depth_mask; }; +struct PitchUnswizzlePushConstants { + std::array origin; + std::array destination; + u32 bytes_per_block_log2; + u32 pitch; +}; + struct QueriesPrefixScanPushConstants { u32 min_accumulation_base; u32 max_accumulation_base; @@ -816,4 +826,55 @@ void BlockLinearUnswizzleImage3DPass::Unswizzle( RecordUnswizzleEndBarrier(scheduler, vk_image, aspect_mask); } +PitchUnswizzlePass::PitchUnswizzlePass(const Device& device_, Scheduler& scheduler_, + DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_) + : ComputePass(device_, scheduler_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, + ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, + COMPUTE_PUSH_CONSTANT_RANGE, + PITCH_UNSWIZZLE_COMP_SPV), + scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, + compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {} + +PitchUnswizzlePass::~PitchUnswizzlePass() = default; + +void PitchUnswizzlePass::Unswizzle(Image& image, const StagingBufferRef& map, + std::span swizzles) { + const u32 bytes_per_block = VideoCore::Surface::BytesPerBlock(image.info.format); + ASSERT(std::has_single_bit(bytes_per_block)); + scheduler.RequestOutsideRenderPassOperationContext(); + const VkImageAspectFlags aspect_mask = image.AspectMask(); + const VkImage vk_image = image.Handle(); + RecordUnswizzleBeginBarrier(scheduler, *pipeline, vk_image, aspect_mask, + image.ExchangeInitialization()); + for (const VideoCommon::SwizzleParameters& swizzle : swizzles) { + const size_t input_offset = swizzle.buffer_offset + map.offset; + const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U); + const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U); + + compute_pass_descriptor_queue.Acquire(scheduler, 2); + compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset, + image.guest_size_bytes - swizzle.buffer_offset); + compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level)); + const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; + + const PitchUnswizzlePushConstants params{ + .origin = {0, 0}, + .destination = {0, 0}, + .bytes_per_block_log2 = static_cast(std::countr_zero(bytes_per_block)), + .pitch = image.info.pitch, + }; + scheduler.Record([this, num_dispatches_x, num_dispatches_y, params, + descriptor_data](vk::CommandBuffer cmdbuf) { + const VkDescriptorSet set = descriptor_allocator.Commit(); + device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); + cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params); + cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, 1); + }); + } + RecordUnswizzleEndBarrier(scheduler, vk_image, aspect_mask); +} + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h index d791554d32..f871648647 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.h +++ b/src/video_core/renderer_vulkan/vk_compute_pass.h @@ -153,6 +153,23 @@ private: MemoryAllocator& memory_allocator; }; +class PitchUnswizzlePass final : public ComputePass { +public: + explicit PitchUnswizzlePass(const Device& device_, Scheduler& scheduler_, + DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_); + ~PitchUnswizzlePass(); + + void Unswizzle(Image& image, const StagingBufferRef& map, + std::span swizzles); + +private: + Scheduler& scheduler; + StagingBufferPool& staging_buffer_pool; + ComputePassDescriptorQueue& compute_pass_descriptor_queue; +}; + class BlockLinearUnswizzleImage3DPass final : public ComputePass { public: explicit BlockLinearUnswizzleImage3DPass( diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index fb64e89574..749a49c898 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -195,7 +195,8 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { } [[nodiscard]] bool WillUseAcceleratedUnswizzle(const Device& device, const ImageInfo& info) { - const bool supported_type = info.type == ImageType::e2D || info.type == ImageType::e3D; + const bool supported_type = info.type == ImageType::e2D || info.type == ImageType::e3D || + info.type == ImageType::Linear; if (!supported_type || info.num_samples > 1) { return false; } @@ -317,7 +318,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { VkFormat format, VkImageViewType view_type) { u32 layer_count = VK_REMAINING_ARRAY_LAYERS; - if (view_type == VK_IMAGE_VIEW_TYPE_3D) { + if (view_type != VK_IMAGE_VIEW_TYPE_2D_ARRAY) { layer_count = 1; } static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{ @@ -1022,6 +1023,8 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched compute_pass_descriptor_queue); bl_unswizzle_3d_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, compute_pass_descriptor_queue); + pitch_unswizzle_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, + compute_pass_descriptor_queue); } if (!device.IsKhrImageFormatListSupported()) { return; @@ -2369,6 +2372,8 @@ VkImageView Image::StorageImageView(s32 level) noexcept { format_info.format = UnswizzleStorageFormat(BytesPerBlock(info.format)); if (info.type == ImageType::e3D) { view_type = VK_IMAGE_VIEW_TYPE_3D; + } else if (info.type == ImageType::Linear) { + view_type = VK_IMAGE_VIEW_TYPE_2D; } } view = MakeStorageView(runtime->device.GetLogical(), level, *original_image, @@ -3180,6 +3185,8 @@ void TextureCacheRuntime::AccelerateImageUpload( astc_decoder_pass->Assemble(image, map, swizzles); } else if (image.info.type == ImageType::e3D) { bl_unswizzle_3d_pass->Unswizzle(image, map, swizzles); + } else if (image.info.type == ImageType::Linear) { + pitch_unswizzle_pass->Unswizzle(image, map, swizzles); } else { bl_unswizzle_2d_pass->Unswizzle(image, map, swizzles); } diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 5023a94b8b..2e6b547e72 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -158,6 +158,7 @@ public: std::optional astc_decoder_pass; std::optional bl_unswizzle_2d_pass; std::optional bl_unswizzle_3d_pass; + std::optional pitch_unswizzle_pass; const Settings::ResolutionScalingInfo& resolution; std::array, VideoCore::Surface::MaxPixelFormat> view_formats; From 8a8c7f4c06ee78f579312d529cbfec8e50543dab Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sat, 19 Sep 2026 13:46:49 -0400 Subject: [PATCH 38/69] addresing MHR regression --- src/video_core/buffer_cache/buffer_cache.h | 5 ----- 1 file changed, 5 deletions(-) diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 6066609e45..eb214af50e 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -1042,11 +1042,6 @@ void BufferCache

::ResolveMultiRangeStorage(Binding& binding, return; } const VirtualSegments segments = *found; - for (const VirtualSegment& segment : segments) { - if (memory_tracker.IsRegionGpuModified(segment.device_addr, segment.size)) { - return; - } - } const u32 first = static_cast(pool.size()); const bool prefer_sparse = runtime.PrefersSparseSources(); for (const VirtualSegment& segment : segments) { From 09371eab6052b916676a5a2a9cc3d7320acca7bd Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sat, 19 Sep 2026 15:33:10 -0400 Subject: [PATCH 39/69] Remove dead code om VMA --- .../renderer_vulkan/vk_buffer_cache.cpp | 1 - .../vulkan_common/vulkan_memory_allocator.cpp | 222 +++--------------- .../vulkan_common/vulkan_memory_allocator.h | 70 ------ 3 files changed, 34 insertions(+), 259 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index f397f9faf4..da2d36e41e 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -263,7 +263,6 @@ protected: StagingBufferPool& staging_pool; vk::Buffer buffer{}; - MemoryCommit memory_commit{}; VkIndexType index_type{}; u32 num_indices = 0; }; diff --git a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp index 4d2d5b1317..31a909ef2f 100644 --- a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp +++ b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp @@ -4,20 +4,8 @@ // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later -#include -#include -#include -#include -#include -#include -#include - -#include "common/alignment.h" -#include "common/assert.h" #include "common/common_types.h" #include "common/literals.h" -#include "common/logging.h" -#include #include "video_core/vulkan_common/vma.h" #include "video_core/vulkan_common/vulkan_device.h" #include "video_core/vulkan_common/vulkan_memory_allocator.h" @@ -53,6 +41,26 @@ namespace { return {}; } + template + [[nodiscard]] VkResult AllocateRelaxing(VmaAllocationCreateInfo& ci, F&& allocate) { + VkResult result = allocate(ci); + if (result == VK_SUCCESS) { + return result; + } + if ((ci.flags & VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT) != 0) { + ci.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT; + result = allocate(ci); + if (result == VK_SUCCESS) { + return result; + } + } + if ((ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT) != 0) { + ci.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; + result = allocate(ci); + } + return result; + } + [[nodiscard]] VmaMemoryUsage MemoryUsageVma(MemoryUsage usage) { switch (usage) { case MemoryUsage::DeviceLocal: @@ -66,106 +74,9 @@ namespace { } } // namespace -//MemoryCommit is now VMA-backed -MemoryCommit::MemoryCommit(VmaAllocator alloc, VmaAllocation a, - const VmaAllocationInfo &info) noexcept - : allocator{alloc}, allocation{a}, memory{info.deviceMemory}, - offset{info.offset}, size{info.size}, mapped_ptr{info.pMappedData} { - // Log GPU memory allocation - if (GPU::Logging::IsActive() && - Settings::values.gpu_log_memory_tracking.GetValue()) { - GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation( - reinterpret_cast(memory), - static_cast(size), - 0 // Memory property flags (not easily available from VMA) - ); - } -} - -MemoryCommit::~MemoryCommit() { Release(); } - -MemoryCommit::MemoryCommit(MemoryCommit &&rhs) noexcept - : allocator{std::exchange(rhs.allocator, nullptr)}, - allocation{std::exchange(rhs.allocation, nullptr)}, - memory{std::exchange(rhs.memory, VK_NULL_HANDLE)}, - offset{std::exchange(rhs.offset, 0)}, - size{std::exchange(rhs.size, 0)}, - mapped_ptr{std::exchange(rhs.mapped_ptr, nullptr)} {} - -MemoryCommit &MemoryCommit::operator=(MemoryCommit &&rhs) noexcept { - if (this != &rhs) { - Release(); - allocator = std::exchange(rhs.allocator, nullptr); - allocation = std::exchange(rhs.allocation, nullptr); - memory = std::exchange(rhs.memory, VK_NULL_HANDLE); - offset = std::exchange(rhs.offset, 0); - size = std::exchange(rhs.size, 0); - mapped_ptr = std::exchange(rhs.mapped_ptr, nullptr); - } - return *this; -} - -std::span MemoryCommit::Map() -{ - if (!allocation) return {}; - if (!mapped_ptr) { - if (vmaMapMemory(allocator, allocation, &mapped_ptr) != VK_SUCCESS) return {}; - } - const size_t n = static_cast(std::min(size, - (std::numeric_limits::max)())); - return std::span{static_cast(mapped_ptr), n}; -} - -std::span MemoryCommit::Map() const -{ - if (!allocation) return {}; - if (!mapped_ptr) { - void *p = nullptr; - if (vmaMapMemory(allocator, allocation, &p) != VK_SUCCESS) return {}; - const_cast(this)->mapped_ptr = p; - } - const size_t n = static_cast(std::min(size, - (std::numeric_limits::max)())); - return std::span{static_cast(mapped_ptr), n}; -} - -void MemoryCommit::Unmap() -{ - if (allocation && mapped_ptr) { - vmaUnmapMemory(allocator, allocation); - mapped_ptr = nullptr; - } -} - -void MemoryCommit::Release() { - if (allocation && allocator) { - // Log GPU memory deallocation - if (GPU::Logging::IsActive() && - Settings::values.gpu_log_memory_tracking.GetValue() && - memory != VK_NULL_HANDLE) { - GPU::Logging::GPULogger::GetInstance().LogMemoryDeallocation( - reinterpret_cast(memory) - ); - } - - if (mapped_ptr) { - vmaUnmapMemory(allocator, allocation); - mapped_ptr = nullptr; - } - vmaFreeMemory(allocator, allocation); - } - allocation = nullptr; - allocator = nullptr; - memory = VK_NULL_HANDLE; - offset = 0; - size = 0; -} - MemoryAllocator::MemoryAllocator(const Device &device_) : device{device_}, allocator{device.GetAllocator()}, - properties{device_.GetPhysical().GetMemoryProperties().memoryProperties}, - buffer_image_granularity{ - device_.GetPhysical().GetProperties().limits.bufferImageGranularity} { + properties{device_.GetPhysical().GetMemoryProperties().memoryProperties} { // Preserve the previous "RenderDoc small heap" trimming behavior that we had in original vma minus the heap bug if (device.HasDebuggingToolAttached()) @@ -187,7 +98,7 @@ MemoryAllocator::~MemoryAllocator() = default; vk::Image MemoryAllocator::CreateImage(const VkImageCreateInfo &ci) const { - const VmaAllocationCreateInfo alloc_ci = { + VmaAllocationCreateInfo alloc_ci = { .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT, .usage = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, .requiredFlags = 0, @@ -201,7 +112,9 @@ vk::Image MemoryAllocator::CreateImage(const VkImageCreateInfo &ci) const VkImage handle{}; VmaAllocation allocation{}; VmaAllocationInfo alloc_info{}; - vk::Check(vmaCreateImage(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info)); + vk::Check(AllocateRelaxing(alloc_ci, [&](const VmaAllocationCreateInfo& retry) { + return vmaCreateImage(allocator, &ci, &retry, &handle, &allocation, &alloc_info); + })); // Log GPU memory allocation for images if (GPU::Logging::IsActive() && @@ -222,7 +135,7 @@ vk::Buffer MemoryAllocator::CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsa auto const anv_flags = (usage == MemoryUsage::Stream && device.GetDriverID() == VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA) ? VK_MEMORY_PROPERTY_HOST_CACHED_BIT : 0; - const VmaAllocationCreateInfo alloc_ci = { + VmaAllocationCreateInfo alloc_ci = { .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage), .usage = MemoryUsageVma(usage), .requiredFlags = 0, @@ -238,7 +151,9 @@ vk::Buffer MemoryAllocator::CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsa VmaAllocation allocation{}; VkMemoryPropertyFlags property_flags{}; - vk::Check(vmaCreateBuffer(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info)); + vk::Check(AllocateRelaxing(alloc_ci, [&](const VmaAllocationCreateInfo& retry) { + return vmaCreateBuffer(allocator, &ci, &retry, &handle, &allocation, &alloc_info); + })); vmaGetAllocationMemoryProperties(allocator, allocation, &property_flags); // Log GPU memory allocation for buffers @@ -278,7 +193,7 @@ vk::Buffer MemoryAllocator::CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsa if (usage == MemoryUsage::Stream) { memory_type_bits = 0u; } - const VmaAllocationCreateInfo alloc_ci = { + VmaAllocationCreateInfo alloc_ci = { .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage), .usage = MemoryUsageVma(usage), .requiredFlags = 0, @@ -294,8 +209,10 @@ vk::Buffer MemoryAllocator::CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsa VmaAllocation allocation{}; VkMemoryPropertyFlags property_flags{}; - vk::Check(vmaCreateBufferWithAlignment(allocator, &ci, &alloc_ci, min_alignment, &handle, - &allocation, &alloc_info)); + vk::Check(AllocateRelaxing(alloc_ci, [&](const VmaAllocationCreateInfo& retry) { + return vmaCreateBufferWithAlignment(allocator, &ci, &retry, min_alignment, &handle, + &allocation, &alloc_info); + })); vmaGetAllocationMemoryProperties(allocator, allocation, &property_flags); u8 *data = reinterpret_cast(alloc_info.pMappedData); @@ -314,75 +231,4 @@ vk::Buffer MemoryAllocator::CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsa location, device.GetDispatchLoader()); } -MemoryCommit MemoryAllocator::Commit(const VkMemoryRequirements &reqs, MemoryUsage usage) -{ - const auto vma_usage = MemoryUsageVma(usage); - VmaAllocationCreateInfo ci{}; - ci.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage); - ci.usage = vma_usage; - ci.memoryTypeBits = reqs.memoryTypeBits & valid_memory_types; - ci.requiredFlags = 0; - ci.preferredFlags = MemoryUsagePreferredVmaFlags(usage); - - VmaAllocation a{}; - VmaAllocationInfo info{}; - - VkResult res = vmaAllocateMemory(allocator, &reqs, &ci, &a, &info); - - if (res != VK_SUCCESS) { - // Relax 1: drop budget constraint - auto ci2 = ci; - ci2.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT; - res = vmaAllocateMemory(allocator, &reqs, &ci2, &a, &info); - - // Relax 2: if we preferred DEVICE_LOCAL, drop that preference - if (res != VK_SUCCESS && (ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) { - auto ci3 = ci2; - ci3.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; - res = vmaAllocateMemory(allocator, &reqs, &ci3, &a, &info); - } - } - - vk::Check(res); - return MemoryCommit(allocator, a, info); -} - -MemoryCommit MemoryAllocator::Commit(const vk::Buffer &buffer, MemoryUsage usage) { - // Allocate memory appropriate for this buffer automatically - const auto vma_usage = MemoryUsageVma(usage); - - VmaAllocationCreateInfo ci{}; - ci.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage); - ci.usage = vma_usage; - ci.requiredFlags = 0; - ci.preferredFlags = MemoryUsagePreferredVmaFlags(usage); - ci.pool = VK_NULL_HANDLE; - ci.pUserData = nullptr; - ci.priority = 0.0f; - - const VkBuffer raw = *buffer; - - VmaAllocation a{}; - VmaAllocationInfo info{}; - - // Let VMA infer memory requirements from the buffer - VkResult res = vmaAllocateMemoryForBuffer(allocator, raw, &ci, &a, &info); - - if (res != VK_SUCCESS) { - auto ci2 = ci; - ci2.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT; - res = vmaAllocateMemoryForBuffer(allocator, raw, &ci2, &a, &info); - - if (res != VK_SUCCESS && (ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) { - auto ci3 = ci2; - ci3.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; - res = vmaAllocateMemoryForBuffer(allocator, raw, &ci3, &a, &info); - } - } - - vk::Check(res); - vk::Check(vmaBindBufferMemory2(allocator, a, 0, raw, nullptr)); - return MemoryCommit(allocator, a, info); -} - } // namespace Vulkan diff --git a/src/video_core/vulkan_common/vulkan_memory_allocator.h b/src/video_core/vulkan_common/vulkan_memory_allocator.h index a0f9081fc0..79a33f0b7b 100644 --- a/src/video_core/vulkan_common/vulkan_memory_allocator.h +++ b/src/video_core/vulkan_common/vulkan_memory_allocator.h @@ -39,51 +39,6 @@ namespace Vulkan { } } -/// Ownership handle of a memory commitment (real VMA allocation). - class MemoryCommit { - public: - MemoryCommit() noexcept = default; - - MemoryCommit(VmaAllocator allocator, VmaAllocation allocation, - const VmaAllocationInfo &info) noexcept; - - ~MemoryCommit(); - - MemoryCommit(const MemoryCommit &) = delete; - - MemoryCommit &operator=(const MemoryCommit &) = delete; - - MemoryCommit(MemoryCommit &&) noexcept; - - MemoryCommit &operator=(MemoryCommit &&) noexcept; - - [[nodiscard]] std::span Map(); - - [[nodiscard]] std::span Map() const; - - void Unmap(); - - explicit operator bool() const noexcept { return allocation != nullptr; } - - VkDeviceMemory Memory() const noexcept { return memory; } - - VkDeviceSize Offset() const noexcept { return offset; } - - VkDeviceSize Size() const noexcept { return size; } - - VmaAllocation Allocation() const noexcept { return allocation; } - - private: - void Release(); - - VmaAllocator allocator{}; ///< VMA allocator - VmaAllocation allocation{}; ///< VMA allocation handle - VkDeviceMemory memory{}; ///< Underlying VkDeviceMemory chosen by VMA - VkDeviceSize offset{}; ///< Offset of this allocation inside VkDeviceMemory - VkDeviceSize size{}; ///< Size of the allocation - void *mapped_ptr{}; ///< Optional persistent mapped pointer - }; - /// Memory allocator container. /// Allocates and releases memory allocations on demand. class MemoryAllocator { @@ -110,35 +65,10 @@ namespace Vulkan { vk::Buffer CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsage usage, VkDeviceSize min_alignment) const; - /** - * Commits a memory with the specified requirements. - * - * @param requirements Requirements returned from a Vulkan call. - * @param usage Indicates how the memory will be used. - * - * @returns A memory commit. - */ - MemoryCommit Commit(const VkMemoryRequirements &requirements, MemoryUsage usage); - - /// Commits memory required by the buffer and binds it (for buffers created outside VMA). - MemoryCommit Commit(const vk::Buffer &buffer, MemoryUsage usage); - private: - static bool IsAutoUsage(VmaMemoryUsage u) noexcept { - switch (u) { - case VMA_MEMORY_USAGE_AUTO: - case VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE: - case VMA_MEMORY_USAGE_AUTO_PREFER_HOST: - return true; - default: - return false; - } - } - const Device &device; ///< Device handle. VmaAllocator allocator; ///< VMA allocator. const VkPhysicalDeviceMemoryProperties properties; ///< Physical device memory properties. - VkDeviceSize buffer_image_granularity; ///< Adjacent buffer/image granularity u32 valid_memory_types{~0u}; }; From 1fbb1a0499eb0e6547404eb2c9197e59113b8792 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sat, 19 Sep 2026 15:47:59 -0400 Subject: [PATCH 40/69] Revert QCOM spoof --- src/video_core/vulkan_common/vulkan_wrapper.cpp | 17 ++--------------- 1 file changed, 2 insertions(+), 15 deletions(-) diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp index d5012ddd83..e4c8f518b9 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.cpp +++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp @@ -468,20 +468,8 @@ Instance Instance::Create(u32 version, Span layers, Span gloom + yellow line glitch appears -#ifdef __ANDROID__ - const VkApplicationInfo application_info{ - .sType = VK_STRUCTURE_TYPE_APPLICATION_INFO, - .pNext = nullptr, - .pApplicationName = "PUBGMobile", - .applicationVersion = VK_MAKE_VERSION(1, 7, 0), - .pEngineName = "UnrealEngine", - .engineVersion = VK_MAKE_VERSION(4, 23, 0), - .apiVersion = VK_API_VERSION_1_3, - }; -#else + // DO NOT TOUCH, breaks RNDA3!! + // Don't know why, but gloom + yellow line glitch appears const VkApplicationInfo application_info{ .sType = VK_STRUCTURE_TYPE_APPLICATION_INFO, .pNext = nullptr, @@ -491,7 +479,6 @@ Instance Instance::Create(u32 version, Span layers, Span Date: Sat, 19 Sep 2026 16:57:03 -0400 Subject: [PATCH 41/69] Add checks on texel buffer alignments --- .../renderer_vulkan/vk_buffer_cache.cpp | 5 +++++ .../vulkan_common/vulkan_device.cpp | 10 ++++++++++ src/video_core/vulkan_common/vulkan_device.h | 20 +++++++++++++++++++ 3 files changed, 35 insertions(+) diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index da2d36e41e..8c93a92046 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -133,6 +133,11 @@ VkBufferView Buffer::View(u32 offset, u32 size, VideoCore::Surface::PixelFormat offset = 0; size = 0; } + const u32 required = + static_cast(device->TexelBufferAlignment(VideoCore::Surface::BytesPerBlock(format))); + const u32 misalign = offset % required; + offset -= misalign; + size += misalign; const auto it{std::ranges::find_if(views, [offset, size, format](const BufferView& view) { return offset == view.offset && size == view.size && format == view.format; })}; diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 1e15f05156..65194f647a 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -1112,6 +1112,16 @@ bool Device::GetSuitability(bool requires_swapchain) { VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_PROPERTIES; SetNext(next, properties.subgroup_size_control); } + properties.texel_buffer_alignment.storageTexelBufferOffsetAlignmentBytes = + properties.properties.limits.minTexelBufferOffsetAlignment; + properties.texel_buffer_alignment.uniformTexelBufferOffsetAlignmentBytes = + properties.properties.limits.minTexelBufferOffsetAlignment; + if (features.texel_buffer_alignment.texelBufferAlignment == VK_TRUE || + instance_version >= VK_API_VERSION_1_3) { + properties.texel_buffer_alignment.sType = + VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TEXEL_BUFFER_ALIGNMENT_PROPERTIES_EXT; + SetNext(next, properties.texel_buffer_alignment); + } if (extensions.transform_feedback) { properties.transform_feedback.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT; diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 8b93aad373..916a78fd56 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -6,6 +6,7 @@ #pragma once +#include #include #include #include @@ -69,6 +70,7 @@ VK_DEFINE_HANDLE(VmaAllocator) primitive_topology_list_restart) \ FEATURE(EXT, ProvokingVertex, PROVOKING_VERTEX, provoking_vertex) \ FEATURE(EXT, Robustness2, ROBUSTNESS_2, robustness2) \ + FEATURE(EXT, TexelBufferAlignment, TEXEL_BUFFER_ALIGNMENT, texel_buffer_alignment) \ FEATURE(EXT, TransformFeedback, TRANSFORM_FEEDBACK, transform_feedback) \ FEATURE(EXT, VertexInputDynamicState, VERTEX_INPUT_DYNAMIC_STATE, vertex_input_dynamic_state) \ FEATURE(KHR, Maintenance5, MAINTENANCE_5, maintenance5) \ @@ -291,6 +293,23 @@ public: return properties.properties.apiVersion; } + VkDeviceSize TexelBufferAlignment(u32 texel_size) const { + VkDeviceSize texel = texel_size; + if (texel % 3 == 0) { + texel /= 3; + } + const auto& limits = properties.texel_buffer_alignment; + VkDeviceSize storage = limits.storageTexelBufferOffsetAlignmentBytes; + if (limits.storageTexelBufferOffsetSingleTexelAlignment != VK_FALSE) { + storage = (std::min)(storage, texel); + } + VkDeviceSize uniform = limits.uniformTexelBufferOffsetAlignmentBytes; + if (limits.uniformTexelBufferOffsetSingleTexelAlignment != VK_FALSE) { + uniform = (std::min)(uniform, texel); + } + return (std::max)((std::max)(storage, uniform), VkDeviceSize{1}); + } + /// Returns the current driver version provided in Vulkan-formatted version numbers. u32 GetDriverVersion() const { return properties.properties.driverVersion; @@ -1161,6 +1180,7 @@ private: VkPhysicalDevicePushDescriptorPropertiesKHR push_descriptor{}; VkPhysicalDeviceDescriptorBufferPropertiesEXT descriptor_buffer{}; VkPhysicalDeviceSubgroupSizeControlProperties subgroup_size_control{}; + VkPhysicalDeviceTexelBufferAlignmentProperties texel_buffer_alignment{}; VkPhysicalDeviceTransformFeedbackPropertiesEXT transform_feedback{}; VkPhysicalDeviceMaintenance5PropertiesKHR maintenance5{}; VkPhysicalDeviceDepthStencilResolveProperties depth_stencil_resolve{}; From d82a9c3ba5e89d3a619c437217495ed26c411b2e Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sat, 19 Sep 2026 18:08:38 -0400 Subject: [PATCH 42/69] Adjust stagging buffer ring --- .../vk_staging_buffer_pool.cpp | 53 +++++++++++++++---- .../renderer_vulkan/vk_staging_buffer_pool.h | 10 +++- .../renderer_vulkan/vk_texture_cache.cpp | 5 +- .../vulkan_common/vulkan_device.cpp | 2 +- .../vulkan_common/vulkan_memory_allocator.cpp | 13 +++-- 5 files changed, 66 insertions(+), 17 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp index b847b66e7c..5f04df2c9e 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp @@ -120,7 +120,14 @@ void StagingBufferPool::FreeDeferred(StagingBufferRef& ref) { } void StagingBufferPool::TickFrame() { - current_delete_level = (current_delete_level + 1) % NUM_LEVELS; + for (size_t step = 0; step < NUM_LEVELS; ++step) { + current_delete_level = (current_delete_level + 1) % NUM_LEVELS; + if (!device_local_cache[current_delete_level].entries.empty() || + !upload_cache[current_delete_level].entries.empty() || + !download_cache[current_delete_level].entries.empty()) { + break; + } + } ReleaseCache(MemoryUsage::DeviceLocal); ReleaseCache(MemoryUsage::Upload); @@ -128,10 +135,12 @@ void StagingBufferPool::TickFrame() { } StagingBufferRef StagingBufferPool::GetStreamBuffer(size_t size) { - if (AreRegionsActive(Region(free_iterator) + 1, - (std::min)(Region(iterator + size) + 1, NUM_SYNCS))) { - // Avoid waiting for the previous usages to be free - return GetStagingBuffer(size, MemoryUsage::Upload); + const size_t wanted_begin = Region(free_iterator) + 1; + const size_t wanted_end = (std::min)(Region(iterator + size) + 1, NUM_SYNCS); + if (AreRegionsActive(wanted_begin, wanted_end)) { + if (auto ref = OverflowStreamBuffer(size, wanted_begin, wanted_end)) { + return *ref; + } } const u64 current_tick = scheduler.CurrentTick(); std::fill(sync_ticks.begin() + Region(used_iterator), sync_ticks.begin() + Region(iterator), @@ -147,8 +156,9 @@ StagingBufferRef StagingBufferPool::GetStreamBuffer(size_t size) { free_iterator = size; if (AreRegionsActive(0, Region(size) + 1)) { - // Avoid waiting for the previous usages to be free - return GetStagingBuffer(size, MemoryUsage::Upload); + if (auto ref = OverflowStreamBuffer(size, 0, Region(size) + 1)) { + return *ref; + } } } const size_t offset = iterator; @@ -164,6 +174,27 @@ StagingBufferRef StagingBufferPool::GetStreamBuffer(size_t size) { }; } +u64 StagingBufferPool::MaxRegionTick(size_t region_begin, size_t region_end) const { + u64 tick = 0; + for (size_t region = region_begin; region < region_end; ++region) { + tick = (std::max)(tick, sync_ticks[region]); + } + return tick; +} + +std::optional StagingBufferPool::OverflowStreamBuffer(size_t size, + size_t region_begin, + size_t region_end) { + if (cache_bytes[static_cast(MemoryUsage::Upload)] < stream_buffer_size) { + return GetStagingBuffer(size, MemoryUsage::Upload); + } + if (auto ref = TryGetReservedBuffer(size, MemoryUsage::Upload, false)) { + return ref; + } + scheduler.Wait(MaxRegionTick(region_begin, region_end)); + return std::nullopt; +} + bool StagingBufferPool::AreRegionsActive(size_t region_begin, size_t region_end) const { const u64 gpu_tick = scheduler.GetMasterSemaphore().KnownGpuTick(); return std::any_of(sync_ticks.begin() + region_begin, sync_ticks.begin() + region_end, @@ -233,6 +264,7 @@ StagingBufferRef StagingBufferPool::CreateStagingBuffer(size_t size, MemoryUsage device.IsBufferDeviceAddressSupported() ? device.GetLogical().GetBufferDeviceAddress(*buffer) : VkDeviceAddress{}; + cache_bytes[static_cast(usage)] += size_t{1} << log2_size; StagingBuffer& entry = GetCache(usage)[log2_size].entries.emplace_back(StagingBuffer{ .buffer = std::move(buffer), .device_address = buffer_address, @@ -261,12 +293,12 @@ StagingBufferPool::StagingBuffersCache& StagingBufferPool::GetCache(MemoryUsage } void StagingBufferPool::ReleaseCache(MemoryUsage usage) { - ReleaseLevel(GetCache(usage), current_delete_level); + ReleaseLevel(usage, current_delete_level); } -void StagingBufferPool::ReleaseLevel(StagingBuffersCache& cache, size_t log2) { +void StagingBufferPool::ReleaseLevel(MemoryUsage usage, size_t log2) { constexpr size_t deletions_per_tick = 16; - auto& staging = cache[log2]; + auto& staging = GetCache(usage)[log2]; auto& entries = staging.entries; const size_t old_size = entries.size(); @@ -280,6 +312,7 @@ void StagingBufferPool::ReleaseLevel(StagingBuffersCache& cache, size_t log2) { entries.erase(std::remove_if(begin, end, is_deletable), end); const size_t new_size = entries.size(); + cache_bytes[static_cast(usage)] -= (old_size - new_size) << log2; staging.delete_index += deletions_per_tick; if (staging.delete_index >= new_size) { staging.delete_index = 0; diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h index 6d967abb41..d684a1289b 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h @@ -7,6 +7,7 @@ #pragma once #include +#include #include #include "common/common_types.h" @@ -88,6 +89,11 @@ private: bool AreRegionsActive(size_t region_begin, size_t region_end) const; + u64 MaxRegionTick(size_t region_begin, size_t region_end) const; + + std::optional OverflowStreamBuffer(size_t size, size_t region_begin, + size_t region_end); + StagingBufferRef GetStagingBuffer(size_t size, MemoryUsage usage, bool deferred = false); std::optional TryGetReservedBuffer(size_t size, MemoryUsage usage, @@ -99,7 +105,7 @@ private: void ReleaseCache(MemoryUsage usage); - void ReleaseLevel(StagingBuffersCache& cache, size_t log2); + void ReleaseLevel(MemoryUsage usage, size_t log2); size_t Region(size_t iter) const noexcept { return iter / region_size; } @@ -119,6 +125,8 @@ private: size_t free_iterator = 0; std::array sync_ticks{}; + std::array cache_bytes{}; + StagingBuffersCache device_local_cache; StagingBuffersCache upload_cache; StagingBuffersCache download_cache; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 749a49c898..33ab5dd57f 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -1091,11 +1091,12 @@ bool TextureCacheRuntime::ShouldReinterpret(Image& dst, Image& src) { } VkBuffer TextureCacheRuntime::GetTemporaryBuffer(size_t needed_size) { - const auto level = (8 * sizeof(size_t)) - std::countl_zero(needed_size - 1ULL); + const size_t wanted_size = (std::max)(needed_size, size_t{1}); + const auto level = (8 * sizeof(size_t)) - std::countl_zero(wanted_size - 1ULL); if (buffers[level]) { return *buffers[level]; } - const auto new_size = Common::NextPow2(needed_size); + const auto new_size = Common::NextPow2(wanted_size); static constexpr VkBufferUsageFlags flags = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT; diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 65194f647a..8fd72cb71f 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -741,7 +741,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR .pHeapSizeLimit = nullptr, .pVulkanFunctions = &functions, .instance = instance, - .vulkanApiVersion = ApiVersion(), + .vulkanApiVersion = VK_API_VERSION_1_1, .pTypeExternalMemoryHandleTypes = nullptr, }; diff --git a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp index 31a909ef2f..ef1143aeb6 100644 --- a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp +++ b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp @@ -26,6 +26,13 @@ namespace { : VkMemoryPropertyFlagBits{}; } + [[nodiscard]] VkMemoryPropertyFlags MemoryUsageRequiredVmaFlags(MemoryUsage usage) { + if (usage == MemoryUsage::Upload || usage == MemoryUsage::Download) { + return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT; + } + return VkMemoryPropertyFlagBits{}; + } + [[nodiscard]] VmaAllocationCreateFlags MemoryUsageVmaFlags(MemoryUsage usage) { switch (usage) { case MemoryUsage::Upload: @@ -103,7 +110,7 @@ vk::Image MemoryAllocator::CreateImage(const VkImageCreateInfo &ci) const .usage = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, .requiredFlags = 0, .preferredFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, - .memoryTypeBits = 0, + .memoryTypeBits = valid_memory_types, .pool = VK_NULL_HANDLE, .pUserData = nullptr, .priority = 0.f, @@ -138,7 +145,7 @@ vk::Buffer MemoryAllocator::CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsa VmaAllocationCreateInfo alloc_ci = { .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage), .usage = MemoryUsageVma(usage), - .requiredFlags = 0, + .requiredFlags = MemoryUsageRequiredVmaFlags(usage), .preferredFlags = MemoryUsagePreferredVmaFlags(usage) | anv_flags, .memoryTypeBits = usage == MemoryUsage::Stream ? 0u : valid_memory_types, .pool = VK_NULL_HANDLE, @@ -196,7 +203,7 @@ vk::Buffer MemoryAllocator::CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsa VmaAllocationCreateInfo alloc_ci = { .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage), .usage = MemoryUsageVma(usage), - .requiredFlags = 0, + .requiredFlags = MemoryUsageRequiredVmaFlags(usage), .preferredFlags = MemoryUsagePreferredVmaFlags(usage) | anv_flags, .memoryTypeBits = memory_type_bits, .pool = VK_NULL_HANDLE, From ffb75c2b68e0d432b40b3ab64adfb13454dd5d6a Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sun, 20 Sep 2026 14:18:28 -0400 Subject: [PATCH 43/69] Revert "[common/core/dynarmic] Optimize page table allocations (#4219)" --- src/common/CMakeLists.txt | 4 +- src/common/fiber.cpp | 1 + src/common/host_memory.cpp | 54 ++--- src/common/host_memory.h | 3 +- src/common/page_table.cpp | 38 +++- src/common/page_table.h | 120 +++++------ src/common/sparse_large_vector.cpp | 143 ------------- src/common/sparse_large_vector.h | 194 ------------------ src/common/virtual_buffer.cpp | 44 ++++ src/common/virtual_buffer.h | 84 ++++++++ src/core/arm/dynarmic/arm_dynarmic_32.cpp | 18 +- src/core/arm/dynarmic/arm_dynarmic_64.cpp | 19 +- src/core/device_memory.h | 10 +- src/core/device_memory_manager.h | 10 +- src/core/device_memory_manager.inc | 49 +++-- src/core/hle/kernel/k_page_table_base.cpp | 110 ++++------ src/core/hle/kernel/k_page_table_base.h | 13 +- src/core/memory.cpp | 81 ++++---- .../backend/arm64/a32_address_space.cpp | 4 +- .../backend/arm64/a64_address_space.cpp | 4 +- .../src/dynarmic/backend/arm64/emit_arm64.h | 4 +- .../backend/arm64/emit_arm64_memory.cpp | 15 +- .../dynarmic/backend/x64/emit_x64_memory.h | 63 ++---- .../src/dynarmic/interface/A32/config.h | 19 +- .../src/dynarmic/interface/A64/config.h | 19 +- src/video_core/memory_manager.cpp | 4 +- src/video_core/memory_manager.h | 4 +- 27 files changed, 394 insertions(+), 737 deletions(-) delete mode 100644 src/common/sparse_large_vector.cpp delete mode 100644 src/common/sparse_large_vector.h create mode 100644 src/common/virtual_buffer.cpp create mode 100644 src/common/virtual_buffer.h diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt index 61e0e64aa8..3a6d2acf8a 100644 --- a/src/common/CMakeLists.txt +++ b/src/common/CMakeLists.txt @@ -108,8 +108,6 @@ add_library( settings_setting.h slot_vector.h socket_types.h - sparse_large_vector.cpp - sparse_large_vector.h spin_lock.h stb.cpp stb.h @@ -139,6 +137,8 @@ add_library( uuid.cpp uuid.h vector_math.h + virtual_buffer.cpp + virtual_buffer.h zstd_compression.cpp zstd_compression.h fs/ryujinx_compat.h fs/ryujinx_compat.cpp diff --git a/src/common/fiber.cpp b/src/common/fiber.cpp index 0e7f28c5b9..69eca732eb 100644 --- a/src/common/fiber.cpp +++ b/src/common/fiber.cpp @@ -9,6 +9,7 @@ #include "common/assert.h" #include "common/fiber.h" +#include "common/virtual_buffer.h" #include diff --git a/src/common/host_memory.cpp b/src/common/host_memory.cpp index 75a9da80e9..dd0780a283 100644 --- a/src/common/host_memory.cpp +++ b/src/common/host_memory.cpp @@ -178,14 +178,6 @@ public: Release(); } - void* Allocate(size_t size) { - auto* ptr = VirtualAlloc(nullptr, size, MEM_RESERVE | MEM_COMMIT, PAGE_READWRITE); - if (ptr == nullptr) { - LOG_CRITICAL(HW_Memory, "Failed to allocate fallback buffer with size {:#x}, error {}", size, GetLastError()); - } - return ptr; - } - void Map(size_t virtual_offset, size_t host_offset, size_t length, MemoryPermission perms) { std::unique_lock lock{placeholder_mutex}; if (!IsNiechePlaceholder(virtual_offset, length)) { @@ -406,10 +398,6 @@ private: // For managarm: see https://github.com/managarm/managarm/issues/1370 #else // ^^^ Windows ^^^ vvv POSIX vvv -#ifndef MAP_NOCORE -#define MAP_NOCORE 0 -#endif - #ifdef ARCHITECTURE_arm64 static void* ChooseVirtualBase(size_t virtual_size) { @@ -434,7 +422,7 @@ static void* ChooseVirtualBase(size_t virtual_size) { // Note: we may be able to take advantage of MAP_FIXED_NOREPLACE here. void* map_pointer = mmap(reinterpret_cast(hint_address), virtual_size, PROT_READ | PROT_WRITE, - MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE | MAP_NOCORE, -1, 0); + MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE, -1, 0); // If we successfully mapped, we're done. if (reinterpret_cast(map_pointer) == hint_address) { @@ -454,11 +442,11 @@ static void* ChooseVirtualBase(size_t virtual_size) { static void* ChooseVirtualBase(size_t virtual_size) { #if defined(__FreeBSD__) || defined(__DragonFly__) || defined(__OpenBSD__) || defined(__sun__) || defined(__HAIKU__) || defined(__managarm__) || defined(__AIX__) - void* virtual_base = mmap(nullptr, virtual_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE | MAP_ALIGNED_SUPER | MAP_NOCORE, -1, 0); + void* virtual_base = mmap(nullptr, virtual_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE | MAP_ALIGNED_SUPER, -1, 0); if (virtual_base != MAP_FAILED) return virtual_base; #endif - return mmap(nullptr, virtual_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE | MAP_NOCORE, -1, 0); + return mmap(nullptr, virtual_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE, -1, 0); } #endif @@ -552,13 +540,13 @@ public: } if (use_anon) { LOG_WARNING(Common_Memory, "Using private mappings instead of shared ones"); - backing_base = static_cast(mmap(nullptr, backing_size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE | MAP_NOCORE, -1, 0)); + backing_base = static_cast(mmap(nullptr, backing_size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0)); if (fd > 0) { fd = -1; close(fd); } } else { - backing_base = static_cast(mmap(nullptr, backing_size, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_NOCORE, fd, 0)); + backing_base = static_cast(mmap(nullptr, backing_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0)); } if (backing_base == MAP_FAILED) { LOG_CRITICAL(HW_Memory, "mmap failed: {}", strerror(errno)); @@ -582,14 +570,6 @@ public: Release(); } - void* Allocate(size_t size) { - auto* ptr = mmap(nullptr, size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0); - if (ptr == MAP_FAILED) { - LOG_CRITICAL(HW_Memory, "Failed to allocate fallback buffer with size {:#x}, {}", size, strerror(errno)); - } - return ptr; - } - void Map(size_t virtual_offset, size_t host_offset, size_t length, MemoryPermission perms) { // Intersect the range with our address space. AdjustMap(&virtual_offset, &length); @@ -710,10 +690,12 @@ HostMemory::HostMemory(size_t backing_size_, size_t virtual_size_) { #if defined(__OPENORBIS__) || defined(__managarm__) LOG_WARNING(HW_Memory, "Platform doesn't support fastmem"); - backing_base = static_cast(mmap(nullptr, backing_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0)); + fallback_buffer.emplace(backing_size); + backing_base = fallback_buffer->data(); virtual_base = nullptr; #else // Try to allocate a fastmem arena. + // The implementation will fail with std::bad_alloc on errors. impl = std::make_unique(AlignUp(backing_size, PageAlignment), AlignUp(virtual_size, PageAlignment) + HugePageSize); if (impl->Init()) { backing_base = impl->backing_base; @@ -724,26 +706,16 @@ HostMemory::HostMemory(size_t backing_size_, size_t virtual_size_) virtual_base_offset = virtual_base - impl->virtual_base; } } else { - LOG_WARNING(HW_Memory, "Platform can support fastmem, but can't create it"); - fallback_buffer = true; - backing_base = static_cast(impl->Allocate(backing_size)); - virtual_base = nullptr; impl.reset(); + LOG_WARNING(HW_Memory, "Platform can support fastmem, but can't create it"); + fallback_buffer.emplace(backing_size); + backing_base = fallback_buffer->data(); + virtual_base = nullptr; } #endif } -HostMemory::~HostMemory() { -#ifdef _WIN32 - if (fallback_buffer) { - VirtualFree(backing_base, backing_size, MEM_RELEASE); - } -#else - if (fallback_buffer) { - munmap(backing_base, backing_size); - } -#endif -} +HostMemory::~HostMemory() = default; HostMemory::HostMemory(HostMemory&&) noexcept = default; diff --git a/src/common/host_memory.h b/src/common/host_memory.h index a7fd0532b7..24f4670732 100644 --- a/src/common/host_memory.h +++ b/src/common/host_memory.h @@ -10,6 +10,7 @@ #include #include "common/common_funcs.h" #include "common/common_types.h" +#include "common/virtual_buffer.h" namespace Common { @@ -85,7 +86,7 @@ private: u8* virtual_base{}; size_t virtual_base_offset{}; // Windows requires it for kernels whom lack proper support for some functions! - bool fallback_buffer{false}; + std::optional> fallback_buffer; }; } // namespace Common diff --git a/src/common/page_table.cpp b/src/common/page_table.cpp index 73278cfc59..d34ba89993 100644 --- a/src/common/page_table.cpp +++ b/src/common/page_table.cpp @@ -1,4 +1,4 @@ -// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project // SPDX-License-Identifier: GPL-3.0-or-later // SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project @@ -13,11 +13,39 @@ PageTable::PageTable() = default; PageTable::~PageTable() noexcept = default; -void PageTable::Resize(std::size_t address_space_width_in_bits, std::size_t page_bits) { - auto const num_page_table_entries = 1ULL << (address_space_width_in_bits - page_bits); - entries.ResizeAndClear(num_page_table_entries); +bool PageTable::BeginTraversal(TraversalEntry* out_entry, TraversalContext* out_context, + Common::ProcessAddress address) const { + out_context->next_offset = GetInteger(address); + out_context->next_page = address / page_size; + + return this->ContinueTraversal(out_entry, out_context); +} + +bool PageTable::ContinueTraversal(TraversalEntry* out_entry, TraversalContext* context) const { + // Setup invalid defaults. + out_entry->phys_addr = 0; + out_entry->block_size = page_size; + // Validate that we can read the actual entry. + if (auto const page = context->next_page; page < entries.size()) { + // Validate that the entry is mapped. + if (auto const paddr = entries[page].addr; paddr != 0) { + // Populate the results. + out_entry->phys_addr = paddr + context->next_offset; + context->next_page += 1; + context->next_offset += page_size; + return true; + } + } + context->next_page += 1; + context->next_offset += page_size; + return false; +} + +void PageTable::Resize(std::size_t address_space_width_in_bits, std::size_t page_size_in_bits) { + auto const num_page_table_entries = 1ULL << (address_space_width_in_bits - page_size_in_bits); + entries.resize(num_page_table_entries); current_address_space_width_in_bits = address_space_width_in_bits; - current_page_bits = page_bits; + page_size = 1ULL << page_size_in_bits; } } // namespace Common diff --git a/src/common/page_table.h b/src/common/page_table.h index 1588eca616..0dfd152331 100644 --- a/src/common/page_table.h +++ b/src/common/page_table.h @@ -9,22 +9,22 @@ #include #include "common/common_types.h" -#include "common/sparse_large_vector.h" #include "common/typed_address.h" +#include "common/virtual_buffer.h" namespace Common { enum class PageType : u8 { /// Page is unmapped and should cause an access error. - Unmapped = 0b00, + Unmapped, /// Page is mapped to regular memory. This is the only type you can get pointers to. - Memory = 0b01, + Memory, /// Page is mapped to regular memory, but inaccessible from CPU fastmem and must use /// the callbacks. - DebugMemory = 0b10, + DebugMemory, /// Page is mapped to regular memory, but also needs to check for rasterizer cache flushing and /// invalidation - RasterizerCachedMemory = 0b11, + RasterizerCachedMemory, }; /** @@ -42,86 +42,57 @@ struct PageTable { u64 next_offset{}; }; - /// Masks out bits reserved for attribute tagging. - static constexpr u64 ATTRIBUTE_MASK = ((1ULL << 44) - 1) << 12; - - /// Specifies sign bit for page table entries. - static constexpr u64 SIGN_BIT = 45 + 12; // 44 bits of data + page offset + /// Number of bits reserved for attribute tagging. + /// This can be at most the guaranteed alignment of the pointers in the page table. + static constexpr int ATTRIBUTE_BITS = 2; /** - * Atomic tuple of host pointer, page type, and block id. - * This uses the lower bits of a given pointer to store the attributes. + * Pair of host pointer and page type attribute. + * This uses the lower bits of a given pointer to store the attribute tag. * Writing and reading the pointer attribute pair is guaranteed to be atomic for the same method * call. In other words, they are guaranteed to be synchronized at all times. */ - class PageEntryData { + class PageInfo { public: - struct Data { - Data(bool marked_, PageType type_, u16 block_, u64 page_) - : marked(static_cast(marked_) & 0b1) - , type(static_cast(type_) & ((1ULL << 2) - 1)) - , block(static_cast(block_) & ((1ULL << 9) - 1)) - , page((page_ >> 12) & ((1ULL << 45) - 1)) - , block2((static_cast(block_) >> 9) & ((1ULL << 7) - 1)) {} - u64 marked : 1; - u64 type : 2; - u64 block : 9; - u64 page : 45; // 44 bits of actual data (64 - page offset (12) - reserved (8)) + a sign bit - u64 block2 : 7; - }; - - [[nodiscard]] Data Raw() const noexcept { - return std::bit_cast(data_raw.load(std::memory_order_relaxed)); - } - /// Returns the page pointer - [[nodiscard]] uintptr_t Pointer(bool ignored_marked = false) const noexcept { - return ExtractPointer(std::bit_cast(data_raw.load(std::memory_order_relaxed)), ignored_marked); + [[nodiscard]] uintptr_t Pointer() const noexcept { + return ExtractPointer(raw.load(std::memory_order_relaxed)); } /// Returns the page type attribute [[nodiscard]] PageType Type() const noexcept { - return static_cast(std::bit_cast(data_raw.load(std::memory_order_relaxed)).type); - } - - /// Returns the block identifier. - [[nodiscard]] u16 Block() const noexcept { - return ExtractBlock(std::bit_cast(data_raw.load(std::memory_order_relaxed))); + return ExtractType(raw.load(std::memory_order_relaxed)); } /// Returns the page pointer and attribute pair, extracted from the same atomic read - [[nodiscard]] std::tuple PointerTypeBlock(bool ignore_marked = false) const noexcept { - const auto non_atomic_raw = std::bit_cast(data_raw.load(std::memory_order_relaxed)); - return {ExtractPointer(non_atomic_raw, ignore_marked), static_cast(non_atomic_raw.type), ExtractBlock(non_atomic_raw)}; + [[nodiscard]] std::pair PointerType() const noexcept { + const uintptr_t non_atomic_raw = raw.load(std::memory_order_relaxed); + return {ExtractPointer(non_atomic_raw), ExtractType(non_atomic_raw)}; } - /// Write page info atomically - constexpr void Store(bool marked, PageType type, u16 block, uintptr_t pointer) noexcept { - data_raw.store(std::bit_cast(Data{marked, type, block, pointer})); + /// Returns the raw representation of the page information. + /// Use ExtractPointer and ExtractType to unpack the value. + [[nodiscard]] uintptr_t Raw() const noexcept { + return raw.load(std::memory_order_relaxed); } - constexpr void MarkRasterizerCached() noexcept { - data_raw.fetch_or(0b111); - } - - constexpr void MarkDebug(u64 ptr, u16 block) noexcept { - Store(true, PageType::DebugMemory, block, ptr); + /// Write a page pointer and type pair atomically + void Store(uintptr_t pointer, PageType type) noexcept { + raw.store(pointer | uintptr_t(type)); } /// Unpack a pointer from a page info raw representation - [[nodiscard]] static uintptr_t ExtractPointer(Data raw, bool ignore_marked = false) noexcept { - return raw.marked && !ignore_marked ? 0 - // shift raw.page's fake sign bit to the actual sign bit, then sign extend - : ((s64)(raw.page << (64 - 44))) >> (64 - 44 - 12); + [[nodiscard]] static uintptr_t ExtractPointer(uintptr_t raw) noexcept { + return raw & (~uintptr_t{0} << ATTRIBUTE_BITS); } - [[nodiscard]] static u16 ExtractBlock(Data raw) noexcept { - return static_cast(raw.block | (raw.block2 << 9)); + /// Unpack a page type from a page info raw representation + [[nodiscard]] static PageType ExtractType(uintptr_t raw) noexcept { + return static_cast(raw & ((uintptr_t{1} << ATTRIBUTE_BITS) - 1)); } private: - std::atomic data_raw; - static_assert(sizeof(Data) == sizeof(std::atomic)); + std::atomic raw; }; PageTable(); @@ -129,8 +100,13 @@ struct PageTable { PageTable(const PageTable&) = delete; PageTable& operator=(const PageTable&) = delete; - PageTable(PageTable&&) noexcept = delete; - PageTable& operator=(PageTable&&) noexcept = delete; + + PageTable(PageTable&&) noexcept = default; + PageTable& operator=(PageTable&&) noexcept = default; + + bool BeginTraversal(TraversalEntry* out_entry, TraversalContext* out_context, + Common::ProcessAddress address) const; + bool ContinueTraversal(TraversalEntry* out_entry, TraversalContext* context) const; /** * Resizes the page table to be able to accommodate enough pages within @@ -145,14 +121,30 @@ struct PageTable { return current_address_space_width_in_bits; } + bool GetPhysicalAddress(Common::PhysicalAddress* out_phys_addr, + Common::ProcessAddress virt_addr) const { + if (virt_addr > (1ULL << this->GetAddressSpaceBits())) { + return false; + } + + *out_phys_addr = entries[virt_addr / page_size].addr + GetInteger(virt_addr); + return true; + } + /// Vector of memory pointers backing each page. An entry can only be non-null if the /// corresponding attribute element is of type `Memory`. - SparseLargeVector entries; - static_assert(sizeof(PageEntryData) == 8); + struct PageEntryData { + PageInfo ptr; + u64 block; + u64 addr; + u64 padding; + }; + VirtualBuffer entries; + static_assert(sizeof(PageEntryData) == 32); u8* fastmem_arena{}; std::size_t current_address_space_width_in_bits{}; - std::size_t current_page_bits{}; + std::size_t page_size{}; }; } // namespace Common diff --git a/src/common/sparse_large_vector.cpp b/src/common/sparse_large_vector.cpp deleted file mode 100644 index ce6455a235..0000000000 --- a/src/common/sparse_large_vector.cpp +++ /dev/null @@ -1,143 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project -// SPDX-License-Identifier: GPL-3.0-or-later - -/* virtual_buffer.cpp */ -// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later - -#ifdef _WIN32 -#include -#include -#else -#include -#endif - -#include "common/alignment.h" -#include "common/assert.h" -#include "common/sparse_large_vector.h" - -namespace Common { - -#ifdef _WIN32 -static std::vector> vector_regions {}; - -// Workaround for handling non-commited memory accessed by Dynarmic; usually result of an error -static LONG WINAPI FakePageFaultHandler(PEXCEPTION_POINTERS info) { - DWORD code = info->ExceptionRecord->ExceptionCode; - u64 exception_addr = reinterpret_cast(info->ExceptionRecord->ExceptionAddress); - - if (code != EXCEPTION_ACCESS_VIOLATION) { - // Not our problem - return EXCEPTION_CONTINUE_SEARCH; - } - - u64 addr = 0, addr2 = 0; - - for (auto region: vector_regions) { - auto addr_shifted = exception_addr >> HostPageBits; - if (region.first <= addr_shifted && addr_shifted <= region.second) { - addr = addr_shifted; - } - - // Page-boundary accesses - if (auto addr_ = (exception_addr + 0x40) >> HostPageBits; addr_ != addr_shifted && region.first <= addr_ && addr_ <= region.second) { - addr2 = addr_; - } - - if (addr != 0 || addr2 != 0) { - break; - } - } - - if (addr == 0 && addr2 == 0) { - // Not our problem - return EXCEPTION_CONTINUE_SEARCH; - } - - LOG_ERROR(HW_Memory, "Accessing an unallocated region of a SparseLargeVector at {:#x}; this shouldn't happen and is likely a Dynarmic error!", exception_addr); - - // Commit this region - if (addr != 0) { - if (!CommitVectorPage(addr << HostPageBits, false)) { - return EXCEPTION_CONTINUE_SEARCH; - } - } - // Commit next region if needed - if (addr2 != 0) { - if (!CommitVectorPage(addr2 << HostPageBits, false)) { - return EXCEPTION_CONTINUE_SEARCH; - } - } - - return EXCEPTION_CONTINUE_EXECUTION; -} - -bool CommitVectorPage(uintptr_t addr, bool write) noexcept { - MEMORY_BASIC_INFORMATION info {}; - auto res = VirtualQuery(reinterpret_cast(addr), &info, sizeof(info)); - if (res == 0) { - LOG_CRITICAL(HW_Memory, "Failed to query large buffer region at {:#x} with error {}, will try committing anyway", addr, GetLastError()); - } else if (info.State != MEM_RESERVE) { - LOG_ERROR(HW_Memory, "Tried to commit an unreserved large buffer region at {:#x} that is not mapped or is already committed (state {:#x})", addr, info.State); - return false; - } - - auto perm = write ? PAGE_READWRITE : PAGE_READONLY; - void* res2 = VirtualAlloc(reinterpret_cast(addr), HostPageSize, MEM_COMMIT, perm); - if (res2 == nullptr) { - LOG_ERROR(HW_Memory, "Failed to commit large buffer region at {:#x}, error {}", addr, GetLastError()); - return false; - } - - return true; -} -#endif - -#ifndef MAP_NOCORE -#define MAP_NOCORE 0 -#endif - -void* AllocateMemoryPages(std::size_t size) noexcept { - if (auto page = HostPageSize; size % page != 0) { - LOG_WARNING(HW_Memory, "Allocating unaligned large vector with size {:#x}; aligning to {} page size", size, page); - size = AlignUp(size, page); - } - -#ifdef _WIN32 - // We will never use this memory entirely so instead of committing it up front let's just reserve it and commit each page individually - void* base = VirtualAlloc(nullptr, size, MEM_RESERVE, PAGE_READWRITE); - - if (base != nullptr) { - vector_regions.emplace_back(reinterpret_cast(base), reinterpret_cast(base) + size); - - static std::once_flag flag; - std::call_once(flag, []() { AddVectoredExceptionHandler(1, FakePageFaultHandler); }); - } else { - // Try committing everything instead?? - LOG_WARNING(HW_Memory, "Failed to reserve large vector region with error {}, trying to commit instead..", GetLastError()); - base = VirtualAlloc(nullptr, size, MEM_COMMIT, PAGE_READWRITE); - } - ASSERT_MSG(base, "Failed to reserve {:#x} sized region with error {}", size, GetLastError()); -#else - void* base = mmap(nullptr, size, PROT_READ, MAP_ANON | MAP_PRIVATE | MAP_NOCORE, -1, 0); - if (base == MAP_FAILED) - base = nullptr; - ASSERT_MSG(base, "Failed to allocate {:#x} sized region with error {}", size, strerror(errno)); -#endif - return base; -} - -void FreeMemoryPages(void* base, [[maybe_unused]] std::size_t size) noexcept { - if (auto page = HostPageSize; size % page != 0) { - size = AlignUp(size, page); - } - if (!base) - return; -#ifdef _WIN32 - ASSERT(VirtualFree(base, 0, MEM_RELEASE)); -#else - ASSERT(munmap(base, size) == 0); -#endif -} - -} // namespace Common diff --git a/src/common/sparse_large_vector.h b/src/common/sparse_large_vector.h deleted file mode 100644 index 827944fb08..0000000000 --- a/src/common/sparse_large_vector.h +++ /dev/null @@ -1,194 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project -// SPDX-License-Identifier: GPL-3.0-or-later - -/* virtual_buffer.h */ -// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later - -#pragma once - -#include -#include -#include -#include - -#ifndef _WIN32 -#include -#include -#endif - -#include "common/alignment.h" -#include "common/assert.h" - -namespace Common { - -#ifdef _WIN32 -constexpr u64 HostPageSize = 0x1000; -constexpr u64 HostPageBits = 12; -constexpr u64 HostPageMask = ~(HostPageSize - 1); -bool CommitVectorPage(uintptr_t addr, bool write) noexcept; -#else -const u64 HostPageSize = sysconf(_SC_PAGESIZE); -const u64 HostPageBits = std::countr_zero(HostPageSize); -const u64 HostPageMask = ~(HostPageSize - 1); -#endif - -void* AllocateMemoryPages(std::size_t size) noexcept; -void FreeMemoryPages(void* base, std::size_t size) noexcept; - -/// A large page-aligned buffer that has optimized memory usage for zero-writes. -template -requires std::is_trivially_copyable_v -class SparseLargeVector final { -public: - constexpr SparseLargeVector() = default; - - explicit SparseLargeVector(std::size_t count) noexcept - : alloc_size{count * sizeof(T)} - { - base_ptr = static_cast(AllocateMemoryPages(alloc_size)); - - // each item in vector holds information for 64 pages - auto denom = HostPageSize * 64; - committed_pages = std::vector>((alloc_size + denom - 1) / denom); - } - - ~SparseLargeVector() noexcept { - FreeMemoryPages(base_ptr, alloc_size); - } - - SparseLargeVector(const SparseLargeVector&) = delete; - SparseLargeVector& operator=(const SparseLargeVector&) = delete; - SparseLargeVector(SparseLargeVector&& other) = delete; - SparseLargeVector& operator=(SparseLargeVector&& other) = delete; - - void ResizeAndClear(std::size_t count) noexcept { - if (auto const new_size = count * sizeof(T); new_size != alloc_size) { - FreeMemoryPages(base_ptr, alloc_size); - alloc_size = new_size; - base_ptr = static_cast(AllocateMemoryPages(alloc_size)); - - auto denom = HostPageSize * 64; - committed_pages = std::vector>((alloc_size + denom - 1) / denom); - } - } - - /// Returns a reference to the value of the requested index and allocates memory if needed. - T& GetAndFault(std::size_t index) noexcept { - if (index > alloc_size / sizeof(T)) { - UNREACHABLE_MSG("Out of bounds RW access on SparseLargeVector @ {}", index); - } - - if (!IsCommittedPage(index)) { - CommitPage(index); - } - return base_ptr[index]; - } - - /// Returns a reference to the value of the requested index if initialized, or will otherwise return a zero-initialized object. - const T& GetOrDefault(std::size_t index) const { -#ifdef _WIN32 - if (!IsCommittedPage(index)) { - return *reinterpret_cast(&default_val); - } -#endif - // On non-Windows, OS page table should optimize this by pointing to a zero page if unallocated. - return base_ptr[index]; - } - - void Set(std::size_t index, const T& value) noexcept { - if (index > alloc_size / sizeof(T)) { - LOG_CRITICAL(Common_Memory, "Out of bounds write on SparseLargeVector @ {}", index); - return; - } - if (!IsCommittedPage(index)) - CommitPage(index); - base_ptr[index] = value; - } - - void ZeroRegion(std::size_t start, std::size_t end_) noexcept { - u64 base = reinterpret_cast(&base_ptr[start]); - const u64 end = reinterpret_cast(&base_ptr[end_]); - - const u64 end_page = AlignUp(base, HostPageSize); - const u64 first_size = (std::min)(end_page, end) - base; - - if (IsCommittedPage(start / sizeof(T))) { - std::memset(reinterpret_cast(base), 0, first_size); - } - - if (end <= end_page) - return; - - base = end_page; - - for (u64 page = base; page < end; page += HostPageSize) { - if (!IsCommittedPage((page - reinterpret_cast(base_ptr)) / sizeof(T))) { - continue; - } - - std::memset(reinterpret_cast(page), 0, (std::min)( HostPageSize, end - page)); - } - } - - constexpr void CommitRegion(size_t index, size_t end_) { - const u64 base = static_cast(index) * sizeof(T); - const u64 end = static_cast(end_) * sizeof(T); - - for (u64 page = AlignDown(base, HostPageSize); page < end; page += HostPageSize) { - if (!IsCommittedPage(page / sizeof(T))) { - CommitPage(page / sizeof(T)); - } - } - } - - constexpr T& GetUnchecked(size_t index) { - return base_ptr[index]; - } - - [[nodiscard]] constexpr const T& operator[](std::size_t index) const noexcept { - return GetOrDefault(index); - } - - [[nodiscard]] constexpr const T* data() const noexcept { - return base_ptr; - } - - [[nodiscard]] constexpr std::size_t size() const noexcept { - return alloc_size / sizeof(T); - } - -private: - [[nodiscard]] constexpr bool IsCommittedPage(std::size_t index) const noexcept { - if (index > alloc_size / sizeof(T)) { - LOG_CRITICAL(Common_Memory, "Out of bounds access on large vector @ {}", index); - return false; - } - - auto page = (index * sizeof(T)) >> HostPageBits; - auto val = committed_pages[page >> 6].load(std::memory_order_acquire); - return (val >> (page & 63)) & 1; - } - - constexpr void CommitPage(std::size_t index) noexcept { - auto page_index = (index * sizeof(T)) >> HostPageBits; - auto page = reinterpret_cast(base_ptr + index) & HostPageMask; -#if defined(_WIN32) - CommitVectorPage(page, true); -#else - mprotect(reinterpret_cast(page), HostPageSize, PROT_READ | PROT_WRITE); -#endif - - committed_pages[page_index >> 6].fetch_or(1ULL << (page_index & 63), std::memory_order_release); - } - - std::size_t alloc_size{}; - T* base_ptr{}; - - std::vector> committed_pages{}; -#ifdef _WIN32 - const std::array default_val{}; -#endif -}; - -} // namespace Common diff --git a/src/common/virtual_buffer.cpp b/src/common/virtual_buffer.cpp new file mode 100644 index 0000000000..8a4265e3e6 --- /dev/null +++ b/src/common/virtual_buffer.cpp @@ -0,0 +1,44 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#ifdef _WIN32 +#include +#else +#include +#endif + +#include "common/assert.h" +#include "common/virtual_buffer.h" + +namespace Common { + +void* AllocateMemoryPages(std::size_t size) noexcept { +#ifdef _WIN32 + void* base = VirtualAlloc(nullptr, size, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE); + if (base == nullptr) { + // Probably failing to reserve is less likely than failing to commit + base = VirtualAlloc(nullptr, size, MEM_COMMIT, PAGE_READWRITE); + } +#else + void* base = mmap(nullptr, size, PROT_READ | PROT_WRITE, MAP_ANON | MAP_PRIVATE, -1, 0); + if (base == MAP_FAILED) + base = nullptr; +#endif + ASSERT(base); + return base; +} + +void FreeMemoryPages(void* base, [[maybe_unused]] std::size_t size) noexcept { + if (!base) + return; +#ifdef _WIN32 + ASSERT(VirtualFree(base, 0, MEM_RELEASE)); +#else + ASSERT(munmap(base, size) == 0); +#endif +} + +} // namespace Common diff --git a/src/common/virtual_buffer.h b/src/common/virtual_buffer.h new file mode 100644 index 0000000000..d6386e2a4d --- /dev/null +++ b/src/common/virtual_buffer.h @@ -0,0 +1,84 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#pragma once + +#include + +namespace Common { + +void* AllocateMemoryPages(std::size_t size) noexcept; +void FreeMemoryPages(void* base, std::size_t size) noexcept; + +template +class VirtualBuffer final { +public: + // TODO: Uncomment this and change Common::PageTable::PageInfo to be trivially constructible + // using std::atomic_ref once libc++ has support for it + // static_assert( + // std::is_trivially_constructible_v, + // "T must be trivially constructible, as non-trivial constructors will not be executed " + // "with the current allocator"); + + constexpr VirtualBuffer() = default; + explicit VirtualBuffer(std::size_t count) noexcept + : alloc_size{count * sizeof(T)} + { + base_ptr = reinterpret_cast(AllocateMemoryPages(alloc_size)); + } + + ~VirtualBuffer() noexcept { + FreeMemoryPages(base_ptr, alloc_size); + } + + VirtualBuffer(const VirtualBuffer&) = delete; + VirtualBuffer& operator=(const VirtualBuffer&) = delete; + + VirtualBuffer(VirtualBuffer&& other) noexcept + : alloc_size{std::exchange(other.alloc_size, 0)} + , base_ptr{std::exchange(other.base_ptr, nullptr)} + {} + + VirtualBuffer& operator=(VirtualBuffer&& other) noexcept { + alloc_size = std::exchange(other.alloc_size, 0); + base_ptr = std::exchange(other.base_ptr, nullptr); + return *this; + } + + void resize(std::size_t count) noexcept { + if (auto const new_size = count * sizeof(T); new_size != alloc_size) { + FreeMemoryPages(base_ptr, alloc_size); + alloc_size = new_size; + base_ptr = reinterpret_cast(AllocateMemoryPages(alloc_size)); + } + } + + [[nodiscard]] constexpr const T& operator[](std::size_t index) const noexcept { + return base_ptr[index]; + } + + [[nodiscard]] constexpr T& operator[](std::size_t index) noexcept { + return base_ptr[index]; + } + + [[nodiscard]] constexpr T* data() noexcept { + return base_ptr; + } + + [[nodiscard]] constexpr const T* data() const noexcept { + return base_ptr; + } + + [[nodiscard]] constexpr std::size_t size() const noexcept { + return alloc_size / sizeof(T); + } + +private: + std::size_t alloc_size{}; + T* base_ptr{}; +}; + +} // namespace Common diff --git a/src/core/arm/dynarmic/arm_dynarmic_32.cpp b/src/core/arm/dynarmic/arm_dynarmic_32.cpp index 80e2b2dca9..97284efa4c 100644 --- a/src/core/arm/dynarmic/arm_dynarmic_32.cpp +++ b/src/core/arm/dynarmic/arm_dynarmic_32.cpp @@ -172,12 +172,12 @@ void ArmDynarmic32::MakeJit(Common::PageTable* page_table) { if (page_table) { constexpr size_t PageBits = 12; constexpr size_t NumPageTableEntries = 1 << (32 - PageBits); + constexpr size_t PageLog2Stride = 5; + static_assert(1 << PageLog2Stride == sizeof(Common::PageTable::PageEntryData)); - // Dynarmic will not write to the page table, const_cast is safe here - config.page_table = reinterpret_cast*>( - const_cast(page_table->entries.data())); - config.page_table_pointer_mask = Common::PageTable::ATTRIBUTE_MASK; - config.page_table_marked_bit = 0; + config.page_table = reinterpret_cast*>(page_table->entries.data()); + config.page_table_pointer_mask_bits = Common::PageTable::ATTRIBUTE_BITS; + config.page_table_log2_stride = PageLog2Stride; config.absolute_offset_page_table = true; config.detect_misaligned_access_via_page_table = 16 | 32 | 64 | 128; config.only_detect_misalignment_via_page_table_on_page_boundary = true; @@ -188,13 +188,6 @@ void ArmDynarmic32::MakeJit(Common::PageTable* page_table) { config.fastmem_exclusive_access = config.fastmem_pointer != std::nullopt; config.recompile_on_exclusive_fastmem_failure = true; - - if (reinterpret_cast(m_system.DeviceMemory().buffer.BackingBasePointer() + - Kernel::Board::Nintendo::Nx::KSystemControl::Init::GetIntendedMemorySize()) < (1ULL << 39)) { - // Systems like FreeBSD allocate memory really low by default, and since we pack our page table entries, - // we have to manually sign extend when our actual pointer is negative. - config.page_table_sign_extension = Common::PageTable::SIGN_BIT; - } } // Multi-process state @@ -427,7 +420,6 @@ void ArmDynarmic32::SignalInterrupt(Kernel::KThread* thread) { } void ArmDynarmic32::ClearInstructionCache() { - m_cb->last_code_addr = u64(-1); m_jit->ClearCache(); } diff --git a/src/core/arm/dynarmic/arm_dynarmic_64.cpp b/src/core/arm/dynarmic/arm_dynarmic_64.cpp index 9662bf1712..0b7f6874a7 100644 --- a/src/core/arm/dynarmic/arm_dynarmic_64.cpp +++ b/src/core/arm/dynarmic/arm_dynarmic_64.cpp @@ -211,12 +211,13 @@ void ArmDynarmic64::MakeJit(Common::PageTable* page_table, std::size_t address_s // Memory if (page_table) { - // Dynarmic will not write to the page table, const_cast is safe here - config.page_table = reinterpret_cast( - const_cast(page_table->entries.data())); + constexpr size_t PageLog2Stride = 5; + static_assert(1 << PageLog2Stride == sizeof(Common::PageTable::PageEntryData)); + + config.page_table = reinterpret_cast(page_table->entries.data()); config.page_table_address_space_bits = std::uint32_t(address_space_bits); - config.page_table_pointer_mask = Common::PageTable::ATTRIBUTE_MASK; - config.page_table_marked_bit = 0; + config.page_table_pointer_mask_bits = Common::PageTable::ATTRIBUTE_BITS; + config.page_table_log2_stride = PageLog2Stride; config.silently_mirror_page_table = false; config.absolute_offset_page_table = true; config.detect_misaligned_access_via_page_table = 16 | 32 | 64 | 128; @@ -230,13 +231,6 @@ void ArmDynarmic64::MakeJit(Common::PageTable* page_table, std::size_t address_s config.fastmem_exclusive_access = config.fastmem_pointer != std::nullopt; config.recompile_on_exclusive_fastmem_failure = true; - - if (reinterpret_cast(m_system.DeviceMemory().buffer.BackingBasePointer() + - Kernel::Board::Nintendo::Nx::KSystemControl::Init::GetIntendedMemorySize()) < (1ULL << 39)) { - // Systems like FreeBSD allocate memory really low by default, and since we pack our page table entries, - // we have to manually sign extend when our actual pointer is negative. - config.page_table_sign_extension = Common::PageTable::SIGN_BIT; - } } // Multi-process state @@ -453,7 +447,6 @@ void ArmDynarmic64::SignalInterrupt(Kernel::KThread* thread) { } void ArmDynarmic64::ClearInstructionCache() { - m_cb->last_code_addr = u64(-1); m_jit->ClearCache(); } diff --git a/src/core/device_memory.h b/src/core/device_memory.h index b5103e23bc..11bf0e3268 100644 --- a/src/core/device_memory.h +++ b/src/core/device_memory.h @@ -1,6 +1,3 @@ -// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project -// SPDX-License-Identifier: GPL-3.0-or-later - // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later @@ -29,11 +26,8 @@ public: template Common::PhysicalAddress GetPhysicalAddr(const T* ptr) const { - return GetPhysicalAddr(reinterpret_cast(ptr)); - } - - Common::PhysicalAddress GetPhysicalAddr(uintptr_t ptr) const { - return (ptr - reinterpret_cast(buffer.BackingBasePointer())) + + return (reinterpret_cast(ptr) - + reinterpret_cast(buffer.BackingBasePointer())) + DramMemoryMap::Base; } diff --git a/src/core/device_memory_manager.h b/src/core/device_memory_manager.h index b4b3b46088..3d97fdcc5c 100644 --- a/src/core/device_memory_manager.h +++ b/src/core/device_memory_manager.h @@ -18,7 +18,7 @@ #include "common/common_types.h" #include "common/range_mutex.h" #include "common/scratch_buffer.h" -#include "common/sparse_large_vector.h" +#include "common/virtual_buffer.h" namespace Core { @@ -178,8 +178,8 @@ private: u32 continuity_tracker; u32 compressed_physical_ptr; }; - Common::SparseLargeVector compressed_device_addr; - Common::SparseLargeVector tracked_entries; + Common::VirtualBuffer compressed_device_addr; + Common::VirtualBuffer tracked_entries; // Process memory interfaces @@ -200,8 +200,8 @@ private: return std::make_pair(asid, address); } - constexpr void InsertCPUBacking(size_t page_index, VAddr address, Asid asid) { - tracked_entries.GetUnchecked(page_index).cpu_backing_address = address | (asid.id << asid_start_bit); + void InsertCPUBacking(size_t page_index, VAddr address, Asid asid) { + tracked_entries[page_index].cpu_backing_address = address | (asid.id << asid_start_bit); } std::array t_slot{}; diff --git a/src/core/device_memory_manager.inc b/src/core/device_memory_manager.inc index 97c2706abb..9866f3a4f7 100644 --- a/src/core/device_memory_manager.inc +++ b/src/core/device_memory_manager.inc @@ -177,6 +177,17 @@ DeviceMemoryManager::DeviceMemoryManager(const DeviceMemory& device_memo { impl = std::make_unique>(); cached_pages = std::make_unique(); + + const size_t total_virtual = device_as_size >> Memory::YUZU_PAGEBITS; + for (size_t i = 0; i < total_virtual; i++) { + tracked_entries[i].compressed_physical_ptr = 0; + tracked_entries[i].continuity_tracker = 1; + tracked_entries[i].cpu_backing_address = 0; + } + const size_t total_phys = 1ULL << ((Settings::values.memory_layout_mode.GetValue() == Settings::MemoryLayout::Memory_4Gb ? physical_min_bits : physical_max_bits) - Memory::YUZU_PAGEBITS); + for (size_t i = 0; i < total_phys; i++) { + compressed_device_addr[i] = 0; + } } template @@ -209,28 +220,26 @@ void DeviceMemoryManager::Map(DAddr address, VAddr virtual_address, size size_t start_page_d = address >> Memory::YUZU_PAGEBITS; size_t num_pages = Common::AlignUp(size, Memory::YUZU_PAGESIZE) >> Memory::YUZU_PAGEBITS; std::scoped_lock lk(mapping_guard); - - tracked_entries.CommitRegion(start_page_d, start_page_d + num_pages); for (size_t i = 0; i < num_pages; i++) { const VAddr new_vaddress = virtual_address + i * Memory::YUZU_PAGESIZE; auto* ptr = process_memory->GetPointerSilent(Common::ProcessAddress(new_vaddress)); if (ptr == nullptr) [[unlikely]] { - tracked_entries.GetUnchecked(start_page_d + i).compressed_physical_ptr = 0; + tracked_entries[start_page_d + i].compressed_physical_ptr = 0; continue; } auto phys_addr = static_cast(GetRawPhysicalAddr(ptr) >> Memory::YUZU_PAGEBITS) + 1U; - tracked_entries.GetUnchecked(start_page_d + i).compressed_physical_ptr = phys_addr; + tracked_entries[start_page_d + i].compressed_physical_ptr = phys_addr; InsertCPUBacking(start_page_d + i, new_vaddress, asid); const u32 base_dev = compressed_device_addr[phys_addr - 1U]; const u32 new_dev = static_cast(start_page_d + i); if (base_dev == 0) [[likely]] { - compressed_device_addr.GetAndFault(phys_addr - 1U) = new_dev; + compressed_device_addr[phys_addr - 1U] = new_dev; continue; } u32 start_id = base_dev & MULTI_MASK; if ((base_dev >> MULTI_FLAG_BITS) == 0) { start_id = impl->multi_dev_address.Register(base_dev); - compressed_device_addr.GetAndFault(phys_addr - 1U) = MULTI_FLAG | start_id; + compressed_device_addr[phys_addr - 1U] = MULTI_FLAG | start_id; } impl->multi_dev_address.Register(new_dev, start_id); } @@ -246,26 +255,24 @@ void DeviceMemoryManager::Unmap(DAddr address, size_t size) { size_t num_pages = Common::AlignUp(size, Memory::YUZU_PAGESIZE) >> Memory::YUZU_PAGEBITS; device_inter->InvalidateRegion(address, size); std::scoped_lock lk(mapping_guard); - - tracked_entries.CommitRegion(start_page_d, start_page_d + num_pages); // should already be committed, but just in case for (size_t i = 0; i < num_pages; i++) { - auto& entry = tracked_entries.GetUnchecked(start_page_d + i); - auto phys_addr = entry.compressed_physical_ptr; - entry.compressed_physical_ptr = 0; - entry.cpu_backing_address = 0; + auto phys_addr = tracked_entries[start_page_d + i].compressed_physical_ptr; + tracked_entries[start_page_d + i].compressed_physical_ptr = 0; + tracked_entries[start_page_d + i].cpu_backing_address = 0; if (phys_addr != 0) [[likely]] { - u32& base_dev = compressed_device_addr.GetAndFault(phys_addr - 1U); + const u32 base_dev = compressed_device_addr[phys_addr - 1U]; if ((base_dev >> MULTI_FLAG_BITS) == 0) [[likely]] { - base_dev = 0; + compressed_device_addr[phys_addr - 1] = 0; continue; } const auto [more_entries, new_start] = impl->multi_dev_address.Unregister( static_cast(start_page_d + i), base_dev & MULTI_MASK); if (!more_entries) { - base_dev = impl->multi_dev_address.ReleaseEntry(new_start); + compressed_device_addr[phys_addr - 1] = + impl->multi_dev_address.ReleaseEntry(new_start); continue; } - base_dev = new_start | MULTI_FLAG; + compressed_device_addr[phys_addr - 1] = new_start | MULTI_FLAG; } } t_slot = {}; @@ -278,8 +285,6 @@ void DeviceMemoryManager::TrackContinuityImpl(DAddr address, VAddr virtu size_t num_pages = Common::AlignUp(size, Memory::YUZU_PAGESIZE) >> Memory::YUZU_PAGEBITS; uintptr_t last_ptr = 0; size_t page_count = 1; - - tracked_entries.CommitRegion(start_page_d, start_page_d + num_pages); for (size_t i = num_pages; i > 0; i--) { size_t index = i - 1; const VAddr new_vaddress = virtual_address + index * Memory::YUZU_PAGESIZE; @@ -291,14 +296,14 @@ void DeviceMemoryManager::TrackContinuityImpl(DAddr address, VAddr virtu page_count = 1; } last_ptr = new_ptr; - tracked_entries.GetUnchecked(start_page_d + index).continuity_tracker = static_cast(page_count) - 1; + tracked_entries[start_page_d + index].continuity_tracker = static_cast(page_count); } } template u8* DeviceMemoryManager::GetSpan(const DAddr src_addr, const std::size_t size) { size_t page_index = src_addr >> page_bits; size_t subbits = src_addr & page_mask; - if ((static_cast(tracked_entries[page_index].continuity_tracker+1) << page_bits) >= size + subbits) { + if ((static_cast(tracked_entries[page_index].continuity_tracker) << page_bits) >= size + subbits) { return GetPointer(src_addr); } return nullptr; @@ -308,7 +313,7 @@ template const u8* DeviceMemoryManager::GetSpan(const DAddr src_addr, const std::size_t size) const { size_t page_index = src_addr >> page_bits; size_t subbits = src_addr & page_mask; - if ((static_cast(tracked_entries[page_index].continuity_tracker+1) << page_bits) >= size + subbits) { + if ((static_cast(tracked_entries[page_index].continuity_tracker) << page_bits) >= size + subbits) { return GetPointer(src_addr); } return nullptr; @@ -378,7 +383,7 @@ void DeviceMemoryManager::WalkBlock(DAddr addr, std::size_t size, auto o std::size_t page_index = addr >> Memory::YUZU_PAGEBITS; std::size_t page_offset = addr & Memory::YUZU_PAGEMASK; while (remaining_size) { - const size_t next_pages = std::size_t(tracked_entries[page_index].continuity_tracker+1); + const size_t next_pages = std::size_t(tracked_entries[page_index].continuity_tracker); const std::size_t copy_amount = (std::min)((next_pages << Memory::YUZU_PAGEBITS) - page_offset, remaining_size); const auto current_vaddr = u64((page_index << Memory::YUZU_PAGEBITS) + page_offset); SCOPE_EXIT{ diff --git a/src/core/hle/kernel/k_page_table_base.cpp b/src/core/hle/kernel/k_page_table_base.cpp index 2a0a7d49d5..a727355a9f 100644 --- a/src/core/hle/kernel/k_page_table_base.cpp +++ b/src/core/hle/kernel/k_page_table_base.cpp @@ -635,36 +635,6 @@ Result KPageTableBase::CheckMemoryState(const KMemoryInfo& info, KMemoryState st R_SUCCEED(); } -bool KPageTableBase::BeginTraversal(const Common::PageTable &impl, TraversalEntry *out_entry, TraversalContext *out_context, - Common::ProcessAddress address) const { - out_context->next_offset = GetInteger(address); - out_context->next_page = GetInteger(address) >> PageBits; - - return ContinueTraversal(impl, out_entry, out_context); -} - -bool KPageTableBase::ContinueTraversal(const Common::PageTable &impl, TraversalEntry *out_entry, - TraversalContext *context) const { - // Setup invalid defaults. - out_entry->phys_addr = 0; - out_entry->block_size = PageSize; - // Validate that we can read the actual entry. - if (auto const page = context->next_page; page < impl.entries.size()) { - // Validate that the entry is mapped. - if (auto const paddr = impl.entries[page].Pointer(true); paddr != 0) { - // Populate the results and return true - out_entry->phys_addr = GetInteger(m_system.DeviceMemory().GetPhysicalAddr(paddr + context->next_offset)); - context->next_page += 1; - context->next_offset += PageSize; - return true; - } - } - context->next_page += 1; - context->next_offset += PageSize; - // Otherwise return false - return false; -} - Result KPageTableBase::CheckMemoryStateContiguous(size_t* out_blocks_needed, KProcessAddress addr, size_t size, KMemoryState state_mask, KMemoryState state, KMemoryPermission perm_mask, @@ -970,7 +940,7 @@ Result KPageTableBase::QueryMappingImpl(KProcessAddress* out, KPhysicalAddress a size_t tot_size = 0; next_valid = - BeginTraversal(impl, std::addressof(next_entry), std::addressof(context), region_start); + impl.BeginTraversal(std::addressof(next_entry), std::addressof(context), region_start); next_entry.block_size = (next_entry.block_size - (GetInteger(region_start) & (next_entry.block_size - 1))); @@ -1006,7 +976,7 @@ Result KPageTableBase::QueryMappingImpl(KProcessAddress* out, KPhysicalAddress a break; } - next_valid = ContinueTraversal(impl, std::addressof(next_entry), std::addressof(context)); + next_valid = impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context)); } // Check the last entry. @@ -1784,7 +1754,7 @@ Result KPageTableBase::MakePageGroup(KPageGroup& pg, KProcessAddress addr, size_ // Begin traversal. TraversalContext context; TraversalEntry next_entry; - R_UNLESS(BeginTraversal(impl, std::addressof(next_entry), std::addressof(context), addr), + R_UNLESS(impl.BeginTraversal(std::addressof(next_entry), std::addressof(context), addr), ResultInvalidCurrentMemory); // Prepare tracking variables. @@ -1794,7 +1764,7 @@ Result KPageTableBase::MakePageGroup(KPageGroup& pg, KProcessAddress addr, size_ // Iterate, adding to group as we go. while (tot_size < size) { - R_UNLESS(ContinueTraversal(impl, std::addressof(next_entry), std::addressof(context)), + R_UNLESS(impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context)), ResultInvalidCurrentMemory); if (next_entry.phys_addr != (cur_addr + cur_size)) { @@ -1858,7 +1828,7 @@ bool KPageTableBase::IsValidPageGroup(const KPageGroup& pg, KProcessAddress addr // Begin traversal. TraversalContext context; TraversalEntry next_entry; - if (!BeginTraversal(impl, std::addressof(next_entry), std::addressof(context), addr)) { + if (!impl.BeginTraversal(std::addressof(next_entry), std::addressof(context), addr)) { return false; } @@ -1869,7 +1839,7 @@ bool KPageTableBase::IsValidPageGroup(const KPageGroup& pg, KProcessAddress addr // Iterate, comparing expected to actual. while (tot_size < size) { - if (!ContinueTraversal(impl, std::addressof(next_entry), std::addressof(context))) { + if (!impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context))) { return false; } @@ -1926,7 +1896,7 @@ Result KPageTableBase::GetContiguousMemoryRangeWithState( // Begin a traversal. TraversalContext context; TraversalEntry cur_entry = {.phys_addr = 0, .block_size = 0}; - R_UNLESS(BeginTraversal(impl, std::addressof(cur_entry), std::addressof(context), address), + R_UNLESS(impl.BeginTraversal(std::addressof(cur_entry), std::addressof(context), address), ResultInvalidCurrentMemory); // Traverse until we have enough size or we aren't contiguous any more. @@ -1935,7 +1905,7 @@ Result KPageTableBase::GetContiguousMemoryRangeWithState( for (contig_size = cur_entry.block_size - (GetInteger(phys_address) & (cur_entry.block_size - 1)); contig_size < size; contig_size += cur_entry.block_size) { - if (!ContinueTraversal(impl, std::addressof(cur_entry), std::addressof(context))) { + if (!impl.ContinueTraversal(std::addressof(cur_entry), std::addressof(context))) { break; } if (cur_entry.phys_addr != phys_address + contig_size) { @@ -2364,7 +2334,7 @@ Result KPageTableBase::QueryPhysicalAddress(Svc::lp64::PhysicalMemoryInfo* out, TraversalContext context; TraversalEntry next_entry; bool traverse_valid = - BeginTraversal(m_impl, std::addressof(next_entry), std::addressof(context), virt_addr); + m_impl.BeginTraversal(std::addressof(next_entry), std::addressof(context), virt_addr); R_UNLESS(traverse_valid, ResultInvalidCurrentMemory); // Set tracking variables. @@ -2375,7 +2345,7 @@ Result KPageTableBase::QueryPhysicalAddress(Svc::lp64::PhysicalMemoryInfo* out, while (true) { // Continue the traversal. traverse_valid = - ContinueTraversal(m_impl, std::addressof(next_entry), std::addressof(context)); + m_impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context)); if (!traverse_valid) { break; } @@ -2597,7 +2567,7 @@ Result KPageTableBase::UnmapIoRegion(KProcessAddress dst_address, KPhysicalAddre TraversalContext context; TraversalEntry next_entry; ASSERT( - BeginTraversal(impl, std::addressof(next_entry), std::addressof(context), dst_address)); + impl.BeginTraversal(std::addressof(next_entry), std::addressof(context), dst_address)); // Check that the physical region matches. R_UNLESS(next_entry.phys_addr == phys_addr, ResultInvalidMemoryRegion); @@ -2607,7 +2577,7 @@ Result KPageTableBase::UnmapIoRegion(KProcessAddress dst_address, KPhysicalAddre next_entry.block_size - (GetInteger(phys_addr) & (next_entry.block_size - 1)); checked_size < size; checked_size += next_entry.block_size) { // Continue the traversal. - ASSERT(ContinueTraversal(impl, std::addressof(next_entry), std::addressof(context))); + ASSERT(impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context))); // Check that the physical region matches. R_UNLESS(next_entry.phys_addr == phys_addr + checked_size, ResultInvalidMemoryRegion); @@ -3059,7 +3029,7 @@ Result KPageTableBase::InvalidateProcessDataCache(KProcessAddress address, size_ TraversalContext context; TraversalEntry next_entry; bool traverse_valid = - BeginTraversal(impl, std::addressof(next_entry), std::addressof(context), address); + impl.BeginTraversal(std::addressof(next_entry), std::addressof(context), address); R_UNLESS(traverse_valid, ResultInvalidCurrentMemory); // Prepare tracking variables. @@ -3071,7 +3041,7 @@ Result KPageTableBase::InvalidateProcessDataCache(KProcessAddress address, size_ while (tot_size < size) { // Continue the traversal. traverse_valid = - ContinueTraversal(impl, std::addressof(next_entry), std::addressof(context)); + impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context)); R_UNLESS(traverse_valid, ResultInvalidCurrentMemory); if (next_entry.phys_addr != (cur_addr + cur_size)) { @@ -3159,7 +3129,7 @@ Result KPageTableBase::ReadDebugMemory(KProcessAddress dst_address, KProcessAddr TraversalContext context; TraversalEntry next_entry; bool traverse_valid = - BeginTraversal(impl, std::addressof(next_entry), std::addressof(context), src_address); + impl.BeginTraversal(std::addressof(next_entry), std::addressof(context), src_address); R_UNLESS(traverse_valid, ResultInvalidCurrentMemory); // Prepare tracking variables. @@ -3197,7 +3167,7 @@ Result KPageTableBase::ReadDebugMemory(KProcessAddress dst_address, KProcessAddr while (tot_size < size) { // Continue the traversal. traverse_valid = - ContinueTraversal(impl, std::addressof(next_entry), std::addressof(context)); + impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context)); ASSERT(traverse_valid); if (next_entry.phys_addr != (cur_addr + cur_size)) { @@ -3255,7 +3225,7 @@ Result KPageTableBase::WriteDebugMemory(KProcessAddress dst_address, KProcessAdd TraversalContext context; TraversalEntry next_entry; bool traverse_valid = - BeginTraversal(impl, std::addressof(next_entry), std::addressof(context), dst_address); + impl.BeginTraversal(std::addressof(next_entry), std::addressof(context), dst_address); R_UNLESS(traverse_valid, ResultInvalidCurrentMemory); // Prepare tracking variables. @@ -3297,7 +3267,7 @@ Result KPageTableBase::WriteDebugMemory(KProcessAddress dst_address, KProcessAdd while (tot_size < size) { // Continue the traversal. traverse_valid = - ContinueTraversal(impl, std::addressof(next_entry), std::addressof(context)); + impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context)); ASSERT(traverse_valid); if (next_entry.phys_addr != (cur_addr + cur_size)) { @@ -3758,7 +3728,7 @@ Result KPageTableBase::CopyMemoryFromLinearToUser( TraversalContext context; TraversalEntry next_entry; bool traverse_valid = - BeginTraversal(impl, std::addressof(next_entry), std::addressof(context), src_addr); + impl.BeginTraversal(std::addressof(next_entry), std::addressof(context), src_addr); ASSERT(traverse_valid); // Prepare tracking variables. @@ -3798,7 +3768,7 @@ Result KPageTableBase::CopyMemoryFromLinearToUser( while (tot_size < size) { // Continue the traversal. traverse_valid = - ContinueTraversal(impl, std::addressof(next_entry), std::addressof(context)); + impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context)); ASSERT(traverse_valid); if (next_entry.phys_addr != (cur_addr + cur_size)) { @@ -3852,7 +3822,7 @@ Result KPageTableBase::CopyMemoryFromLinearToKernel( TraversalContext context; TraversalEntry next_entry; bool traverse_valid = - BeginTraversal(impl, std::addressof(next_entry), std::addressof(context), src_addr); + impl.BeginTraversal(std::addressof(next_entry), std::addressof(context), src_addr); ASSERT(traverse_valid); // Prepare tracking variables. @@ -3875,7 +3845,7 @@ Result KPageTableBase::CopyMemoryFromLinearToKernel( while (tot_size < size) { // Continue the traversal. traverse_valid = - ContinueTraversal(impl, std::addressof(next_entry), std::addressof(context)); + impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context)); ASSERT(traverse_valid); if (next_entry.phys_addr != (cur_addr + cur_size)) { @@ -3932,7 +3902,7 @@ Result KPageTableBase::CopyMemoryFromUserToLinear( TraversalContext context; TraversalEntry next_entry; bool traverse_valid = - BeginTraversal(impl, std::addressof(next_entry), std::addressof(context), dst_addr); + impl.BeginTraversal(std::addressof(next_entry), std::addressof(context), dst_addr); ASSERT(traverse_valid); // Prepare tracking variables. @@ -3971,7 +3941,7 @@ Result KPageTableBase::CopyMemoryFromUserToLinear( while (tot_size < size) { // Continue the traversal. traverse_valid = - ContinueTraversal(impl, std::addressof(next_entry), std::addressof(context)); + impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context)); ASSERT(traverse_valid); if (next_entry.phys_addr != (cur_addr + cur_size)) { @@ -4027,7 +3997,7 @@ Result KPageTableBase::CopyMemoryFromKernelToLinear(KProcessAddress dst_addr, si TraversalContext context; TraversalEntry next_entry; bool traverse_valid = - BeginTraversal(impl, std::addressof(next_entry), std::addressof(context), dst_addr); + impl.BeginTraversal(std::addressof(next_entry), std::addressof(context), dst_addr); ASSERT(traverse_valid); // Prepare tracking variables. @@ -4050,7 +4020,7 @@ Result KPageTableBase::CopyMemoryFromKernelToLinear(KProcessAddress dst_addr, si while (tot_size < size) { // Continue the traversal. traverse_valid = - ContinueTraversal(impl, std::addressof(next_entry), std::addressof(context)); + impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context)); ASSERT(traverse_valid); if (next_entry.phys_addr != (cur_addr + cur_size)) { @@ -4119,10 +4089,10 @@ Result KPageTableBase::CopyMemoryFromHeapToHeap( bool traverse_valid; // Begin traversal. - traverse_valid = BeginTraversal(src_impl, std::addressof(src_next_entry), + traverse_valid = src_impl.BeginTraversal(std::addressof(src_next_entry), std::addressof(src_context), src_addr); ASSERT(traverse_valid); - traverse_valid = BeginTraversal(dst_impl, std::addressof(dst_next_entry), + traverse_valid = dst_impl.BeginTraversal(std::addressof(dst_next_entry), std::addressof(dst_context), dst_addr); ASSERT(traverse_valid); @@ -4157,7 +4127,7 @@ Result KPageTableBase::CopyMemoryFromHeapToHeap( if (ofs + cur_copy_size != size) { if (cur_src_addr + cur_min_size == cur_src_block_addr + cur_src_size) { // Continue the src traversal. - traverse_valid = ContinueTraversal(src_impl, std::addressof(src_next_entry), + traverse_valid = src_impl.ContinueTraversal(std::addressof(src_next_entry), std::addressof(src_context)); ASSERT(traverse_valid); @@ -4168,7 +4138,7 @@ Result KPageTableBase::CopyMemoryFromHeapToHeap( if (cur_dst_addr + cur_min_size == dst_next_entry.phys_addr + dst_next_entry.block_size) { // Continue the dst traversal. - traverse_valid = ContinueTraversal(dst_impl, std::addressof(dst_next_entry), + traverse_valid = dst_impl.ContinueTraversal(std::addressof(dst_next_entry), std::addressof(dst_context)); ASSERT(traverse_valid); @@ -4253,10 +4223,10 @@ Result KPageTableBase::CopyMemoryFromHeapToHeapWithoutCheckDestination( bool traverse_valid; // Begin traversal. - traverse_valid = BeginTraversal(src_impl, std::addressof(src_next_entry), + traverse_valid = src_impl.BeginTraversal(std::addressof(src_next_entry), std::addressof(src_context), src_addr); ASSERT(traverse_valid); - traverse_valid = BeginTraversal(dst_impl, std::addressof(dst_next_entry), + traverse_valid = dst_impl.BeginTraversal(std::addressof(dst_next_entry), std::addressof(dst_context), dst_addr); ASSERT(traverse_valid); @@ -4291,7 +4261,7 @@ Result KPageTableBase::CopyMemoryFromHeapToHeapWithoutCheckDestination( if (ofs + cur_copy_size != size) { if (cur_src_addr + cur_min_size == cur_src_block_addr + cur_src_size) { // Continue the src traversal. - traverse_valid = ContinueTraversal(src_impl, std::addressof(src_next_entry), + traverse_valid = src_impl.ContinueTraversal(std::addressof(src_next_entry), std::addressof(src_context)); ASSERT(traverse_valid); @@ -4302,7 +4272,7 @@ Result KPageTableBase::CopyMemoryFromHeapToHeapWithoutCheckDestination( if (cur_dst_addr + cur_min_size == dst_next_entry.phys_addr + dst_next_entry.block_size) { // Continue the dst traversal. - traverse_valid = ContinueTraversal(dst_impl, std::addressof(dst_next_entry), + traverse_valid = dst_impl.ContinueTraversal(std::addressof(dst_next_entry), std::addressof(dst_context)); ASSERT(traverse_valid); @@ -4577,7 +4547,7 @@ Result KPageTableBase::SetupForIpcServer(KProcessAddress* out_addr, size_t size, // Begin traversal. TraversalContext context; TraversalEntry next_entry; - bool traverse_valid = BeginTraversal(src_impl, std::addressof(next_entry), + bool traverse_valid = src_impl.BeginTraversal(std::addressof(next_entry), std::addressof(context), aligned_src_start); ASSERT(traverse_valid); @@ -4627,7 +4597,7 @@ Result KPageTableBase::SetupForIpcServer(KProcessAddress* out_addr, size_t size, // If the block's size was one page, we may need to continue traversal. if (cur_block_size == 0 && aligned_src_size > PageSize) { traverse_valid = - ContinueTraversal(src_impl, std::addressof(next_entry), std::addressof(context)); + src_impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context)); ASSERT(traverse_valid); cur_block_addr = next_entry.phys_addr; @@ -4640,7 +4610,7 @@ Result KPageTableBase::SetupForIpcServer(KProcessAddress* out_addr, size_t size, while (aligned_src_start + tot_block_size < mapping_src_end) { // Continue the traversal. traverse_valid = - ContinueTraversal(src_impl, std::addressof(next_entry), std::addressof(context)); + src_impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context)); ASSERT(traverse_valid); // Process the block. @@ -4683,7 +4653,7 @@ Result KPageTableBase::SetupForIpcServer(KProcessAddress* out_addr, size_t size, if (mapped_block_end + cur_block_size < aligned_src_end && cur_block_size == last_block_size) { traverse_valid = - ContinueTraversal(src_impl, std::addressof(next_entry), std::addressof(context)); + src_impl.ContinueTraversal(std::addressof(next_entry), std::addressof(context)); ASSERT(traverse_valid); cur_block_addr = next_entry.phys_addr; @@ -5631,7 +5601,7 @@ Result KPageTableBase::UnmapProcessMemory(KProcessAddress dst_address, size_t si ContiguousRangeInfo(KPageTableBase& pt, KProcessAddress address, size_t size) : m_pt(pt), m_remaining_size(size) { // Begin a traversal. - ASSERT(m_pt.BeginTraversal(m_pt.GetImpl(), std::addressof(m_entry), + ASSERT(m_pt.GetImpl().BeginTraversal(std::addressof(m_entry), std::addressof(m_context), address)); // Setup tracking fields. @@ -5662,7 +5632,7 @@ Result KPageTableBase::UnmapProcessMemory(KProcessAddress dst_address, size_t si void DetermineContiguousBlockExtents() { // Continue traversing until we're not contiguous, or we have enough. while (m_cur_size < m_remaining_size) { - ASSERT(m_pt.ContinueTraversal(m_pt.GetImpl(), std::addressof(m_entry), + ASSERT(m_pt.GetImpl().ContinueTraversal(std::addressof(m_entry), std::addressof(m_context))); // If we're not contiguous, we're done. diff --git a/src/core/hle/kernel/k_page_table_base.h b/src/core/hle/kernel/k_page_table_base.h index 0c222910fc..1d4fa85f04 100644 --- a/src/core/hle/kernel/k_page_table_base.h +++ b/src/core/hle/kernel/k_page_table_base.h @@ -370,10 +370,6 @@ private: size_t num_pages, size_t alignment, size_t offset, size_t guard_pages) const; - bool BeginTraversal(const Common::PageTable& impl, TraversalEntry* out_entry, TraversalContext* out_context, - Common::ProcessAddress address) const; - bool ContinueTraversal(const Common::PageTable& impl, TraversalEntry* out_entry, TraversalContext* context) const; - Result CheckMemoryStateContiguous(size_t* out_blocks_needed, KProcessAddress addr, size_t size, KMemoryState state_mask, KMemoryState state, KMemoryPermission perm_mask, KMemoryPermission perm, @@ -478,14 +474,7 @@ private: // Validate pre-conditions. ASSERT(this->IsLockedByCurrentThread()); - if (virt_addr > (1ULL << m_address_space_width)) { - return false; - } - - *out = m_system.DeviceMemory().GetPhysicalAddr( - this->GetImpl().entries[GetInteger(virt_addr) >> PageBits].Pointer(true) + GetInteger(virt_addr)); - - return true; + return this->GetImpl().GetPhysicalAddress(out, virt_addr); } public: diff --git a/src/core/memory.cpp b/src/core/memory.cpp index 837016de1d..a5e6b8592c 100644 --- a/src/core/memory.cpp +++ b/src/core/memory.cpp @@ -101,10 +101,8 @@ struct Memory::Impl { } u64 protect_bytes = 0, protect_begin = 0; - - current_page_table->entries.CommitRegion(vaddr >> YUZU_PAGEBITS, (vaddr + size) >> YUZU_PAGEBITS); for (u64 addr = vaddr; addr < vaddr + size; addr += YUZU_PAGESIZE) { - const Common::PageType page_type = current_page_table->entries.GetUnchecked(addr >> YUZU_PAGEBITS).Type(); + const Common::PageType page_type = current_page_table->entries[addr >> YUZU_PAGEBITS].ptr.Type(); switch (page_type) { case Common::PageType::RasterizerCachedMemory: if (protect_bytes > 0) { @@ -125,14 +123,16 @@ struct Memory::Impl { } [[nodiscard]] u8* GetPointerFromRasterizerCachedMemory(u64 vaddr) const { - if (u64 paddr = current_page_table->entries[vaddr >> YUZU_PAGEBITS].Pointer(true); paddr) - return reinterpret_cast(paddr) + vaddr; + Common::PhysicalAddress const paddr = current_page_table->entries[vaddr >> YUZU_PAGEBITS].addr; + if (paddr) + return system.DeviceMemory().GetPointer(paddr + vaddr); return {}; } [[nodiscard]] u8* GetPointerFromDebugMemory(u64 vaddr) const { - if (u64 paddr = current_page_table->entries[vaddr >> YUZU_PAGEBITS].Pointer(true); paddr) - return reinterpret_cast(paddr) + vaddr; + const Common::PhysicalAddress paddr = current_page_table->entries[vaddr >> YUZU_PAGEBITS].addr; + if (paddr != 0) + return system.DeviceMemory().GetPointer(paddr + vaddr); return {}; } @@ -243,12 +243,10 @@ struct Memory::Impl { std::size_t page_index = addr >> YUZU_PAGEBITS; std::size_t page_offset = addr & YUZU_PAGEMASK; bool user_accessible = true; - - current_page_table->entries.CommitRegion(page_index, page_index + (size >> YUZU_PAGEBITS) + 1); while (remaining_size != 0) { const std::size_t copy_amount = (std::min)(std::size_t(YUZU_PAGESIZE) - page_offset, remaining_size); const auto current_vaddr = u64((page_index << YUZU_PAGEBITS) + page_offset); - const auto [pointer, type, _] = current_page_table->entries.GetUnchecked(page_index).PointerTypeBlock(); + const auto [pointer, type] = current_page_table->entries[page_index].ptr.PointerType(); switch (type) { case Common::PageType::Unmapped: { user_accessible = false; @@ -299,10 +297,10 @@ struct Memory::Impl { } [[nodiscard]] inline const u8* GetSpan(const VAddr addr, const std::size_t size) const noexcept { - return (current_page_table->entries[addr >> YUZU_PAGEBITS].Block() == current_page_table->entries[(addr + size) >> YUZU_PAGEBITS].Block()) ? GetPointerSilent(addr) : nullptr; + return (current_page_table->entries[addr >> YUZU_PAGEBITS].block == current_page_table->entries[(addr + size) >> YUZU_PAGEBITS].block) ? GetPointerSilent(addr) : nullptr; } [[nodiscard]] inline u8* GetSpan(const VAddr addr, const std::size_t size) noexcept { - return (current_page_table->entries[addr >> YUZU_PAGEBITS].Block() == current_page_table->entries[(addr + size) >> YUZU_PAGEBITS].Block()) ? GetPointerSilent(addr) : nullptr; + return (current_page_table->entries[addr >> YUZU_PAGEBITS].block == current_page_table->entries[(addr + size) >> YUZU_PAGEBITS].block) ? GetPointerSilent(addr) : nullptr; } bool WriteBlockImpl(const Common::ProcessAddress addr, const void* buffer, const std::size_t size, bool unsafe) { @@ -406,14 +404,11 @@ struct Memory::Impl { // The region is at a granularity of CPU pages. const u64 num_pages = ((vaddr + size - 1) >> YUZU_PAGEBITS) - (vaddr >> YUZU_PAGEBITS) + 1; - - current_page_table->entries.CommitRegion(vaddr >> YUZU_PAGEBITS, (vaddr >> YUZU_PAGEBITS) + num_pages); for (u64 i = 0; i < num_pages; ++i, vaddr += YUZU_PAGESIZE) { - auto& entry = current_page_table->entries.GetUnchecked(vaddr >> YUZU_PAGEBITS); - const auto [pointer, type, block] = entry.PointerTypeBlock(true); + const Common::PageType page_type = current_page_table->entries[vaddr >> YUZU_PAGEBITS].ptr.Type(); if (debug) { // Switch page type to debug if now debug - switch (type) { + switch (page_type) { case Common::PageType::Unmapped: ASSERT(false && "Attempted to mark unmapped pages as debug"); break; @@ -422,14 +417,14 @@ struct Memory::Impl { // Page is already marked. break; case Common::PageType::Memory: - entry.MarkDebug(pointer, block); + current_page_table->entries[vaddr >> YUZU_PAGEBITS].ptr.Store(0, Common::PageType::DebugMemory); break; default: UNREACHABLE(); } } else { // Switch page type to non-debug if now non-debug - switch (type) { + switch (page_type) { case Common::PageType::Unmapped: ASSERT(false && "Attempted to mark unmapped pages as non-debug"); break; @@ -438,7 +433,8 @@ struct Memory::Impl { // Don't mess with already non-debug or rasterizer memory. break; case Common::PageType::DebugMemory: { - entry.Store(false, Common::PageType::Memory, block, pointer); + u8* const pointer = GetPointerFromDebugMemory(vaddr & ~YUZU_PAGEMASK); + current_page_table->entries[vaddr >> YUZU_PAGEBITS].ptr.Store(uintptr_t(pointer) - (vaddr & ~YUZU_PAGEMASK), Common::PageType::Memory); break; } default: @@ -470,10 +466,8 @@ struct Memory::Impl { // is different). This assumes the specified GPU address region is contiguous as well. const u64 num_pages = ((vaddr + size - 1) >> YUZU_PAGEBITS) - (vaddr >> YUZU_PAGEBITS) + 1; - current_page_table->entries.CommitRegion(vaddr >> YUZU_PAGEBITS, (vaddr >> YUZU_PAGEBITS) + num_pages); for (u64 i = 0; i < num_pages; ++i, vaddr += YUZU_PAGESIZE) { - auto& entry = current_page_table->entries.GetUnchecked(vaddr >> YUZU_PAGEBITS); - const Common::PageType page_type = entry.Type(); + const Common::PageType page_type= current_page_table->entries[vaddr >> YUZU_PAGEBITS].ptr.Type(); if (cached) { // Switch page type to cached if now cached switch (page_type) { @@ -483,7 +477,7 @@ struct Memory::Impl { break; case Common::PageType::DebugMemory: case Common::PageType::Memory: - entry.MarkRasterizerCached(); + current_page_table->entries[vaddr >> YUZU_PAGEBITS].ptr.Store(0, Common::PageType::RasterizerCachedMemory); break; case Common::PageType::RasterizerCachedMemory: // There can be more than one GPU region mapped per CPU region, so it's common @@ -505,13 +499,13 @@ struct Memory::Impl { // that this area is already unmarked as cached. break; case Common::PageType::RasterizerCachedMemory: { - if (auto [ptr, _, block] = entry.PointerTypeBlock(true); ptr == 0) { + if (u8* const pointer = GetPointerFromRasterizerCachedMemory(vaddr & ~YUZU_PAGEMASK); pointer == nullptr) { // It's possible that this function has been called while updating the // pagetable after unmapping a VMA. In that case the underlying VMA will no // longer exist, and we should just leave the pagetable entry blank. - entry.Store(false, Common::PageType::Unmapped, block, 0); + current_page_table->entries[vaddr >> YUZU_PAGEBITS].ptr.Store(0, Common::PageType::Unmapped); } else { - entry.Store(false, Common::PageType::Memory, block, ptr); + current_page_table->entries[vaddr >> YUZU_PAGEBITS].ptr.Store(uintptr_t(pointer) - (vaddr & ~YUZU_PAGEMASK), Common::PageType::Memory); } break; } @@ -545,18 +539,22 @@ struct Memory::Impl { ASSERT_MSG(type != Common::PageType::Memory, "Mapping memory page without a pointer @ {:016x}", base * YUZU_PAGESIZE); - page_table.entries.ZeroRegion(base, end); - } else { - auto current_block = block_count.fetch_add(1, std::memory_order_relaxed); - ASSERT(current_block != 65535); - - page_table.entries.CommitRegion(base, end); while (base != end) { - auto host_ptr = reinterpret_cast(system.DeviceMemory().GetPointer(target)) - (base << YUZU_PAGEBITS);; - auto& entry = page_table.entries.GetUnchecked(base); + page_table.entries[base].ptr.Store(0, type); + page_table.entries[base].addr = 0; + page_table.entries[base].block = 0; + base += 1; + } + } else { + auto orig_base = base; + while (base != end) { + auto host_ptr = uintptr_t(system.DeviceMemory().GetPointer(target)) - (base << YUZU_PAGEBITS); + auto backing = GetInteger(target) - (base << YUZU_PAGEBITS); + page_table.entries[base].ptr.Store(host_ptr, type); + page_table.entries[base].addr = backing; + page_table.entries[base].block = orig_base << YUZU_PAGEBITS; - entry.Store(false, type, current_block, host_ptr); - ASSERT_MSG(page_table.entries[base].Pointer(), + ASSERT_MSG(page_table.entries[base].ptr.Pointer(), "memory mapping base yield a nullptr within the table"); base += 1; @@ -571,11 +569,11 @@ struct Memory::Impl { vaddr &= 0xffffffffffffULL; if (AddressSpaceContains(*current_page_table, vaddr, 1)) [[likely]] { // Avoid adding any extra logic to this fast-path block - const auto raw = current_page_table->entries[vaddr >> YUZU_PAGEBITS].Raw(); - if (auto pointer = Common::PageTable::PageEntryData::ExtractPointer(raw); pointer) [[likely]] { + const uintptr_t raw_pointer = current_page_table->entries[vaddr >> YUZU_PAGEBITS].ptr.Raw(); + if (const uintptr_t pointer = Common::PageTable::PageInfo::ExtractPointer(raw_pointer)) [[likely]] { return reinterpret_cast(pointer + vaddr); } else { - switch (static_cast(raw.type)) { + switch (Common::PageTable::PageInfo::ExtractType(raw_pointer)) { case Common::PageType::Memory: ASSERT_MSG(false, "Mapped memory page without a pointer @ {:#016x}", vaddr); return nullptr; @@ -775,7 +773,6 @@ struct Memory::Impl { #else Common::HostMemory* host_buffer{}; #endif - std::atomic block_count = 0; }; Memory::Memory(Core::System& system_) : system{system_} { @@ -814,7 +811,7 @@ bool Memory::IsValidVirtualAddress(const Common::ProcessAddress vaddr) const { if (page >= page_table.entries.size()) { return false; } - const auto [pointer, type, _] = page_table.entries[page].PointerTypeBlock(); + const auto [pointer, type] = page_table.entries[page].ptr.PointerType(); return pointer != 0 || type == Common::PageType::RasterizerCachedMemory || type == Common::PageType::DebugMemory; } diff --git a/src/dynarmic/src/dynarmic/backend/arm64/a32_address_space.cpp b/src/dynarmic/src/dynarmic/backend/arm64/a32_address_space.cpp index 8b6ec2506e..da51220e9c 100644 --- a/src/dynarmic/src/dynarmic/backend/arm64/a32_address_space.cpp +++ b/src/dynarmic/src/dynarmic/backend/arm64/a32_address_space.cpp @@ -371,10 +371,8 @@ EmitConfig A32AddressSpace::GetEmitConfig() { .page_table_pointer = std::bit_cast(conf.page_table), .page_table_address_space_bits = 32, - .page_table_pointer_mask = conf.page_table_pointer_mask, + .page_table_pointer_mask_bits = conf.page_table_pointer_mask_bits, .page_table_log2_stride = conf.page_table_log2_stride, - .page_table_marked_bit = conf.page_table_marked_bit, - .page_table_sign_extension = conf.page_table_sign_extension, .silently_mirror_page_table = true, .absolute_offset_page_table = conf.absolute_offset_page_table, .detect_misaligned_access_via_page_table = conf.detect_misaligned_access_via_page_table, diff --git a/src/dynarmic/src/dynarmic/backend/arm64/a64_address_space.cpp b/src/dynarmic/src/dynarmic/backend/arm64/a64_address_space.cpp index 2dbb7d2e1d..2c71ffe282 100644 --- a/src/dynarmic/src/dynarmic/backend/arm64/a64_address_space.cpp +++ b/src/dynarmic/src/dynarmic/backend/arm64/a64_address_space.cpp @@ -545,10 +545,8 @@ EmitConfig A64AddressSpace::GetEmitConfig() { .page_table_pointer = std::bit_cast(conf.page_table), .page_table_address_space_bits = conf.page_table_address_space_bits, - .page_table_pointer_mask = conf.page_table_pointer_mask, + .page_table_pointer_mask_bits = conf.page_table_pointer_mask_bits, .page_table_log2_stride = conf.page_table_log2_stride, - .page_table_marked_bit = conf.page_table_marked_bit, - .page_table_sign_extension = conf.page_table_sign_extension, .silently_mirror_page_table = conf.silently_mirror_page_table, .absolute_offset_page_table = conf.absolute_offset_page_table, .detect_misaligned_access_via_page_table = conf.detect_misaligned_access_via_page_table, diff --git a/src/dynarmic/src/dynarmic/backend/arm64/emit_arm64.h b/src/dynarmic/src/dynarmic/backend/arm64/emit_arm64.h index 693f71ea39..f2c89bbfc0 100644 --- a/src/dynarmic/src/dynarmic/backend/arm64/emit_arm64.h +++ b/src/dynarmic/src/dynarmic/backend/arm64/emit_arm64.h @@ -128,10 +128,8 @@ struct EmitConfig { // Page table u64 page_table_pointer; std::size_t page_table_address_space_bits; - u64 page_table_pointer_mask; + int page_table_pointer_mask_bits; std::size_t page_table_log2_stride; - std::optional page_table_marked_bit; - std::optional page_table_sign_extension; bool silently_mirror_page_table; bool absolute_offset_page_table; u8 detect_misaligned_access_via_page_table; diff --git a/src/dynarmic/src/dynarmic/backend/arm64/emit_arm64_memory.cpp b/src/dynarmic/src/dynarmic/backend/arm64/emit_arm64_memory.cpp index 170fcf9613..143244d60b 100644 --- a/src/dynarmic/src/dynarmic/backend/arm64/emit_arm64_memory.cpp +++ b/src/dynarmic/src/dynarmic/backend/arm64/emit_arm64_memory.cpp @@ -273,18 +273,9 @@ std::pair InlinePageTableEmitVAddrLookup(oaknut::Cod // load x0 = *<(u8*)pagetable + index> code.LDR(Xscratch0, Xpagetable, Xscratch0); - if (ctx.conf.page_table_marked_bit) { - code.TST(Xscratch0, 1ULL << *ctx.conf.page_table_marked_bit); - code.B(NE, *fallback); - } - - if (ctx.conf.page_table_pointer_mask != 0) { - code.AND(Xscratch0, Xscratch0, ctx.conf.page_table_pointer_mask); - } - - // TODO: combine this with page_table_pointer_mask - if (ctx.conf.page_table_sign_extension) { - code.SBFM(Xscratch0, Xscratch0, 0, *ctx.conf.page_table_sign_extension); + if (ctx.conf.page_table_pointer_mask_bits != 0) { + const u64 mask = u64(~u64(0)) << ctx.conf.page_table_pointer_mask_bits; + code.AND(Xscratch0, Xscratch0, mask); } code.CBZ(Xscratch0, *fallback); diff --git a/src/dynarmic/src/dynarmic/backend/x64/emit_x64_memory.h b/src/dynarmic/src/dynarmic/backend/x64/emit_x64_memory.h index 54ad638ada..3ac078f1d7 100644 --- a/src/dynarmic/src/dynarmic/backend/x64/emit_x64_memory.h +++ b/src/dynarmic/src/dynarmic/backend/x64/emit_x64_memory.h @@ -9,7 +9,6 @@ #pragma once #include -#include #include "dynarmic/backend/x64/xbyak.h" #include "dynarmic/backend/x64/a32_emit_x64.h" @@ -79,46 +78,27 @@ Xbyak::RegExp EmitVAddrLookup(BlockOfCode& code, EmitContext& ctx, size_t bitsiz template<> [[maybe_unused]] Xbyak::RegExp EmitVAddrLookup(BlockOfCode& code, A32EmitContext& ctx, size_t bitsize, Xbyak::Label& abort, Xbyak::Reg64 vaddr) { const Xbyak::Reg64 page = ctx.reg_alloc.ScratchGpr(code); - const Xbyak::Reg64 tmp = ctx.conf.absolute_offset_page_table && ctx.conf.page_table_pointer_mask == 0 ? page : ctx.reg_alloc.ScratchGpr(code); + const Xbyak::Reg32 tmp = ctx.conf.absolute_offset_page_table ? page.cvt32() : ctx.reg_alloc.ScratchGpr(code).cvt32(); - EmitDetectMisalignedVAddr(code, ctx, bitsize, abort, vaddr, tmp); + EmitDetectMisalignedVAddr(code, ctx, bitsize, abort, vaddr, tmp.cvt64()); - code.mov(tmp, vaddr); + // TODO: This code assumes vaddr has been zext from 32-bits to 64-bits. + + code.mov(tmp, vaddr.cvt32()); code.shr(tmp, int(page_table_const_bits)); - - if (ctx.conf.page_table_log2_stride > 3) { - code.shl(tmp, int(ctx.conf.page_table_log2_stride)); - code.mov(page, qword[r14 + tmp.cvt64()]); - } else { - code.mov(page, qword[r14 + tmp.cvt64() * int(ctx.conf.page_table_log2_stride)]); - } - - // check for marked bit, use as unmapped if marked - if (ctx.conf.page_table_marked_bit) { - code.bt(page, *ctx.conf.page_table_marked_bit); - code.jc(abort, code.T_NEAR); - } - // mask away attributes - if (ctx.conf.page_table_pointer_mask == 0) { + code.shl(tmp, int(ctx.conf.page_table_log2_stride)); + code.mov(page, qword[r14 + tmp.cvt64()]); + if (ctx.conf.page_table_pointer_mask_bits == 0) { code.test(page, page); - } else if (std::in_range(ctx.conf.page_table_pointer_mask)) { - code.and_(page, ctx.conf.page_table_pointer_mask); } else { - code.mov(tmp, ctx.conf.page_table_pointer_mask); - code.and_(page, tmp); + code.and_(page, ~u32(0) << ctx.conf.page_table_pointer_mask_bits); } - // check for sign bit, apply sign extension as needed - if (ctx.conf.page_table_sign_extension) { - code.shl(page, 63 - int(*ctx.conf.page_table_sign_extension)); - code.sar(page, 63 - int(*ctx.conf.page_table_sign_extension)); - } - code.jz(abort, code.T_NEAR); if (ctx.conf.absolute_offset_page_table) { return page + vaddr; } - code.mov(tmp, vaddr); - code.and_(tmp, u32(page_table_const_mask)); + code.mov(tmp, vaddr.cvt32()); + code.and_(tmp, static_cast(page_table_const_mask)); return page + tmp.cvt64(); } @@ -128,7 +108,7 @@ template<> const size_t unused_top_bits = 64 - ctx.conf.page_table_address_space_bits; const Xbyak::Reg64 page = ctx.reg_alloc.ScratchGpr(code); - const Xbyak::Reg64 tmp = ctx.conf.absolute_offset_page_table && ctx.conf.page_table_pointer_mask == 0 ? page : ctx.reg_alloc.ScratchGpr(code); + const Xbyak::Reg64 tmp = ctx.conf.absolute_offset_page_table ? page : ctx.reg_alloc.ScratchGpr(code); EmitDetectMisalignedVAddr(code, ctx, bitsize, abort, vaddr, tmp); @@ -163,26 +143,11 @@ template<> code.shl(tmp, int(ctx.conf.page_table_log2_stride)); code.mov(page, qword[r14 + tmp]); - - // check for marked bit, use as unmapped if marked - if (ctx.conf.page_table_marked_bit) { - code.bt(page, *ctx.conf.page_table_marked_bit); - code.jc(abort, code.T_NEAR); - } - // mask away attributes - if (ctx.conf.page_table_pointer_mask == 0) { + if (ctx.conf.page_table_pointer_mask_bits == 0) { code.test(page, page); - } else if (std::in_range(ctx.conf.page_table_pointer_mask)) { - code.and_(page, ctx.conf.page_table_pointer_mask); } else { - code.mov(tmp, ctx.conf.page_table_pointer_mask); - code.and_(page, tmp); + code.and_(page, ~u32(0) << ctx.conf.page_table_pointer_mask_bits); } - if (ctx.conf.page_table_sign_extension) { - code.shl(page, *ctx.conf.page_table_sign_extension); - code.sar(page, *ctx.conf.page_table_sign_extension); - } - code.jz(abort, code.T_NEAR); if (ctx.conf.absolute_offset_page_table) { return page + vaddr; diff --git a/src/dynarmic/src/dynarmic/interface/A32/config.h b/src/dynarmic/src/dynarmic/interface/A32/config.h index b93b116164..5a97fb69f3 100644 --- a/src/dynarmic/src/dynarmic/interface/A32/config.h +++ b/src/dynarmic/src/dynarmic/interface/A32/config.h @@ -159,23 +159,14 @@ struct UserConfig { /// Maximum size is limited by the maximum length of a x86_64 / arm64 jump. std::uint32_t code_cache_size = 128 * 1024 * 1024; // bytes - /// Applies a bit mask to the bits in host pointers from the page table. + /// Masks out the first N bits in host pointers from the page table. /// The intention behind this is to allow users of Dynarmic to pack attributes in the /// same integer and update the pointer attribute pair atomically. - /// If the configured value is ~(0b111ULL), all pointers will be forcefully aligned to 8 bytes. - std::uint64_t page_table_pointer_mask = 0; + /// If the configured value is 3, all pointers will be forcefully aligned to 8 bytes. + std::int32_t page_table_pointer_mask_bits = 0; - /// Log2 of the size per page entry, value should be either 3 or 4 - std::uint32_t page_table_log2_stride = 3; - - /// Setting this value has Dynarmic check the specified bit of the page pointer provided by page table. - /// If the bit is set to 1, Dynarmic will treat it as unmapped. - /// This bit should be included as part of `page_table_pointer_mask_bits`. - std::optional page_table_marked_bit = std::nullopt; - - /// If this value is set, Dynarmic will sign extend the page table pointer by this bit. - /// Useful for compacting bits into the page table and should be used as part of `page_table_pointer_mask`. - std::optional page_table_sign_extension = std::nullopt; + // Log2 of the size per page entry, value should be either 3 or 4 + std::size_t page_table_log2_stride = 3; /// Select the architecture version to use. /// There are minor behavioural differences between versions. diff --git a/src/dynarmic/src/dynarmic/interface/A64/config.h b/src/dynarmic/src/dynarmic/interface/A64/config.h index 818aed73e7..83c1593fd8 100644 --- a/src/dynarmic/src/dynarmic/interface/A64/config.h +++ b/src/dynarmic/src/dynarmic/interface/A64/config.h @@ -173,23 +173,14 @@ struct UserConfig { /// This is only used if page_table is not nullptr. std::uint32_t page_table_address_space_bits = 36; - /// Applies a bit mask to the bits in host pointers from the page table. + /// Masks out the first N bits in host pointers from the page table. /// The intention behind this is to allow users of Dynarmic to pack attributes in the /// same integer and update the pointer attribute pair atomically. - /// If the configured value is ~(0b111ULL), all pointers will be forcefully aligned to 8 bytes. - std::uint64_t page_table_pointer_mask = 0; + /// If the configured value is 3, all pointers will be forcefully aligned to 8 bytes. + std::int32_t page_table_pointer_mask_bits = 0; - /// Log2 of the size per page entry, value should be either 3 or 4 - std::uint32_t page_table_log2_stride = 3; - - /// Setting this value has Dynarmic check the specified bit of the page pointer provided by page table. - /// If the bit is set to 1, Dynarmic will treat it as unmapped. - /// This bit should be included as part of `page_table_pointer_mask`. - std::optional page_table_marked_bit = std::nullopt; - - /// If this value is set, Dynarmic will sign extend the page table pointer by this bit. - /// Useful for compacting bits into the page table and should be used as part of `page_table_pointer_mask`. - std::optional page_table_sign_extension = std::nullopt; + // Log2 of the size per page entry, value should be either 3 or 4 + std::size_t page_table_log2_stride = 3; /// Counter-timer frequency register. The value of the register is not interpreted by /// dynarmic. diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp index d5838a387d..b9b136f6b8 100644 --- a/src/video_core/memory_manager.cpp +++ b/src/video_core/memory_manager.cpp @@ -46,8 +46,8 @@ MemoryManager::MemoryManager(Core::System& system_, MaxwellDeviceMemoryManager& page_table_mask = page_table_size - 1; big_page_table_mask = big_page_table_size - 1; - big_page_table_dev.ResizeAndClear(big_page_table_size); big_entries.resize(big_page_table_size / 32, 0); + big_page_table_dev.resize(big_page_table_size); big_page_continuous.resize(big_page_table_size / continuous_bits, 0); entries.resize(page_table_size / 32, 0); } @@ -143,7 +143,7 @@ GPUVAddr MemoryManager::BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr const DAddr current_dev_addr = dev_addr + offset; const auto index = PageEntryIndex(current_gpu_addr, true); const u32 sub_value = static_cast(current_dev_addr >> cpu_page_bits); - big_page_table_dev.Set(index, sub_value); + big_page_table_dev[index] = sub_value; const bool is_continuous = ([&] { uintptr_t base_ptr{ reinterpret_cast(memory.GetPointer(current_dev_addr))}; diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h index 08978567d2..f9fddd177a 100644 --- a/src/video_core/memory_manager.h +++ b/src/video_core/memory_manager.h @@ -17,7 +17,7 @@ #include "common/multi_level_page_table.h" #include "common/range_map.h" #include "common/scratch_buffer.h" -#include "common/sparse_large_vector.h" +#include "common/virtual_buffer.h" #include "video_core/invalidation_accumulator.h" #include "video_core/cache_types.h" #include "video_core/host1x/gpu_device_memory_manager.h" @@ -214,7 +214,7 @@ private: Common::MultiLevelPageTable page_table; Common::RangeMap kind_map; - Common::SparseLargeVector big_page_table_dev; + Common::VirtualBuffer big_page_table_dev; std::vector big_page_continuous; boost::container::small_vector, 32> page_stash{}; From 36b06254b48511e99a907aede600cb069f69bb30 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sun, 20 Sep 2026 15:50:04 -0400 Subject: [PATCH 44/69] Fix licenses --- src/common/page_table.cpp | 2 +- src/core/device_memory.h | 3 +++ 2 files changed, 4 insertions(+), 1 deletion(-) diff --git a/src/common/page_table.cpp b/src/common/page_table.cpp index d34ba89993..33fff39002 100644 --- a/src/common/page_table.cpp +++ b/src/common/page_table.cpp @@ -1,4 +1,4 @@ -// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project // SPDX-License-Identifier: GPL-3.0-or-later // SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project diff --git a/src/core/device_memory.h b/src/core/device_memory.h index 11bf0e3268..7603e94698 100644 --- a/src/core/device_memory.h +++ b/src/core/device_memory.h @@ -1,3 +1,6 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later From 8d8e67cd7fd657bc461fe00fc4fce25782178f00 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sun, 20 Sep 2026 18:08:27 -0400 Subject: [PATCH 45/69] Just an small experiment to simply interpreter --- src/core/CMakeLists.txt | 3 +- src/core/arm/nce/interpreter_visitor.cpp | 920 ++----- src/core/arm/nce/interpreter_visitor.h | 136 +- src/core/arm/nce/visitor_base.h | 2783 ---------------------- 4 files changed, 323 insertions(+), 3519 deletions(-) delete mode 100644 src/core/arm/nce/visitor_base.h diff --git a/src/core/CMakeLists.txt b/src/core/CMakeLists.txt index 74391b5e65..9cc2792a8c 100644 --- a/src/core/CMakeLists.txt +++ b/src/core/CMakeLists.txt @@ -1238,8 +1238,7 @@ if (HAS_NCE) arm/nce/interpreter_visitor.cpp arm/nce/interpreter_visitor.h arm/nce/patcher.cpp - arm/nce/patcher.h - arm/nce/visitor_base.h) + arm/nce/patcher.h) target_link_libraries(core PRIVATE merry::oaknut) endif() diff --git a/src/core/arm/nce/interpreter_visitor.cpp b/src/core/arm/nce/interpreter_visitor.cpp index 2ccc176323..11b759edaf 100644 --- a/src/core/arm/nce/interpreter_visitor.cpp +++ b/src/core/arm/nce/interpreter_visitor.cpp @@ -5,779 +5,387 @@ // SPDX-FileCopyrightText: Copyright 2023 merryhime // SPDX-License-Identifier: GPL-2.0-or-later -#include +#include + #include "core/arm/nce/interpreter_visitor.h" namespace Core { namespace { -// Prefetch tuning parameters -[[maybe_unused]] constexpr size_t CACHE_LINE_SIZE = 64; -[[maybe_unused]] constexpr size_t PREFETCH_STRIDE = 128; // 2 cache lines ahead -[[maybe_unused]] constexpr size_t SIMD_PREFETCH_THRESHOLD = 32; // Bytes + +u64 SignExtend(u64 value, size_t bitsize, size_t regsize) { + const u64 mask = u64{1} << (bitsize - 1); + const u64 extended = ((value & ((u64{1} << bitsize) - 1)) ^ mask) - mask; + if (regsize == 64) { + return extended; + } + return static_cast(extended); +} + +u128 VectorGetElement(u128 value, size_t bitsize) { + if (bitsize >= 128) { + return value; + } + if (bitsize >= 64) { + return {value[0], 0}; + } + return {value[0] & ((u64{1} << bitsize) - 1), 0}; +} + } // namespace -template -u64 SignExtendToLong(u64 value) { - u64 mask = 1ULL << (BitSize - 1); - value &= (1ULL << BitSize) - 1; - return (value ^ mask) - mask; -} - -static u64 SignExtendToLong(u64 value, u64 bitsize) { - switch (bitsize) { - case 8: - return SignExtendToLong<8>(value); - case 16: - return SignExtendToLong<16>(value); - case 32: - return SignExtendToLong<32>(value); - default: - return value; - } -} - -template -u32 SignExtendToWord(u32 value) { - u32 mask = 1ULL << (BitSize - 1); - value &= (1ULL << BitSize) - 1; - return (value ^ mask) - mask; -} - -static u32 SignExtendToWord(u32 value, u64 bitsize) { - switch (bitsize) { - case 8: - return SignExtendToWord<8>(value); - case 16: - return SignExtendToWord<16>(value); - default: - return value; - } -} - -static u64 SignExtend(u64 value, u64 bitsize, u64 regsize) { - if (regsize == 64) { - return SignExtendToLong(value, bitsize); - } else { - return SignExtendToWord(static_cast(value), bitsize); - } -} - -static u128 VectorGetElement(u128 value, u64 bitsize) { - switch (bitsize) { - case 8: - return {value[0] & ((1ULL << 8) - 1), 0}; - case 16: - return {value[0] & ((1ULL << 16) - 1), 0}; - case 32: - return {value[0] & ((1ULL << 32) - 1), 0}; - case 64: - return {value[0], 0}; - default: - return value; - } -} - -u64 InterpreterVisitor::ExtendReg(size_t bitsize, Reg reg, Imm<3> option, u8 shift) { - ASSERT(shift <= 4); - ASSERT(bitsize == 32 || bitsize == 64); +u64 InterpreterVisitor::ExtendReg(Reg reg, Imm<3> option, u8 shift) { + const size_t len = size_t{8} << option.Bits<0, 1, size_t>(); u64 val = this->GetReg(reg); - size_t len; - u64 extended; - bool signed_extend; - - switch (option.ZeroExtend()) { - case 0b000: { // UXTB - val &= ((1ULL << 8) - 1); - len = 8; - signed_extend = false; - break; + if (len < 64) { + val &= (u64{1} << len) - 1; + if (option.Bit<2>()) { + val = SignExtend(val, len, 64); + } } - case 0b001: { // UXTH - val &= ((1ULL << 16) - 1); - len = 16; - signed_extend = false; - break; - } - case 0b010: { // UXTW - val &= ((1ULL << 32) - 1); - len = 32; - signed_extend = false; - break; - } - case 0b011: { // UXTX - len = 64; - signed_extend = false; - break; - } - case 0b100: { // SXTB - val &= ((1ULL << 8) - 1); - len = 8; - signed_extend = true; - break; - } - case 0b101: { // SXTH - val &= ((1ULL << 16) - 1); - len = 16; - signed_extend = true; - break; - } - case 0b110: { // SXTW - val &= ((1ULL << 32) - 1); - len = 32; - signed_extend = true; - break; - } - case 0b111: { // SXTX - len = 64; - signed_extend = true; - break; - } - default: - UNREACHABLE(); - } - - if (len < bitsize && signed_extend) { - extended = SignExtend(val, len, bitsize); - } else { - extended = val; - } - - return extended << shift; + return val << shift; } -u128 InterpreterVisitor::GetVec(Vec v) { - return m_fpsimd_regs[static_cast(v)]; -} - -u64 InterpreterVisitor::GetReg(Reg r) { - return m_regs[static_cast(r)]; -} - -u64 InterpreterVisitor::GetSp() { - return m_sp; -} - -u64 InterpreterVisitor::GetPc() { - return m_pc; -} - -void InterpreterVisitor::SetVec(Vec v, u128 value) { - m_fpsimd_regs[static_cast(v)] = value; -} - -void InterpreterVisitor::SetReg(Reg r, u64 value) { - m_regs[static_cast(r)] = value; -} - -void InterpreterVisitor::SetSp(u64 value) { - m_sp = value; -} - -bool InterpreterVisitor::Ordered(size_t size, bool L, bool o0, Reg Rn, Reg Rt) { - const auto memop = L ? MemOp::Load : MemOp::Store; - const size_t elsize = 8 << size; - const size_t datasize = elsize; - const size_t dbytes = datasize / 8; - - u64 address = (Rn == Reg::SP) ? this->GetSp() : this->GetReg(Rn); - switch (memop) { - case MemOp::Store: { - std::atomic_thread_fence(std::memory_order_seq_cst); - u64 value = this->GetReg(Rt); - m_memory.WriteBlock(address, &value, dbytes); - std::atomic_thread_fence(std::memory_order_seq_cst); - break; - } - case MemOp::Load: { +bool InterpreterVisitor::Ordered(size_t size, bool load, Reg Rn, Reg Rt) { + const size_t dbytes = size_t{1} << size; + const u64 address = this->GetRegSp(Rn); + if (load) { u64 value = 0; m_memory.ReadBlock(address, &value, dbytes); this->SetReg(Rt, value); std::atomic_thread_fence(std::memory_order_seq_cst); - break; - } - default: - UNREACHABLE(); + return true; } + std::atomic_thread_fence(std::memory_order_seq_cst); + u64 value = this->GetReg(Rt); + m_memory.WriteBlock(address, &value, dbytes); + std::atomic_thread_fence(std::memory_order_seq_cst); return true; } -bool InterpreterVisitor::STLLR(Imm<2> sz, Reg Rn, Reg Rt) { - const size_t size = sz.ZeroExtend(); - const bool L = 0; - const bool o0 = 0; - return this->Ordered(size, L, o0, Rn, Rt); -} - -bool InterpreterVisitor::STLR(Imm<2> sz, Reg Rn, Reg Rt) { - const size_t size = sz.ZeroExtend(); - const bool L = 0; - const bool o0 = 1; - return this->Ordered(size, L, o0, Rn, Rt); -} - -bool InterpreterVisitor::LDLAR(Imm<2> sz, Reg Rn, Reg Rt) { - const size_t size = sz.ZeroExtend(); - const bool L = 1; - const bool o0 = 0; - return this->Ordered(size, L, o0, Rn, Rt); -} - -bool InterpreterVisitor::LDAR(Imm<2> sz, Reg Rn, Reg Rt) { - const size_t size = sz.ZeroExtend(); - const bool L = 1; - const bool o0 = 1; - return this->Ordered(size, L, o0, Rn, Rt); -} - -bool InterpreterVisitor::LDR_lit_gen(bool opc_0, Imm<19> imm19, Reg Rt) { - const size_t size = opc_0 == 0 ? 4 : 8; - const s64 offset = Dynarmic::concatenate(imm19, Imm<2>{0}).SignExtend(); - const u64 address = this->GetPc() + offset; - +bool InterpreterVisitor::LoadLiteral(bool wide, Imm<19> imm19, Reg Rt) { + size_t size = 4; + if (wide) { + size = 8; + } u64 data = 0; - m_memory.ReadBlock(address, &data, size); - + m_memory.ReadBlock(m_pc + (imm19.SignExtend() << 2), &data, size); this->SetReg(Rt, data); return true; } -bool InterpreterVisitor::LDR_lit_fpsimd(Imm<2> opc, Imm<19> imm19, Vec Vt) { +bool InterpreterVisitor::LoadLiteralSimd(Imm<2> opc, Imm<19> imm19, Vec Vt) { if (opc == 0b11) { - // Unallocated encoding return false; } - - // Size in bytes - const u64 size = 4 << opc.ZeroExtend(); - const u64 offset = imm19.SignExtend() << 2; - const u64 address = this->GetPc() + offset; - u128 data{}; - m_memory.ReadBlock(address, &data, size); + m_memory.ReadBlock(m_pc + (imm19.SignExtend() << 2), &data, + size_t{4} << opc.ZeroExtend()); this->SetVec(Vt, data); return true; } -bool InterpreterVisitor::STP_LDP_gen(Imm<2> opc, bool not_postindex, bool wback, Imm<1> L, - Imm<7> imm7, Reg Rt2, Reg Rn, Reg Rt) { - if ((L == 0 && opc.Bit<0>() == 1) || opc == 0b11) { - // Unallocated encoding +bool InterpreterVisitor::Pair(Imm<2> opc, bool not_postindex, bool wback, bool load, Imm<7> imm7, + Reg Rt2, Reg Rn, Reg Rt) { + if (!not_postindex && !wback) { + return false; + } + const bool signed_ = opc.Bit<0>(); + if (opc == 0b11 || (!load && signed_)) { + return false; + } + if (load && (Rt == Rt2 || (wback && (Rt == Rn || Rt2 == Rn) && Rn != Reg::R31))) { + return false; + } + if (!load && wback && (Rt == Rn || Rt2 == Rn) && Rn != Reg::R31) { return false; } - const auto memop = L == 1 ? MemOp::Load : MemOp::Store; - if (memop == MemOp::Load && wback && (Rt == Rn || Rt2 == Rn) && Rn != Reg::R31) { - // Unpredictable instruction - return false; - } - if (memop == MemOp::Store && wback && (Rt == Rn || Rt2 == Rn) && Rn != Reg::R31) { - // Unpredictable instruction - return false; - } - if (memop == MemOp::Load && Rt == Rt2) { - // Unpredictable instruction - return false; - } - - u64 address; - if (Rn == Reg::SP) { - address = this->GetSp(); - } else { - address = this->GetReg(Rn); - } - - const bool postindex = !not_postindex; - const bool signed_ = opc.Bit<0>() != 0; const size_t scale = 2 + opc.Bit<1>(); - const size_t datasize = 8 << scale; + const size_t dbytes = size_t{1} << scale; const u64 offset = imm7.SignExtend() << scale; - - if (!postindex) { + u64 address = this->GetRegSp(Rn); + if (not_postindex) { address += offset; } - - const size_t dbytes = datasize / 8; - switch (memop) { - case MemOp::Store: { - u64 data1 = this->GetReg(Rt); - u64 data2 = this->GetReg(Rt2); - m_memory.WriteBlock(address, &data1, dbytes); - m_memory.WriteBlock(address + dbytes, &data2, dbytes); - break; - } - case MemOp::Load: { + if (load) { u64 data1 = 0, data2 = 0; m_memory.ReadBlock(address, &data1, dbytes); m_memory.ReadBlock(address + dbytes, &data2, dbytes); if (signed_) { - this->SetReg(Rt, SignExtend(data1, datasize, 64)); - this->SetReg(Rt2, SignExtend(data2, datasize, 64)); - } else { - this->SetReg(Rt, data1); - this->SetReg(Rt2, data2); + data1 = SignExtend(data1, dbytes * 8, 64); + data2 = SignExtend(data2, dbytes * 8, 64); } - break; + this->SetReg(Rt, data1); + this->SetReg(Rt2, data2); + } else { + u64 data1 = this->GetReg(Rt); + u64 data2 = this->GetReg(Rt2); + m_memory.WriteBlock(address, &data1, dbytes); + m_memory.WriteBlock(address + dbytes, &data2, dbytes); } - default: - UNREACHABLE(); - } - if (wback) { - if (postindex) { + if (!not_postindex) { address += offset; } - - if (Rn == Reg::SP) { - this->SetSp(address); - } else { - this->SetReg(Rn, address); - } + this->SetRegSp(Rn, address); } - return true; } -bool InterpreterVisitor::STP_LDP_fpsimd(Imm<2> opc, bool not_postindex, bool wback, Imm<1> L, - Imm<7> imm7, Vec Vt2, Reg Rn, Vec Vt) { - if (opc == 0b11) { - // Unallocated encoding +bool InterpreterVisitor::PairSimd(Imm<2> opc, bool not_postindex, bool wback, bool load, + Imm<7> imm7, Vec Vt2, Reg Rn, Vec Vt) { + if (!not_postindex && !wback) { + return false; + } + if (opc == 0b11 || (load && Vt == Vt2)) { return false; } - const auto memop = L == 1 ? MemOp::Load : MemOp::Store; - if (memop == MemOp::Load && Vt == Vt2) { - // Unpredictable instruction - return false; - } - - u64 address; - if (Rn == Reg::SP) { - address = this->GetSp(); - } else { - address = this->GetReg(Rn); - } - - const bool postindex = !not_postindex; const size_t scale = 2 + opc.ZeroExtend(); - const size_t datasize = 8 << scale; + const size_t dbytes = size_t{1} << scale; const u64 offset = imm7.SignExtend() << scale; - const size_t dbytes = datasize / 8; - - if (!postindex) { + u64 address = this->GetRegSp(Rn); + if (not_postindex) { address += offset; } - - switch (memop) { - case MemOp::Store: { - u128 data1 = VectorGetElement(this->GetVec(Vt), datasize); - u128 data2 = VectorGetElement(this->GetVec(Vt2), datasize); - m_memory.WriteBlock(address, &data1, dbytes); - m_memory.WriteBlock(address + dbytes, &data2, dbytes); - break; - } - case MemOp::Load: { + if (load) { u128 data1{}, data2{}; m_memory.ReadBlock(address, &data1, dbytes); m_memory.ReadBlock(address + dbytes, &data2, dbytes); this->SetVec(Vt, data1); this->SetVec(Vt2, data2); - break; + } else { + u128 data1 = VectorGetElement(this->GetVec(Vt), dbytes * 8); + u128 data2 = VectorGetElement(this->GetVec(Vt2), dbytes * 8); + m_memory.WriteBlock(address, &data1, dbytes); + m_memory.WriteBlock(address + dbytes, &data2, dbytes); } - default: - UNREACHABLE(); + if (wback) { + if (!not_postindex) { + address += offset; + } + this->SetRegSp(Rn, address); + } + return true; +} + +bool InterpreterVisitor::RegisterImmediate(bool wback, bool postindex, u64 offset, Imm<2> size, + Imm<2> opc, Reg Rn, Reg Rt) { + const bool signed_ = opc.Bit<1>(); + if (signed_ && (size == 0b11 || (size == 0b10 && opc.Bit<0>()))) { + return false; } + const size_t datasize = size_t{8} << size.ZeroExtend(); + size_t regsize = 32; + if (size == 0b11 || (signed_ && !opc.Bit<0>())) { + regsize = 64; + } + u64 address = this->GetRegSp(Rn); + if (!postindex) { + address += offset; + } + if (signed_ || opc.Bit<0>()) { + u64 data = 0; + m_memory.ReadBlock(address, &data, datasize / 8); + if (signed_) { + data = SignExtend(data, datasize, regsize); + } + this->SetReg(Rt, data); + } else { + u64 data = this->GetReg(Rt); + m_memory.WriteBlock(address, &data, datasize / 8); + } if (wback) { if (postindex) { address += offset; } - - if (Rn == Reg::SP) { - this->SetSp(address); - } else { - this->SetReg(Rn, address); - } + this->SetRegSp(Rn, address); } - return true; } -bool InterpreterVisitor::RegisterImmediate(bool wback, bool postindex, size_t scale, u64 offset, - Imm<2> size, Imm<2> opc, Reg Rn, Reg Rt) { - MemOp memop; - bool signed_ = false; - size_t regsize = 0; - const size_t datasize = 8 << scale; - - if (opc.Bit<1>() == 0) { - memop = opc.Bit<0>() ? MemOp::Load : MemOp::Store; - regsize = size == 0b11 ? 64 : 32; - } else if (size == 0b11) { - memop = MemOp::Prefetch; - ASSERT(!opc.Bit<0>()); - } else { - memop = MemOp::Load; - ASSERT(!(size == 0b10 && opc.Bit<0>() == 1)); - regsize = opc.Bit<0>() ? 32 : 64; - signed_ = true; +bool InterpreterVisitor::RegisterOffset(bool S, Imm<2> size, Imm<1> opc_1, Imm<1> opc_0, Reg Rm, + Imm<3> option, Reg Rn, Reg Rt) { + if (!option.Bit<1>()) { + return false; + } + const bool high = opc_1 == 1; + const bool wide = size == 0b11; + if (high && opc_0 == 1 && (wide || size == 0b10)) { + return false; + } + if (high && wide) { + return true; } - u64 address = (Rn == Reg::SP) ? this->GetSp() : this->GetReg(Rn); - if (!postindex) - address += offset; - - switch (memop) { - case MemOp::Store: { - u64 data = this->GetReg(Rt); - m_memory.WriteBlock(address, &data, datasize / 8); - break; + const size_t scale = size.ZeroExtend(); + const size_t datasize = size_t{8} << scale; + size_t regsize = 32; + if (wide || (high && opc_0 == 0)) { + regsize = 64; } - case MemOp::Load: { + u8 shift = 0; + if (S) { + shift = static_cast(scale); + } + const u64 address = this->GetRegSp(Rn) + this->ExtendReg(Rm, option, shift); + if (high || opc_0 == 1) { u64 data = 0; m_memory.ReadBlock(address, &data, datasize / 8); - if (signed_) { - this->SetReg(Rt, SignExtend(data, datasize, regsize)); - } else { - this->SetReg(Rt, data); + if (high) { + data = SignExtend(data, datasize, regsize); } - break; - } - case MemOp::Prefetch: - break; - } - - if (wback) { - if (postindex) - address += offset; - if (Rn == Reg::SP) - this->SetSp(address); - else - this->SetReg(Rn, address); + this->SetReg(Rt, data); + return true; } + u64 data = this->GetReg(Rt); + m_memory.WriteBlock(address, &data, datasize / 8); return true; } -bool InterpreterVisitor::STRx_LDRx_imm_1(Imm<2> size, Imm<2> opc, Imm<9> imm9, bool not_postindex, - Reg Rn, Reg Rt) { - const bool wback = true; - const bool postindex = !not_postindex; - const size_t scale = size.ZeroExtend(); - const u64 offset = imm9.SignExtend(); +bool InterpreterVisitor::SimdImmediate(bool wback, bool postindex, size_t scale, u64 offset, + bool load, Reg Rn, Vec Vt) { + if (scale > 4) { + return false; + } - return this->RegisterImmediate(wback, postindex, scale, offset, size, opc, Rn, Rt); -} - -bool InterpreterVisitor::STRx_LDRx_imm_2(Imm<2> size, Imm<2> opc, Imm<12> imm12, Reg Rn, Reg Rt) { - const bool wback = false; - const bool postindex = false; - const size_t scale = size.ZeroExtend(); - const u64 offset = imm12.ZeroExtend() << scale; - - return this->RegisterImmediate(wback, postindex, scale, offset, size, opc, Rn, Rt); -} - -bool InterpreterVisitor::STURx_LDURx(Imm<2> size, Imm<2> opc, Imm<9> imm9, Reg Rn, Reg Rt) { - const bool wback = false; - const bool postindex = false; - const size_t scale = size.ZeroExtend(); - const u64 offset = imm9.SignExtend(); - - return this->RegisterImmediate(wback, postindex, scale, offset, size, opc, Rn, Rt); -} - -bool InterpreterVisitor::SIMDImmediate(bool wback, bool postindex, size_t scale, u64 offset, - MemOp memop, Reg Rn, Vec Vt) { - const size_t datasize = 8 << scale; - u64 address = (Rn == Reg::SP) ? this->GetSp() : this->GetReg(Rn); - if (!postindex) + const size_t dbytes = size_t{1} << scale; + u64 address = this->GetRegSp(Rn); + if (!postindex) { address += offset; - - switch (memop) { - case MemOp::Store: { - u128 data = VectorGetElement(this->GetVec(Vt), datasize); - m_memory.WriteBlock(address, &data, datasize / 8); - break; } - case MemOp::Load: { + if (load) { u128 data{}; - m_memory.ReadBlock(address, &data, datasize / 8); + m_memory.ReadBlock(address, &data, dbytes); this->SetVec(Vt, data); - break; + } else { + u128 data = VectorGetElement(this->GetVec(Vt), dbytes * 8); + m_memory.WriteBlock(address, &data, dbytes); } - default: - UNREACHABLE(); - } - if (wback) { - if (postindex) + if (postindex) { address += offset; - if (Rn == Reg::SP) - this->SetSp(address); - else - this->SetReg(Rn, address); + } + this->SetRegSp(Rn, address); } return true; } -bool InterpreterVisitor::STR_imm_fpsimd_1(Imm<2> size, Imm<1> opc_1, Imm<9> imm9, - bool not_postindex, Reg Rn, Vec Vt) { - const size_t scale = Dynarmic::concatenate(opc_1, size).ZeroExtend(); - if (scale > 4) { - // Unallocated encoding +bool InterpreterVisitor::SimdOffset(size_t scale, bool S, bool load, Reg Rm, Imm<3> option, Reg Rn, + Vec Vt) { + if (scale > 4 || !option.Bit<1>()) { return false; } - const bool wback = true; - const bool postindex = !not_postindex; - const u64 offset = imm9.SignExtend(); - - return this->SIMDImmediate(wback, postindex, scale, offset, MemOp::Store, Rn, Vt); -} - -bool InterpreterVisitor::STR_imm_fpsimd_2(Imm<2> size, Imm<1> opc_1, Imm<12> imm12, Reg Rn, - Vec Vt) { - const size_t scale = Dynarmic::concatenate(opc_1, size).ZeroExtend(); - if (scale > 4) { - // Unallocated encoding - return false; + const size_t dbytes = size_t{1} << scale; + u8 shift = 0; + if (S) { + shift = static_cast(scale); } - - const bool wback = false; - const bool postindex = false; - const u64 offset = imm12.ZeroExtend() << scale; - - return this->SIMDImmediate(wback, postindex, scale, offset, MemOp::Store, Rn, Vt); -} - -bool InterpreterVisitor::LDR_imm_fpsimd_1(Imm<2> size, Imm<1> opc_1, Imm<9> imm9, - bool not_postindex, Reg Rn, Vec Vt) { - const size_t scale = Dynarmic::concatenate(opc_1, size).ZeroExtend(); - if (scale > 4) { - // Unallocated encoding - return false; - } - - const bool wback = true; - const bool postindex = !not_postindex; - const u64 offset = imm9.SignExtend(); - - return this->SIMDImmediate(wback, postindex, scale, offset, MemOp::Load, Rn, Vt); -} - -bool InterpreterVisitor::LDR_imm_fpsimd_2(Imm<2> size, Imm<1> opc_1, Imm<12> imm12, Reg Rn, - Vec Vt) { - const size_t scale = Dynarmic::concatenate(opc_1, size).ZeroExtend(); - if (scale > 4) { - // Unallocated encoding - return false; - } - - const bool wback = false; - const bool postindex = false; - const u64 offset = imm12.ZeroExtend() << scale; - - return this->SIMDImmediate(wback, postindex, scale, offset, MemOp::Load, Rn, Vt); -} - -bool InterpreterVisitor::STUR_fpsimd(Imm<2> size, Imm<1> opc_1, Imm<9> imm9, Reg Rn, Vec Vt) { - const size_t scale = Dynarmic::concatenate(opc_1, size).ZeroExtend(); - if (scale > 4) { - // Unallocated encoding - return false; - } - - const bool wback = false; - const bool postindex = false; - const u64 offset = imm9.SignExtend(); - - return this->SIMDImmediate(wback, postindex, scale, offset, MemOp::Store, Rn, Vt); -} - -bool InterpreterVisitor::LDUR_fpsimd(Imm<2> size, Imm<1> opc_1, Imm<9> imm9, Reg Rn, Vec Vt) { - const size_t scale = Dynarmic::concatenate(opc_1, size).ZeroExtend(); - if (scale > 4) { - // Unallocated encoding - return false; - } - - const bool wback = false; - const bool postindex = false; - const u64 offset = imm9.SignExtend(); - - return this->SIMDImmediate(wback, postindex, scale, offset, MemOp::Load, Rn, Vt); -} - -bool InterpreterVisitor::RegisterOffset(size_t scale, u8 shift, Imm<2> size, Imm<1> opc_1, - Imm<1> opc_0, Reg Rm, Imm<3> option, Reg Rn, Reg Rt) { - MemOp memop; - size_t regsize = 64; - bool signed_ = false; - - if (opc_1 == 0) { - memop = opc_0 == 1 ? MemOp::Load : MemOp::Store; - regsize = size == 0b11 ? 64 : 32; - signed_ = false; - } else if (size == 0b11) { - memop = MemOp::Prefetch; - if (opc_0 == 1) { - // Unallocated encoding - return false; - } - } else { - memop = MemOp::Load; - if (size == 0b10 && opc_0 == 1) { - // Unallocated encoding - return false; - } - regsize = opc_0 == 1 ? 32 : 64; - signed_ = true; - } - - const size_t datasize = 8 << scale; - - // Operation - const u64 offset = this->ExtendReg(64, Rm, option, shift); - - u64 address; - if (Rn == Reg::SP) { - address = this->GetSp(); - } else { - address = this->GetReg(Rn); - } - address += offset; - - switch (memop) { - case MemOp::Store: { - u64 data = this->GetReg(Rt); - m_memory.WriteBlock(address, &data, datasize / 8); - break; - } - case MemOp::Load: { - u64 data = 0; - m_memory.ReadBlock(address, &data, datasize / 8); - if (signed_) { - this->SetReg(Rt, SignExtend(data, datasize, regsize)); - } else { - this->SetReg(Rt, data); - } - break; - } - case MemOp::Prefetch: - break; - } - - return true; -} - -bool InterpreterVisitor::STRx_reg(Imm<2> size, Imm<1> opc_1, Reg Rm, Imm<3> option, bool S, Reg Rn, - Reg Rt) { - const Imm<1> opc_0{0}; - const size_t scale = size.ZeroExtend(); - const u8 shift = S ? static_cast(scale) : 0; - if (!option.Bit<1>()) { - // Unallocated encoding - return false; - } - return this->RegisterOffset(scale, shift, size, opc_1, opc_0, Rm, option, Rn, Rt); -} - -bool InterpreterVisitor::LDRx_reg(Imm<2> size, Imm<1> opc_1, Reg Rm, Imm<3> option, bool S, Reg Rn, - Reg Rt) { - const Imm<1> opc_0{1}; - const size_t scale = size.ZeroExtend(); - const u8 shift = S ? static_cast(scale) : 0; - if (!option.Bit<1>()) { - // Unallocated encoding - return false; - } - return this->RegisterOffset(scale, shift, size, opc_1, opc_0, Rm, option, Rn, Rt); -} - -bool InterpreterVisitor::SIMDOffset(size_t scale, u8 shift, Imm<1> opc_0, Reg Rm, Imm<3> option, - Reg Rn, Vec Vt) { - const auto memop = opc_0 == 1 ? MemOp::Load : MemOp::Store; - const size_t datasize = 8 << scale; - - // Operation - const u64 offset = this->ExtendReg(64, Rm, option, shift); - - u64 address; - if (Rn == Reg::SP) { - address = this->GetSp(); - } else { - address = this->GetReg(Rn); - } - address += offset; - - switch (memop) { - case MemOp::Store: { - u128 data = VectorGetElement(this->GetVec(Vt), datasize); - m_memory.WriteBlock(address, &data, datasize / 8); - break; - } - case MemOp::Load: { + const u64 address = this->GetRegSp(Rn) + this->ExtendReg(Rm, option, shift); + if (load) { u128 data{}; - m_memory.ReadBlock(address, &data, datasize / 8); + m_memory.ReadBlock(address, &data, dbytes); this->SetVec(Vt, data); - break; + return true; } - default: - UNREACHABLE(); - } - + u128 data = VectorGetElement(this->GetVec(Vt), dbytes * 8); + m_memory.WriteBlock(address, &data, dbytes); return true; } -bool InterpreterVisitor::STR_reg_fpsimd(Imm<2> size, Imm<1> opc_1, Reg Rm, Imm<3> option, bool S, - Reg Rn, Vec Vt) { - const Imm<1> opc_0{0}; - const size_t scale = Dynarmic::concatenate(opc_1, size).ZeroExtend(); - if (scale > 4) { - // Unallocated encoding - return false; +bool InterpreterVisitor::Execute(u32 inst) { + const Imm<2> size{inst >> 30}; + const Imm<2> opc{(inst >> 22) & 3}; + const Imm<3> option{(inst >> 13) & 7}; + const Imm<19> imm19{(inst >> 5) & 0x7FFFF}; + const Reg Rn = static_cast((inst >> 5) & 31); + const Reg Rt = static_cast(inst & 31); + const Vec Vt = static_cast(inst & 31); + const bool load = (inst & (1U << 22)) != 0; + const bool not_postindex = (inst & (1U << 11)) != 0; + const bool scaled = (inst & (1U << 12)) != 0; + const size_t vscale = ((inst >> 21) & 4) | (inst >> 30); + const u64 imm9 = Imm<9>{(inst >> 12) & 0x1FF}.SignExtend(); + + switch ((inst >> 24) & 0x3F) { + case 0b001000: + if ((inst & 0x00BF7C00) != 0x009F7C00) { + return false; + } + return this->Ordered(size.ZeroExtend(), load, Rn, Rt); + case 0b011000: + if ((inst & 0x80000000) != 0) { + return false; + } + return this->LoadLiteral((inst & 0x40000000) != 0, imm19, Rt); + case 0b011100: + return this->LoadLiteralSimd(size, imm19, Vt); + case 0b101000: + case 0b101001: + return this->Pair(size, (inst & (1U << 24)) != 0, (inst & (1U << 23)) != 0, load, + Imm<7>{(inst >> 15) & 0x7F}, static_cast((inst >> 10) & 31), Rn, Rt); + case 0b101100: + case 0b101101: + return this->PairSimd(size, (inst & (1U << 24)) != 0, (inst & (1U << 23)) != 0, load, + Imm<7>{(inst >> 15) & 0x7F}, static_cast((inst >> 10) & 31), Rn, + Vt); + case 0b111000: + if ((inst & (1U << 21)) != 0) { + if ((inst & 0x00000C00) != 0x00000800) { + return false; + } + return this->RegisterOffset(scaled, size, Imm<1>{(inst >> 23) & 1}, + Imm<1>{(inst >> 22) & 1}, + static_cast((inst >> 16) & 31), option, Rn, Rt); + } + if ((inst & (1U << 10)) != 0) { + return this->RegisterImmediate(true, !not_postindex, imm9, size, opc, Rn, Rt); + } + if (not_postindex) { + return false; + } + return this->RegisterImmediate(false, false, imm9, size, opc, Rn, Rt); + case 0b111001: + return this->RegisterImmediate(false, false, + u64{(inst >> 10) & 0xFFF} << size.ZeroExtend(), size, + opc, Rn, Rt); + case 0b111100: + if ((inst & (1U << 21)) != 0) { + if ((inst & 0x00000C00) != 0x00000800) { + return false; + } + return this->SimdOffset(vscale, scaled, load, static_cast((inst >> 16) & 31), + option, Rn, Vt); + } + if ((inst & (1U << 10)) != 0) { + return this->SimdImmediate(true, !not_postindex, vscale, imm9, load, Rn, Vt); + } + if (not_postindex) { + return false; + } + return this->SimdImmediate(false, false, vscale, imm9, load, Rn, Vt); + case 0b111101: + return this->SimdImmediate(false, false, vscale, u64{(inst >> 10) & 0xFFF} << vscale, load, + Rn, Vt); } - const u8 shift = S ? static_cast(scale) : 0; - if (!option.Bit<1>()) { - // Unallocated encoding - return false; - } - return this->SIMDOffset(scale, shift, opc_0, Rm, option, Rn, Vt); + return false; } -bool InterpreterVisitor::LDR_reg_fpsimd(Imm<2> size, Imm<1> opc_1, Reg Rm, Imm<3> option, bool S, - Reg Rn, Vec Vt) { - const Imm<1> opc_0{1}; - const size_t scale = Dynarmic::concatenate(opc_1, size).ZeroExtend(); - if (scale > 4) { - // Unallocated encoding - return false; - } - const u8 shift = S ? static_cast(scale) : 0; - if (!option.Bit<1>()) { - // Unallocated encoding - return false; - } - return this->SIMDOffset(scale, shift, opc_0, Rm, option, Rn, Vt); -} - -std::optional MatchAndExecuteOneInstruction(Core::Memory::Memory& memory, mcontext_t* context, fpsimd_context* fpsimd_context) { +std::optional MatchAndExecuteOneInstruction(Core::Memory::Memory& memory, mcontext_t* context, + fpsimd_context* fpsimd_context) { std::span regs(reinterpret_cast(context->regs), 31); std::span vregs(reinterpret_cast(fpsimd_context->vregs), 32); u64& sp = *reinterpret_cast(&context->sp); const u64& pc = *reinterpret_cast(&context->pc); InterpreterVisitor visitor(memory, regs, vregs, sp, pc); - u32 instruction = memory.Read32(pc); - bool was_executed = false; - - auto decoder = Dynarmic::A64::Decode(visitor, instruction); - if (decoder) { - was_executed = *decoder; - } else { - was_executed = false; + if (!visitor.Execute(memory.Read32(pc))) { + return std::nullopt; } - return was_executed ? std::optional(pc + 4) : std::nullopt; + return pc + 4; } } // namespace Core diff --git a/src/core/arm/nce/interpreter_visitor.h b/src/core/arm/nce/interpreter_visitor.h index 131b69cd7d..667ea24ac8 100644 --- a/src/core/arm/nce/interpreter_visitor.h +++ b/src/core/arm/nce/interpreter_visitor.h @@ -7,97 +7,77 @@ #pragma once -#include -#include -#include +#include #include +#include + +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wshadow" +#include +#include +#pragma GCC diagnostic pop -#include "core/hle/kernel/k_thread.h" #include "core/memory.h" -#include "common/logging.h" -#include "core/arm/nce/visitor_base.h" namespace Core { -namespace Memory { -class Memory; -} - -class InterpreterVisitor final : public VisitorBase { +class InterpreterVisitor { public: explicit InterpreterVisitor(Core::Memory::Memory& memory, std::span regs, std::span fpsimd_regs, u64& sp, const u64& pc) : m_memory(memory), m_regs(regs), m_fpsimd_regs(fpsimd_regs), m_sp(sp), m_pc(pc) {} - ~InterpreterVisitor() override = default; - enum class MemOp { - Load, - Store, - Prefetch, - }; - - u128 GetVec(Vec v); - u64 GetReg(Reg r); - u64 GetSp(); - u64 GetPc(); - - void SetVec(Vec v, u128 value); - void SetReg(Reg r, u64 value); - void SetSp(u64 value); - - u64 ExtendReg(size_t bitsize, Reg reg, Imm<3> option, u8 shift); - - // Loads and stores - Load/Store Exclusive - bool Ordered(size_t size, bool L, bool o0, Reg Rn, Reg Rt); - bool STLLR(Imm<2> size, Reg Rn, Reg Rt) override; - bool STLR(Imm<2> size, Reg Rn, Reg Rt) override; - bool LDLAR(Imm<2> size, Reg Rn, Reg Rt) override; - bool LDAR(Imm<2> size, Reg Rn, Reg Rt) override; - - // Loads and stores - Load register (literal) - bool LDR_lit_gen(bool opc_0, Imm<19> imm19, Reg Rt) override; - bool LDR_lit_fpsimd(Imm<2> opc, Imm<19> imm19, Vec Vt) override; - - // Loads and stores - Load/Store register pair - bool STP_LDP_gen(Imm<2> opc, bool not_postindex, bool wback, Imm<1> L, Imm<7> imm7, Reg Rt2, - Reg Rn, Reg Rt) override; - bool STP_LDP_fpsimd(Imm<2> opc, bool not_postindex, bool wback, Imm<1> L, Imm<7> imm7, Vec Vt2, - Reg Rn, Vec Vt) override; - - // Loads and stores - Load/Store register (immediate) - bool RegisterImmediate(bool wback, bool postindex, size_t scale, u64 offset, Imm<2> size, - Imm<2> opc, Reg Rn, Reg Rt); - bool STRx_LDRx_imm_1(Imm<2> size, Imm<2> opc, Imm<9> imm9, bool not_postindex, Reg Rn, - Reg Rt) override; - bool STRx_LDRx_imm_2(Imm<2> size, Imm<2> opc, Imm<12> imm12, Reg Rn, Reg Rt) override; - bool STURx_LDURx(Imm<2> size, Imm<2> opc, Imm<9> imm9, Reg Rn, Reg Rt) override; - - bool SIMDImmediate(bool wback, bool postindex, size_t scale, u64 offset, MemOp memop, Reg Rn, - Vec Vt); - bool STR_imm_fpsimd_1(Imm<2> size, Imm<1> opc_1, Imm<9> imm9, bool not_postindex, Reg Rn, - Vec Vt) override; - bool STR_imm_fpsimd_2(Imm<2> size, Imm<1> opc_1, Imm<12> imm12, Reg Rn, Vec Vt) override; - bool LDR_imm_fpsimd_1(Imm<2> size, Imm<1> opc_1, Imm<9> imm9, bool not_postindex, Reg Rn, - Vec Vt) override; - bool LDR_imm_fpsimd_2(Imm<2> size, Imm<1> opc_1, Imm<12> imm12, Reg Rn, Vec Vt) override; - bool STUR_fpsimd(Imm<2> size, Imm<1> opc_1, Imm<9> imm9, Reg Rn, Vec Vt) override; - bool LDUR_fpsimd(Imm<2> size, Imm<1> opc_1, Imm<9> imm9, Reg Rn, Vec Vt) override; - - // Loads and stores - Load/Store register (register offset) - bool RegisterOffset(size_t scale, u8 shift, Imm<2> size, Imm<1> opc_1, Imm<1> opc_0, Reg Rm, - Imm<3> option, Reg Rn, Reg Rt); - bool STRx_reg(Imm<2> size, Imm<1> opc_1, Reg Rm, Imm<3> option, bool S, Reg Rn, - Reg Rt) override; - bool LDRx_reg(Imm<2> size, Imm<1> opc_1, Reg Rm, Imm<3> option, bool S, Reg Rn, - Reg Rt) override; - - bool SIMDOffset(size_t scale, u8 shift, Imm<1> opc_0, Reg Rm, Imm<3> option, Reg Rn, Vec Vt); - bool STR_reg_fpsimd(Imm<2> size, Imm<1> opc_1, Reg Rm, Imm<3> option, bool S, Reg Rn, - Vec Vt) override; - bool LDR_reg_fpsimd(Imm<2> size, Imm<1> opc_1, Reg Rm, Imm<3> option, bool S, Reg Rn, - Vec Vt) override; + bool Execute(u32 inst); private: + template + using Imm = Dynarmic::Imm; + using Reg = Dynarmic::A64::Reg; + using Vec = Dynarmic::A64::Vec; + + u128 GetVec(Vec v) const { + return m_fpsimd_regs[static_cast(v)]; + } + void SetVec(Vec v, u128 value) { + m_fpsimd_regs[static_cast(v)] = value; + } + u64 GetReg(Reg r) const { + return m_regs[static_cast(r)]; + } + void SetReg(Reg r, u64 value) { + m_regs[static_cast(r)] = value; + } + u64 GetRegSp(Reg r) const { + if (r == Reg::SP) { + return m_sp; + } + return m_regs[static_cast(r)]; + } + void SetRegSp(Reg r, u64 value) { + if (r == Reg::SP) { + m_sp = value; + return; + } + m_regs[static_cast(r)] = value; + } + + u64 ExtendReg(Reg reg, Imm<3> option, u8 shift); + + bool Ordered(size_t size, bool load, Reg Rn, Reg Rt); + bool LoadLiteral(bool wide, Imm<19> imm19, Reg Rt); + bool LoadLiteralSimd(Imm<2> opc, Imm<19> imm19, Vec Vt); + bool Pair(Imm<2> opc, bool not_postindex, bool wback, bool load, Imm<7> imm7, Reg Rt2, Reg Rn, + Reg Rt); + bool PairSimd(Imm<2> opc, bool not_postindex, bool wback, bool load, Imm<7> imm7, Vec Vt2, + Reg Rn, Vec Vt); + bool RegisterImmediate(bool wback, bool postindex, u64 offset, Imm<2> size, Imm<2> opc, Reg Rn, + Reg Rt); + bool RegisterOffset(bool S, Imm<2> size, Imm<1> opc_1, Imm<1> opc_0, Reg Rm, Imm<3> option, + Reg Rn, Reg Rt); + bool SimdImmediate(bool wback, bool postindex, size_t scale, u64 offset, bool load, Reg Rn, + Vec Vt); + bool SimdOffset(size_t scale, bool S, bool load, Reg Rm, Imm<3> option, Reg Rn, Vec Vt); + Core::Memory::Memory& m_memory; std::span m_regs; std::span m_fpsimd_regs; diff --git a/src/core/arm/nce/visitor_base.h b/src/core/arm/nce/visitor_base.h deleted file mode 100644 index 6a2be3d9bc..0000000000 --- a/src/core/arm/nce/visitor_base.h +++ /dev/null @@ -1,2783 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project -// SPDX-FileCopyrightText: Copyright 2023 merryhime -// SPDX-License-Identifier: GPL-2.0-or-later - -#pragma once - -#pragma GCC diagnostic push -#pragma GCC diagnostic ignored "-Wshadow" - -#include -#include -#include - -#pragma GCC diagnostic pop - -namespace Core { - -class VisitorBase { -public: - using instruction_return_type = bool; - - template - using Imm = Dynarmic::Imm; - using Reg = Dynarmic::A64::Reg; - using Vec = Dynarmic::A64::Vec; - using Cond = Dynarmic::A64::Cond; - - virtual ~VisitorBase() {} - - virtual bool UnallocatedEncoding() { - return false; - } - - // Data processing - Immediate - PC relative addressing - virtual bool ADR(Imm<2> immlo, Imm<19> immhi, Reg Rd) { - return false; - } - virtual bool ADRP(Imm<2> immlo, Imm<19> immhi, Reg Rd) { - return false; - } - - // Data processing - Immediate - Add/Sub (with tag) - virtual bool ADDG(Imm<6> offset_imm, Imm<4> tag_offset, Reg Rn, Reg Rd) { - return false; - } - virtual bool SUBG(Imm<6> offset_imm, Imm<4> tag_offset, Reg Rn, Reg Rd) { - return false; - } - - // Data processing - Immediate - Add/Sub - virtual bool ADD_imm(bool sf, Imm<2> shift, Imm<12> imm12, Reg Rn, Reg Rd) { - return false; - } - virtual bool ADDS_imm(bool sf, Imm<2> shift, Imm<12> imm12, Reg Rn, Reg Rd) { - return false; - } - virtual bool SUB_imm(bool sf, Imm<2> shift, Imm<12> imm12, Reg Rn, Reg Rd) { - return false; - } - virtual bool SUBS_imm(bool sf, Imm<2> shift, Imm<12> imm12, Reg Rn, Reg Rd) { - return false; - } - - // Data processing - Immediate - Logical - virtual bool AND_imm(bool sf, bool N, Imm<6> immr, Imm<6> imms, Reg Rn, Reg Rd) { - return false; - } - virtual bool ORR_imm(bool sf, bool N, Imm<6> immr, Imm<6> imms, Reg Rn, Reg Rd) { - return false; - } - virtual bool EOR_imm(bool sf, bool N, Imm<6> immr, Imm<6> imms, Reg Rn, Reg Rd) { - return false; - } - virtual bool ANDS_imm(bool sf, bool N, Imm<6> immr, Imm<6> imms, Reg Rn, Reg Rd) { - return false; - } - - // Data processing - Immediate - Move Wide - virtual bool MOVN(bool sf, Imm<2> hw, Imm<16> imm16, Reg Rd) { - return false; - } - virtual bool MOVZ(bool sf, Imm<2> hw, Imm<16> imm16, Reg Rd) { - return false; - } - virtual bool MOVK(bool sf, Imm<2> hw, Imm<16> imm16, Reg Rd) { - return false; - } - - // Data processing - Immediate - Bitfield - virtual bool SBFM(bool sf, bool N, Imm<6> immr, Imm<6> imms, Reg Rn, Reg Rd) { - return false; - } - virtual bool BFM(bool sf, bool N, Imm<6> immr, Imm<6> imms, Reg Rn, Reg Rd) { - return false; - } - virtual bool UBFM(bool sf, bool N, Imm<6> immr, Imm<6> imms, Reg Rn, Reg Rd) { - return false; - } - virtual bool ASR_1(Imm<5> immr, Reg Rn, Reg Rd) { - return false; - } - virtual bool ASR_2(Imm<6> immr, Reg Rn, Reg Rd) { - return false; - } - virtual bool SXTB_1(Reg Rn, Reg Rd) { - return false; - } - virtual bool SXTB_2(Reg Rn, Reg Rd) { - return false; - } - virtual bool SXTH_1(Reg Rn, Reg Rd) { - return false; - } - virtual bool SXTH_2(Reg Rn, Reg Rd) { - return false; - } - virtual bool SXTW(Reg Rn, Reg Rd) { - return false; - } - - // Data processing - Immediate - Extract - virtual bool EXTR(bool sf, bool N, Reg Rm, Imm<6> imms, Reg Rn, Reg Rd) { - return false; - } - - // Conditional branch - virtual bool B_cond(Imm<19> imm19, Cond cond) { - return false; - } - - // Exception generation - virtual bool SVC(Imm<16> imm16) { - return false; - } - virtual bool HVC(Imm<16> imm16) { - return false; - } - virtual bool SMC(Imm<16> imm16) { - return false; - } - virtual bool BRK(Imm<16> imm16) { - return false; - } - virtual bool HLT(Imm<16> imm16) { - return false; - } - virtual bool DCPS1(Imm<16> imm16) { - return false; - } - virtual bool DCPS2(Imm<16> imm16) { - return false; - } - virtual bool DCPS3(Imm<16> imm16) { - return false; - } - - // System - virtual bool MSR_imm(Imm<3> op1, Imm<4> CRm, Imm<3> op2) { - return false; - } - virtual bool HINT(Imm<4> CRm, Imm<3> op2) { - return false; - } - virtual bool NOP() { - return false; - } - virtual bool YIELD() { - return false; - } - virtual bool WFE() { - return false; - } - virtual bool WFI() { - return false; - } - virtual bool SEV() { - return false; - } - virtual bool SEVL() { - return false; - } - virtual bool XPAC_1(bool D, Reg Rd) { - return false; - } - virtual bool XPAC_2() { - return false; - } - virtual bool PACIA_1(bool Z, Reg Rn, Reg Rd) { - return false; - } - virtual bool PACIA_2() { - return false; - } - virtual bool PACIB_1(bool Z, Reg Rn, Reg Rd) { - return false; - } - virtual bool PACIB_2() { - return false; - } - virtual bool AUTIA_1(bool Z, Reg Rn, Reg Rd) { - return false; - } - virtual bool AUTIA_2() { - return false; - } - virtual bool AUTIB_1(bool Z, Reg Rn, Reg Rd) { - return false; - } - virtual bool AUTIB_2() { - return false; - } - virtual bool BTI(Imm<2> upper_op2) { - return false; - } - virtual bool ESB() { - return false; - } - virtual bool PSB() { - return false; - } - virtual bool TSB() { - return false; - } - virtual bool CSDB() { - return false; - } - virtual bool CLREX(Imm<4> CRm) { - return false; - } - virtual bool DSB(Imm<4> CRm) { - return false; - } - virtual bool SSBB() { - return false; - } - virtual bool PSSBB() { - return false; - } - virtual bool DMB(Imm<4> CRm) { - return false; - } - virtual bool ISB(Imm<4> CRm) { - return false; - } - virtual bool SYS(Imm<3> op1, Imm<4> CRn, Imm<4> CRm, Imm<3> op2, Reg Rt) { - return false; - } - virtual bool SB() { - return false; - } - virtual bool MSR_reg(Imm<1> o0, Imm<3> op1, Imm<4> CRn, Imm<4> CRm, Imm<3> op2, Reg Rt) { - return false; - } - virtual bool SYSL(Imm<3> op1, Imm<4> CRn, Imm<4> CRm, Imm<3> op2, Reg Rt) { - return false; - } - virtual bool MRS(Imm<1> o0, Imm<3> op1, Imm<4> CRn, Imm<4> CRm, Imm<3> op2, Reg Rt) { - return false; - } - - // System - Flag manipulation instructions - virtual bool CFINV() { - return false; - } - virtual bool RMIF(Imm<6> lsb, Reg Rn, Imm<4> mask) { - return false; - } - virtual bool SETF8(Reg Rn) { - return false; - } - virtual bool SETF16(Reg Rn) { - return false; - } - - // System - Flag format instructions - virtual bool XAFlag() { - return false; - } - virtual bool AXFlag() { - return false; - } - - // SYS: Data Cache - virtual bool DC_IVAC(Reg Rt) { - return false; - } - virtual bool DC_ISW(Reg Rt) { - return false; - } - virtual bool DC_CSW(Reg Rt) { - return false; - } - virtual bool DC_CISW(Reg Rt) { - return false; - } - virtual bool DC_ZVA(Reg Rt) { - return false; - } - virtual bool DC_CVAC(Reg Rt) { - return false; - } - virtual bool DC_CVAU(Reg Rt) { - return false; - } - virtual bool DC_CVAP(Reg Rt) { - return false; - } - virtual bool DC_CIVAC(Reg Rt) { - return false; - } - - // SYS: Instruction Cache - virtual bool IC_IALLU() { - return false; - } - virtual bool IC_IALLUIS() { - return false; - } - virtual bool IC_IVAU(Reg Rt) { - return false; - } - - // Unconditional branch (Register) - virtual bool BR(Reg Rn) { - return false; - } - virtual bool BRA(bool Z, bool M, Reg Rn, Reg Rm) { - return false; - } - virtual bool BLR(Reg Rn) { - return false; - } - virtual bool BLRA(bool Z, bool M, Reg Rn, Reg Rm) { - return false; - } - virtual bool RET(Reg Rn) { - return false; - } - virtual bool RETA(bool M) { - return false; - } - virtual bool ERET() { - return false; - } - virtual bool ERETA(bool M) { - return false; - } - virtual bool DRPS() { - return false; - } - - // Unconditional branch (immediate) - virtual bool B_uncond(Imm<26> imm26) { - return false; - } - virtual bool BL(Imm<26> imm26) { - return false; - } - - // Compare and branch (immediate) - virtual bool CBZ(bool sf, Imm<19> imm19, Reg Rt) { - return false; - } - virtual bool CBNZ(bool sf, Imm<19> imm19, Reg Rt) { - return false; - } - virtual bool TBZ(Imm<1> b5, Imm<5> b40, Imm<14> imm14, Reg Rt) { - return false; - } - virtual bool TBNZ(Imm<1> b5, Imm<5> b40, Imm<14> imm14, Reg Rt) { - return false; - } - - // Loads and stores - Advanced SIMD Load/Store multiple structures - virtual bool STx_mult_1(bool Q, Imm<4> opcode, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool STx_mult_2(bool Q, Reg Rm, Imm<4> opcode, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool LDx_mult_1(bool Q, Imm<4> opcode, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool LDx_mult_2(bool Q, Reg Rm, Imm<4> opcode, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - - // Loads and stores - Advanced SIMD Load/Store single structures - virtual bool ST1_sngl_1(bool Q, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool ST1_sngl_2(bool Q, Reg Rm, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, - Vec Vt) { - return false; - } - virtual bool ST3_sngl_1(bool Q, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool ST3_sngl_2(bool Q, Reg Rm, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, - Vec Vt) { - return false; - } - virtual bool ST2_sngl_1(bool Q, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool ST2_sngl_2(bool Q, Reg Rm, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, - Vec Vt) { - return false; - } - virtual bool ST4_sngl_1(bool Q, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool ST4_sngl_2(bool Q, Reg Rm, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, - Vec Vt) { - return false; - } - virtual bool LD1_sngl_1(bool Q, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool LD1_sngl_2(bool Q, Reg Rm, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, - Vec Vt) { - return false; - } - virtual bool LD3_sngl_1(bool Q, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool LD3_sngl_2(bool Q, Reg Rm, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, - Vec Vt) { - return false; - } - virtual bool LD1R_1(bool Q, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool LD1R_2(bool Q, Reg Rm, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool LD3R_1(bool Q, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool LD3R_2(bool Q, Reg Rm, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool LD2_sngl_1(bool Q, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool LD2_sngl_2(bool Q, Reg Rm, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, - Vec Vt) { - return false; - } - virtual bool LD4_sngl_1(bool Q, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool LD4_sngl_2(bool Q, Reg Rm, Imm<2> upper_opcode, bool S, Imm<2> size, Reg Rn, - Vec Vt) { - return false; - } - virtual bool LD2R_1(bool Q, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool LD2R_2(bool Q, Reg Rm, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool LD4R_1(bool Q, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - virtual bool LD4R_2(bool Q, Reg Rm, Imm<2> size, Reg Rn, Vec Vt) { - return false; - } - - // Loads and stores - Load/Store Exclusive - virtual bool STXR(Imm<2> size, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool STLXR(Imm<2> size, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool STXP(Imm<1> size, Reg Rs, Reg Rt2, Reg Rn, Reg Rt) { - return false; - } - virtual bool STLXP(Imm<1> size, Reg Rs, Reg Rt2, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDXR(Imm<2> size, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDAXR(Imm<2> size, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDXP(Imm<1> size, Reg Rt2, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDAXP(Imm<1> size, Reg Rt2, Reg Rn, Reg Rt) { - return false; - } - virtual bool STLLR(Imm<2> size, Reg Rn, Reg Rt) { - return false; - } - virtual bool STLR(Imm<2> size, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDLAR(Imm<2> size, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDAR(Imm<2> size, Reg Rn, Reg Rt) { - return false; - } - virtual bool CASP(bool sz, bool L, Reg Rs, bool o0, Reg Rn, Reg Rt) { - return false; - } - virtual bool CASB(bool L, Reg Rs, bool o0, Reg Rn, Reg Rt) { - return false; - } - virtual bool CASH(bool L, Reg Rs, bool o0, Reg Rn, Reg Rt) { - return false; - } - virtual bool CAS(bool sz, bool L, Reg Rs, bool o0, Reg Rn, Reg Rt) { - return false; - } - - // Loads and stores - Load register (literal) - virtual bool LDR_lit_gen(bool opc_0, Imm<19> imm19, Reg Rt) { - return false; - } - virtual bool LDR_lit_fpsimd(Imm<2> opc, Imm<19> imm19, Vec Vt) { - return false; - } - virtual bool LDRSW_lit(Imm<19> imm19, Reg Rt) { - return false; - } - virtual bool PRFM_lit(Imm<19> imm19, Imm<5> prfop) { - return false; - } - - // Loads and stores - Load/Store no-allocate pair - virtual bool STNP_LDNP_gen(Imm<1> upper_opc, Imm<1> L, Imm<7> imm7, Reg Rt2, Reg Rn, Reg Rt) { - return false; - } - virtual bool STNP_LDNP_fpsimd(Imm<2> opc, Imm<1> L, Imm<7> imm7, Vec Vt2, Reg Rn, Vec Vt) { - return false; - } - - // Loads and stores - Load/Store register pair - virtual bool STP_LDP_gen(Imm<2> opc, bool not_postindex, bool wback, Imm<1> L, Imm<7> imm7, - Reg Rt2, Reg Rn, Reg Rt) { - return false; - } - virtual bool STP_LDP_fpsimd(Imm<2> opc, bool not_postindex, bool wback, Imm<1> L, Imm<7> imm7, - Vec Vt2, Reg Rn, Vec Vt) { - return false; - } - virtual bool STGP_1(Imm<7> offset_imm, Reg Rt2, Reg Rn, Reg Rt) { - return false; - } - virtual bool STGP_2(Imm<7> offset_imm, Reg Rt2, Reg Rn, Reg Rt) { - return false; - } - virtual bool STGP_3(Imm<7> offset_imm, Reg Rt2, Reg Rn, Reg Rt) { - return false; - } - - // Loads and stores - Load/Store register (immediate) - virtual bool STRx_LDRx_imm_1(Imm<2> size, Imm<2> opc, Imm<9> imm9, bool not_postindex, Reg Rn, - Reg Rt) { - return false; - } - virtual bool STRx_LDRx_imm_2(Imm<2> size, Imm<2> opc, Imm<12> imm12, Reg Rn, Reg Rt) { - return false; - } - virtual bool STURx_LDURx(Imm<2> size, Imm<2> opc, Imm<9> imm9, Reg Rn, Reg Rt) { - return false; - } - virtual bool PRFM_imm(Imm<12> imm12, Reg Rn, Reg Rt) { - return false; - } - virtual bool PRFM_unscaled_imm(Imm<9> imm9, Reg Rn, Reg Rt) { - return false; - } - virtual bool STR_imm_fpsimd_1(Imm<2> size, Imm<1> opc_1, Imm<9> imm9, bool not_postindex, - Reg Rn, Vec Vt) { - return false; - } - virtual bool STR_imm_fpsimd_2(Imm<2> size, Imm<1> opc_1, Imm<12> imm12, Reg Rn, Vec Vt) { - return false; - } - virtual bool LDR_imm_fpsimd_1(Imm<2> size, Imm<1> opc_1, Imm<9> imm9, bool not_postindex, - Reg Rn, Vec Vt) { - return false; - } - virtual bool LDR_imm_fpsimd_2(Imm<2> size, Imm<1> opc_1, Imm<12> imm12, Reg Rn, Vec Vt) { - return false; - } - virtual bool STUR_fpsimd(Imm<2> size, Imm<1> opc_1, Imm<9> imm9, Reg Rn, Vec Vt) { - return false; - } - virtual bool LDUR_fpsimd(Imm<2> size, Imm<1> opc_1, Imm<9> imm9, Reg Rn, Vec Vt) { - return false; - } - - // Loads and stores - Load/Store register (unprivileged) - virtual bool STTRB(Imm<9> imm9, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDTRB(Imm<9> imm9, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDTRSB(Imm<2> opc, Imm<9> imm9, Reg Rn, Reg Rt) { - return false; - } - virtual bool STTRH(Imm<9> imm9, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDTRH(Imm<9> imm9, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDTRSH(Imm<2> opc, Imm<9> imm9, Reg Rn, Reg Rt) { - return false; - } - virtual bool STTR(Imm<2> size, Imm<9> imm9, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDTR(Imm<2> size, Imm<9> imm9, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDTRSW(Imm<9> imm9, Reg Rn, Reg Rt) { - return false; - } - - // Loads and stores - Atomic memory options - virtual bool LDADDB(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDCLRB(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDEORB(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDSETB(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDSMAXB(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDSMINB(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDUMAXB(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDUMINB(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool SWPB(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDAPRB(Reg Rn, Reg Rt) { - return false; - } - virtual bool LDADDH(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDCLRH(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDEORH(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDSETH(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDSMAXH(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDSMINH(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDUMAXH(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDUMINH(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool SWPH(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDAPRH(Reg Rn, Reg Rt) { - return false; - } - virtual bool LDADD(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDCLR(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDEOR(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDSET(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDSMAX(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDSMIN(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDUMAX(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDUMIN(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool SWP(bool A, bool R, Reg Rs, Reg Rn, Reg Rt) { - return false; - } - virtual bool LDAPR(Reg Rn, Reg Rt) { - return false; - } - - // Loads and stores - Load/Store register (register offset) - virtual bool STRx_reg(Imm<2> size, Imm<1> opc_1, Reg Rm, Imm<3> option, bool S, Reg Rn, - Reg Rt) { - return false; - } - virtual bool LDRx_reg(Imm<2> size, Imm<1> opc_1, Reg Rm, Imm<3> option, bool S, Reg Rn, - Reg Rt) { - return false; - } - virtual bool STR_reg_fpsimd(Imm<2> size, Imm<1> opc_1, Reg Rm, Imm<3> option, bool S, Reg Rn, - Vec Vt) { - return false; - } - virtual bool LDR_reg_fpsimd(Imm<2> size, Imm<1> opc_1, Reg Rm, Imm<3> option, bool S, Reg Rn, - Vec Vt) { - return false; - } - - // Loads and stores - Load/Store memory tags - virtual bool STG_1(Imm<9> imm9, Reg Rn) { - return false; - } - virtual bool STG_2(Imm<9> imm9, Reg Rn) { - return false; - } - virtual bool STG_3(Imm<9> imm9, Reg Rn) { - return false; - } - virtual bool LDG(Imm<9> offset_imm, Reg Rn, Reg Rt) { - return false; - } - virtual bool STZG_1(Imm<9> offset_imm, Reg Rn) { - return false; - } - virtual bool STZG_2(Imm<9> offset_imm, Reg Rn) { - return false; - } - virtual bool STZG_3(Imm<9> offset_imm, Reg Rn) { - return false; - } - virtual bool ST2G_1(Imm<9> offset_imm, Reg Rn) { - return false; - } - virtual bool ST2G_2(Imm<9> offset_imm, Reg Rn) { - return false; - } - virtual bool ST2G_3(Imm<9> offset_imm, Reg Rn) { - return false; - } - virtual bool STGV(Reg Rn, Reg Rt) { - return false; - } - virtual bool STZ2G_1(Imm<9> offset_imm, Reg Rn) { - return false; - } - virtual bool STZ2G_2(Imm<9> offset_imm, Reg Rn) { - return false; - } - virtual bool STZ2G_3(Imm<9> offset_imm, Reg Rn) { - return false; - } - virtual bool LDGV(Reg Rn, Reg Rt) { - return false; - } - - // Loads and stores - Load/Store register (pointer authentication) - virtual bool LDRA(bool M, bool S, Imm<9> imm9, bool W, Reg Rn, Reg Rt) { - return false; - } - - // Data Processing - Register - 2 source - virtual bool UDIV(bool sf, Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool SDIV(bool sf, Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool LSLV(bool sf, Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool LSRV(bool sf, Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool ASRV(bool sf, Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool RORV(bool sf, Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool CRC32(bool sf, Reg Rm, Imm<2> sz, Reg Rn, Reg Rd) { - return false; - } - virtual bool CRC32C(bool sf, Reg Rm, Imm<2> sz, Reg Rn, Reg Rd) { - return false; - } - virtual bool PACGA(Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool SUBP(Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool IRG(Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool GMI(Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool SUBPS(Reg Rm, Reg Rn, Reg Rd) { - return false; - } - - // Data Processing - Register - 1 source - virtual bool RBIT_int(bool sf, Reg Rn, Reg Rd) { - return false; - } - virtual bool REV16_int(bool sf, Reg Rn, Reg Rd) { - return false; - } - virtual bool REV(bool sf, bool opc_0, Reg Rn, Reg Rd) { - return false; - } - virtual bool CLZ_int(bool sf, Reg Rn, Reg Rd) { - return false; - } - virtual bool CLS_int(bool sf, Reg Rn, Reg Rd) { - return false; - } - virtual bool REV32_int(Reg Rn, Reg Rd) { - return false; - } - virtual bool PACDA(bool Z, Reg Rn, Reg Rd) { - return false; - } - virtual bool PACDB(bool Z, Reg Rn, Reg Rd) { - return false; - } - virtual bool AUTDA(bool Z, Reg Rn, Reg Rd) { - return false; - } - virtual bool AUTDB(bool Z, Reg Rn, Reg Rd) { - return false; - } - - // Data Processing - Register - Logical (shifted register) - virtual bool AND_shift(bool sf, Imm<2> shift, Reg Rm, Imm<6> imm6, Reg Rn, Reg Rd) { - return false; - } - virtual bool BIC_shift(bool sf, Imm<2> shift, Reg Rm, Imm<6> imm6, Reg Rn, Reg Rd) { - return false; - } - virtual bool ORR_shift(bool sf, Imm<2> shift, Reg Rm, Imm<6> imm6, Reg Rn, Reg Rd) { - return false; - } - virtual bool ORN_shift(bool sf, Imm<2> shift, Reg Rm, Imm<6> imm6, Reg Rn, Reg Rd) { - return false; - } - virtual bool EOR_shift(bool sf, Imm<2> shift, Reg Rm, Imm<6> imm6, Reg Rn, Reg Rd) { - return false; - } - virtual bool EON(bool sf, Imm<2> shift, Reg Rm, Imm<6> imm6, Reg Rn, Reg Rd) { - return false; - } - virtual bool ANDS_shift(bool sf, Imm<2> shift, Reg Rm, Imm<6> imm6, Reg Rn, Reg Rd) { - return false; - } - virtual bool BICS(bool sf, Imm<2> shift, Reg Rm, Imm<6> imm6, Reg Rn, Reg Rd) { - return false; - } - - // Data Processing - Register - Add/Sub (shifted register) - virtual bool ADD_shift(bool sf, Imm<2> shift, Reg Rm, Imm<6> imm6, Reg Rn, Reg Rd) { - return false; - } - virtual bool ADDS_shift(bool sf, Imm<2> shift, Reg Rm, Imm<6> imm6, Reg Rn, Reg Rd) { - return false; - } - virtual bool SUB_shift(bool sf, Imm<2> shift, Reg Rm, Imm<6> imm6, Reg Rn, Reg Rd) { - return false; - } - virtual bool SUBS_shift(bool sf, Imm<2> shift, Reg Rm, Imm<6> imm6, Reg Rn, Reg Rd) { - return false; - } - - // Data Processing - Register - Add/Sub (shifted register) - virtual bool ADD_ext(bool sf, Reg Rm, Imm<3> option, Imm<3> imm3, Reg Rn, Reg Rd) { - return false; - } - virtual bool ADDS_ext(bool sf, Reg Rm, Imm<3> option, Imm<3> imm3, Reg Rn, Reg Rd) { - return false; - } - virtual bool SUB_ext(bool sf, Reg Rm, Imm<3> option, Imm<3> imm3, Reg Rn, Reg Rd) { - return false; - } - virtual bool SUBS_ext(bool sf, Reg Rm, Imm<3> option, Imm<3> imm3, Reg Rn, Reg Rd) { - return false; - } - - // Data Processing - Register - Add/Sub (with carry) - virtual bool ADC(bool sf, Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool ADCS(bool sf, Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool SBC(bool sf, Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool SBCS(bool sf, Reg Rm, Reg Rn, Reg Rd) { - return false; - } - - // Data Processing - Register - Conditional compare - virtual bool CCMN_reg(bool sf, Reg Rm, Cond cond, Reg Rn, Imm<4> nzcv) { - return false; - } - virtual bool CCMP_reg(bool sf, Reg Rm, Cond cond, Reg Rn, Imm<4> nzcv) { - return false; - } - virtual bool CCMN_imm(bool sf, Imm<5> imm5, Cond cond, Reg Rn, Imm<4> nzcv) { - return false; - } - virtual bool CCMP_imm(bool sf, Imm<5> imm5, Cond cond, Reg Rn, Imm<4> nzcv) { - return false; - } - - // Data Processing - Register - Conditional select - virtual bool CSEL(bool sf, Reg Rm, Cond cond, Reg Rn, Reg Rd) { - return false; - } - virtual bool CSINC(bool sf, Reg Rm, Cond cond, Reg Rn, Reg Rd) { - return false; - } - virtual bool CSINV(bool sf, Reg Rm, Cond cond, Reg Rn, Reg Rd) { - return false; - } - virtual bool CSNEG(bool sf, Reg Rm, Cond cond, Reg Rn, Reg Rd) { - return false; - } - - // Data Processing - Register - 3 source - virtual bool MADD(bool sf, Reg Rm, Reg Ra, Reg Rn, Reg Rd) { - return false; - } - virtual bool MSUB(bool sf, Reg Rm, Reg Ra, Reg Rn, Reg Rd) { - return false; - } - virtual bool SMADDL(Reg Rm, Reg Ra, Reg Rn, Reg Rd) { - return false; - } - virtual bool SMSUBL(Reg Rm, Reg Ra, Reg Rn, Reg Rd) { - return false; - } - virtual bool SMULH(Reg Rm, Reg Rn, Reg Rd) { - return false; - } - virtual bool UMADDL(Reg Rm, Reg Ra, Reg Rn, Reg Rd) { - return false; - } - virtual bool UMSUBL(Reg Rm, Reg Ra, Reg Rn, Reg Rd) { - return false; - } - virtual bool UMULH(Reg Rm, Reg Rn, Reg Rd) { - return false; - } - - // Data Processing - FP and SIMD - AES - virtual bool AESE(Vec Vn, Vec Vd) { - return false; - } - virtual bool AESD(Vec Vn, Vec Vd) { - return false; - } - virtual bool AESMC(Vec Vn, Vec Vd) { - return false; - } - virtual bool AESIMC(Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SHA - virtual bool SHA1C(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHA1P(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHA1M(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHA1SU0(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHA256H(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHA256H2(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHA256SU1(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHA1H(Vec Vn, Vec Vd) { - return false; - } - virtual bool SHA1SU1(Vec Vn, Vec Vd) { - return false; - } - virtual bool SHA256SU0(Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Scalar copy - virtual bool DUP_elt_1(Imm<5> imm5, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Scalar three - virtual bool FMULX_vec_1(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMULX_vec_2(bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMEQ_reg_1(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMEQ_reg_2(bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRECPS_1(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRECPS_2(bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRSQRTS_1(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRSQRTS_2(bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGE_reg_1(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGE_reg_2(bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FACGE_1(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FACGE_2(bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FABD_1(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FABD_2(bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGT_reg_1(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGT_reg_2(bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FACGT_1(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FACGT_2(bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Two register misc FP16 - virtual bool FCVTNS_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTMS_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTAS_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool SCVTF_int_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGT_zero_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMEQ_zero_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMLT_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTPS_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTZS_int_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FRECPE_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FRECPX_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTNU_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTMU_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTAU_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool UCVTF_int_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGE_zero_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMLE_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTPU_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTZU_int_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FRSQRTE_1(Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Two register misc - virtual bool FCVTNS_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTMS_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTAS_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool SCVTF_int_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGT_zero_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMEQ_zero_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMLT_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTPS_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTZS_int_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRECPE_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRECPX_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTNU_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTMU_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTAU_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool UCVTF_int_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGE_zero_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMLE_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTPU_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTZU_int_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRSQRTE_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Scalar two register misc FP16 - virtual bool FCVTNS_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTMS_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTAS_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool SCVTF_int_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGT_zero_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMEQ_zero_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMLT_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTPS_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTZS_int_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRECPE_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTNU_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTMU_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTAU_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool UCVTF_int_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGE_zero_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMLE_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTPU_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTZU_int_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRSQRTE_3(bool Q, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Scalar two register misc - virtual bool FCVTNS_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTMS_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTAS_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool SCVTF_int_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGT_zero_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMEQ_zero_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMLT_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTPS_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTZS_int_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRECPE_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTNU_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTMU_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTAU_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool UCVTF_int_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGE_zero_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMLE_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTPU_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTZU_int_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRSQRTE_4(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Scalar three same extra - virtual bool SQRDMLAH_vec_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQRDMLAH_vec_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQRDMLSH_vec_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQRDMLSH_vec_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Scalar two-register misc - virtual bool SUQADD_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQABS_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMGT_zero_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMEQ_zero_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMLT_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool ABS_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQXTN_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool USQADD_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQNEG_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMGE_zero_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMLE_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool NEG_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQXTUN_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQXTN_1(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTXN_1(bool sz, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD Scalar pairwise - virtual bool ADDP_pair(Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXNMP_pair_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXNMP_pair_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FADDP_pair_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FADDP_pair_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXP_pair_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXP_pair_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINNMP_pair_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINNMP_pair_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINP_pair_1(Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINP_pair_2(bool sz, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD Scalar three different - virtual bool SQDMLAL_vec_1(Imm<2> size, Reg Rm, Reg Rn, Vec Vd) { - return false; - } - virtual bool SQDMLSL_vec_1(Imm<2> size, Reg Rm, Reg Rn, Vec Vd) { - return false; - } - virtual bool SQDMULL_vec_1(Imm<2> size, Reg Rm, Reg Rn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD Scalar three same - virtual bool SQADD_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQSUB_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMGT_reg_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMGE_reg_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SSHL_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQSHL_reg_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SRSHL_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQRSHL_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool ADD_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMTST_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQDMULH_vec_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQADD_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQSUB_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMHI_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMHS_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool USHL_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQSHL_reg_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool URSHL_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQRSHL_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SUB_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMEQ_reg_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQRDMULH_vec_1(Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD Scalar shift by immediate - virtual bool SSHR_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SSRA_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SRSHR_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SRSRA_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHL_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQSHL_imm_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQSHRN_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQRSHRN_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SCVTF_fix_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTZS_fix_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool USHR_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool USRA_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool URSHR_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool URSRA_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SRI_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SLI_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQSHLU_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQSHL_imm_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQSHRUN_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQRSHRUN_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQSHRN_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQRSHRN_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool UCVTF_fix_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTZU_fix_1(Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD Scalar x indexed element - virtual bool SQDMLAL_elt_1(Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool SQDMLSL_elt_1(Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool SQDMULL_elt_1(Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool SQDMULH_elt_1(Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool SQRDMULH_elt_1(Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool FMLA_elt_1(Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMLA_elt_2(bool sz, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMLS_elt_1(Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMLS_elt_2(bool sz, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMUL_elt_1(Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMUL_elt_2(bool sz, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQRDMLAH_elt_1(Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool SQRDMLSH_elt_1(Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool FMULX_elt_1(Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMULX_elt_2(bool sz, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD Table Lookup - virtual bool TBL(bool Q, Vec Vm, Imm<2> len, size_t Vn, Vec Vd) { - return false; - } - virtual bool TBX(bool Q, Vec Vm, Imm<2> len, size_t Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD Permute - virtual bool UZP1(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool TRN1(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool ZIP1(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UZP2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool TRN2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool ZIP2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD Extract - virtual bool EXT(bool Q, Vec Vm, Imm<4> imm4, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD Copy - virtual bool DUP_elt_2(bool Q, Imm<5> imm5, Vec Vn, Vec Vd) { - return false; - } - virtual bool DUP_gen(bool Q, Imm<5> imm5, Reg Rn, Vec Vd) { - return false; - } - virtual bool SMOV(bool Q, Imm<5> imm5, Vec Vn, Reg Rd) { - return false; - } - virtual bool UMOV(bool Q, Imm<5> imm5, Vec Vn, Reg Rd) { - return false; - } - virtual bool INS_gen(Imm<5> imm5, Reg Rn, Vec Vd) { - return false; - } - virtual bool INS_elt(Imm<5> imm5, Imm<4> imm4, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD Three same - virtual bool FMULX_vec_3(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMEQ_reg_3(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRECPS_3(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRSQRTS_3(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGE_reg_3(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FACGE_3(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FABD_3(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGT_reg_3(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FACGT_3(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXNM_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMLA_vec_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FADD_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAX_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINNM_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMLS_vec_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FSUB_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMIN_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXNMP_vec_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FADDP_vec_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMUL_vec_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXP_vec_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FDIV_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINNMP_vec_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINP_vec_1(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD Three same extra - virtual bool SDOT_vec(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UDOT_vec(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMLA_vec(bool Q, Imm<2> size, Vec Vm, Imm<2> rot, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCADD_vec(bool Q, Imm<2> size, Vec Vm, Imm<1> rot, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD Two register misc - virtual bool REV64_asimd(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool REV16_asimd(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool SADDLP(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool CLS_asimd(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool CNT(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool SADALP(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool XTN(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTN(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTL(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool URECPE(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool REV32_asimd(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool UADDLP(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool CLZ_asimd(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool UADALP(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHLL(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool NOT(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool RBIT_asimd(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool URSQRTE(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool SUQADD_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQABS_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMGT_zero_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMEQ_zero_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMLT_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool ABS_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQXTN_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool USQADD_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQNEG_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMGE_zero_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMLE_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool NEG_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQXTUN_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQXTN_2(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTXN_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTN_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTN_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTM_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTM_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FABS_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FABS_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTP_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTP_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTZ_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTZ_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTA_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTA_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTX_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTX_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FNEG_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FNEG_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTI_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTI_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FSQRT_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FSQRT_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINT32X_1(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINT64X_1(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINT32Z_1(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINT64Z_1(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD across lanes - virtual bool SADDLV(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool SMAXV(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool SMINV(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool ADDV(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXNMV_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXNMV_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXV_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXV_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINNMV_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINNMV_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINV_1(bool Q, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINV_2(bool Q, bool sz, Vec Vn, Vec Vd) { - return false; - } - virtual bool UADDLV(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool UMAXV(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - virtual bool UMINV(bool Q, Imm<2> size, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD three different - virtual bool SADDL(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SADDW(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SSUBL(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SSUBW(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool ADDHN(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SABAL(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SUBHN(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SABDL(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SMLAL_vec(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SMLSL_vec(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SMULL_vec(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool PMULL(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UADDL(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UADDW(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool USUBL(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool USUBW(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool RADDHN(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UABAL(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool RSUBHN(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UABDL(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UMLAL_vec(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UMLSL_vec(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UMULL_vec(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQDMLAL_vec_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQDMLSL_vec_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQDMULL_vec_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD three same - virtual bool SHADD(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SRHADD(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHSUB(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SMAX(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SMIN(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SABD(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SABA(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool MLA_vec(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool MUL_vec(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SMAXP(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SMINP(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool ADDP_vec(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMLAL_vec_1(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMLAL_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool AND_asimd(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool BIC_asimd_reg(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMLSL_vec_1(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMLSL_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool ORR_asimd_reg(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool ORN_asimd(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UHADD(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool URHADD(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UHSUB(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UMAX(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UMIN(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UABD(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UABA(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool MLS_vec(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool PMUL(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UMAXP(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UMINP(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool EOR_asimd(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool BSL(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool BIT(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool BIF(bool Q, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXNM_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMLA_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FADD_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAX_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINNM_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMLS_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FSUB_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMIN_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXNMP_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FADDP_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMUL_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXP_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FDIV_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINNMP_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINP_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMULX_vec_4(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMEQ_reg_4(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRECPS_4(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRSQRTS_4(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGE_reg_4(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FACGE_4(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FABD_4(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCMGT_reg_4(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FACGT_4(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQADD_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQSUB_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMGT_reg_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMGE_reg_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SSHL_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQSHL_reg_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SRSHL_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQRSHL_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool ADD_vector(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMTST_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQDMULH_vec_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQADD_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQSUB_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMHI_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMHS_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool USHL_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQSHL_reg_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool URSHL_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQRSHL_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SUB_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool CMEQ_reg_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQRDMULH_vec_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD modified immediate - virtual bool MOVI(bool Q, bool op, Imm<1> a, Imm<1> b, Imm<1> c, Imm<4> cmode, Imm<1> d, - Imm<1> e, Imm<1> f, Imm<1> g, Imm<1> h, Vec Vd) { - return false; - } - virtual bool FMOV_2(bool Q, bool op, Imm<1> a, Imm<1> b, Imm<1> c, Imm<1> d, Imm<1> e, Imm<1> f, - Imm<1> g, Imm<1> h, Vec Vd) { - return false; - } - virtual bool FMOV_3(bool Q, Imm<1> a, Imm<1> b, Imm<1> c, Imm<1> d, Imm<1> e, Imm<1> f, - Imm<1> g, Imm<1> h, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD Shift by immediate - virtual bool SSHR_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SSRA_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SRSHR_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SRSRA_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHL_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQSHL_imm_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHRN(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool RSHRN(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQSHRN_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQRSHRN_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SSHLL(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SCVTF_fix_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTZS_fix_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool USHR_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool USRA_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool URSHR_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool URSRA_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SRI_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SLI_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQSHLU_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQSHL_imm_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQSHRUN_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool SQRSHRUN_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQSHRN_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool UQRSHRN_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool USHLL(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool UCVTF_fix_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVTZU_fix_2(bool Q, Imm<4> immh, Imm<3> immb, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SIMD vector x indexed element - virtual bool SMLAL_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool SQDMLAL_elt_2(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, - Vec Vn, Vec Vd) { - return false; - } - virtual bool SMLSL_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool SQDMLSL_elt_2(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, - Vec Vn, Vec Vd) { - return false; - } - virtual bool MUL_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool SMULL_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vm, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool SQDMULL_elt_2(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, - Vec Vn, Vec Vd) { - return false; - } - virtual bool SQDMULH_elt_2(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, - Vec Vn, Vec Vd) { - return false; - } - virtual bool SQRDMULH_elt_2(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, - Vec Vn, Vec Vd) { - return false; - } - virtual bool SDOT_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool FMLA_elt_3(bool Q, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMLA_elt_4(bool Q, bool sz, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool FMLS_elt_3(bool Q, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMLS_elt_4(bool Q, bool sz, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool FMUL_elt_3(bool Q, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMUL_elt_4(bool Q, bool sz, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool FMLAL_elt_1(bool Q, bool sz, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool FMLAL_elt_2(bool Q, bool sz, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool FMLSL_elt_1(bool Q, bool sz, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool FMLSL_elt_2(bool Q, bool sz, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool MLA_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool UMLAL_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool MLS_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool UMLSL_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool UMULL_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool SQRDMLAH_elt_2(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, - Vec Vn, Vec Vd) { - return false; - } - virtual bool UDOT_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool SQRDMLSH_elt_2(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, - Vec Vn, Vec Vd) { - return false; - } - virtual bool FMULX_elt_3(bool Q, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMULX_elt_4(bool Q, bool sz, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, - Vec Vd) { - return false; - } - virtual bool FCMLA_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<2> rot, - Imm<1> H, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Cryptographic three register - virtual bool SM3TT1A(Vec Vm, Imm<2> imm2, Vec Vn, Vec Vd) { - return false; - } - virtual bool SM3TT1B(Vec Vm, Imm<2> imm2, Vec Vn, Vec Vd) { - return false; - } - virtual bool SM3TT2A(Vec Vm, Imm<2> imm2, Vec Vn, Vec Vd) { - return false; - } - virtual bool SM3TT2B(Vec Vm, Imm<2> imm2, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SHA512 three register - virtual bool SHA512H(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHA512H2(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SHA512SU1(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool RAX1(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool XAR(Vec Vm, Imm<6> imm6, Vec Vn, Vec Vd) { - return false; - } - virtual bool SM3PARTW1(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SM3PARTW2(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool SM4EKEY(Vec Vm, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Cryptographic four register - virtual bool EOR3(Vec Vm, Vec Va, Vec Vn, Vec Vd) { - return false; - } - virtual bool BCAX(Vec Vm, Vec Va, Vec Vn, Vec Vd) { - return false; - } - virtual bool SM3SS1(Vec Vm, Vec Va, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - SHA512 two register - virtual bool SHA512SU0(Vec Vn, Vec Vd) { - return false; - } - virtual bool SM4E(Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Conversion between floating point and fixed point - virtual bool SCVTF_float_fix(bool sf, Imm<2> type, Imm<6> scale, Reg Rn, Vec Vd) { - return false; - } - virtual bool UCVTF_float_fix(bool sf, Imm<2> type, Imm<6> scale, Reg Rn, Vec Vd) { - return false; - } - virtual bool FCVTZS_float_fix(bool sf, Imm<2> type, Imm<6> scale, Vec Vn, Reg Rd) { - return false; - } - virtual bool FCVTZU_float_fix(bool sf, Imm<2> type, Imm<6> scale, Vec Vn, Reg Rd) { - return false; - } - - // Data Processing - FP and SIMD - Conversion between floating point and integer - virtual bool FCVTNS_float(bool sf, Imm<2> type, Vec Vn, Reg Rd) { - return false; - } - virtual bool FCVTNU_float(bool sf, Imm<2> type, Vec Vn, Reg Rd) { - return false; - } - virtual bool SCVTF_float_int(bool sf, Imm<2> type, Reg Rn, Vec Vd) { - return false; - } - virtual bool UCVTF_float_int(bool sf, Imm<2> type, Reg Rn, Vec Vd) { - return false; - } - virtual bool FCVTAS_float(bool sf, Imm<2> type, Vec Vn, Reg Rd) { - return false; - } - virtual bool FCVTAU_float(bool sf, Imm<2> type, Vec Vn, Reg Rd) { - return false; - } - virtual bool FMOV_float_gen(bool sf, Imm<2> type, Imm<1> rmode_0, Imm<1> opc_0, size_t n, - size_t d) { - return false; - } - virtual bool FCVTPS_float(bool sf, Imm<2> type, Vec Vn, Reg Rd) { - return false; - } - virtual bool FCVTPU_float(bool sf, Imm<2> type, Vec Vn, Reg Rd) { - return false; - } - virtual bool FCVTMS_float(bool sf, Imm<2> type, Vec Vn, Reg Rd) { - return false; - } - virtual bool FCVTMU_float(bool sf, Imm<2> type, Vec Vn, Reg Rd) { - return false; - } - virtual bool FCVTZS_float_int(bool sf, Imm<2> type, Vec Vn, Reg Rd) { - return false; - } - virtual bool FCVTZU_float_int(bool sf, Imm<2> type, Vec Vn, Reg Rd) { - return false; - } - virtual bool FJCVTZS(Vec Vn, Reg Rd) { - return false; - } - - // Data Processing - FP and SIMD - Floating point data processing - virtual bool FMOV_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FABS_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FNEG_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FSQRT_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FCVT_float(Imm<2> type, Imm<2> opc, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTN_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTP_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTM_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTZ_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTA_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTX_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINTI_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINT32X_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINT64X_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINT32Z_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - virtual bool FRINT64Z_float(Imm<2> type, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Floating point compare - virtual bool FCMP_float(Imm<2> type, Vec Vm, Vec Vn, bool cmp_with_zero) { - return false; - } - virtual bool FCMPE_float(Imm<2> type, Vec Vm, Vec Vn, bool cmp_with_zero) { - return false; - } - - // Data Processing - FP and SIMD - Floating point immediate - virtual bool FMOV_float_imm(Imm<2> type, Imm<8> imm8, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Floating point conditional compare - virtual bool FCCMP_float(Imm<2> type, Vec Vm, Cond cond, Vec Vn, Imm<4> nzcv) { - return false; - } - virtual bool FCCMPE_float(Imm<2> type, Vec Vm, Cond cond, Vec Vn, Imm<4> nzcv) { - return false; - } - - // Data Processing - FP and SIMD - Floating point data processing two register - virtual bool FMUL_float(Imm<2> type, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FDIV_float(Imm<2> type, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FADD_float(Imm<2> type, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FSUB_float(Imm<2> type, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAX_float(Imm<2> type, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMIN_float(Imm<2> type, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMAXNM_float(Imm<2> type, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMINNM_float(Imm<2> type, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - virtual bool FNMUL_float(Imm<2> type, Vec Vm, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Floating point conditional select - virtual bool FCSEL_float(Imm<2> type, Vec Vm, Cond cond, Vec Vn, Vec Vd) { - return false; - } - - // Data Processing - FP and SIMD - Floating point data processing three register - virtual bool FMADD_float(Imm<2> type, Vec Vm, Vec Va, Vec Vn, Vec Vd) { - return false; - } - virtual bool FMSUB_float(Imm<2> type, Vec Vm, Vec Va, Vec Vn, Vec Vd) { - return false; - } - virtual bool FNMADD_float(Imm<2> type, Vec Vm, Vec Va, Vec Vn, Vec Vd) { - return false; - } - virtual bool FNMSUB_float(Imm<2> type, Vec Vm, Vec Va, Vec Vn, Vec Vd) { - return false; - } -}; - -} // namespace Core From 3ab215fd30dd3cb4481d8d58c748a1b9520f8c1a Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sun, 20 Sep 2026 19:16:06 -0400 Subject: [PATCH 46/69] experiment to address some TPIDR_EL0 stalls --- src/core/arm/nce/patcher.cpp | 30 +++++++----------------------- 1 file changed, 7 insertions(+), 23 deletions(-) diff --git a/src/core/arm/nce/patcher.cpp b/src/core/arm/nce/patcher.cpp index 7d2041d4d2..b1569ceee6 100644 --- a/src/core/arm/nce/patcher.cpp +++ b/src/core/arm/nce/patcher.cpp @@ -371,14 +371,9 @@ size_t Patcher::GetPreSectionSize() const noexcept { } void Patcher::WriteLoadContext(oaknut::VectorCodeGenerator& cg) { - // This function was called, which modifies X30, so use that as a scratch register. - // SP contains the guest X30, so save our return X30 to SP + 8, since we have allocated 16 bytes - // of stack. cg.STR(X30, SP, 8); - cg.MRS(X30, oaknut::SystemReg::TPIDR_EL0); - cg.LDR(X30, X30, offsetof(NativeExecutionParameters, native_context)); + cg.LDR(X30, SP, 0); - // Load system registers. cg.LDR(W0, X30, offsetof(GuestContext, fpsr)); cg.MSR(oaknut::SystemReg::FPSR, X0); cg.LDR(W0, X30, offsetof(GuestContext, fpcr)); @@ -386,7 +381,9 @@ void Patcher::WriteLoadContext(oaknut::VectorCodeGenerator& cg) { cg.LDR(W0, X30, offsetof(GuestContext, nzcv)); cg.MSR(oaknut::SystemReg::NZCV, X0); - // Load all vector registers. + cg.LDR(X0, X30, 8 * 30); + cg.STR(X0, SP, 0); + static constexpr size_t VEC_OFF = offsetof(GuestContext, vector_registers); for (int i = 0; i <= 30; i += 2) { cg.LDP(oaknut::QReg{i}, oaknut::QReg{i + 1}, X30, VEC_OFF + 16 * i); @@ -436,7 +433,7 @@ void Patcher::WriteSaveContext(oaknut::VectorCodeGenerator& cg) { cg.STR(W0, X30, offsetof(GuestContext, nzcv)); cg.LDR(X0, SP, POST_INDEXED, 16); - // Reload our return X30 from the stack, and return. + cg.MOV(X1, X30); cg.LDR(X30, SP, 8); cg.RET(); } @@ -457,8 +454,6 @@ void Patcher::WriteSvcTrampoline(ModuleDestLabel module_dest, u32 svc_id, oaknut // Now that we've saved all registers, we can use any registers as scratch. // Store PC + 4 to arm interface, since we know the instruction offset from the entry point. oaknut::Label pc_after_svc; - cg.MRS(X1, oaknut::SystemReg::TPIDR_EL0); - cg.LDR(X1, X1, offsetof(NativeExecutionParameters, native_context)); cg.LDR(X2, pc_after_svc); cg.STR(X2, X1, offsetof(GuestContext, pc)); @@ -514,24 +509,13 @@ void Patcher::WriteSvcTrampoline(ModuleDestLabel module_dest, u32 svc_id, oaknut // Host called this location. Save the return address so we can // unwind the stack properly when jumping back. - cg.MRS(X2, oaknut::SystemReg::TPIDR_EL0); - cg.LDR(X2, X2, offsetof(NativeExecutionParameters, native_context)); - cg.ADD(X0, X2, offsetof(GuestContext, host_ctx)); + cg.ADD(X0, X1, offsetof(GuestContext, host_ctx)); cg.STR(X30, X0, offsetof(HostContext, host_saved_regs) + 11 * sizeof(u64)); - // Reload all guest registers except X30 and PC. - // The function also expects 16 bytes of stack already allocated. - cg.STR(X30, SP, PRE_INDEXED, -16); + cg.STR(X1, SP, PRE_INDEXED, -16); cg.BL(load_ctx); cg.LDR(X30, SP, POST_INDEXED, 16); - // Use X1 as a scratch register to restore X30. - cg.STR(X1, SP, PRE_INDEXED, -16); - cg.MRS(X1, oaknut::SystemReg::TPIDR_EL0); - cg.LDR(X1, X1, offsetof(NativeExecutionParameters, native_context)); - cg.LDR(X30, X1, offsetof(GuestContext, cpu_registers) + sizeof(u64) * 30); - cg.LDR(X1, SP, POST_INDEXED, 16); - // Unlock the context. this->UnlockContext(cg); From fa9e069907d78ffaeef7dacc02f3697743d444fb Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Tue, 22 Sep 2026 16:27:47 -0400 Subject: [PATCH 47/69] Adjust buffer stream size + reduce descriptor loads --- src/video_core/renderer_vulkan/vk_blit_screen.cpp | 2 +- src/video_core/renderer_vulkan/vk_descriptor_pool.cpp | 1 - src/video_core/renderer_vulkan/vk_descriptor_pool.h | 2 +- src/video_core/renderer_vulkan/vk_master_semaphore.cpp | 2 +- src/video_core/renderer_vulkan/vk_rasterizer.cpp | 2 +- src/video_core/renderer_vulkan/vk_scheduler.cpp | 9 +-------- src/video_core/renderer_vulkan/vk_scheduler.h | 2 -- .../renderer_vulkan/vk_staging_buffer_pool.cpp | 9 --------- src/video_core/renderer_vulkan/vk_update_descriptor.h | 4 ++-- src/video_core/vulkan_common/vulkan_device.h | 1 - 10 files changed, 7 insertions(+), 27 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp index 205382f81a..242d60c5e1 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp +++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp @@ -33,8 +33,8 @@ BlitScreen::BlitScreen(Tegra::MaxwellDeviceMemoryManager& device_memory_, const BlitScreen::~BlitScreen() = default; void BlitScreen::WaitIdle(const Device& device) { - present_manager.WaitPresent(); scheduler.Finish(); + present_manager.WaitPresent(); device.GetLogical().WaitIdle(); } diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp index b020d99b93..b41e755871 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp @@ -125,7 +125,6 @@ vk::DescriptorSets DescriptorAllocator::AllocateDescriptors(size_t count) { throw vk::Exception(VK_ERROR_OUT_OF_POOL_MEMORY); } -DescriptorPool::DescriptorPool(const Device& device_, Scheduler& scheduler) {} DescriptorPool::~DescriptorPool() = default; DescriptorAllocator DescriptorPool::Allocator(const Device& device, Scheduler& scheduler, VkDescriptorSetLayout layout, std::span infos) { diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.h b/src/video_core/renderer_vulkan/vk_descriptor_pool.h index 4efeb60373..45d3acdb55 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_pool.h +++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.h @@ -65,7 +65,7 @@ private: class DescriptorPool { public: - explicit DescriptorPool(const Device& device, Scheduler& scheduler); + DescriptorPool() = default; ~DescriptorPool(); DescriptorPool& operator=(const DescriptorPool&) = delete; diff --git a/src/video_core/renderer_vulkan/vk_master_semaphore.cpp b/src/video_core/renderer_vulkan/vk_master_semaphore.cpp index be5a1144ce..9ca215b425 100644 --- a/src/video_core/renderer_vulkan/vk_master_semaphore.cpp +++ b/src/video_core/renderer_vulkan/vk_master_semaphore.cpp @@ -353,7 +353,7 @@ void MasterSemaphore::WaitThread(std::stop_token token) { free_queue.push_front(std::move(fence)); gpu_tick.store(host_tick, std::memory_order_release); } - gpu_tick.notify_one(); + gpu_tick.notify_all(); } } diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index fd95ea288b..aa36aa9093 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -206,7 +206,7 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra StateTracker& state_tracker_, Scheduler& scheduler_) : gpu{gpu_}, device_memory{device_memory_}, device{device_}, memory_allocator{memory_allocator_}, state_tracker{state_tracker_}, scheduler{scheduler_}, - staging_pool(device, memory_allocator, scheduler), descriptor_pool(device, scheduler), + staging_pool(device, memory_allocator, scheduler), guest_descriptor_queue(device, UpdateDescriptorQueue::GUEST_FRAME_PAYLOAD_SIZE, device.IsExtDescriptorBufferSupported()), compute_pass_descriptor_queue(device, UpdateDescriptorQueue::COMPUTE_FRAME_PAYLOAD_SIZE), diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 4a7a7d5e78..74379e8063 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -56,9 +56,7 @@ Scheduler::~Scheduler() = default; u64 Scheduler::Flush(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) { // When flushing, we only send data to the worker thread; no waiting is necessary. - const u64 signal_value = SubmitExecution(signal_semaphore, wait_semaphore); - AllocateNewContext(); - return signal_value; + return SubmitExecution(signal_semaphore, wait_semaphore); } void Scheduler::Finish(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) { @@ -66,7 +64,6 @@ void Scheduler::Finish(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) const u64 presubmit_tick = CurrentTick(); SubmitExecution(signal_semaphore, wait_semaphore); Wait(presubmit_tick); - AllocateNewContext(); } void Scheduler::WaitWorker() { @@ -384,10 +381,6 @@ u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_se return signal_value; } -void Scheduler::AllocateNewContext() { - // Enable counters once again. These are disabled when a command buffer is finished. -} - void Scheduler::InvalidateState() { state.graphics_pipeline = nullptr; state.rescaling_defined = false; diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index d4bc8d0ec5..08ae0bfca1 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -287,8 +287,6 @@ private: u64 SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore); - void AllocateNewContext(); - void EndPendingOperations(); void EndRenderPass(); diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp index 5f04df2c9e..e2bc1ad3a4 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp @@ -28,16 +28,7 @@ using namespace Common::Literals; // Maximum potential alignment of a Vulkan buffer constexpr VkDeviceSize MAX_ALIGNMENT = 256; -// Stream buffer size in bytes -// *NIX drivers are more sensitive to increased buffers for streaming. -// Windows ones however, can intake bigger buffers and generally do not OOM. -// - GTX 960 on Windows will not OOM with 256mib -// - GT 1030 on ^NIX will OOM with 256mib -#if defined(__FreeBSD__) constexpr VkDeviceSize MAX_STREAM_BUFFER_SIZE = 128_MiB; -#else -constexpr VkDeviceSize MAX_STREAM_BUFFER_SIZE = 256_MiB; -#endif size_t GetStreamBufferSize(const Device& device) { if (!device.HasDebuggingToolAttached()) { diff --git a/src/video_core/renderer_vulkan/vk_update_descriptor.h b/src/video_core/renderer_vulkan/vk_update_descriptor.h index 9f30e65dbd..04d76a8386 100644 --- a/src/video_core/renderer_vulkan/vk_update_descriptor.h +++ b/src/video_core/renderer_vulkan/vk_update_descriptor.h @@ -40,8 +40,8 @@ class UpdateDescriptorQueue final { static constexpr size_t FRAMES_IN_FLIGHT = 8; public: - static constexpr size_t GUEST_FRAME_PAYLOAD_SIZE = 0x80000; - static constexpr size_t COMPUTE_FRAME_PAYLOAD_SIZE = 0x20000; + static constexpr size_t GUEST_FRAME_PAYLOAD_SIZE = 0x20000; + static constexpr size_t COMPUTE_FRAME_PAYLOAD_SIZE = 0x8000; explicit UpdateDescriptorQueue(const Device& device_, size_t frame_payload_size_, bool supports_descriptor_buffer_ = false); diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 916a78fd56..0927ca8a05 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -1011,7 +1011,6 @@ FN_MAX_LIMIT_LIST return features2.features.multiViewport; } - /// Returns true if the device supports VK_KHR_maintenance5. /// Returns true if the device supports VK_KHR_maintenance4. bool IsKhrMaintenance4Supported() const { return extensions.maintenance4; From e6fe763afa00b30be9e8f468048e6ec0578c5850 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Tue, 22 Sep 2026 16:40:50 -0400 Subject: [PATCH 48/69] Correction on the descriptor pool --- src/video_core/renderer_vulkan/vk_descriptor_pool.cpp | 1 + src/video_core/renderer_vulkan/vk_descriptor_pool.h | 2 +- 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp index b41e755871..2f4af7c330 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp @@ -125,6 +125,7 @@ vk::DescriptorSets DescriptorAllocator::AllocateDescriptors(size_t count) { throw vk::Exception(VK_ERROR_OUT_OF_POOL_MEMORY); } +DescriptorPool::DescriptorPool() = default; DescriptorPool::~DescriptorPool() = default; DescriptorAllocator DescriptorPool::Allocator(const Device& device, Scheduler& scheduler, VkDescriptorSetLayout layout, std::span infos) { diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.h b/src/video_core/renderer_vulkan/vk_descriptor_pool.h index 45d3acdb55..6a80f8e107 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_pool.h +++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.h @@ -65,7 +65,7 @@ private: class DescriptorPool { public: - DescriptorPool() = default; + DescriptorPool(); ~DescriptorPool(); DescriptorPool& operator=(const DescriptorPool&) = delete; From 6ffd65adaf83292f370e150bf98eb378de4e0108 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Tue, 22 Sep 2026 18:35:26 -0400 Subject: [PATCH 49/69] another intent to remove sync1 and replace it with sync2 completely --- src/video_core/renderer_vulkan/blit_image.cpp | 122 ++--- src/video_core/renderer_vulkan/blit_image.h | 18 +- .../renderer_vulkan/present/frame_gen.cpp | 38 +- .../renderer_vulkan/present/layer.cpp | 33 +- .../renderer_vulkan/present/lsfg_common.cpp | 31 +- .../renderer_vulkan/present/lsfg_common.h | 4 +- .../renderer_vulkan/present/lsfg_generate.cpp | 21 +- .../renderer_vulkan/present/util.cpp | 50 +- .../renderer_vulkan/vk_buffer_cache.cpp | 91 ++-- .../renderer_vulkan/vk_compute_pass.cpp | 137 ++--- .../renderer_vulkan/vk_descriptor_pool.h | 3 +- .../renderer_vulkan/vk_master_semaphore.cpp | 276 ++++------ .../renderer_vulkan/vk_present_manager.cpp | 128 +++-- .../renderer_vulkan/vk_query_cache.cpp | 92 ++-- .../renderer_vulkan/vk_rasterizer.cpp | 51 +- .../renderer_vulkan/vk_render_pass_cache.cpp | 187 +++---- .../renderer_vulkan/vk_scheduler.cpp | 64 +-- .../renderer_vulkan/vk_texture_cache.cpp | 478 ++++++++++-------- .../renderer_vulkan/vk_turbo_mode.cpp | 24 +- .../vulkan_common/vulkan_device.cpp | 6 +- src/video_core/vulkan_common/vulkan_device.h | 8 +- .../vulkan_common/vulkan_wrapper.cpp | 12 +- src/video_core/vulkan_common/vulkan_wrapper.h | 164 ++---- 23 files changed, 1005 insertions(+), 1033 deletions(-) diff --git a/src/video_core/renderer_vulkan/blit_image.cpp b/src/video_core/renderer_vulkan/blit_image.cpp index f9d9460c37..928e7bfb9c 100644 --- a/src/video_core/renderer_vulkan/blit_image.cpp +++ b/src/video_core/renderer_vulkan/blit_image.cpp @@ -454,12 +454,15 @@ VkExtent2D GetConversionExtent(const ImageView& src_image_view) { void TransitionImageLayout(vk::CommandBuffer& cmdbuf, VkImage image, VkImageLayout target_layout, VkImageLayout source_layout = VK_IMAGE_LAYOUT_GENERAL) { - constexpr VkFlags flags{VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_SHADER_READ_BIT}; - const VkImageMemoryBarrier barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + constexpr VkAccessFlags2 flags{VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | + VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_SHADER_READ_BIT}; + const VkImageMemoryBarrier2 barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, .srcAccessMask = flags, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, .dstAccessMask = flags, .oldLayout = source_layout, .newLayout = target_layout, @@ -474,8 +477,7 @@ void TransitionImageLayout(vk::CommandBuffer& cmdbuf, VkImage image, VkImageLayo .layerCount = 1, }, }; - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - 0, barrier); + cmdbuf.PipelineBarrier(barrier); } void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view) { @@ -483,14 +485,22 @@ void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view) const VkImageSubresourceRange subresource_range = SubresourceRangeFromView(image_view); scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([image, subresource_range](vk::CommandBuffer cmdbuf) { - const VkImageMemoryBarrier barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | + VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | + VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | + VK_PIPELINE_STAGE_2_TRANSFER_BIT | + VK_PIPELINE_STAGE_2_EARLY_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT, + .srcAccessMask = VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_SHADER_WRITE_BIT | + VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | + VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -498,17 +508,7 @@ void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view) .image = image, .subresourceRange = subresource_range, }; - cmdbuf.PipelineBarrier( - VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | - VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | - VK_PIPELINE_STAGE_TRANSFER_BIT | - VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT, - VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - 0, - barrier); + cmdbuf.PipelineBarrier(barrier); }); } @@ -991,10 +991,12 @@ void BlitImageHelper::CopyMSAAImpl(VkRenderPass renderpass, VkPipeline pipeline, .layerCount = VK_REMAINING_ARRAY_LAYERS, }; const std::array pre_barriers{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = aspect_info.pre_src_stages, .srcAccessMask = aspect_info.pre_src_access, + .dstStageMask = aspect_info.pre_dst_stages, .dstAccessMask = aspect_info.pre_src_dst_access, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1003,10 +1005,12 @@ void BlitImageHelper::CopyMSAAImpl(VkRenderPass renderpass, VkPipeline pipeline, .image = src, .subresourceRange = barrier_range, }, - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = aspect_info.pre_src_stages, .srcAccessMask = aspect_info.pre_src_access, + .dstStageMask = aspect_info.pre_dst_stages, .dstAccessMask = aspect_info.pre_dst_dst_access, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1016,8 +1020,7 @@ void BlitImageHelper::CopyMSAAImpl(VkRenderPass renderpass, VkPipeline pipeline, .subresourceRange = barrier_range, }, }; - cmdbuf.PipelineBarrier(aspect_info.pre_src_stages, aspect_info.pre_dst_stages, 0, - nullptr, nullptr, pre_barriers); + cmdbuf.PipelineBarrier(0, {}, {}, pre_barriers); const VkRenderPassBeginInfo renderpass_bi{ .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO, .pNext = nullptr, @@ -1053,10 +1056,12 @@ void BlitImageHelper::CopyMSAAImpl(VkRenderPass renderpass, VkPipeline pipeline, cmdbuf.PushConstants(layout, VK_SHADER_STAGE_FRAGMENT_BIT, push_constants); cmdbuf.Draw(3, 1, 0, 0); cmdbuf.EndRenderPass(); - const VkImageMemoryBarrier post_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 post_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = aspect_info.post_src_stages, .srcAccessMask = aspect_info.post_src_access, + .dstStageMask = aspect_info.post_dst_stages, .dstAccessMask = aspect_info.post_dst_access, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1065,8 +1070,7 @@ void BlitImageHelper::CopyMSAAImpl(VkRenderPass renderpass, VkPipeline pipeline, .image = dst, .subresourceRange = barrier_range, }; - cmdbuf.PipelineBarrier(aspect_info.post_src_stages, aspect_info.post_dst_stages, 0, - post_barrier); + cmdbuf.PipelineBarrier(post_barrier); }); msaa_copy_resources.push_back(MSAACopyResources{ .tick = scheduler.CurrentTick(), @@ -1115,21 +1119,21 @@ void BlitImageHelper::CopyMSAA(RenderPassCache& render_pass_cache, VkImage dst_i .src_view_aspect = VK_IMAGE_ASPECT_COLOR_BIT, .attachment_aspect = VK_IMAGE_ASPECT_COLOR_BIT, .barrier_aspect = VK_IMAGE_ASPECT_COLOR_BIT, - .pre_src_access = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_TRANSFER_WRITE_BIT, - .pre_src_dst_access = VK_ACCESS_SHADER_READ_BIT, + .pre_src_access = VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_2_SHADER_WRITE_BIT | + VK_ACCESS_2_TRANSFER_WRITE_BIT, + .pre_src_dst_access = VK_ACCESS_2_SHADER_READ_BIT, .pre_dst_dst_access = - VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, - .pre_src_stages = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | - VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT, + VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT, + .pre_src_stages = VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | + VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | + VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_2_TRANSFER_BIT, .pre_dst_stages = - VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, - .post_src_access = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, - .post_dst_access = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT, - .post_src_stages = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, - .post_dst_stages = VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT, + VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT, + .post_src_access = VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT, + .post_dst_access = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_TRANSFER_READ_BIT, + .post_src_stages = VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT, + .post_dst_stages = VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | + VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_2_TRANSFER_BIT, }; const VkFormat src_vk_format = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, true, src_format).format; @@ -1625,19 +1629,19 @@ void BlitImageHelper::CopyMSAADepth(RenderPassCache& render_pass_cache, VkImage .attachment_aspect = attachment_aspect, .barrier_aspect = attachment_aspect, .pre_src_access = - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, - .pre_src_dst_access = VK_ACCESS_SHADER_READ_BIT, - .pre_dst_dst_access = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, - .pre_src_stages = VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_TRANSFER_BIT, + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, + .pre_src_dst_access = VK_ACCESS_2_SHADER_READ_BIT, + .pre_dst_dst_access = VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, + .pre_src_stages = VK_PIPELINE_STAGE_2_EARLY_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_2_TRANSFER_BIT, .pre_dst_stages = - VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT, - .post_src_access = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, - .post_dst_access = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT, - .post_src_stages = VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT, + VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_2_EARLY_FRAGMENT_TESTS_BIT, + .post_src_access = VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, + .post_dst_access = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_TRANSFER_READ_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT, + .post_src_stages = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT, .post_dst_stages = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, }; const VkFormat src_vk_format = diff --git a/src/video_core/renderer_vulkan/blit_image.h b/src/video_core/renderer_vulkan/blit_image.h index 62e6743a23..895d0f8de4 100644 --- a/src/video_core/renderer_vulkan/blit_image.h +++ b/src/video_core/renderer_vulkan/blit_image.h @@ -140,15 +140,15 @@ private: VkImageAspectFlags src_view_aspect; VkImageAspectFlags attachment_aspect; VkImageAspectFlags barrier_aspect; - VkAccessFlags pre_src_access; - VkAccessFlags pre_src_dst_access; - VkAccessFlags pre_dst_dst_access; - VkPipelineStageFlags pre_src_stages; - VkPipelineStageFlags pre_dst_stages; - VkAccessFlags post_src_access; - VkAccessFlags post_dst_access; - VkPipelineStageFlags post_src_stages; - VkPipelineStageFlags post_dst_stages; + VkAccessFlags2 pre_src_access; + VkAccessFlags2 pre_src_dst_access; + VkAccessFlags2 pre_dst_dst_access; + VkPipelineStageFlags2 pre_src_stages; + VkPipelineStageFlags2 pre_dst_stages; + VkAccessFlags2 post_src_access; + VkAccessFlags2 post_dst_access; + VkPipelineStageFlags2 post_src_stages; + VkPipelineStageFlags2 post_dst_stages; }; void BlitImpl(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, diff --git a/src/video_core/renderer_vulkan/present/frame_gen.cpp b/src/video_core/renderer_vulkan/present/frame_gen.cpp index e6d857323b..d5916d6170 100644 --- a/src/video_core/renderer_vulkan/present/frame_gen.cpp +++ b/src/video_core/renderer_vulkan/present/frame_gen.cpp @@ -113,13 +113,17 @@ void WriteColorPpm(const std::filesystem::path& path, VkExtent2D extent, WritePortablePixmap(path, "P6", extent, rgb); } -VkImageMemoryBarrier MakeTransitionBarrier(VkImage image, VkAccessFlags src_access, - VkAccessFlags dst_access, VkImageLayout old_layout, - VkImageLayout new_layout) { - return VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, +VkImageMemoryBarrier2 MakeTransitionBarrier(VkImage image, VkPipelineStageFlags2 src_stage, + VkAccessFlags2 src_access, + VkPipelineStageFlags2 dst_stage, + VkAccessFlags2 dst_access, VkImageLayout old_layout, + VkImageLayout new_layout) { + return VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = src_stage, .srcAccessMask = src_access, + .dstStageMask = dst_stage, .dstAccessMask = dst_access, .oldLayout = old_layout, .newLayout = new_layout, @@ -160,29 +164,31 @@ void CopyPresentedFrame(vk::CommandBuffer cmdbuf, VkImage source, LsfgImage& des VkExtent2D extent) { const auto make_barrier = MakeTransitionBarrier; + static constexpr VkPipelineStageFlags2 present_stage = + VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT; + const std::array before{ - make_barrier(source, VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, VK_ACCESS_TRANSFER_READ_BIT, + make_barrier(source, present_stage, VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT, + VK_PIPELINE_STAGE_2_TRANSFER_BIT, VK_ACCESS_2_TRANSFER_READ_BIT, VK_IMAGE_LAYOUT_GENERAL, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL), - make_barrier(destination.Handle(), VK_ACCESS_SHADER_READ_BIT, VK_ACCESS_TRANSFER_WRITE_BIT, + make_barrier(destination.Handle(), present_stage, VK_ACCESS_2_SHADER_READ_BIT, + VK_PIPELINE_STAGE_2_TRANSFER_BIT, VK_ACCESS_2_TRANSFER_WRITE_BIT, destination.Layout(), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL), }; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - VK_PIPELINE_STAGE_TRANSFER_BIT, 0, {}, {}, before); + cmdbuf.PipelineBarrier(0, {}, {}, before); cmdbuf.CopyImage(source, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, destination.Handle(), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, MakeCopyRegion(extent)); const std::array after{ - make_barrier(source, VK_ACCESS_TRANSFER_READ_BIT, VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, + make_barrier(source, VK_PIPELINE_STAGE_2_TRANSFER_BIT, VK_ACCESS_2_TRANSFER_READ_BIT, + present_stage, VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, VK_IMAGE_LAYOUT_GENERAL), - make_barrier(destination.Handle(), VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT, + make_barrier(destination.Handle(), VK_PIPELINE_STAGE_2_TRANSFER_BIT, + VK_ACCESS_2_TRANSFER_WRITE_BIT, present_stage, VK_ACCESS_2_SHADER_READ_BIT, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_IMAGE_LAYOUT_GENERAL), }; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - 0, {}, {}, after); + cmdbuf.PipelineBarrier(0, {}, {}, after); destination.SetLayout(VK_IMAGE_LAYOUT_GENERAL); } diff --git a/src/video_core/renderer_vulkan/present/layer.cpp b/src/video_core/renderer_vulkan/present/layer.cpp index 1868142f51..562928b953 100644 --- a/src/video_core/renderer_vulkan/present/layer.cpp +++ b/src/video_core/renderer_vulkan/present/layer.cpp @@ -364,11 +364,13 @@ void Layer::UpdateRawImage(const Tegra::FramebufferConfig& framebuffer, size_t i }; scheduler.Record([this, copy, index = image_index](vk::CommandBuffer cmdbuf) { const VkImage image = *raw_images[index]; - const VkImageMemoryBarrier base_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 base_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = 0, - .dstAccessMask = 0, + .srcStageMask = VK_PIPELINE_STAGE_2_NONE, + .srcAccessMask = VK_ACCESS_2_NONE, + .dstStageMask = VK_PIPELINE_STAGE_2_NONE, + .dstAccessMask = VK_ACCESS_2_NONE, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -382,23 +384,24 @@ void Layer::UpdateRawImage(const Tegra::FramebufferConfig& framebuffer, size_t i .layerCount = 1, }, }; - VkImageMemoryBarrier read_barrier = base_barrier; - read_barrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; + VkImageMemoryBarrier2 read_barrier = base_barrier; + read_barrier.srcStageMask = VK_PIPELINE_STAGE_2_HOST_BIT; + read_barrier.dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT; + read_barrier.dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT; read_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED; read_barrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL; - VkImageMemoryBarrier write_barrier = base_barrier; - write_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; - write_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; + VkImageMemoryBarrier2 write_barrier = base_barrier; + write_barrier.srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT; + write_barrier.srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT; + write_barrier.dstStageMask = VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | + VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT; + write_barrier.dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT; write_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, - read_barrier); + cmdbuf.PipelineBarrier(read_barrier); cmdbuf.CopyBufferToImage(*buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copy); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - 0, write_barrier); + cmdbuf.PipelineBarrier(write_barrier); }); } diff --git a/src/video_core/renderer_vulkan/present/lsfg_common.cpp b/src/video_core/renderer_vulkan/present/lsfg_common.cpp index d2df2755e5..74e5d446d8 100644 --- a/src/video_core/renderer_vulkan/present/lsfg_common.cpp +++ b/src/video_core/renderer_vulkan/present/lsfg_common.cpp @@ -67,12 +67,14 @@ vk::Buffer CreateUniformBuffer(MemoryAllocator& memory_allocator, VkDeviceSize s return memory_allocator.CreateBuffer(buffer_ci, MemoryUsage::Upload); } -VkImageMemoryBarrier MakeBarrier(const LsfgImage& image, VkAccessFlags src_access, - VkAccessFlags dst_access) { - return VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, +VkImageMemoryBarrier2 MakeBarrier(const LsfgImage& image, VkAccessFlags2 src_access, + VkAccessFlags2 dst_access) { + return VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, .srcAccessMask = src_access, + .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, .dstAccessMask = dst_access, .oldLayout = image.Layout(), .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -98,19 +100,19 @@ LsfgImage::LsfgImage(const Device& device, MemoryAllocator& memory_allocator, Vk view = CreateWrappedImageView(device, image, format); } -LsfgBarriers& LsfgBarriers::Push(LsfgImage& image, VkAccessFlags src_access, - VkAccessFlags dst_access) { +LsfgBarriers& LsfgBarriers::Push(LsfgImage& image, VkAccessFlags2 src_access, + VkAccessFlags2 dst_access) { barriers.push_back(MakeBarrier(image, src_access, dst_access)); image.SetLayout(VK_IMAGE_LAYOUT_GENERAL); return *this; } LsfgBarriers& LsfgBarriers::WriteToRead(LsfgImage& image) { - return Push(image, VK_ACCESS_SHADER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT); + return Push(image, VK_ACCESS_2_SHADER_WRITE_BIT, VK_ACCESS_2_SHADER_READ_BIT); } LsfgBarriers& LsfgBarriers::ReadToWrite(LsfgImage& image) { - return Push(image, VK_ACCESS_SHADER_READ_BIT, VK_ACCESS_SHADER_WRITE_BIT); + return Push(image, VK_ACCESS_2_SHADER_READ_BIT, VK_ACCESS_2_SHADER_WRITE_BIT); } LsfgBarriers& LsfgBarriers::WriteToRead(LsfgImage* image) { @@ -122,11 +124,13 @@ LsfgBarriers& LsfgBarriers::ReadToWrite(LsfgImage* image) { } LsfgBarriers& LsfgBarriers::DiscardToWrite(VkImage image) { - barriers.push_back(VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + barriers.push_back(VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = 0, - .dstAccessMask = VK_ACCESS_SHADER_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .srcAccessMask = VK_ACCESS_2_NONE, + .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .dstAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -199,8 +203,7 @@ void LsfgBarriers::Build() { if (barriers.empty()) { return; } - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, {}, {}, barriers); + cmdbuf.PipelineBarrier(0, {}, {}, barriers); barriers.clear(); } diff --git a/src/video_core/renderer_vulkan/present/lsfg_common.h b/src/video_core/renderer_vulkan/present/lsfg_common.h index d1e3271a6d..bc538f4d1b 100644 --- a/src/video_core/renderer_vulkan/present/lsfg_common.h +++ b/src/video_core/renderer_vulkan/present/lsfg_common.h @@ -147,10 +147,10 @@ public: void Build(); private: - LsfgBarriers& Push(LsfgImage& image, VkAccessFlags src_access, VkAccessFlags dst_access); + LsfgBarriers& Push(LsfgImage& image, VkAccessFlags2 src_access, VkAccessFlags2 dst_access); vk::CommandBuffer cmdbuf; - std::vector barriers; + std::vector barriers; }; class LsfgDescriptorWriter { diff --git a/src/video_core/renderer_vulkan/present/lsfg_generate.cpp b/src/video_core/renderer_vulkan/present/lsfg_generate.cpp index 4473b59a2f..2aa470bdb0 100644 --- a/src/video_core/renderer_vulkan/present/lsfg_generate.cpp +++ b/src/video_core/renderer_vulkan/present/lsfg_generate.cpp @@ -22,12 +22,15 @@ constexpr u32 DISPATCH_TILE_SHIFT = 4; return (size + (1u << DISPATCH_TILE_SHIFT) - 1) >> DISPATCH_TILE_SHIFT; } -VkImageMemoryBarrier MakeTargetBarrier(VkImage image, VkAccessFlags src_access, - VkAccessFlags dst_access, VkImageLayout old_layout) { - return VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, +VkImageMemoryBarrier2 MakeTargetBarrier(VkImage image, VkPipelineStageFlags2 src_stage, + VkAccessFlags2 src_access, VkPipelineStageFlags2 dst_stage, + VkAccessFlags2 dst_access, VkImageLayout old_layout) { + return VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = src_stage, .srcAccessMask = src_access, + .dstStageMask = dst_stage, .dstAccessMask = dst_access, .oldLayout = old_layout, .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -118,13 +121,11 @@ void LsfgGenerate::Dispatch(vk::CommandBuffer cmdbuf, u64 frame_count, size_t sl cmdbuf.Dispatch(GroupCount(extent.width), GroupCount(extent.height), 1); const std::array after{MakeTargetBarrier( - image, VK_ACCESS_SHADER_WRITE_BIT, - VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT, + image, VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, VK_ACCESS_2_SHADER_WRITE_BIT, + VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_2_TRANSFER_BIT, + VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_2_TRANSFER_READ_BIT, VK_IMAGE_LAYOUT_GENERAL)}; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, {}, {}, after); + cmdbuf.PipelineBarrier(0, {}, {}, after); } } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/present/util.cpp b/src/video_core/renderer_vulkan/present/util.cpp index 11250687d6..131973bdd6 100644 --- a/src/video_core/renderer_vulkan/present/util.cpp +++ b/src/video_core/renderer_vulkan/present/util.cpp @@ -49,12 +49,15 @@ vk::Image CreateWrappedImage(MemoryAllocator& allocator, VkExtent2D dimensions, void TransitionImageLayout(vk::CommandBuffer& cmdbuf, VkImage image, VkImageLayout target_layout, VkImageLayout source_layout) { - constexpr VkFlags flags{VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_SHADER_READ_BIT}; - const VkImageMemoryBarrier barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + constexpr VkAccessFlags2 flags{VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | + VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_SHADER_READ_BIT}; + const VkImageMemoryBarrier2 barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, .srcAccessMask = flags, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, .dstAccessMask = flags, .oldLayout = source_layout, .newLayout = target_layout, @@ -69,8 +72,7 @@ void TransitionImageLayout(vk::CommandBuffer& cmdbuf, VkImage image, VkImageLayo .layerCount = 1, }, }; - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - 0, barrier); + cmdbuf.PipelineBarrier(barrier); } void UploadImage(const Device& device, MemoryAllocator& allocator, Scheduler& scheduler, @@ -116,11 +118,13 @@ void UploadImage(const Device& device, MemoryAllocator& allocator, Scheduler& sc void DownloadColorImage(vk::CommandBuffer& cmdbuf, VkImage image, VkBuffer buffer, VkExtent3D extent) { - const VkImageMemoryBarrier read_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 read_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -134,11 +138,13 @@ void DownloadColorImage(vk::CommandBuffer& cmdbuf, VkImage image, VkBuffer buffe .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }; - const VkImageMemoryBarrier image_write_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 image_write_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = 0, - .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_NONE, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -152,11 +158,13 @@ void DownloadColorImage(vk::CommandBuffer& cmdbuf, VkImage image, VkBuffer buffe .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }; - static constexpr VkMemoryBarrier memory_write_barrier{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 memory_write_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, }; const VkBufferImageCopy copy{ .bufferOffset = 0, @@ -171,11 +179,9 @@ void DownloadColorImage(vk::CommandBuffer& cmdbuf, VkImage image, VkBuffer buffe .imageOffset{.x = 0, .y = 0, .z = 0}, .imageExtent{extent}, }; - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, - read_barrier); + cmdbuf.PipelineBarrier(read_barrier); cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, buffer, copy); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, 0, - memory_write_barrier, nullptr, image_write_barrier); + cmdbuf.PipelineBarrier(0, memory_write_barrier, {}, image_write_barrier); } vk::ImageView CreateWrappedImageView(const Device& device, vk::Image& image, VkFormat format) { diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 8c93a92046..56f0e7b870 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -227,11 +227,13 @@ public: .dstOffset = 0, .size = size_bytes, }; - const VkBufferMemoryBarrier write_barrier{ - .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, + const VkBufferMemoryBarrier2 write_barrier{ + .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_INDEX_READ_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_VERTEX_INPUT_BIT, + .dstAccessMask = VK_ACCESS_2_INDEX_READ_BIT, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .buffer = dst_buffer, @@ -239,8 +241,7 @@ public: .size = size_bytes, }; cmdbuf.CopyBuffer(src_buffer, dst_buffer, copy); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, write_barrier); + cmdbuf.PipelineBarrier(write_barrier); }); } else { buffer.Flush(); @@ -456,17 +457,21 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer, if (dst_buffer == VK_NULL_HANDLE || src_buffer == VK_NULL_HANDLE) { return; } - static constexpr VkMemoryBarrier READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, }; - static constexpr VkMemoryBarrier WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, }; // Measuring a popular game, this number never exceeds the specified size once data is warmed up @@ -483,42 +488,42 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer, scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([src_buffer, dst_buffer, vk_copies, barrier](vk::CommandBuffer cmdbuf) { if (barrier) { - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - VK_PIPELINE_STAGE_TRANSFER_BIT, 0, READ_BARRIER); + cmdbuf.PipelineBarrier(READ_BARRIER); } cmdbuf.CopyBuffer(src_buffer, dst_buffer, VideoCommon::FixSmallVectorADL(vk_copies)); if (barrier) { - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, - vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, 0, WRITE_BARRIER); + cmdbuf.PipelineBarrier(WRITE_BARRIER); } }); } void BufferCacheRuntime::PreCopyBarrier() { - static constexpr VkMemoryBarrier READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, }; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, READ_BARRIER); + cmdbuf.PipelineBarrier(READ_BARRIER); }); } void BufferCacheRuntime::PostCopyBarrier() { - static constexpr VkMemoryBarrier WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, }; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - 0, WRITE_BARRIER); + cmdbuf.PipelineBarrier(WRITE_BARRIER); }); } @@ -526,26 +531,28 @@ void BufferCacheRuntime::ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t si if (dest_buffer == VK_NULL_HANDLE) { return; } - static constexpr VkMemoryBarrier READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, }; - static constexpr VkMemoryBarrier WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, }; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([dest_buffer, offset, size, value](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, READ_BARRIER); + cmdbuf.PipelineBarrier(READ_BARRIER); cmdbuf.FillBuffer(dest_buffer, offset, size, value); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - 0, WRITE_BARRIER); + cmdbuf.PipelineBarrier(WRITE_BARRIER); }); } diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 4dd65f0269..ca5a267c6a 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -322,19 +322,20 @@ std::pair Uint8Pass::Assemble(u32 num_vertices, VkBuffer scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([this, descriptor_data, num_vertices](vk::CommandBuffer cmdbuf) { static constexpr u32 DISPATCH_SIZE = 1024; - static constexpr VkMemoryBarrier WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_VERTEX_INPUT_BIT, + .dstAccessMask = VK_ACCESS_2_VERTEX_ATTRIBUTE_READ_BIT, }; const VkDescriptorSet set = descriptor_allocator.Commit(); device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline); cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); cmdbuf.Dispatch(Common::DivCeil(num_vertices, DISPATCH_SIZE), 1, 1); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, WRITE_BARRIER); + cmdbuf.PipelineBarrier(WRITE_BARRIER); }); return {staging.buffer, staging.offset}; } @@ -391,11 +392,13 @@ std::pair QuadIndexedPass::Assemble( scheduler.Record([this, descriptor_data, num_tri_vertices, base_vertex, index_shift, is_strip](vk::CommandBuffer cmdbuf) { static constexpr u32 DISPATCH_SIZE = 1024; - static constexpr VkMemoryBarrier WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_INDEX_READ_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_VERTEX_INPUT_BIT, + .dstAccessMask = VK_ACCESS_2_INDEX_READ_BIT, }; const std::array push_constants{base_vertex, index_shift, is_strip ? 1u : 0u}; const VkDescriptorSet set = descriptor_allocator.Commit(); @@ -405,8 +408,7 @@ std::pair QuadIndexedPass::Assemble( cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(push_constants), &push_constants); cmdbuf.Dispatch(Common::DivCeil(num_tri_vertices, DISPATCH_SIZE), 1, 1); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, WRITE_BARRIER); + cmdbuf.PipelineBarrier(WRITE_BARRIER); }); return {staging.buffer, staging.offset}; } @@ -441,11 +443,13 @@ std::pair IndirectQuadsPass::Assemble(u32 num_draws, u32 scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([this, descriptor_data, num_draws, src_stride](vk::CommandBuffer cmdbuf) { static constexpr u32 DISPATCH_SIZE = 32; - static constexpr VkMemoryBarrier WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_INDIRECT_COMMAND_READ_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_DRAW_INDIRECT_BIT, + .dstAccessMask = VK_ACCESS_2_INDIRECT_COMMAND_READ_BIT, }; const std::array push_constants{num_draws, src_stride}; const VkDescriptorSet set = descriptor_allocator.Commit(); @@ -455,8 +459,7 @@ std::pair IndirectQuadsPass::Assemble(u32 num_draws, u32 cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(push_constants), &push_constants); cmdbuf.Dispatch(Common::DivCeil(num_draws, DISPATCH_SIZE), 1, 1); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT, 0, WRITE_BARRIER); + cmdbuf.PipelineBarrier(WRITE_BARRIER); }); return {staging.buffer, staging.offset}; } @@ -484,17 +487,21 @@ void ConditionalRenderingResolvePass::Resolve(VkBuffer dst_buffer, VkBuffer src_ scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([this, descriptor_data, compare_to_zero](vk::CommandBuffer cmdbuf) { - static constexpr VkMemoryBarrier read_barrier{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 read_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_SHADER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT | VK_ACCESS_2_SHADER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT, }; - static constexpr VkMemoryBarrier write_barrier{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 write_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_CONDITIONAL_RENDERING_READ_BIT_EXT, + .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_CONDITIONAL_RENDERING_BIT_EXT, + .dstAccessMask = VK_ACCESS_2_CONDITIONAL_RENDERING_READ_BIT_EXT, }; const ConditionalRenderingResolvePushConstants uniforms{ .compare_to_zero = compare_to_zero ? 1U : 0U, @@ -502,15 +509,12 @@ void ConditionalRenderingResolvePass::Resolve(VkBuffer dst_buffer, VkBuffer src_ const VkDescriptorSet set = descriptor_allocator.Commit(); device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, read_barrier); + cmdbuf.PipelineBarrier(read_barrier); cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline); cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, uniforms); cmdbuf.Dispatch(1, 1, 1); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - VK_PIPELINE_STAGE_CONDITIONAL_RENDERING_BIT_EXT, 0, - write_barrier); + cmdbuf.PipelineBarrier(write_barrier); }); } @@ -532,14 +536,14 @@ QueriesPrefixScanPass::QueriesPrefixScanPass( void QueriesPrefixScanPass::Run(VkBuffer accumulation_buffer, VkBuffer dst_buffer, VkBuffer src_buffer, size_t number_of_sums, size_t min_accumulation_limit, size_t max_accumulation_limit) { - constexpr VkAccessFlags BASE_DST_ACCESS = VK_ACCESS_SHADER_READ_BIT | - VK_ACCESS_TRANSFER_READ_BIT | - VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | - VK_ACCESS_INDIRECT_COMMAND_READ_BIT | - VK_ACCESS_INDEX_READ_BIT | - VK_ACCESS_UNIFORM_READ_BIT; - const VkAccessFlags conditional_access = - device.IsExtConditionalRendering() ? VK_ACCESS_CONDITIONAL_RENDERING_READ_BIT_EXT : 0; + constexpr VkAccessFlags2 BASE_DST_ACCESS = VK_ACCESS_2_SHADER_READ_BIT | + VK_ACCESS_2_TRANSFER_READ_BIT | + VK_ACCESS_2_VERTEX_ATTRIBUTE_READ_BIT | + VK_ACCESS_2_INDIRECT_COMMAND_READ_BIT | + VK_ACCESS_2_INDEX_READ_BIT | + VK_ACCESS_2_UNIFORM_READ_BIT; + const VkAccessFlags2 conditional_access = + device.IsExtConditionalRendering() ? VK_ACCESS_2_CONDITIONAL_RENDERING_READ_BIT_EXT : 0; size_t current_runs = number_of_sums; size_t offset = 0; while (current_runs != 0) { @@ -557,16 +561,20 @@ void QueriesPrefixScanPass::Run(VkBuffer accumulation_buffer, VkBuffer dst_buffe scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([this, descriptor_data, min_accumulation_limit, max_accumulation_limit, runs_to_do, used_offset, conditional_access](vk::CommandBuffer cmdbuf) { - static constexpr VkMemoryBarrier read_barrier{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 read_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT, }; - const VkMemoryBarrier write_barrier{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + const VkMemoryBarrier2 write_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, .dstAccessMask = BASE_DST_ACCESS | conditional_access, }; const QueriesPrefixScanPushConstants uniforms{ @@ -578,14 +586,12 @@ void QueriesPrefixScanPass::Run(VkBuffer accumulation_buffer, VkBuffer dst_buffe const VkDescriptorSet set = descriptor_allocator.Commit(); device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, read_barrier); + cmdbuf.PipelineBarrier(read_barrier); cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline); cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, uniforms); cmdbuf.Dispatch(1, 1, 1); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, 0, write_barrier); + cmdbuf.PipelineBarrier(write_barrier); }); } } @@ -594,21 +600,23 @@ namespace { void RecordUnswizzleBeginBarrier(Scheduler& scheduler, VkPipeline vk_pipeline, VkImage vk_image, VkImageAspectFlags aspect_mask, bool is_initialized) { - VkAccessFlags src_access = VK_ACCESS_NONE; + VkAccessFlags2 src_access = VK_ACCESS_2_NONE; VkImageLayout old_layout = VK_IMAGE_LAYOUT_UNDEFINED; - VkPipelineStageFlags src_stage = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT; + VkPipelineStageFlags2 src_stage = VK_PIPELINE_STAGE_2_TOP_OF_PIPE_BIT; if (is_initialized) { - src_access = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT; + src_access = VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT; old_layout = VK_IMAGE_LAYOUT_GENERAL; src_stage = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER; } scheduler.Record([vk_pipeline, vk_image, aspect_mask, src_access, old_layout, src_stage](vk::CommandBuffer cmdbuf) { - const VkImageMemoryBarrier image_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 image_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = src_stage, .srcAccessMask = src_access, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT, .oldLayout = old_layout, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -622,7 +630,7 @@ void RecordUnswizzleBeginBarrier(Scheduler& scheduler, VkPipeline vk_pipeline, V .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }; - cmdbuf.PipelineBarrier(src_stage, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier); + cmdbuf.PipelineBarrier(image_barrier); cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline); }); } @@ -630,12 +638,14 @@ void RecordUnswizzleBeginBarrier(Scheduler& scheduler, VkPipeline vk_pipeline, V void RecordUnswizzleEndBarrier(Scheduler& scheduler, VkImage vk_image, VkImageAspectFlags aspect_mask) { scheduler.Record([vk_image, aspect_mask](vk::CommandBuffer cmdbuf) { - const VkImageMemoryBarrier image_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 image_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT | + VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -649,8 +659,7 @@ void RecordUnswizzleEndBarrier(Scheduler& scheduler, VkImage vk_image, .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, 0, image_barrier); + cmdbuf.PipelineBarrier(image_barrier); }); } diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.h b/src/video_core/renderer_vulkan/vk_descriptor_pool.h index 6a80f8e107..a524c2a043 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_pool.h +++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.h @@ -65,7 +65,8 @@ private: class DescriptorPool { public: - DescriptorPool(); + +DescriptorPool(); ~DescriptorPool(); DescriptorPool& operator=(const DescriptorPool&) = delete; diff --git a/src/video_core/renderer_vulkan/vk_master_semaphore.cpp b/src/video_core/renderer_vulkan/vk_master_semaphore.cpp index 9ca215b425..e52e079969 100644 --- a/src/video_core/renderer_vulkan/vk_master_semaphore.cpp +++ b/src/video_core/renderer_vulkan/vk_master_semaphore.cpp @@ -121,8 +121,9 @@ VkResult MasterSemaphore::SubmitQueue(vk::CommandBuffer& cmdbuf, vk::CommandBuff } } -static constexpr VkPipelineStageFlags wait_stage_mask = VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | - VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT; +static constexpr VkPipelineStageFlags2 wait_stage_mask = + VK_PIPELINE_STAGE_2_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | + VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT; VkResult MasterSemaphore::SubmitQueueTimeline(vk::CommandBuffer& cmdbuf, vk::CommandBuffer& upload_cmdbuf, @@ -130,130 +131,35 @@ VkResult MasterSemaphore::SubmitQueueTimeline(vk::CommandBuffer& cmdbuf, VkSemaphore wait_semaphore, u64 host_tick) { const VkSemaphore timeline_semaphore = *semaphore; - if (device.HasSynchronization2()) { - const std::array cmdbuffer_infos{{ - { - .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO, - .pNext = nullptr, - .commandBuffer = *upload_cmdbuf, - .deviceMask = 0, - }, - { - .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO, - .pNext = nullptr, - .commandBuffer = *cmdbuf, - .deviceMask = 0, - }, - }}; + const std::array cmdbuffer_infos{{ + { + .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO, + .pNext = nullptr, + .commandBuffer = *upload_cmdbuf, + .deviceMask = 0, + }, + { + .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO, + .pNext = nullptr, + .commandBuffer = *cmdbuf, + .deviceMask = 0, + }, + }}; - std::array signal_infos{{ - { - .sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO, - .pNext = nullptr, - .semaphore = timeline_semaphore, - .value = host_tick, - .stageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, - .deviceIndex = 0, - }, - {}, - }}; - u32 num_signal_semaphores = 1; - if (signal_semaphore) { - signal_infos[1] = VkSemaphoreSubmitInfo{ - .sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO, - .pNext = nullptr, - .semaphore = signal_semaphore, - .value = 0, - .stageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, - .deviceIndex = 0, - }; - num_signal_semaphores = 2; - } - - const u32 num_wait_semaphores = wait_semaphore ? 1 : 0; - const VkSemaphoreSubmitInfo wait_info{ + std::array signal_infos{{ + { .sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO, .pNext = nullptr, - .semaphore = wait_semaphore, - .value = 0, - .stageMask = static_cast(wait_stage_mask), + .semaphore = timeline_semaphore, + .value = host_tick, + .stageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, .deviceIndex = 0, - }; - - const VkSubmitInfo2 submit_info2{ - .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO_2, - .pNext = nullptr, - .flags = 0, - .waitSemaphoreInfoCount = num_wait_semaphores, - .pWaitSemaphoreInfos = num_wait_semaphores ? &wait_info : nullptr, - .commandBufferInfoCount = static_cast(cmdbuffer_infos.size()), - .pCommandBufferInfos = cmdbuffer_infos.data(), - .signalSemaphoreInfoCount = num_signal_semaphores, - .pSignalSemaphoreInfos = signal_infos.data(), - }; - return device.GetGraphicsQueue().Submit2(submit_info2); - } - - const u32 num_signal_semaphores = signal_semaphore ? 2 : 1; - const std::array signal_values{host_tick, u64(0)}; - const std::array signal_semaphores{timeline_semaphore, signal_semaphore}; - - const std::array cmdbuffers{*upload_cmdbuf, *cmdbuf}; - - const u32 num_wait_semaphores = wait_semaphore ? 1 : 0; - // Pointers must be null when the count is zero (best-practices) - const VkSemaphore* p_wait_sems = - (num_wait_semaphores > 0) ? &wait_semaphore : nullptr; - const VkPipelineStageFlags* p_wait_masks = - (num_wait_semaphores > 0) ? &wait_stage_mask : nullptr; - const VkSemaphore* p_signal_sems = - (num_signal_semaphores > 0) ? signal_semaphores.data() : nullptr; - const u64 wait_zero = 0; // dummy for binary wait - const VkTimelineSemaphoreSubmitInfo timeline_si{ - .sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO, - .pNext = nullptr, - .waitSemaphoreValueCount = num_wait_semaphores, - .pWaitSemaphoreValues = num_wait_semaphores ? &wait_zero : nullptr, - .signalSemaphoreValueCount = num_signal_semaphores, - .pSignalSemaphoreValues = signal_values.data(), - }; - const VkSubmitInfo submit_info{ - .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO, - .pNext = &timeline_si, - .waitSemaphoreCount = num_wait_semaphores, - .pWaitSemaphores = p_wait_sems, - .pWaitDstStageMask = p_wait_masks, - .commandBufferCount = static_cast(cmdbuffers.size()), - .pCommandBuffers = cmdbuffers.data(), - .signalSemaphoreCount = num_signal_semaphores, - .pSignalSemaphores = p_signal_sems, - }; - - return device.GetGraphicsQueue().Submit(submit_info); -} - -VkResult MasterSemaphore::SubmitQueueFence(vk::CommandBuffer& cmdbuf, - vk::CommandBuffer& upload_cmdbuf, - VkSemaphore signal_semaphore, VkSemaphore wait_semaphore, - u64 host_tick) { - if (device.HasSynchronization2()) { - const std::array cmdbuffer_infos{{ - { - .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO, - .pNext = nullptr, - .commandBuffer = *upload_cmdbuf, - .deviceMask = 0, - }, - { - .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO, - .pNext = nullptr, - .commandBuffer = *cmdbuf, - .deviceMask = 0, - }, - }}; - - const u32 num_signal_semaphores = signal_semaphore ? 1 : 0; - const VkSemaphoreSubmitInfo signal_info{ + }, + {}, + }}; + u32 num_signal_semaphores = 1; + if (signal_semaphore) { + signal_infos[1] = VkSemaphoreSubmitInfo{ .sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO, .pNext = nullptr, .semaphore = signal_semaphore, @@ -261,66 +167,86 @@ VkResult MasterSemaphore::SubmitQueueFence(vk::CommandBuffer& cmdbuf, .stageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, .deviceIndex = 0, }; - - const u32 num_wait_semaphores = wait_semaphore ? 1 : 0; - const VkSemaphoreSubmitInfo wait_info{ - .sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO, - .pNext = nullptr, - .semaphore = wait_semaphore, - .value = 0, - .stageMask = static_cast(wait_stage_mask), - .deviceIndex = 0, - }; - - const VkSubmitInfo2 submit_info2{ - .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO_2, - .pNext = nullptr, - .flags = 0, - .waitSemaphoreInfoCount = num_wait_semaphores, - .pWaitSemaphoreInfos = num_wait_semaphores ? &wait_info : nullptr, - .commandBufferInfoCount = static_cast(cmdbuffer_infos.size()), - .pCommandBufferInfos = cmdbuffer_infos.data(), - .signalSemaphoreInfoCount = num_signal_semaphores, - .pSignalSemaphoreInfos = num_signal_semaphores ? &signal_info : nullptr, - }; - - auto fence = GetFreeFence(); - auto result = device.GetGraphicsQueue().Submit2(submit_info2, *fence); - - if (result == VK_SUCCESS) { - std::scoped_lock lock{wait_mutex}; - wait_queue.emplace(host_tick, std::move(fence)); - wait_cv.notify_one(); - } - - return result; + num_signal_semaphores = 2; } - const u32 num_signal_semaphores = signal_semaphore ? 1 : 0; const u32 num_wait_semaphores = wait_semaphore ? 1 : 0; - - const VkSemaphore* p_wait_sems = - (num_wait_semaphores > 0) ? &wait_semaphore : nullptr; - const VkPipelineStageFlags* p_wait_masks = - (num_wait_semaphores > 0) ? &wait_stage_mask : nullptr; - const VkSemaphore* p_signal_sems = - (num_signal_semaphores > 0) ? &signal_semaphore : nullptr; - const std::array cmdbuffers{*upload_cmdbuf, *cmdbuf}; - - const VkSubmitInfo submit_info{ - .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO, + const VkSemaphoreSubmitInfo wait_info{ + .sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO, .pNext = nullptr, - .waitSemaphoreCount = num_wait_semaphores, - .pWaitSemaphores = p_wait_sems, - .pWaitDstStageMask = p_wait_masks, - .commandBufferCount = static_cast(cmdbuffers.size()), - .pCommandBuffers = cmdbuffers.data(), - .signalSemaphoreCount = num_signal_semaphores, - .pSignalSemaphores = p_signal_sems, + .semaphore = wait_semaphore, + .value = 0, + .stageMask = wait_stage_mask, + .deviceIndex = 0, + }; + + const VkSubmitInfo2 submit_info2{ + .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO_2, + .pNext = nullptr, + .flags = 0, + .waitSemaphoreInfoCount = num_wait_semaphores, + .pWaitSemaphoreInfos = num_wait_semaphores ? &wait_info : nullptr, + .commandBufferInfoCount = static_cast(cmdbuffer_infos.size()), + .pCommandBufferInfos = cmdbuffer_infos.data(), + .signalSemaphoreInfoCount = num_signal_semaphores, + .pSignalSemaphoreInfos = signal_infos.data(), + }; + return device.GetGraphicsQueue().Submit(submit_info2); +} + +VkResult MasterSemaphore::SubmitQueueFence(vk::CommandBuffer& cmdbuf, + vk::CommandBuffer& upload_cmdbuf, + VkSemaphore signal_semaphore, VkSemaphore wait_semaphore, + u64 host_tick) { + const std::array cmdbuffer_infos{{ + { + .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO, + .pNext = nullptr, + .commandBuffer = *upload_cmdbuf, + .deviceMask = 0, + }, + { + .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO, + .pNext = nullptr, + .commandBuffer = *cmdbuf, + .deviceMask = 0, + }, + }}; + + const u32 num_signal_semaphores = signal_semaphore ? 1 : 0; + const VkSemaphoreSubmitInfo signal_info{ + .sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO, + .pNext = nullptr, + .semaphore = signal_semaphore, + .value = 0, + .stageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, + .deviceIndex = 0, + }; + + const u32 num_wait_semaphores = wait_semaphore ? 1 : 0; + const VkSemaphoreSubmitInfo wait_info{ + .sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO, + .pNext = nullptr, + .semaphore = wait_semaphore, + .value = 0, + .stageMask = wait_stage_mask, + .deviceIndex = 0, + }; + + const VkSubmitInfo2 submit_info2{ + .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO_2, + .pNext = nullptr, + .flags = 0, + .waitSemaphoreInfoCount = num_wait_semaphores, + .pWaitSemaphoreInfos = num_wait_semaphores ? &wait_info : nullptr, + .commandBufferInfoCount = static_cast(cmdbuffer_infos.size()), + .pCommandBufferInfos = cmdbuffer_infos.data(), + .signalSemaphoreInfoCount = num_signal_semaphores, + .pSignalSemaphoreInfos = num_signal_semaphores ? &signal_info : nullptr, }; auto fence = GetFreeFence(); - auto result = device.GetGraphicsQueue().Submit(submit_info, *fence); + auto result = device.GetGraphicsQueue().Submit(submit_info2, *fence); if (result == VK_SUCCESS) { std::scoped_lock lock{wait_mutex}; diff --git a/src/video_core/renderer_vulkan/vk_present_manager.cpp b/src/video_core/renderer_vulkan/vk_present_manager.cpp index 9298764546..4db267f724 100644 --- a/src/video_core/renderer_vulkan/vk_present_manager.cpp +++ b/src/video_core/renderer_vulkan/vk_present_manager.cpp @@ -378,18 +378,24 @@ void PresentManager::SetImageCount() { } void PresentManager::DiscardFrame(Frame* frame) { - static constexpr VkPipelineStageFlags wait_stage = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT; - const VkSemaphore render_ready = *frame->render_ready; - const VkSubmitInfo submit_info{ - .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO, + const VkSemaphoreSubmitInfo wait_info{ + .sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO, .pNext = nullptr, - .waitSemaphoreCount = 1U, - .pWaitSemaphores = &render_ready, - .pWaitDstStageMask = &wait_stage, - .commandBufferCount = 0U, - .pCommandBuffers = nullptr, - .signalSemaphoreCount = 0U, - .pSignalSemaphores = nullptr, + .semaphore = *frame->render_ready, + .value = 0, + .stageMask = VK_PIPELINE_STAGE_2_TOP_OF_PIPE_BIT, + .deviceIndex = 0, + }; + const VkSubmitInfo2 submit_info{ + .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO_2, + .pNext = nullptr, + .flags = 0, + .waitSemaphoreInfoCount = 1U, + .pWaitSemaphoreInfos = &wait_info, + .commandBufferInfoCount = 0U, + .pCommandBufferInfos = nullptr, + .signalSemaphoreInfoCount = 0U, + .pSignalSemaphoreInfos = nullptr, }; std::scoped_lock submit_lock{scheduler.submit_mutex}; @@ -449,11 +455,13 @@ void PresentManager::CopyToSwapchainImpl(Frame* frame) { const VkImage image{swapchain.CurrentImage()}; const VkExtent2D extent = swapchain.GetExtent(); const std::array pre_barriers{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = 0, - .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_NONE, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -467,11 +475,13 @@ void PresentManager::CopyToSwapchainImpl(Frame* frame) { .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }, - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -487,11 +497,13 @@ void PresentManager::CopyToSwapchainImpl(Frame* frame) { }, }; const std::array post_barriers{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_ALL_GRAPHICS_BIT, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -505,11 +517,13 @@ void PresentManager::CopyToSwapchainImpl(Frame* frame) { .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }, - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_ALL_GRAPHICS_BIT, + .dstAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -525,8 +539,7 @@ void PresentManager::CopyToSwapchainImpl(Frame* frame) { }, }; - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, VK_PIPELINE_STAGE_TRANSFER_BIT, {}, - {}, {}, pre_barriers); + cmdbuf.PipelineBarrier(0, {}, {}, pre_barriers); if (blit_supported) { cmdbuf.BlitImage(*frame->image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, image, @@ -539,30 +552,55 @@ void PresentManager::CopyToSwapchainImpl(Frame* frame) { MakeImageCopy(frame->width, frame->height, extent.width, extent.height)); } - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, {}, - {}, {}, post_barriers); + cmdbuf.PipelineBarrier(0, {}, {}, post_barriers); cmdbuf.End(); const VkSemaphore present_semaphore = swapchain.CurrentPresentSemaphore(); const VkSemaphore render_semaphore = swapchain.CurrentRenderSemaphore(); - const std::array wait_semaphores = {present_semaphore, *frame->render_ready}; - - static constexpr std::array wait_stage_masks{ - VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_TRANSFER_BIT, + const std::array wait_infos{{ + { + .sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO, + .pNext = nullptr, + .semaphore = present_semaphore, + .value = 0, + .stageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .deviceIndex = 0, + }, + { + .sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO, + .pNext = nullptr, + .semaphore = *frame->render_ready, + .value = 0, + .stageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .deviceIndex = 0, + }, + }}; + const VkCommandBufferSubmitInfo cmdbuf_info{ + .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO, + .pNext = nullptr, + .commandBuffer = *cmdbuf, + .deviceMask = 0, + }; + const VkSemaphoreSubmitInfo signal_info{ + .sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO, + .pNext = nullptr, + .semaphore = render_semaphore, + .value = 0, + .stageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, + .deviceIndex = 0, }; - const VkSubmitInfo submit_info{ - .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO, + const VkSubmitInfo2 submit_info{ + .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO_2, .pNext = nullptr, - .waitSemaphoreCount = 2U, - .pWaitSemaphores = wait_semaphores.data(), - .pWaitDstStageMask = wait_stage_masks.data(), - .commandBufferCount = 1, - .pCommandBuffers = cmdbuf.address(), - .signalSemaphoreCount = 1U, - .pSignalSemaphores = &render_semaphore, + .flags = 0, + .waitSemaphoreInfoCount = 2U, + .pWaitSemaphoreInfos = wait_infos.data(), + .commandBufferInfoCount = 1U, + .pCommandBufferInfos = &cmdbuf_info, + .signalSemaphoreInfoCount = 1U, + .pSignalSemaphoreInfos = &signal_info, }; // Submit the image copy/blit to the swapchain diff --git a/src/video_core/renderer_vulkan/vk_query_cache.cpp b/src/video_core/renderer_vulkan/vk_query_cache.cpp index 2b5aa24053..7bdb5738c8 100644 --- a/src/video_core/renderer_vulkan/vk_query_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_query_cache.cpp @@ -260,11 +260,13 @@ public: scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([start, amount, base_offset, query_pool, buffer = *resolve_buffer](vk::CommandBuffer cmdbuf) { - const VkBufferMemoryBarrier copy_query_pool_barrier{ - .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, + const VkBufferMemoryBarrier2 copy_query_pool_barrier{ + .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .buffer = buffer, @@ -276,8 +278,7 @@ public: query_pool, static_cast(start), static_cast(amount), buffer, static_cast(base_offset), SamplesQueryBank::QUERY_SIZE, VK_QUERY_RESULT_WAIT_BIT | VK_QUERY_RESULT_64_BIT); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_TRANSFER_BIT, 0, copy_query_pool_barrier); + cmdbuf.PipelineBarrier(copy_query_pool_barrier); }); offsets[bank_id] = {start, base_offset}; base_offset += amount * SamplesQueryBank::QUERY_SIZE; @@ -862,16 +863,17 @@ public: offset_base += TFBQueryBank::QUERY_SIZE; bank.CloseReference(); } - static constexpr VkMemoryBarrier WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_HOST, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, }; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, - vk::PIPELINE_STAGE_HOST, 0, WRITE_BARRIER); + cmdbuf.PipelineBarrier(WRITE_BARRIER); }); std::scoped_lock lk(flush_guard); @@ -928,16 +930,16 @@ private: }); return; } - static constexpr VkMemoryBarrier COUNTER_RESUME_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 COUNTER_RESUME_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_WRITE_BIT_EXT, - .dstAccessMask = VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_READ_BIT_EXT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT, + .srcAccessMask = VK_ACCESS_2_TRANSFORM_FEEDBACK_COUNTER_WRITE_BIT_EXT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT, + .dstAccessMask = VK_ACCESS_2_TRANSFORM_FEEDBACK_COUNTER_READ_BIT_EXT, }; scheduler.Record([this, total = static_cast(buffers_count)](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFORM_FEEDBACK_BIT_EXT, - VK_PIPELINE_STAGE_TRANSFORM_FEEDBACK_BIT_EXT, 0, - COUNTER_RESUME_BARRIER); + cmdbuf.PipelineBarrier(COUNTER_RESUME_BARRIER); cmdbuf.BeginTransformFeedbackEXT(0, total, counter_buffers.data(), offsets.data()); }); } @@ -1024,33 +1026,35 @@ private: const size_t slot = other; // workaround to compile bug. current_bank->AddReference(); - static constexpr VkMemoryBarrier READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_WRITE_BIT_EXT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT, + .srcAccessMask = VK_ACCESS_2_TRANSFORM_FEEDBACK_COUNTER_WRITE_BIT_EXT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, }; - static constexpr VkMemoryBarrier WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT, }; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([dst_buffer = current_bank->GetBuffer(), src_buffer = counter_buffers[slot_index], src_offset = offsets[slot_index], slot](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFORM_FEEDBACK_BIT_EXT, - VK_PIPELINE_STAGE_TRANSFER_BIT, 0, READ_BARRIER); + cmdbuf.PipelineBarrier(READ_BARRIER); std::array copy{VkBufferCopy{ .srcOffset = src_offset, .dstOffset = slot * TFBQueryBank::QUERY_SIZE, .size = TFBQueryBank::QUERY_SIZE, }}; cmdbuf.CopyBuffer(src_buffer, dst_buffer, copy); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, WRITE_BARRIER); + cmdbuf.PipelineBarrier(WRITE_BARRIER); }); return {current_bank_id, slot}; } @@ -1573,28 +1577,30 @@ VideoCommon::StreamerInterface* QueryCacheRuntime::GetStreamerInterface(QueryTyp } void QueryCacheRuntime::Barriers(bool is_prebarrier) { - static constexpr VkMemoryBarrier READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, }; - static constexpr VkMemoryBarrier WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER_HOST, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, }; impl->scheduler.RequestOutsideRenderPassOperationContext(); if (is_prebarrier) { impl->scheduler.Record([](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - VK_PIPELINE_STAGE_TRANSFER_BIT, 0, READ_BARRIER); + cmdbuf.PipelineBarrier(READ_BARRIER); }); } else { impl->scheduler.Record([](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, - vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER_HOST, 0, WRITE_BARRIER); + cmdbuf.PipelineBarrier(WRITE_BARRIER); }); } } diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index aa36aa9093..52227ab290 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -649,14 +649,15 @@ void RasterizerVulkan::DispatchCompute() { return; } scheduler.RequestOutsideRenderPassOperationContext(); - static constexpr VkMemoryBarrier READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT, }; - scheduler.Record([](vk::CommandBuffer cmdbuf) { cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - 0, READ_BARRIER); }); + scheduler.Record([](vk::CommandBuffer cmdbuf) { cmdbuf.PipelineBarrier(READ_BARRIER); }); scheduler.Record([pipeline, dim](vk::CommandBuffer cmdbuf) { if (!pipeline->IsBound()) { return; @@ -882,22 +883,42 @@ void RasterizerVulkan::FlushAndInvalidateRegion(DAddr addr, u64 size, void RasterizerVulkan::WaitForIdle() { // Everything but wait pixel operations. This intentionally includes FRAGMENT_SHADER_BIT because // fragment shaders can still write storage buffers. - VkPipelineStageFlags flags = - VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT | VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | - VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_TESSELLATION_CONTROL_SHADER_BIT | - VK_PIPELINE_STAGE_TESSELLATION_EVALUATION_SHADER_BIT | - VK_PIPELINE_STAGE_GEOMETRY_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT; + VkPipelineStageFlags2 flags = + VK_PIPELINE_STAGE_2_DRAW_INDIRECT_BIT | VK_PIPELINE_STAGE_2_VERTEX_INPUT_BIT | + VK_PIPELINE_STAGE_2_VERTEX_SHADER_BIT | + VK_PIPELINE_STAGE_2_TESSELLATION_CONTROL_SHADER_BIT | + VK_PIPELINE_STAGE_2_TESSELLATION_EVALUATION_SHADER_BIT | + VK_PIPELINE_STAGE_2_GEOMETRY_SHADER_BIT | VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | + VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_2_TRANSFER_BIT; if (device.IsExtTransformFeedbackSupported()) { - flags |= VK_PIPELINE_STAGE_TRANSFORM_FEEDBACK_BIT_EXT; + flags |= VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT; } query_cache.NotifyWFI(); scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([event = *wfi_event, flags](vk::CommandBuffer cmdbuf) { - cmdbuf.SetEvent(event, flags); - cmdbuf.WaitEvents(event, flags, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, {}, {}, {}); + const VkMemoryBarrier2 barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = flags, + .srcAccessMask = VK_ACCESS_2_NONE, + .dstStageMask = VK_PIPELINE_STAGE_2_TOP_OF_PIPE_BIT, + .dstAccessMask = VK_ACCESS_2_NONE, + }; + const VkDependencyInfo dependency_info{ + .sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO, + .pNext = nullptr, + .dependencyFlags = 0, + .memoryBarrierCount = 1, + .pMemoryBarriers = &barrier, + .bufferMemoryBarrierCount = 0, + .pBufferMemoryBarriers = nullptr, + .imageMemoryBarrierCount = 0, + .pImageMemoryBarriers = nullptr, + }; + cmdbuf.SetEvent(event, dependency_info); + cmdbuf.WaitEvents(event, dependency_info); }); fence_manager.SignalOrdering(); } diff --git a/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp b/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp index 4d8fa15d16..d9aa7698f7 100644 --- a/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp @@ -43,10 +43,10 @@ using VideoCore::Surface::SurfaceType; } } - VkAttachmentDescription AttachmentDescription(const Device& device, PixelFormat format, - VkSampleCountFlagBits samples, - VkAttachmentLoadOp load_op, - VkAttachmentStoreOp store_op) { + VkAttachmentDescription2 AttachmentDescription(const Device& device, PixelFormat format, + VkSampleCountFlagBits samples, + VkAttachmentLoadOp load_op, + VkAttachmentStoreOp store_op) { using MaxwellToVK::SurfaceFormat; const SurfaceType surface_type = GetSurfaceType(format); @@ -54,6 +54,8 @@ using VideoCore::Surface::SurfaceType; surface_type == SurfaceType::Stencil; return { + .sType = VK_STRUCTURE_TYPE_ATTACHMENT_DESCRIPTION_2, + .pNext = nullptr, .flags = {}, .format = SurfaceFormat(device, FormatType::Optimal, true, format).format, .samples = samples, @@ -141,16 +143,19 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { } static constexpr size_t MAX_ATTACHMENTS = 2 * std::tuple_size_v + 2; - boost::container::static_vector descriptions; - std::array references{}; + boost::container::static_vector descriptions; + std::array references{}; u32 num_attachments{}; u32 num_colors{}; for (size_t index = 0; index < key.color_formats.size(); ++index) { const PixelFormat format{key.color_formats[index]}; const bool is_valid{format != PixelFormat::Invalid}; - references[index] = VkAttachmentReference{ + references[index] = VkAttachmentReference2{ + .sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2, + .pNext = nullptr, .attachment = is_valid ? num_colors : VK_ATTACHMENT_UNUSED, .layout = VK_IMAGE_LAYOUT_GENERAL, + .aspectMask = 0, }; if (is_valid) { const VkAttachmentLoadOp load_op = (key.color_clear_mask & (1u << index)) != 0 @@ -166,11 +171,14 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { } } const bool has_depth{key.depth_format != PixelFormat::Invalid}; - VkAttachmentReference depth_reference{}; - if (key.depth_format != PixelFormat::Invalid) { - depth_reference = VkAttachmentReference{ + VkAttachmentReference2 depth_reference{}; + if (has_depth) { + depth_reference = VkAttachmentReference2{ + .sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2, + .pNext = nullptr, .attachment = num_colors, .layout = VK_IMAGE_LAYOUT_GENERAL, + .aspectMask = 0, }; const VkAttachmentLoadOp depth_load_op = key.depth_stencil_clear ? VK_ATTACHMENT_LOAD_OP_CLEAR @@ -181,19 +189,23 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { descriptions.push_back(AttachmentDescription(*device, key.depth_format, key.samples, depth_load_op, depth_store_op)); } - std::array resolve_references{}; + std::array resolve_references{}; const bool do_resolve_color = key.resolve_color && key.samples != VK_SAMPLE_COUNT_1_BIT && num_colors > 0; if (do_resolve_color) { for (size_t index = 0; index < key.color_formats.size(); ++index) { const PixelFormat format{key.color_formats[index]}; const bool is_valid{format != PixelFormat::Invalid}; - resolve_references[index] = VkAttachmentReference{ - .attachment = is_valid ? static_cast(descriptions.size()) : VK_ATTACHMENT_UNUSED, + resolve_references[index] = VkAttachmentReference2{ + .sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2, + .pNext = nullptr, + .attachment = is_valid ? static_cast(descriptions.size()) + : VK_ATTACHMENT_UNUSED, .layout = VK_IMAGE_LAYOUT_GENERAL, + .aspectMask = 0, }; if (is_valid) { - VkAttachmentDescription resolve_desc = + VkAttachmentDescription2 resolve_desc = AttachmentDescription(*device, format, VK_SAMPLE_COUNT_1_BIT, VK_ATTACHMENT_LOAD_OP_DONT_CARE, VK_ATTACHMENT_STORE_OP_STORE); @@ -205,21 +217,35 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { const bool do_resolve_depth_stencil = key.resolve_depth_stencil && has_depth && key.samples != VK_SAMPLE_COUNT_1_BIT && SupportsDepthStencilResolve(*device, key.depth_format); - VkAttachmentReference depth_resolve_reference{}; + VkAttachmentReference2 depth_resolve_reference{}; if (do_resolve_depth_stencil) { - depth_resolve_reference = VkAttachmentReference{ + depth_resolve_reference = VkAttachmentReference2{ + .sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2, + .pNext = nullptr, .attachment = static_cast(descriptions.size()), .layout = VK_IMAGE_LAYOUT_GENERAL, + .aspectMask = 0, }; - VkAttachmentDescription resolve_desc = + VkAttachmentDescription2 resolve_desc = AttachmentDescription(*device, key.depth_format, VK_SAMPLE_COUNT_1_BIT, VK_ATTACHMENT_LOAD_OP_DONT_CARE, VK_ATTACHMENT_STORE_OP_STORE); resolve_desc.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; descriptions.push_back(resolve_desc); } - const VkSubpassDescription subpass{ + const ResolveModes resolve_modes = PickResolveModes(*device, key.depth_format); + const VkSubpassDescriptionDepthStencilResolve depth_stencil_resolve{ + .sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_DEPTH_STENCIL_RESOLVE, + .pNext = nullptr, + .depthResolveMode = resolve_modes.depth, + .stencilResolveMode = resolve_modes.stencil, + .pDepthStencilResolveAttachment = &depth_resolve_reference, + }; + const VkSubpassDescription2 subpass{ + .sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_2, + .pNext = do_resolve_depth_stencil ? &depth_stencil_resolve : nullptr, .flags = 0, .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS, + .viewMask = 0, .inputAttachmentCount = 0, .pInputAttachments = nullptr, .colorAttachmentCount = num_attachments, @@ -229,106 +255,31 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { .preserveAttachmentCount = 0, .pPreserveAttachments = nullptr, }; - const VkSubpassDependency dependency{ - .srcSubpass = 0, // Current subpass - .dstSubpass = 0, // Same subpass (self-dependency) - .srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT, - .dstStageMask = VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, - .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT, - .dependencyFlags = VK_DEPENDENCY_BY_REGION_BIT + const VkMemoryBarrier2 dependency_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | + VK_PIPELINE_STAGE_2_EARLY_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT, + .srcAccessMask = VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT, + .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT, }; - - if (device->IsKhrCreateRenderPass2Supported()) { - boost::container::static_vector descriptions2; - for (const VkAttachmentDescription& description : descriptions) { - descriptions2.push_back(VkAttachmentDescription2{ - .sType = VK_STRUCTURE_TYPE_ATTACHMENT_DESCRIPTION_2, - .pNext = nullptr, - .flags = description.flags, - .format = description.format, - .samples = description.samples, - .loadOp = description.loadOp, - .storeOp = description.storeOp, - .stencilLoadOp = description.stencilLoadOp, - .stencilStoreOp = description.stencilStoreOp, - .initialLayout = description.initialLayout, - .finalLayout = description.finalLayout, - }); - } - const auto promote = [](const VkAttachmentReference& reference) { - return VkAttachmentReference2{ - .sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2, - .pNext = nullptr, - .attachment = reference.attachment, - .layout = reference.layout, - .aspectMask = 0, - }; - }; - std::array references2{}; - std::array resolve_references2{}; - for (size_t index = 0; index < references.size(); ++index) { - references2[index] = promote(references[index]); - resolve_references2[index] = promote(resolve_references[index]); - } - const VkAttachmentReference2 depth_reference2 = promote(depth_reference); - const VkAttachmentReference2 depth_resolve_reference2 = promote(depth_resolve_reference); - const ResolveModes resolve_modes = PickResolveModes(*device, key.depth_format); - const VkSubpassDescriptionDepthStencilResolve depth_stencil_resolve{ - .sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_DEPTH_STENCIL_RESOLVE, - .pNext = nullptr, - .depthResolveMode = resolve_modes.depth, - .stencilResolveMode = resolve_modes.stencil, - .pDepthStencilResolveAttachment = &depth_resolve_reference2, - }; - const VkSubpassDescription2 subpass2{ - .sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_2, - .pNext = do_resolve_depth_stencil ? &depth_stencil_resolve : nullptr, - .flags = 0, - .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS, - .viewMask = 0, - .inputAttachmentCount = 0, - .pInputAttachments = nullptr, - .colorAttachmentCount = num_attachments, - .pColorAttachments = references2.data(), - .pResolveAttachments = do_resolve_color ? resolve_references2.data() : nullptr, - .pDepthStencilAttachment = has_depth ? &depth_reference2 : nullptr, - .preserveAttachmentCount = 0, - .pPreserveAttachments = nullptr, - }; - const VkSubpassDependency2 dependency2{ - .sType = VK_STRUCTURE_TYPE_SUBPASS_DEPENDENCY_2, - .pNext = nullptr, - .srcSubpass = dependency.srcSubpass, - .dstSubpass = dependency.dstSubpass, - .srcStageMask = dependency.srcStageMask, - .dstStageMask = dependency.dstStageMask, - .srcAccessMask = dependency.srcAccessMask, - .dstAccessMask = dependency.dstAccessMask, - .dependencyFlags = dependency.dependencyFlags, - .viewOffset = 0, - }; - pair->second = device->GetLogical().CreateRenderPass2({ - .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2, - .pNext = nullptr, - .flags = 0, - .attachmentCount = static_cast(descriptions2.size()), - .pAttachments = descriptions2.empty() ? nullptr : descriptions2.data(), - .subpassCount = 1, - .pSubpasses = &subpass2, - .dependencyCount = 1, - .pDependencies = &dependency2, - .correlatedViewMaskCount = 0, - .pCorrelatedViewMasks = nullptr, - }); - return *pair->second; - } - - pair->second = device->GetLogical().CreateRenderPass({ - .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO, + const VkSubpassDependency2 dependency{ + .sType = VK_STRUCTURE_TYPE_SUBPASS_DEPENDENCY_2, + .pNext = &dependency_barrier, + .srcSubpass = 0, + .dstSubpass = 0, + .srcStageMask = 0, + .dstStageMask = 0, + .srcAccessMask = 0, + .dstAccessMask = 0, + .dependencyFlags = VK_DEPENDENCY_BY_REGION_BIT, + .viewOffset = 0, + }; + pair->second = device->GetLogical().CreateRenderPass2({ + .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2, .pNext = nullptr, .flags = 0, .attachmentCount = static_cast(descriptions.size()), @@ -337,6 +288,8 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { .pSubpasses = &subpass, .dependencyCount = 1, .pDependencies = &dependency, + .correlatedViewMaskCount = 0, + .pCorrelatedViewMasks = nullptr, }); return *pair->second; } diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 74379e8063..5c07c99828 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -343,13 +343,15 @@ u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_se const u64 signal_value = master_semaphore->NextTick(); RecordWithUploadBuffer([signal_semaphore, wait_semaphore, signal_value, this](vk::CommandBuffer cmdbuf, vk::CommandBuffer upload_cmdbuf) { - static constexpr VkMemoryBarrier WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, }; - upload_cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, WRITE_BARRIER); + upload_cmdbuf.PipelineBarrier(WRITE_BARRIER); upload_cmdbuf.End(); cmdbuf.End(); @@ -365,7 +367,7 @@ u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_se if (GPU::Logging::IsActive() && Settings::values.gpu_log_vulkan_calls.GetValue()) { GPU::Logging::GPULogger::GetInstance().LogVulkanCall( - "vkQueueSubmit", "", VK_SUCCESS); + "vkQueueSubmit2", "", VK_SUCCESS); } break; case VK_ERROR_DEVICE_LOST: @@ -416,7 +418,7 @@ void Scheduler::EndRenderPass() ranges = renderpass_image_ranges, has_transform_feedback = device.IsExtTransformFeedbackSupported()]( vk::CommandBuffer cmdbuf) { - std::array barriers; + std::array barriers; for (size_t i = 0; i < num_images; ++i) { const VkImageSubresourceRange& range = ranges[i]; const bool is_color = (range.aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) != 0; @@ -424,25 +426,29 @@ void Scheduler::EndRenderPass() & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) !=0; - VkAccessFlags src_access = 0; + VkAccessFlags2 src_access = 0; if (is_color) - src_access |= VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; + src_access |= VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT; else if (is_depth_stencil) - src_access |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; + src_access |= VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; else - src_access |= VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT - | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; + src_access |= VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT + | VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; - barriers[i] = VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + barriers[i] = VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_EARLY_FRAGMENT_TESTS_BIT + | VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT + | VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT, .srcAccessMask = src_access, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT - | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT - | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT - | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT - | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT + | VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT + | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT + | VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT + | VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -452,19 +458,19 @@ void Scheduler::EndRenderPass() }; } cmdbuf.EndRenderPass(); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - 0, nullptr, nullptr, vk::Span(barriers.data(), num_images)); + cmdbuf.PipelineBarrier(0, {}, {}, vk::Span(barriers.data(), num_images)); if (has_transform_feedback) { - static constexpr VkMemoryBarrier XFB_OUTPUT_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 XFB_OUTPUT_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFORM_FEEDBACK_WRITE_BIT_EXT, - .dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT, + .srcAccessMask = VK_ACCESS_2_TRANSFORM_FEEDBACK_WRITE_BIT_EXT, + .dstStageMask = VK_PIPELINE_STAGE_2_VERTEX_INPUT_BIT + | VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_VERTEX_ATTRIBUTE_READ_BIT + | VK_ACCESS_2_TRANSFER_READ_BIT, }; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFORM_FEEDBACK_BIT_EXT, - VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, XFB_OUTPUT_BARRIER); + cmdbuf.PipelineBarrier(XFB_OUTPUT_BARRIER); } }); diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 33ab5dd57f..18d38f3761 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -689,12 +689,12 @@ struct RangedBarrierRange { void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage image, VkImageAspectFlags aspect_mask, bool is_initialized, std::span copies) { - static constexpr VkAccessFlags WRITE_ACCESS_FLAGS = - VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; - static constexpr VkAccessFlags READ_ACCESS_FLAGS = VK_ACCESS_SHADER_READ_BIT | - VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT; + static constexpr VkAccessFlags2 WRITE_ACCESS_FLAGS = + VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; + static constexpr VkAccessFlags2 READ_ACCESS_FLAGS = VK_ACCESS_2_SHADER_READ_BIT | + VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT; // Compute exact mip/layer range being written to RangedBarrierRange range; @@ -703,11 +703,15 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im } const VkImageSubresourceRange subresource_range = range.SubresourceRange(aspect_mask); - const VkImageMemoryBarrier read_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 read_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | + VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, .srcAccessMask = WRITE_ACCESS_FLAGS, - .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -716,10 +720,14 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im .subresourceRange = subresource_range, }; - const VkImageMemoryBarrier write_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 write_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | + VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT, .dstAccessMask = WRITE_ACCESS_FLAGS | READ_ACCESS_FLAGS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -729,18 +737,10 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im .subresourceRange = subresource_range, }; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, - read_barrier); + cmdbuf.PipelineBarrier(read_barrier); cmdbuf.CopyBufferToImage(src_buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copies); // TODO: Move this to another API - cmdbuf.PipelineBarrier( - VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | - VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, - 0, nullptr, nullptr, write_barrier); + cmdbuf.PipelineBarrier(0, {}, {}, write_barrier); } [[nodiscard]] VkImageBlit MakeImageBlit(const Region2D& dst_region, const Region2D& src_region, @@ -940,11 +940,13 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const .layerCount = VK_REMAINING_ARRAY_LAYERS, }; const std::array read_barriers{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -952,13 +954,15 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const .image = src_image, .subresourceRange = subresource_range, }, - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED, // Discard contents .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -968,11 +972,13 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const }, }; const std::array write_barriers{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = 0, - .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT | VK_ACCESS_MEMORY_READ_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT | VK_ACCESS_2_MEMORY_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -980,11 +986,13 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const .image = src_image, .subresourceRange = subresource_range, }, - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT | VK_ACCESS_MEMORY_READ_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT | VK_ACCESS_2_MEMORY_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -993,12 +1001,10 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const .subresourceRange = subresource_range, }, }; - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, nullptr, nullptr, read_barriers); + cmdbuf.PipelineBarrier(0, {}, {}, read_barriers); cmdbuf.BlitImage(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, regions, vk_filter); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - 0, nullptr, nullptr, write_barriers); + cmdbuf.PipelineBarrier(0, {}, {}, write_barriers); }); } } // Anonymous namespace @@ -1253,26 +1259,32 @@ void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src, for (const VkBufferImageCopy& copy : vk_out_copies) { dst_range.AddLayers(copy.imageSubresource); } - static constexpr VkMemoryBarrier READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, }; - static constexpr VkMemoryBarrier WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, }; const std::array pre_barriers{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1282,10 +1294,12 @@ void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src, }, }; const std::array middle_in_barrier{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = 0, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, .dstAccessMask = 0, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1296,13 +1310,15 @@ void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src, }, }; const std::array middle_out_barrier{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1312,16 +1328,18 @@ void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src, }, }; const std::array post_barriers{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT | + VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | + VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1330,19 +1348,15 @@ void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src, .subresourceRange = dst_range.SubresourceRange(dst_aspect_mask), }, }; - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, {}, {}, pre_barriers); + cmdbuf.PipelineBarrier(0, {}, {}, pre_barriers); cmdbuf.CopyImageToBuffer(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, copy_buffer, vk_in_copies); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - 0, WRITE_BARRIER, nullptr, middle_in_barrier); + cmdbuf.PipelineBarrier(0, WRITE_BARRIER, {}, middle_in_barrier); - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, READ_BARRIER, {}, middle_out_barrier); + cmdbuf.PipelineBarrier(0, READ_BARRIER, {}, middle_out_barrier); cmdbuf.CopyBufferToImage(copy_buffer, dst_image, VK_IMAGE_LAYOUT_GENERAL, vk_out_copies); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - 0, {}, {}, post_barriers); + cmdbuf.PipelineBarrier(0, {}, {}, post_barriers); }); } @@ -1421,13 +1435,15 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst scheduler.Record([filter, dst_region, src_region, dst_image, src_image, dst_layers, src_layers, aspect_mask, is_resolve](vk::CommandBuffer cmdbuf) { const std::array read_barriers{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1441,13 +1457,15 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }, - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1462,14 +1480,16 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst }, }, }; - VkImageMemoryBarrier write_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2 write_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1483,8 +1503,7 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }; - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, nullptr, nullptr, read_barriers); + cmdbuf.PipelineBarrier(0, {}, {}, read_barriers); if (is_resolve) { cmdbuf.ResolveImage(src_image, VK_IMAGE_LAYOUT_GENERAL, dst_image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, @@ -1496,8 +1515,7 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst src_image, VK_IMAGE_LAYOUT_GENERAL, dst_image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, MakeImageBlit(dst_region, src_region, dst_layers, src_layers), vk_filter); } - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - 0, write_barrier); + cmdbuf.PipelineBarrier(write_barrier); }); } @@ -1661,13 +1679,18 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, src_range.AddLayers(copy.srcSubresource); } const std::array pre_barriers{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | + VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | + VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1675,13 +1698,18 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, .image = src_image, .subresourceRange = src_range.SubresourceRange(aspect_mask), }, - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | + VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | + VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1691,10 +1719,15 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, }, }; const std::array post_barriers{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = 0, + .dstStageMask = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | + VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | + VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = 0, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1703,16 +1736,21 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, .image = src_image, .subresourceRange = src_range.SubresourceRange(aspect_mask), }, - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | + VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | + VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT | + VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | + VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1721,18 +1759,10 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, .subresourceRange = dst_range.SubresourceRange(aspect_mask), }, }; - cmdbuf.PipelineBarrier( - VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | - VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, nullptr, nullptr, pre_barriers); + cmdbuf.PipelineBarrier(0, {}, {}, pre_barriers); cmdbuf.CopyImage(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VideoCommon::FixSmallVectorADL(vk_copies)); - cmdbuf.PipelineBarrier( - VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | - VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, nullptr, nullptr, post_barriers); + cmdbuf.PipelineBarrier(0, {}, {}, post_barriers); }); } @@ -1752,16 +1782,16 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src, static_cast(copy.src_subresource.base_layer + copy.src_subresource.num_layers) <= shadow->layers) { const VkImageAspectFlags aspect_mask = shadow->aspect_mask; - VkPipelineStageFlags attachment_stage = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT; - VkAccessFlags attachment_write = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; - VkAccessFlags attachment_read_write = - VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; + VkPipelineStageFlags2 attachment_stage = VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT; + VkAccessFlags2 attachment_write = VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT; + VkAccessFlags2 attachment_read_write = + VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT; if ((aspect_mask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) != 0) { - attachment_stage = VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT; - attachment_write = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; - attachment_read_write = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; + attachment_stage = VK_PIPELINE_STAGE_2_EARLY_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT; + attachment_write = VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; + attachment_read_write = VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; } const VkImage shadow_image = *shadow->image; const VkImage dst_image = dst.Handle(); @@ -1787,11 +1817,13 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src, attachment_write, attachment_read_write](vk::CommandBuffer cmdbuf) { const std::array pre_barriers{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = attachment_stage | VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = attachment_write, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1800,12 +1832,14 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src, .subresourceRange{aspect_mask, 0, VK_REMAINING_MIP_LEVELS, 0, VK_REMAINING_ARRAY_LAYERS}, }, - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | attachment_write | - VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = attachment_stage | VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | attachment_write | + VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1816,10 +1850,12 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src, }, }; const std::array post_barriers{ - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = 0, + .dstStageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, .dstAccessMask = 0, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1829,12 +1865,14 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src, .subresourceRange{aspect_mask, 0, VK_REMAINING_MIP_LEVELS, 0, VK_REMAINING_ARRAY_LAYERS}, }, - VkImageMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | attachment_read_write | - VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, + .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | attachment_read_write | + VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1844,14 +1882,10 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src, VK_REMAINING_ARRAY_LAYERS}, }, }; - cmdbuf.PipelineBarrier(attachment_stage | VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_TRANSFER_BIT, 0, nullptr, nullptr, - pre_barriers); + cmdbuf.PipelineBarrier(0, {}, {}, pre_barriers); cmdbuf.CopyImage(shadow_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, region); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, nullptr, nullptr, - post_barriers); + cmdbuf.PipelineBarrier(0, {}, {}, post_barriers); }); return; } @@ -2148,23 +2182,25 @@ void Image::DownloadMemory(std::span buffers_span, std::span o const VkImage temp_vk_image = runtime->AcquireMsaaScratchImage(image_ci); const VkImageAspectFlags temp_aspect_mask = aspect_mask; - const VkAccessFlags attachment_access = - msaa_download_is_depth ? (VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT) - : (VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT); - const VkPipelineStageFlags attachment_stage = - msaa_download_is_depth ? (VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT) - : VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT; + const VkAccessFlags2 attachment_access = + msaa_download_is_depth ? (VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT) + : (VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | + VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT); + const VkPipelineStageFlags2 attachment_stage = + msaa_download_is_depth ? (VK_PIPELINE_STAGE_2_EARLY_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT) + : VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT; scheduler->RequestOutsideRenderPassOperationContext(); scheduler->Record([temp_vk_image, temp_aspect_mask, attachment_access, attachment_stage](vk::CommandBuffer cmdbuf) { - const VkImageMemoryBarrier init_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 init_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_TOP_OF_PIPE_BIT, .srcAccessMask = 0, + .dstStageMask = attachment_stage, .dstAccessMask = attachment_access, .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED, .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -2179,8 +2215,7 @@ void Image::DownloadMemory(std::span buffers_span, std::span o .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, attachment_stage, 0, - init_barrier); + cmdbuf.PipelineBarrier(init_barrier); }); boost::container::small_vector image_copies; @@ -2220,11 +2255,13 @@ void Image::DownloadMemory(std::span buffers_span, std::span o scheduler->RequestOutsideRenderPassOperationContext(); scheduler->Record([buffers = std::move(buffers_vector), image = temp_vk_image, aspect_mask_ = aspect_mask, vk_copies](vk::CommandBuffer cmdbuf) { - const VkImageMemoryBarrier read_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 read_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -2238,25 +2275,28 @@ void Image::DownloadMemory(std::span buffers_span, std::span o .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }; - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, read_barrier); + cmdbuf.PipelineBarrier(read_barrier); for (size_t index = 0; index < buffers.size(); index++) { cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, buffers[index], vk_copies[index]); } - const VkMemoryBarrier memory_write_barrier{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + const VkMemoryBarrier2 memory_write_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, }; - const VkImageMemoryBarrier image_write_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 image_write_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = 0, - .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -2270,8 +2310,7 @@ void Image::DownloadMemory(std::span buffers_span, std::span o .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - 0, memory_write_barrier, nullptr, image_write_barrier); + cmdbuf.PipelineBarrier(0, memory_write_barrier, {}, image_write_barrier); }); runtime->ReleaseMsaaScratchImage(temp_vk_image); } @@ -2291,11 +2330,13 @@ void Image::DownloadMemory(std::span buffers_span, std::span o scheduler->RequestOutsideRenderPassOperationContext(); scheduler->Record([buffers = std::move(buffers_vector), image = *original_image, aspect_mask_ = aspect_mask, vk_copies](vk::CommandBuffer cmdbuf) { - const VkImageMemoryBarrier read_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 read_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -2309,25 +2350,28 @@ void Image::DownloadMemory(std::span buffers_span, std::span o .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }; - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, read_barrier); + cmdbuf.PipelineBarrier(read_barrier); for (size_t index = 0; index < buffers.size(); index++) { cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, buffers[index], vk_copies[index]); } - const VkMemoryBarrier memory_write_barrier{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + const VkMemoryBarrier2 memory_write_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, }; - const VkImageMemoryBarrier image_write_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + const VkImageMemoryBarrier2 image_write_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = 0, - .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -2341,8 +2385,7 @@ void Image::DownloadMemory(std::span buffers_span, std::span o .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - 0, memory_write_barrier, nullptr, image_write_barrier); + cmdbuf.PipelineBarrier(0, memory_write_barrier, {}, image_write_barrier); }); } @@ -3198,11 +3241,13 @@ void TextureCacheRuntime::AccelerateImageUpload( void TextureCacheRuntime::TransitionImageLayout(Image& image) { if (!image.ExchangeInitialization()) { - VkImageMemoryBarrier barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + VkImageMemoryBarrier2 barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_NONE, - .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .srcAccessMask = VK_ACCESS_2_NONE, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -3218,8 +3263,7 @@ void TextureCacheRuntime::TransitionImageLayout(Image& image) { }; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([barrier](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, 0, barrier); + cmdbuf.PipelineBarrier(barrier); }); } } diff --git a/src/video_core/renderer_vulkan/vk_turbo_mode.cpp b/src/video_core/renderer_vulkan/vk_turbo_mode.cpp index 095908b6cf..8b18ae7d51 100644 --- a/src/video_core/renderer_vulkan/vk_turbo_mode.cpp +++ b/src/video_core/renderer_vulkan/vk_turbo_mode.cpp @@ -208,16 +208,22 @@ void TurboMode::Run(std::stop_token stop_token) { // Finish. cmdbuf.End(); - const VkSubmitInfo submit_info{ - .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO, + const VkCommandBufferSubmitInfo cmdbuf_info{ + .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO, .pNext = nullptr, - .waitSemaphoreCount = 0, - .pWaitSemaphores = nullptr, - .pWaitDstStageMask = nullptr, - .commandBufferCount = 1, - .pCommandBuffers = cmdbuf.address(), - .signalSemaphoreCount = 0, - .pSignalSemaphores = nullptr, + .commandBuffer = *cmdbuf, + .deviceMask = 0, + }; + const VkSubmitInfo2 submit_info{ + .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO_2, + .pNext = nullptr, + .flags = 0, + .waitSemaphoreInfoCount = 0, + .pWaitSemaphoreInfos = nullptr, + .commandBufferInfoCount = 1, + .pCommandBufferInfos = &cmdbuf_info, + .signalSemaphoreInfoCount = 0, + .pSignalSemaphoreInfos = nullptr, }; m_device.GetGraphicsQueue().Submit(std::array{submit_info}, *fence); diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 8fd72cb71f..4a90b1a21d 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -1427,10 +1427,10 @@ void Device::RemoveUnsuitableExtensions() { VK_KHR_MAINTENANCE_5_EXTENSION_NAME); - // VK_KHR_synchronization2 extensions.synchronization2 = features.synchronization2.synchronization2; - RemoveExtensionFeatureIfUnsuitable(extensions.synchronization2, features.synchronization2, - VK_KHR_SYNCHRONIZATION_2_EXTENSION_NAME); + if (!extensions.synchronization2 || !IsKhrCreateRenderPass2Supported()) { + throw vk::Exception(VK_ERROR_FEATURE_NOT_PRESENT); + } } void Device::SetupFamilies(VkSurfaceKHR surface) { diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 0927ca8a05..23dceceb26 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -640,8 +640,7 @@ FN_MAX_LIMIT_LIST /// Returns true if the device supports VK_KHR_depth_stencil_resolve. bool IsKhrDepthStencilResolveSupported() const { - return (extensions.depth_stencil_resolve || instance_version >= VK_API_VERSION_1_2) && - IsKhrCreateRenderPass2Supported(); + return extensions.depth_stencil_resolve || instance_version >= VK_API_VERSION_1_2; } /// Returns the supported resolve modes for the depth aspect. @@ -905,11 +904,6 @@ FN_MAX_LIMIT_LIST bool HasTimelineSemaphore() const; - /// Returns true if the device supports VK_KHR_synchronization2. - bool HasSynchronization2() const { - return extensions.synchronization2; - } - /// Returns the minimum supported version of SPIR-V. u32 SupportedSpirvVersion() const { if (instance_version >= VK_API_VERSION_1_3) { diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp index e4c8f518b9..01a19e8cef 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.cpp +++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp @@ -123,7 +123,6 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCmdEndTransformFeedbackEXT); X(vkCmdEndDebugUtilsLabelEXT); X(vkCmdFillBuffer); - X(vkCmdPipelineBarrier); X(vkCmdPipelineBarrier2); X(vkCmdPushConstants); X(vkCmdPushDescriptorSetWithTemplateKHR); @@ -131,13 +130,13 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCmdSetDepthBias); X(vkCmdSetDepthBias2EXT); X(vkCmdSetDepthBounds); - X(vkCmdSetEvent); + X(vkCmdSetEvent2); X(vkCmdSetScissor); X(vkCmdSetStencilCompareMask); X(vkCmdSetStencilReference); X(vkCmdSetStencilWriteMask); X(vkCmdSetViewport); - X(vkCmdWaitEvents); + X(vkCmdWaitEvents2); X(vkCmdBindVertexBuffers2EXT); X(vkCmdSetCullModeEXT); X(vkCmdSetDepthBoundsTestEnableEXT); @@ -232,7 +231,6 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkGetSemaphoreCounterValue); X(vkMapMemory); X(vkQueueBindSparse); - X(vkQueueSubmit); X(vkQueueSubmit2); X(vkResetFences); X(vkResetQueryPool); @@ -280,6 +278,12 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { if (!dld.vkQueueSubmit2) { Proc(dld.vkQueueSubmit2, dld, "vkQueueSubmit2KHR", device); } + if (!dld.vkCmdSetEvent2) { + Proc(dld.vkCmdSetEvent2, dld, "vkCmdSetEvent2KHR", device); + } + if (!dld.vkCmdWaitEvents2) { + Proc(dld.vkCmdWaitEvents2, dld, "vkCmdWaitEvents2KHR", device); + } if (!dld.vkCreateRenderPass2) { Proc(dld.vkCreateRenderPass2, dld, "vkCreateRenderPass2KHR", device); diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index 5c2d78bc2c..bfa56f084a 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -147,16 +147,16 @@ inline VkResult Filter(VkResult result) { return result; } -inline constexpr VkPipelineStageFlags PIPELINE_STAGE_GRAPHICS_COMPUTE = - VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; +inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_GRAPHICS_COMPUTE = + VK_PIPELINE_STAGE_2_ALL_GRAPHICS_BIT | VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT; -inline constexpr VkPipelineStageFlags PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER = - PIPELINE_STAGE_GRAPHICS_COMPUTE | VK_PIPELINE_STAGE_TRANSFER_BIT; +inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER = + PIPELINE_STAGE_GRAPHICS_COMPUTE | VK_PIPELINE_STAGE_2_TRANSFER_BIT; -inline constexpr VkPipelineStageFlags PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER_HOST = - PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER | VK_PIPELINE_STAGE_HOST_BIT; +inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER_HOST = + PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER | VK_PIPELINE_STAGE_2_HOST_BIT; -inline constexpr VkPipelineStageFlags PIPELINE_STAGE_HOST = VK_PIPELINE_STAGE_HOST_BIT; +inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_HOST = VK_PIPELINE_STAGE_2_HOST_BIT; /// Table holding Vulkan instance function pointers. @@ -239,7 +239,6 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdEndRenderPass vkCmdEndRenderPass{}; PFN_vkCmdEndTransformFeedbackEXT vkCmdEndTransformFeedbackEXT{}; PFN_vkCmdFillBuffer vkCmdFillBuffer{}; - PFN_vkCmdPipelineBarrier vkCmdPipelineBarrier{}; PFN_vkCmdPipelineBarrier2 vkCmdPipelineBarrier2{}; PFN_vkCmdPushConstants vkCmdPushConstants{}; PFN_vkCmdPushDescriptorSetWithTemplateKHR vkCmdPushDescriptorSetWithTemplateKHR{}; @@ -265,7 +264,7 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdSetDepthBiasEnableEXT vkCmdSetDepthBiasEnableEXT{}; PFN_vkCmdSetLogicOpEnableEXT vkCmdSetLogicOpEnableEXT{}; PFN_vkCmdSetDepthClampEnableEXT vkCmdSetDepthClampEnableEXT{}; - PFN_vkCmdSetEvent vkCmdSetEvent{}; + PFN_vkCmdSetEvent2 vkCmdSetEvent2{}; PFN_vkCmdSetFrontFaceEXT vkCmdSetFrontFaceEXT{}; PFN_vkCmdSetPatchControlPointsEXT vkCmdSetPatchControlPointsEXT{}; PFN_vkCmdSetLogicOpEXT vkCmdSetLogicOpEXT{}; @@ -283,7 +282,7 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdSetColorWriteEnableEXT vkCmdSetColorWriteEnableEXT{}; PFN_vkCmdSetColorBlendEnableEXT vkCmdSetColorBlendEnableEXT{}; PFN_vkCmdSetColorBlendEquationEXT vkCmdSetColorBlendEquationEXT{}; - PFN_vkCmdWaitEvents vkCmdWaitEvents{}; + PFN_vkCmdWaitEvents2 vkCmdWaitEvents2{}; PFN_vkCreateBuffer vkCreateBuffer{}; PFN_vkCreateBufferView vkCreateBufferView{}; PFN_vkCreateCommandPool vkCreateCommandPool{}; @@ -348,7 +347,6 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkGetSemaphoreCounterValue vkGetSemaphoreCounterValue{}; PFN_vkMapMemory vkMapMemory{}; PFN_vkQueueBindSparse vkQueueBindSparse{}; - PFN_vkQueueSubmit vkQueueSubmit{}; PFN_vkQueueSubmit2 vkQueueSubmit2{}; PFN_vkResetFences vkResetFences{}; PFN_vkResetQueryPool vkResetQueryPool{}; @@ -848,13 +846,7 @@ public: constexpr Queue(VkQueue queue_, const DeviceDispatch& dld_) noexcept : queue{queue_}, dld{&dld_} {} - VkResult Submit(Span submit_infos, - VkFence fence = VK_NULL_HANDLE) const noexcept { - return dld->vkQueueSubmit(queue, submit_infos.size(), submit_infos.data(), fence); - } - - /// Submits using VK_KHR_synchronization2 / Vulkan 1.3 vkQueueSubmit2. - VkResult Submit2(Span submit_infos, + VkResult Submit(Span submit_infos, VkFence fence = VK_NULL_HANDLE) const noexcept { return dld->vkQueueSubmit2(queue, submit_infos.size(), submit_infos.data(), fence); } @@ -1371,103 +1363,50 @@ public: dld->vkCmdDispatchIndirect(handle, indirect_buffer, offset); } - void PipelineBarrier(VkPipelineStageFlags src_stage_mask, VkPipelineStageFlags dst_stage_mask, - VkDependencyFlags dependency_flags, Span memory_barriers, - Span buffer_barriers, - Span image_barriers) const noexcept { - static constexpr u32 MaxBarriers = 16; - if (dld->vkCmdPipelineBarrier2 && memory_barriers.size() <= MaxBarriers && - buffer_barriers.size() <= MaxBarriers && image_barriers.size() <= MaxBarriers) { - const auto src_stage_mask2 = static_cast(src_stage_mask); - const auto dst_stage_mask2 = static_cast(dst_stage_mask); - - std::array memory_barriers2; - for (u32 i = 0; i < memory_barriers.size(); ++i) { - memory_barriers2[i] = VkMemoryBarrier2{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = src_stage_mask2, - .srcAccessMask = static_cast(memory_barriers[i].srcAccessMask), - .dstStageMask = dst_stage_mask2, - .dstAccessMask = static_cast(memory_barriers[i].dstAccessMask), - }; - } - std::array buffer_barriers2; - for (u32 i = 0; i < buffer_barriers.size(); ++i) { - const auto& barrier = buffer_barriers[i]; - buffer_barriers2[i] = VkBufferMemoryBarrier2{ - .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = src_stage_mask2, - .srcAccessMask = static_cast(barrier.srcAccessMask), - .dstStageMask = dst_stage_mask2, - .dstAccessMask = static_cast(barrier.dstAccessMask), - .srcQueueFamilyIndex = barrier.srcQueueFamilyIndex, - .dstQueueFamilyIndex = barrier.dstQueueFamilyIndex, - .buffer = barrier.buffer, - .offset = barrier.offset, - .size = barrier.size, - }; - } - std::array image_barriers2; - for (u32 i = 0; i < image_barriers.size(); ++i) { - const auto& barrier = image_barriers[i]; - image_barriers2[i] = VkImageMemoryBarrier2{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = src_stage_mask2, - .srcAccessMask = static_cast(barrier.srcAccessMask), - .dstStageMask = dst_stage_mask2, - .dstAccessMask = static_cast(barrier.dstAccessMask), - .oldLayout = barrier.oldLayout, - .newLayout = barrier.newLayout, - .srcQueueFamilyIndex = barrier.srcQueueFamilyIndex, - .dstQueueFamilyIndex = barrier.dstQueueFamilyIndex, - .image = barrier.image, - .subresourceRange = barrier.subresourceRange, - }; - } - const VkDependencyInfo dependency_info{ - .sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO, - .pNext = nullptr, - .dependencyFlags = dependency_flags, - .memoryBarrierCount = memory_barriers.size(), - .pMemoryBarriers = memory_barriers2.data(), - .bufferMemoryBarrierCount = buffer_barriers.size(), - .pBufferMemoryBarriers = buffer_barriers2.data(), - .imageMemoryBarrierCount = image_barriers.size(), - .pImageMemoryBarriers = image_barriers2.data(), - }; - dld->vkCmdPipelineBarrier2(handle, &dependency_info); - return; - } - dld->vkCmdPipelineBarrier(handle, src_stage_mask, dst_stage_mask, dependency_flags, - memory_barriers.size(), memory_barriers.data(), - buffer_barriers.size(), buffer_barriers.data(), - image_barriers.size(), image_barriers.data()); + void PipelineBarrier(VkDependencyFlags dependency_flags, + Span memory_barriers, + Span buffer_barriers, + Span image_barriers) const noexcept { + const VkDependencyInfo dependency_info{ + .sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO, + .pNext = nullptr, + .dependencyFlags = dependency_flags, + .memoryBarrierCount = memory_barriers.size(), + .pMemoryBarriers = memory_barriers.data(), + .bufferMemoryBarrierCount = buffer_barriers.size(), + .pBufferMemoryBarriers = buffer_barriers.data(), + .imageMemoryBarrierCount = image_barriers.size(), + .pImageMemoryBarriers = image_barriers.data(), + }; + dld->vkCmdPipelineBarrier2(handle, &dependency_info); } - void PipelineBarrier(VkPipelineStageFlags src_stage_mask, VkPipelineStageFlags dst_stage_mask, + void PipelineBarrier(VkPipelineStageFlags2 src_stage_mask, VkPipelineStageFlags2 dst_stage_mask, VkDependencyFlags dependency_flags = 0) const noexcept { - PipelineBarrier(src_stage_mask, dst_stage_mask, dependency_flags, {}, {}, {}); + const VkMemoryBarrier2 barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = src_stage_mask, + .srcAccessMask = VK_ACCESS_2_NONE, + .dstStageMask = dst_stage_mask, + .dstAccessMask = VK_ACCESS_2_NONE, + }; + PipelineBarrier(dependency_flags, barrier, {}, {}); } - void PipelineBarrier(VkPipelineStageFlags src_stage_mask, VkPipelineStageFlags dst_stage_mask, - VkDependencyFlags dependency_flags, - const VkMemoryBarrier& memory_barrier) const noexcept { - PipelineBarrier(src_stage_mask, dst_stage_mask, dependency_flags, memory_barrier, {}, {}); + void PipelineBarrier(const VkMemoryBarrier2& memory_barrier, + VkDependencyFlags dependency_flags = 0) const noexcept { + PipelineBarrier(dependency_flags, memory_barrier, {}, {}); } - void PipelineBarrier(VkPipelineStageFlags src_stage_mask, VkPipelineStageFlags dst_stage_mask, - VkDependencyFlags dependency_flags, - const VkBufferMemoryBarrier& buffer_barrier) const noexcept { - PipelineBarrier(src_stage_mask, dst_stage_mask, dependency_flags, {}, buffer_barrier, {}); + void PipelineBarrier(const VkBufferMemoryBarrier2& buffer_barrier, + VkDependencyFlags dependency_flags = 0) const noexcept { + PipelineBarrier(dependency_flags, {}, buffer_barrier, {}); } - void PipelineBarrier(VkPipelineStageFlags src_stage_mask, VkPipelineStageFlags dst_stage_mask, - VkDependencyFlags dependency_flags, - const VkImageMemoryBarrier& image_barrier) const noexcept { - PipelineBarrier(src_stage_mask, dst_stage_mask, dependency_flags, {}, {}, image_barrier); + void PipelineBarrier(const VkImageMemoryBarrier2& image_barrier, + VkDependencyFlags dependency_flags = 0) const noexcept { + PipelineBarrier(dependency_flags, {}, {}, image_barrier); } void BindDescriptorBuffersEXT(Span bindings) const noexcept { @@ -1572,17 +1511,12 @@ public: dld->vkCmdSetDepthBounds(handle, min_depth_bounds, max_depth_bounds); } - void SetEvent(VkEvent event, VkPipelineStageFlags stage_flags) const noexcept { - dld->vkCmdSetEvent(handle, event, stage_flags); + void SetEvent(VkEvent event, const VkDependencyInfo& dependency_info) const noexcept { + dld->vkCmdSetEvent2(handle, event, &dependency_info); } - void WaitEvents(Span events, VkPipelineStageFlags src_stage_mask, - VkPipelineStageFlags dst_stage_mask, Span memory_barriers, - Span buffer_barriers, - Span image_barriers) const noexcept { - dld->vkCmdWaitEvents(handle, events.size(), events.data(), src_stage_mask, dst_stage_mask, - memory_barriers.size(), memory_barriers.data(), buffer_barriers.size(), - buffer_barriers.data(), image_barriers.size(), image_barriers.data()); + void WaitEvents(Span events, const VkDependencyInfo& dependency_info) const noexcept { + dld->vkCmdWaitEvents2(handle, events.size(), events.data(), &dependency_info); } void BindVertexBuffers2EXT(u32 first_binding, u32 binding_count, const VkBuffer* buffers, From f71a80671ca49b182af6134ccf8c9576f97e447b Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Wed, 23 Sep 2026 02:37:57 -0400 Subject: [PATCH 50/69] Remove final references to sync1 --- .../renderer_vulkan/present/post_process.cpp | 20 ++++++-- .../renderer_vulkan/present/util.cpp | 47 ++++++++++++++----- .../vulkan_common/vulkan_wrapper.cpp | 7 --- src/video_core/vulkan_common/vulkan_wrapper.h | 3 -- 4 files changed, 50 insertions(+), 27 deletions(-) diff --git a/src/video_core/renderer_vulkan/present/post_process.cpp b/src/video_core/renderer_vulkan/present/post_process.cpp index 07d75b7025..f5164e488e 100644 --- a/src/video_core/renderer_vulkan/present/post_process.cpp +++ b/src/video_core/renderer_vulkan/present/post_process.cpp @@ -134,7 +134,9 @@ vk::RenderPass CreateFxRenderPass(const Device& device, VkFormat format, bool cl initial_layout = VK_IMAGE_LAYOUT_UNDEFINED; } - const VkAttachmentDescription attachment{ + const VkAttachmentDescription2 attachment{ + .sType = VK_STRUCTURE_TYPE_ATTACHMENT_DESCRIPTION_2, + .pNext = nullptr, .flags = 0, .format = format, .samples = VK_SAMPLE_COUNT_1_BIT, @@ -146,14 +148,20 @@ vk::RenderPass CreateFxRenderPass(const Device& device, VkFormat format, bool cl .finalLayout = VK_IMAGE_LAYOUT_GENERAL, }; - const VkAttachmentReference reference{ + static constexpr VkAttachmentReference2 reference{ + .sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2, + .pNext = nullptr, .attachment = 0, .layout = VK_IMAGE_LAYOUT_GENERAL, + .aspectMask = 0, }; - const VkSubpassDescription subpass{ + const VkSubpassDescription2 subpass{ + .sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_2, + .pNext = nullptr, .flags = 0, .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS, + .viewMask = 0, .inputAttachmentCount = 0, .pInputAttachments = nullptr, .colorAttachmentCount = 1, @@ -164,8 +172,8 @@ vk::RenderPass CreateFxRenderPass(const Device& device, VkFormat format, bool cl .pPreserveAttachments = nullptr, }; - return device.GetLogical().CreateRenderPass(VkRenderPassCreateInfo{ - .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO, + return device.GetLogical().CreateRenderPass2(VkRenderPassCreateInfo2{ + .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2, .pNext = nullptr, .flags = 0, .attachmentCount = 1, @@ -174,6 +182,8 @@ vk::RenderPass CreateFxRenderPass(const Device& device, VkFormat format, bool cl .pSubpasses = &subpass, .dependencyCount = 0, .pDependencies = nullptr, + .correlatedViewMaskCount = 0, + .pCorrelatedViewMasks = nullptr, }); } diff --git a/src/video_core/renderer_vulkan/present/util.cpp b/src/video_core/renderer_vulkan/present/util.cpp index 131973bdd6..80d5e19578 100644 --- a/src/video_core/renderer_vulkan/present/util.cpp +++ b/src/video_core/renderer_vulkan/present/util.cpp @@ -203,7 +203,9 @@ vk::ImageView CreateWrappedImageView(const Device& device, vk::Image& image, VkF vk::RenderPass CreateWrappedRenderPass(const Device& device, VkFormat format, VkImageLayout initial_layout) { - const VkAttachmentDescription attachment{ + const VkAttachmentDescription2 attachment{ + .sType = VK_STRUCTURE_TYPE_ATTACHMENT_DESCRIPTION_2, + .pNext = nullptr, .flags = VK_ATTACHMENT_DESCRIPTION_MAY_ALIAS_BIT, .format = format, .samples = VK_SAMPLE_COUNT_1_BIT, @@ -216,14 +218,20 @@ vk::RenderPass CreateWrappedRenderPass(const Device& device, VkFormat format, .finalLayout = VK_IMAGE_LAYOUT_GENERAL, }; - constexpr VkAttachmentReference color_attachment_ref{ + static constexpr VkAttachmentReference2 color_attachment_ref{ + .sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2, + .pNext = nullptr, .attachment = 0, .layout = VK_IMAGE_LAYOUT_GENERAL, + .aspectMask = 0, }; - const VkSubpassDescription subpass_description{ + const VkSubpassDescription2 subpass_description{ + .sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_2, + .pNext = nullptr, .flags = 0, .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS, + .viewMask = 0, .inputAttachmentCount = 0, .pInputAttachments = nullptr, .colorAttachmentCount = 1, @@ -234,18 +242,31 @@ vk::RenderPass CreateWrappedRenderPass(const Device& device, VkFormat format, .pPreserveAttachments = nullptr, }; - constexpr VkSubpassDependency dependency{ - .srcSubpass = VK_SUBPASS_EXTERNAL, - .dstSubpass = 0, - .srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, - .dstStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, + static constexpr VkMemoryBarrier2 dependency_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT, .srcAccessMask = 0, - .dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, - .dependencyFlags = 0, + .dstStageMask = VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT, + .dstAccessMask = + VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT, }; - return device.GetLogical().CreateRenderPass(VkRenderPassCreateInfo{ - .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO, + static constexpr VkSubpassDependency2 dependency{ + .sType = VK_STRUCTURE_TYPE_SUBPASS_DEPENDENCY_2, + .pNext = &dependency_barrier, + .srcSubpass = VK_SUBPASS_EXTERNAL, + .dstSubpass = 0, + .srcStageMask = 0, + .dstStageMask = 0, + .srcAccessMask = 0, + .dstAccessMask = 0, + .dependencyFlags = 0, + .viewOffset = 0, + }; + + return device.GetLogical().CreateRenderPass2(VkRenderPassCreateInfo2{ + .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2, .pNext = nullptr, .flags = 0, .attachmentCount = 1, @@ -254,6 +275,8 @@ vk::RenderPass CreateWrappedRenderPass(const Device& device, VkFormat format, .pSubpasses = &subpass_description, .dependencyCount = 1, .pDependencies = &dependency, + .correlatedViewMaskCount = 0, + .pCorrelatedViewMasks = nullptr, }); } diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp index 01a19e8cef..a2ca3b7fa6 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.cpp +++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp @@ -183,7 +183,6 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCreatePipelineCache); X(vkCreatePipelineLayout); X(vkCreateQueryPool); - X(vkCreateRenderPass); X(vkCreateRenderPass2); X(vkCreateSampler); X(vkCreateSemaphore); @@ -737,12 +736,6 @@ DescriptorPool Device::CreateDescriptorPool(const VkDescriptorPoolCreateInfo& ci return DescriptorPool(object, handle, *dld); } -RenderPass Device::CreateRenderPass(const VkRenderPassCreateInfo& ci) const { - VkRenderPass object; - Check(dld->vkCreateRenderPass(handle, &ci, nullptr, &object)); - return RenderPass(object, handle, *dld); -} - RenderPass Device::CreateRenderPass2(const VkRenderPassCreateInfo2& ci) const { VkRenderPass object; Check(dld->vkCreateRenderPass2(handle, &ci, nullptr, &object)); diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index bfa56f084a..9c60f03fd2 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -299,7 +299,6 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCreatePipelineCache vkCreatePipelineCache{}; PFN_vkCreatePipelineLayout vkCreatePipelineLayout{}; PFN_vkCreateQueryPool vkCreateQueryPool{}; - PFN_vkCreateRenderPass vkCreateRenderPass{}; PFN_vkCreateRenderPass2 vkCreateRenderPass2{}; PFN_vkCreateSampler vkCreateSampler{}; PFN_vkCreateSemaphore vkCreateSemaphore{}; @@ -1057,8 +1056,6 @@ public: [[nodiscard]] DescriptorPool CreateDescriptorPool(const VkDescriptorPoolCreateInfo& ci) const; - [[nodiscard]] RenderPass CreateRenderPass(const VkRenderPassCreateInfo& ci) const; - [[nodiscard]] RenderPass CreateRenderPass2(const VkRenderPassCreateInfo2& ci) const; [[nodiscard]] DescriptorSetLayout CreateDescriptorSetLayout( From fd8da678dfaaa35c19a6864bb2cb2ae1a95f6803 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Wed, 23 Sep 2026 15:33:27 -0400 Subject: [PATCH 51/69] Re-adjust dispatch draw + masks --- .../renderer_vulkan/vk_buffer_cache.cpp | 72 +++++++++++++------ .../renderer_vulkan/vk_buffer_cache.h | 3 + .../renderer_vulkan/vk_rasterizer.cpp | 17 +++-- .../renderer_vulkan/vk_scheduler.cpp | 3 +- src/video_core/renderer_vulkan/vk_scheduler.h | 2 +- 5 files changed, 64 insertions(+), 33 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 56f0e7b870..dec26f328d 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -24,6 +24,18 @@ namespace Vulkan { namespace { +constexpr u32 COMPACT_VERTEX_BINDINGS = 8; + +template +struct VertexBindings { + std::array buffers; + std::array offsets; + std::array sizes; + std::array strides; + u32 first; + u32 count; +}; + VkBufferCopy MakeBufferCopy(const VideoCommon::BufferCopy& copy) { return VkBufferCopy{ .srcOffset = copy.src_offset, @@ -670,36 +682,52 @@ void BufferCacheRuntime::BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset } } -void BufferCacheRuntime::BindVertexBuffers(VideoCommon::HostBindings& bindings) { - boost::container::static_vector buffer_handles(bindings.buffers.size()); - for (u32 i = 0; i < bindings.buffers.size(); ++i) { - auto handle = bindings.buffers[i]->Handle(); - if (handle == VK_NULL_HANDLE) { - bindings.offsets[i] = 0; - bindings.sizes[i] = VK_WHOLE_SIZE; +template +void BufferCacheRuntime::RecordVertexBuffers(const VideoCommon::HostBindings& bindings, + u32 count) { + VertexBindings vertex{}; + vertex.first = bindings.min_index; + vertex.count = count; + for (u32 i = 0; i < count; ++i) { + vertex.buffers[i] = bindings.buffers[i]->Handle(); + vertex.offsets[i] = bindings.offsets[i]; + vertex.sizes[i] = bindings.sizes[i]; + vertex.strides[i] = bindings.strides[i]; + if (vertex.buffers[i] == VK_NULL_HANDLE) { + vertex.offsets[i] = 0; + vertex.sizes[i] = VK_WHOLE_SIZE; if (!device.HasNullDescriptor()) { ReserveNullBuffer(); - handle = *null_buffer; + vertex.buffers[i] = *null_buffer; } } - buffer_handles[i] = handle; - } - const u32 device_max = device.GetMaxVertexInputBindings(); - const u32 min_binding = (std::min)(bindings.min_index, device_max); - const u32 max_binding = (std::min)(bindings.max_index, device_max); - const u32 binding_count = max_binding - min_binding; - if (binding_count == 0) { - return; } if (device.IsExtExtendedDynamicStateSupported()) { - scheduler.Record([bindings_ = std::move(bindings), buffer_handles_ = std::move(buffer_handles), binding_count](vk::CommandBuffer cmdbuf) { - cmdbuf.BindVertexBuffers2EXT(bindings_.min_index, binding_count, buffer_handles_.data(), bindings_.offsets.data(), bindings_.sizes.data(), bindings_.strides.data()); - }); - } else { - scheduler.Record([bindings_ = std::move(bindings), buffer_handles_ = std::move(buffer_handles), binding_count](vk::CommandBuffer cmdbuf) { - cmdbuf.BindVertexBuffers(bindings_.min_index, binding_count, buffer_handles_.data(), bindings_.offsets.data()); + scheduler.Record([vertex](vk::CommandBuffer cmdbuf) { + cmdbuf.BindVertexBuffers2EXT(vertex.first, vertex.count, vertex.buffers.data(), + vertex.offsets.data(), vertex.sizes.data(), + vertex.strides.data()); }); + return; } + scheduler.Record([vertex](vk::CommandBuffer cmdbuf) { + cmdbuf.BindVertexBuffers(vertex.first, vertex.count, vertex.buffers.data(), + vertex.offsets.data()); + }); +} + +void BufferCacheRuntime::BindVertexBuffers(VideoCommon::HostBindings& bindings) { + const u32 device_max = device.GetMaxVertexInputBindings(); + const u32 count = (std::min)(bindings.max_index, device_max) - + (std::min)(bindings.min_index, device_max); + if (count == 0) { + return; + } + if (count <= COMPACT_VERTEX_BINDINGS) { + RecordVertexBuffers(bindings, count); + return; + } + RecordVertexBuffers(bindings, count); } void BufferCacheRuntime::BindTransformFeedbackBuffer(u32 index, VkBuffer buffer, u32 offset, diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h index fbd8c2e753..582f665b4e 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.h +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h @@ -244,6 +244,9 @@ private: VkFormat TexelBufferFormat(VideoCore::Surface::PixelFormat format) const; + template + void RecordVertexBuffers(const VideoCommon::HostBindings& bindings, u32 count); + void ReserveNullBuffer(); vk::Buffer CreateNullBuffer(); diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 52227ab290..c32cbd2a0d 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -1032,23 +1032,22 @@ void RasterizerVulkan::LoadDiskResources(u64 title_id, std::stop_token stop_load void RasterizerVulkan::FlushWork() { #ifdef __ANDROID__ - static constexpr u32 DRAWS_TO_DISPATCH = 512; - static constexpr u32 CHECK_MASK = 3; + static constexpr u32 DRAWS_TO_DISPATCH = 1024; #else static constexpr u32 DRAWS_TO_DISPATCH = 4096; - static constexpr u32 CHECK_MASK = 7; #endif // __ANDROID__ + static constexpr u32 CHECK_MASK = 7; - static_assert(DRAWS_TO_DISPATCH % (CHECK_MASK + 1) == 0); - if ((++draw_counter & CHECK_MASK) != CHECK_MASK) { + if (++draw_counter >= DRAWS_TO_DISPATCH && + (!scheduler.IsRenderPassActive() || + maxwell3d->dirty.flags[VideoCommon::Dirty::RenderTargets])) { + scheduler.Flush(); + draw_counter = 0; return; } - if (draw_counter < DRAWS_TO_DISPATCH) { + if ((draw_counter & CHECK_MASK) == CHECK_MASK) { scheduler.DispatchWork(); - return; } - scheduler.Flush(); - draw_counter = 0; } AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_, TextureCache& texture_cache_, diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 5c07c99828..33b03ea8f2 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -13,6 +13,7 @@ #include "video_core/renderer_vulkan/vk_query_cache.h" +#include "common/make_unique_for_overwrite.h" #include "common/settings.h" #include "common/thread.h" #include "video_core/gpu_logging/gpu_logging.h" @@ -484,7 +485,7 @@ void Scheduler::AcquireNewChunk() { if (chunk_reserve.empty()) { // If we don't have anything reserved, we need to make a new chunk. - chunk = std::make_unique(); + chunk = Common::make_unique_for_overwrite(); } else { // Otherwise, we can just take from the reserve. chunk = std::move(chunk_reserve.back()); diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index 08ae0bfca1..e82b0414fc 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -251,7 +251,7 @@ private: size_t command_offset = 0; bool submit = false; - alignas(std::max_align_t) std::array data{}; + alignas(std::max_align_t) std::array data; }; struct State { From 3ff00afd5988d7cc894f0b2d03f16fde31009197 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 24 Sep 2026 13:53:54 -0400 Subject: [PATCH 52/69] Some other dead code removal --- .../renderer_vulkan/present/lsfg_common.cpp | 1 - .../renderer_vulkan/present/lsfg_common.h | 5 - .../renderer_vulkan/present/post_process.cpp | 2 - .../renderer_vulkan/present/post_process.h | 3 - .../renderer_vulkan/renderer_vulkan.h | 7 -- .../renderer_vulkan/vk_rasterizer.h | 1 - .../renderer_vulkan/vk_render_pass_cache.cpp | 27 +---- .../renderer_vulkan/vk_state_tracker.cpp | 2 - .../renderer_vulkan/vk_state_tracker.h | 10 -- src/video_core/renderer_vulkan/vk_swapchain.h | 10 -- .../renderer_vulkan/vk_texture_cache.cpp | 6 - .../renderer_vulkan/vk_texture_cache.h | 15 --- .../renderer_vulkan/vk_update_descriptor.h | 4 - .../vulkan_common/vulkan_device.cpp | 17 --- src/video_core/vulkan_common/vulkan_device.h | 47 -------- .../vulkan_common/vulkan_wrapper.cpp | 112 +----------------- src/video_core/vulkan_common/vulkan_wrapper.h | 86 -------------- 17 files changed, 4 insertions(+), 351 deletions(-) diff --git a/src/video_core/renderer_vulkan/present/lsfg_common.cpp b/src/video_core/renderer_vulkan/present/lsfg_common.cpp index 74e5d446d8..e05c7a8b02 100644 --- a/src/video_core/renderer_vulkan/present/lsfg_common.cpp +++ b/src/video_core/renderer_vulkan/present/lsfg_common.cpp @@ -286,7 +286,6 @@ LsfgPass::LsfgPass(const Device& device, const LsfgShaders& shaders, u32 shader_ for (const auto& [count, type] : bindings) { types.insert(types.end(), count, type); } - descriptor_count = static_cast(types.size()); descriptor_set_layout = CreateWrappedDescriptorSetLayout( device, std::span{types}, VK_SHADER_STAGE_COMPUTE_BIT); diff --git a/src/video_core/renderer_vulkan/present/lsfg_common.h b/src/video_core/renderer_vulkan/present/lsfg_common.h index bc538f4d1b..33547b1346 100644 --- a/src/video_core/renderer_vulkan/present/lsfg_common.h +++ b/src/video_core/renderer_vulkan/present/lsfg_common.h @@ -204,10 +204,6 @@ public: return *descriptor_set_layout; } - [[nodiscard]] u32 DescriptorCount() const { - return descriptor_count; - } - void Bind(vk::CommandBuffer cmdbuf, VkDescriptorSet set) const; void BindPipeline(vk::CommandBuffer cmdbuf) const; void BindSet(vk::CommandBuffer cmdbuf, VkDescriptorSet set) const; @@ -216,7 +212,6 @@ private: vk::DescriptorSetLayout descriptor_set_layout; vk::PipelineLayout pipeline_layout; vk::Pipeline pipeline; - u32 descriptor_count{}; }; [[nodiscard]] vk::DescriptorPool CreateLsfgDescriptorPool(const Device& device, u32 max_sets); diff --git a/src/video_core/renderer_vulkan/present/post_process.cpp b/src/video_core/renderer_vulkan/present/post_process.cpp index f5164e488e..2027f75828 100644 --- a/src/video_core/renderer_vulkan/present/post_process.cpp +++ b/src/video_core/renderer_vulkan/present/post_process.cpp @@ -611,8 +611,6 @@ bool PostProcessChain::BuildEffects(const Device& device, MemoryAllocator& alloc *vertex_shader->second, *fragment_shader->second, pass); if (out.writes_backbuffer) { - out.backbuffer_slot = static_cast(effect.backbuffer_pass_count % 2); - ++effect.backbuffer_pass_count; for (u32 image = 0; image < m_image_count; ++image) { for (size_t slot = 0; slot < 2; ++slot) { out.framebuffers.push_back(CreateWrappedFramebuffer( diff --git a/src/video_core/renderer_vulkan/present/post_process.h b/src/video_core/renderer_vulkan/present/post_process.h index 9d42bd4751..9238ef0c39 100644 --- a/src/video_core/renderer_vulkan/present/post_process.h +++ b/src/video_core/renderer_vulkan/present/post_process.h @@ -89,7 +89,6 @@ private: u32 num_vertices{3}; bool clear{}; bool writes_backbuffer{}; - u32 backbuffer_slot{}; }; struct Effect { @@ -105,8 +104,6 @@ private: vk::DescriptorSetLayout uniform_layout{}; vk::DescriptorPool descriptor_pool{}; vk::DescriptorSets uniform_sets{}; - size_t backbuffer_pass_count{}; - u32 backbuffer_slots{1}; }; struct FrameImages { diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.h b/src/video_core/renderer_vulkan/renderer_vulkan.h index 87422dd935..b0ceda5905 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.h +++ b/src/video_core/renderer_vulkan/renderer_vulkan.h @@ -59,14 +59,7 @@ public: return device.GetDriverName(); } - // Enhanced platform-specific initialization - void InitializePlatformSpecific(); - private: - void InterpolateFrames(Frame* prev_frame, Frame* curr_frame); - Frame* previous_frame = nullptr; // Store the previous frame for interpolation - VkCommandBuffer BeginSingleTimeCommands(); - void EndSingleTimeCommands(VkCommandBuffer command_buffer); void Report() const; vk::Buffer RenderToBuffer(std::span framebuffers, diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index bb7c5ff09e..ea98948df7 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -180,7 +180,6 @@ private: void UpdateRasterizerDiscardEnable(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateConservativeRasterizationMode(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateLineStippleEnable(Tegra::Engines::Maxwell3D::Regs& regs); - void UpdateLineStipple(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateLineRasterizationMode(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateDepthBiasEnable(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateLogicOpEnable(Tegra::Engines::Maxwell3D::Regs& regs); diff --git a/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp b/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp index d9aa7698f7..8c0921e9e8 100644 --- a/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp @@ -255,29 +255,6 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { .preserveAttachmentCount = 0, .pPreserveAttachments = nullptr, }; - const VkMemoryBarrier2 dependency_barrier{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_2_EARLY_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT, - .srcAccessMask = VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT, - .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT, - }; - const VkSubpassDependency2 dependency{ - .sType = VK_STRUCTURE_TYPE_SUBPASS_DEPENDENCY_2, - .pNext = &dependency_barrier, - .srcSubpass = 0, - .dstSubpass = 0, - .srcStageMask = 0, - .dstStageMask = 0, - .srcAccessMask = 0, - .dstAccessMask = 0, - .dependencyFlags = VK_DEPENDENCY_BY_REGION_BIT, - .viewOffset = 0, - }; pair->second = device->GetLogical().CreateRenderPass2({ .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2, .pNext = nullptr, @@ -286,8 +263,8 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { .pAttachments = descriptions.empty() ? nullptr : descriptions.data(), .subpassCount = 1, .pSubpasses = &subpass, - .dependencyCount = 1, - .pDependencies = &dependency, + .dependencyCount = 0, + .pDependencies = nullptr, .correlatedViewMaskCount = 0, .pCorrelatedViewMasks = nullptr, }); diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.cpp b/src/video_core/renderer_vulkan/vk_state_tracker.cpp index 0b132c97d6..49b5dded33 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.cpp +++ b/src/video_core/renderer_vulkan/vk_state_tracker.cpp @@ -63,7 +63,6 @@ Maxwell3D::DirtyState::Flags MakeInvalidationFlags() { BlendEnable, ConservativeRasterizationMode, LineStippleEnable, - LineStippleParams, }; Maxwell3D::DirtyState::Flags flags{}; for (const int flag : INVALIDATION_FLAGS) { @@ -228,7 +227,6 @@ void SetupDirtyVertexBindings(Maxwell3D::DirtyState::Tables& tables) { void SetupRasterModes(Maxwell3D::DirtyState::Tables &tables) { auto& table = tables[0]; - table[OFF(line_stipple_params)] = LineStippleParams; table[OFF(conservative_raster_enable)] = ConservativeRasterizationMode; table[OFF(line_anti_alias_enable)] = LineRasterizationMode; } diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.h b/src/video_core/renderer_vulkan/vk_state_tracker.h index 6b47ba4176..eebd25a81e 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.h +++ b/src/video_core/renderer_vulkan/vk_state_tracker.h @@ -56,7 +56,6 @@ enum : u8 { ConservativeRasterizationMode, LineRasterizationMode, LineStippleEnable, - LineStippleParams, DepthBiasEnable, StateEnable, LogicOp, @@ -89,14 +88,6 @@ public: stencil_reset = true; } - void InvalidateViewports() { - (*flags)[Dirty::Viewports] = true; - } - - void InvalidateScissors() { - (*flags)[Dirty::Scissors] = true; - } - void InvalidateStateEnableFlag() { (*flags)[Dirty::StateEnable] = true; } @@ -220,7 +211,6 @@ public: bool TouchLineStippleEnable() { return Exchange(Dirty::LineStippleEnable, false); } - bool TouchLineStipple() { return Exchange(Dirty::LineStippleParams, false); } bool TouchDepthBiasEnable() { return Exchange(Dirty::DepthBiasEnable, false); } diff --git a/src/video_core/renderer_vulkan/vk_swapchain.h b/src/video_core/renderer_vulkan/vk_swapchain.h index d926cc118a..62559355a9 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.h +++ b/src/video_core/renderer_vulkan/vk_swapchain.h @@ -47,11 +47,6 @@ public: return IsSubOptimal() || NeedsPresentModeUpdate(); } - /// Returns true when the swapchain is outdated. - bool IsOutDated() const { - return is_outdated; - } - /// Returns true when the swapchain is suboptimal. bool IsSubOptimal() const { return is_suboptimal; @@ -69,10 +64,6 @@ public: return image_index; } - std::size_t GetFrameIndex() const { - return frame_index; - } - VkImage GetImageIndex(std::size_t index) const { return images[index]; } @@ -112,7 +103,6 @@ public: private: void CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities); void CreateSemaphores(); - void CreateImageViews(); void Destroy(); diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 18d38f3761..32d4c5a52d 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -2613,13 +2613,9 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI if (device->ApiVersion() >= VK_API_VERSION_1_3) { const VkFormatProperties3 properties3 = device->GetPhysical().GetFormatProperties3(format_info.format); - supports_depth_comparison = - (properties3.optimalTilingFeatures & - VK_FORMAT_FEATURE_2_SAMPLED_IMAGE_DEPTH_COMPARISON_BIT) != 0; supports_minmax_filter = (properties3.optimalTilingFeatures & VK_FORMAT_FEATURE_2_SAMPLED_IMAGE_FILTER_MINMAX_BIT) != 0; } else { - supports_depth_comparison = true; supports_minmax_filter = (device->GetPhysical().GetFormatProperties(format_info.format).optimalTilingFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_MINMAX_BIT) != 0; @@ -2888,7 +2884,6 @@ Sampler::Sampler(TextureCacheRuntime& runtime, const Tegra::Texture::TSCEntry& t has_added_anisotropy = max_anisotropy > default_anisotropy; has_linear_filtering = mag_filter == VK_FILTER_LINEAR || min_filter == VK_FILTER_LINEAR || mipmap_mode == VK_SAMPLER_MIPMAP_MODE_LINEAR; - has_depth_comparison = tsc.depth_compare_enabled != 0; has_minmax_reduction = reduction_mode != VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT; has_srgb_border_color = tsc.srgb_conversion != 0 && srgb_border_color != border_color; if (has_minmax_reduction && !device.IsExtSamplerFilterMinmaxSupported()) { @@ -3165,7 +3160,6 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime, resolve_shadow_images[num_resolve_shadows++] = depth_image; } - num_color_buffers = static_cast(num_colors); framebuffer = runtime.device.GetLogical().CreateFramebuffer({ .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO, .pNext = nullptr, diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 2e6b547e72..67cf67872b 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -89,10 +89,6 @@ public: void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view); - bool CanAccelerateImageUpload(Image&) const noexcept { - return false; - } - bool CanUploadMSAA() const noexcept { return true; } @@ -234,10 +230,6 @@ public: return samples; } - [[nodiscard]] u32 NumColorBuffers() const noexcept { - return num_color_buffers; - } - [[nodiscard]] u32 NumImages() const noexcept { return num_images; } @@ -285,7 +277,6 @@ private: VkRenderPass renderpass{}; VkExtent2D render_area{}; VkSampleCountFlagBits samples = VK_SAMPLE_COUNT_1_BIT; - u32 num_color_buffers = 0; u32 num_images = 0; std::array images{}; std::array image_ranges{}; @@ -428,10 +419,6 @@ public: return samples; } - [[nodiscard]] bool SupportsDepthComparison() const noexcept { - return supports_depth_comparison; - } - [[nodiscard]] bool RequiresBorderColorFormat() const noexcept { return requires_border_color_format; } @@ -482,7 +469,6 @@ private: VkComponentMapping swizzle_mapping{}; - bool supports_depth_comparison = false; bool requires_border_color_format = false; bool supports_minmax_filter = false; bool has_identity_swizzle = true; @@ -559,7 +545,6 @@ private: bool has_added_anisotropy{}; bool has_linear_filtering{}; - bool has_depth_comparison{}; bool has_minmax_reduction{}; bool has_custom_border_colors{}; bool has_srgb_border_color{}; diff --git a/src/video_core/renderer_vulkan/vk_update_descriptor.h b/src/video_core/renderer_vulkan/vk_update_descriptor.h index 04d76a8386..1efe39180f 100644 --- a/src/video_core/renderer_vulkan/vk_update_descriptor.h +++ b/src/video_core/renderer_vulkan/vk_update_descriptor.h @@ -47,10 +47,6 @@ public: bool supports_descriptor_buffer_ = false); ~UpdateDescriptorQueue(); - [[nodiscard]] bool UsesDescriptorBuffer() const noexcept { - return use_descriptor_buffer; - } - void TickFrame(); void Acquire(Scheduler& scheduler, size_t required_entries = 0, bool use_descriptor_buffer_ = false); diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 4a90b1a21d..639552341e 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -487,9 +487,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR properties.subgroup_size_control.maxSubgroupSize > GuestWarpSize; is_integrated = properties.properties.deviceType == VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU; - is_virtual = properties.properties.deviceType == VK_PHYSICAL_DEVICE_TYPE_VIRTUAL_GPU; - is_non_gpu = properties.properties.deviceType == VK_PHYSICAL_DEVICE_TYPE_OTHER || - properties.properties.deviceType == VK_PHYSICAL_DEVICE_TYPE_CPU; supports_d24_depth = IsFormatSupported(VK_FORMAT_D24_UNORM_S8_UINT, @@ -896,10 +893,6 @@ bool Device::GetSuitability(bool requires_swapchain) { // Assume we will be suitable. bool suitable = true; - // Configure properties. - VkPhysicalDeviceVulkan12Features features_1_2{}; - VkPhysicalDeviceVulkan13Features features_1_3{}; - // Configure properties. properties.properties = physical.GetProperties(); @@ -996,16 +989,6 @@ bool Device::GetSuitability(bool requires_swapchain) { // Set next pointer. void** next = &features2.pNext; - // Vulkan 1.2 and 1.3 features - if (instance_version >= VK_API_VERSION_1_2) { - features_1_2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES; - features_1_3.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_3_FEATURES; - - features_1_2.pNext = &features_1_3; - - *next = &features_1_2; - } - // Test all features we know about. If the feature is not available in core at our // current API version, and was not enabled by an extension, skip testing the feature. // We set the structure sType explicitly here as it is zeroed by the constructor. diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 23dceceb26..fba45d2385 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -128,7 +128,6 @@ VK_DEFINE_HANDLE(VmaAllocator) EXTENSION_NAME(VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_EXTENDED_DYNAMIC_STATE_3_EXTENSION_NAME) \ - EXTENSION_NAME(VK_EXT_EXTERNAL_MEMORY_HOST_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_4444_FORMATS_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_LINE_RASTERIZATION_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_ROBUSTNESS_2_EXTENSION_NAME) \ @@ -364,11 +363,6 @@ public: return {count[0], count[1], count[2]}; } - /// Returns the maximum size for push constants. - VkDeviceSize GetMaxPushConstantsSize() const { - return properties.properties.limits.maxPushConstantsSize; - } - #define FN_MAX_LIMIT_LIST \ FN_MAX_LIMIT_ELEM(ComputeSharedMemorySize) \ FN_MAX_LIMIT_ELEM(PerStageDescriptorSampledImages) \ @@ -658,13 +652,6 @@ FN_MAX_LIMIT_LIST return properties.depth_stencil_resolve.independentResolveNone == VK_TRUE; } - /// Returns true if depth/stencil operations can be performed efficiently. - /// Either through shader export or hardware blits. - bool CanPerformDepthStencilOperations() const { - return extensions.shader_stencil_export || is_blit_depth24_stencil8_supported || - is_blit_depth32_stencil8_supported; - } - /// Returns true if the device supports VK_EXT_depth_range_unrestricted. bool IsExtDepthRangeUnrestrictedSupported() const { return extensions.depth_range_unrestricted; @@ -705,12 +692,6 @@ FN_MAX_LIMIT_LIST return extensions.transform_feedback && properties.transform_feedback.transformFeedbackDraw; } - /// Returns true if transform feedback query types are supported. - bool IsTransformFeedbackQueriesSupported() const { - return extensions.transform_feedback && - properties.transform_feedback.transformFeedbackQueries; - } - /// Returns true if the device supports VK_EXT_transform_feedback properly. bool AreTransformFeedbackGeometryStreamsSupported() const { return features.transform_feedback.geometryStreams; @@ -745,11 +726,6 @@ FN_MAX_LIMIT_LIST !features.border_color_swizzle.borderColorSwizzleFromImage; } - /// Returns true if borderColorSwizzleFromImage is available. - bool IsBorderColorSwizzleFromImageSupported() const { - return features.border_color_swizzle.borderColorSwizzleFromImage; - } - /// Returns true if the device supports VK_EXT_extended_dynamic_state. bool IsExtExtendedDynamicStateSupported() const { return extensions.extended_dynamic_state; @@ -764,11 +740,6 @@ FN_MAX_LIMIT_LIST return features.extended_dynamic_state2.extendedDynamicState2LogicOp; } - /// Returns true if the device supports VK_EXT_extended_dynamic_state3. - bool IsExtExtendedDynamicState3Supported() const { - return extensions.extended_dynamic_state3; - } - /// Returns true if the device supports VK_EXT_4444_formats. bool IsExt4444FormatsSupported() const { return features.format_a4b4g4r4.formatA4B4G4R4; @@ -973,10 +944,6 @@ FN_MAX_LIMIT_LIST bool MustEmulateBGR565() const; - bool HasExactDepthBiasControl() const { - return features.depth_bias_control.depthBiasExact; - } - u32 GetMaxVertexInputAttributes() const { return properties.properties.limits.maxVertexInputAttributes; } @@ -1029,23 +996,11 @@ FN_MAX_LIMIT_LIST return extensions.maintenance5; } - /// Returns true if polygon mode POINT supports gl_PointSize. - bool SupportsPolygonModePointSize() const { - return extensions.maintenance5 && properties.maintenance5.polygonModePointSize; - } - /// Returns true if depth/stencil swizzle ONE is supported. bool SupportsDepthStencilSwizzleOne() const { return extensions.maintenance5 && properties.maintenance5.depthStencilSwizzleOneSupport; } - /// Returns true if early fragment tests optimizations are available. - bool SupportsEarlyFragmentTests() const { - return extensions.maintenance5 && - properties.maintenance5.earlyFragmentMultisampleCoverageAfterSampleCounting && - properties.maintenance5.earlyFragmentSampleMaskTestBeforeSampleCounting; - } - /// Returns true if the device supports UINT8 index buffer conversion via compute shader. bool SupportsUint8Indices() const { return features.bit8_storage.storageBuffer8BitAccess && @@ -1197,8 +1152,6 @@ private: bool is_blit_depth32_stencil8_supported{}; ///< Support for blitting from and to D32S8. bool is_warp_potentially_bigger{}; ///< Host warp size can be bigger than guest. bool is_integrated{}; ///< Is GPU an iGPU. - bool is_virtual{}; ///< Is GPU a virtual GPU. - bool is_non_gpu{}; ///< Is SoftwareRasterizer, FPGA, non-GPU device. bool has_broken_compute{}; ///< Compute shaders can cause crashes bool has_broken_cube_compatibility{}; ///< Has broken cube compatibility bit bool has_broken_descriptor_aliasing{}; ///< Miscompiles descriptors aliased on one binding diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp index a2ca3b7fa6..25c2218c53 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.cpp +++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp @@ -85,15 +85,11 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkAcquireNextImageKHR); X(vkAllocateCommandBuffers); X(vkAllocateDescriptorSets); - X(vkAllocateMemory); X(vkBeginCommandBuffer); - X(vkBindBufferMemory); - X(vkBindImageMemory); X(vkCmdBeginConditionalRenderingEXT); X(vkCmdBeginQuery); X(vkCmdBeginRenderPass); X(vkCmdBeginTransformFeedbackEXT); - X(vkCmdBeginDebugUtilsLabelEXT); X(vkCmdBindDescriptorSets); X(vkCmdBindIndexBuffer); X(vkCmdBindIndexBuffer2KHR); @@ -121,7 +117,6 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCmdEndQuery); X(vkCmdEndRenderPass); X(vkCmdEndTransformFeedbackEXT); - X(vkCmdEndDebugUtilsLabelEXT); X(vkCmdFillBuffer); X(vkCmdPipelineBarrier2); X(vkCmdPushConstants); @@ -155,7 +150,6 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCmdSetDepthClampEnableEXT); X(vkCmdSetFrontFaceEXT); X(vkCmdSetLogicOpEXT); - X(vkCmdSetPatchControlPointsEXT); X(vkCmdSetLineWidth); X(vkCmdSetPrimitiveTopologyEXT); X(vkCmdSetStencilOpEXT); @@ -178,7 +172,6 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCreateFence); X(vkCreateFramebuffer); X(vkCreateGraphicsPipelines); - X(vkCreateImage); X(vkCreateImageView); X(vkCreatePipelineCache); X(vkCreatePipelineLayout); @@ -212,30 +205,21 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkEndCommandBuffer); X(vkFreeCommandBuffers); X(vkFreeDescriptorSets); - X(vkFreeMemory); X(vkGetBufferMemoryRequirements2); X(vkGetDeviceBufferMemoryRequirements); X(vkGetDeviceQueue); X(vkGetEventStatus); X(vkGetFenceStatus); - X(vkGetImageMemoryRequirements); X(vkGetPipelineCacheData); - X(vkGetMemoryFdKHR); -#ifdef _WIN32 - X(vkGetMemoryWin32HandleKHR); -#endif X(vkGetQueryPoolResults); X(vkGetPipelineExecutablePropertiesKHR); X(vkGetPipelineExecutableStatisticsKHR); X(vkGetSemaphoreCounterValue); - X(vkMapMemory); X(vkQueueBindSparse); X(vkQueueSubmit2); X(vkResetFences); X(vkResetQueryPool); X(vkSetDebugUtilsObjectNameEXT); - X(vkSetDebugUtilsObjectTagEXT); - X(vkUnmapMemory); X(vkUpdateDescriptorSetWithTemplate); X(vkUpdateDescriptorSets); X(vkGetBufferDeviceAddress); @@ -319,9 +303,7 @@ bool Load(VkInstance instance, InstanceDispatch& dld) noexcept { // These functions may fail to load depending on the enabled extensions. // Don't return a failure on these. X(vkCreateDebugUtilsMessengerEXT); - X(vkCreateDebugReportCallbackEXT); X(vkDestroyDebugUtilsMessengerEXT); - X(vkDestroyDebugReportCallbackEXT); X(vkDestroySurfaceKHR); X(vkGetPhysicalDeviceFeatures2); X(vkGetPhysicalDeviceFormatProperties2); @@ -334,10 +316,10 @@ bool Load(VkInstance instance, InstanceDispatch& dld) noexcept { X(vkGetSwapchainImagesKHR); X(vkQueuePresentKHR); - return X(vkCreateDevice) && X(vkDestroyDevice) && X(vkDestroyDevice) && + return X(vkCreateDevice) && X(vkDestroyDevice) && X(vkEnumerateDeviceExtensionProperties) && X(vkEnumeratePhysicalDevices) && X(vkGetDeviceProcAddr) && X(vkGetPhysicalDeviceFormatProperties) && - X(vkGetPhysicalDeviceMemoryProperties) && X(vkGetPhysicalDeviceMemoryProperties2) && + X(vkGetPhysicalDeviceMemoryProperties2) && X(vkGetPhysicalDeviceProperties) && X(vkGetPhysicalDeviceQueueFamilyProperties); #undef X } @@ -379,10 +361,6 @@ void Destroy(VkDevice device, VkDescriptorUpdateTemplate handle, dld.vkDestroyDescriptorUpdateTemplate(device, handle, nullptr); } -void Destroy(VkDevice device, VkDeviceMemory handle, const DeviceDispatch& dld) noexcept { - dld.vkFreeMemory(device, handle, nullptr); -} - void Destroy(VkDevice device, VkEvent handle, const DeviceDispatch& dld) noexcept { dld.vkDestroyEvent(device, handle, nullptr); } @@ -444,11 +422,6 @@ void Destroy(VkInstance instance, VkDebugUtilsMessengerEXT handle, dld.vkDestroyDebugUtilsMessengerEXT(instance, handle, nullptr); } -void Destroy(VkInstance instance, VkDebugReportCallbackEXT handle, - const InstanceDispatch& dld) noexcept { - dld.vkDestroyDebugReportCallbackEXT(instance, handle, nullptr); -} - void Destroy(VkInstance instance, VkSurfaceKHR handle, const InstanceDispatch& dld) noexcept { dld.vkDestroySurfaceKHR(instance, handle, nullptr); } @@ -518,13 +491,6 @@ DebugUtilsMessenger Instance::CreateDebugUtilsMessenger( return DebugUtilsMessenger(object, handle, *dld); } -DebugReportCallback Instance::CreateDebugReportCallback( - const VkDebugReportCallbackCreateInfoEXT& create_info) const { - VkDebugReportCallbackEXT object; - Check(dld->vkCreateDebugReportCallbackEXT(handle, &create_info, nullptr, &object)); - return DebugReportCallback(object, handle, *dld); -} - void Image::SetObjectNameEXT(const char* name) const { SetObjectName(dld, owner, handle, VK_OBJECT_TYPE_IMAGE, name); } @@ -565,36 +531,6 @@ void ImageView::SetObjectNameEXT(const char* name) const { SetObjectName(dld, owner, handle, VK_OBJECT_TYPE_IMAGE_VIEW, name); } -int DeviceMemory::GetMemoryFdKHR() const { - const VkMemoryGetFdInfoKHR get_fd_info{ - .sType = VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR, - .pNext = nullptr, - .memory = handle, - .handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR, - }; - int fd; - Check(dld->vkGetMemoryFdKHR(owner, &get_fd_info, &fd)); - return fd; -} - -#ifdef _WIN32 -HANDLE DeviceMemory::GetMemoryWin32HandleKHR() const { - const VkMemoryGetWin32HandleInfoKHR get_win32_handle_info{ - .sType = VK_STRUCTURE_TYPE_MEMORY_GET_WIN32_HANDLE_INFO_KHR, - .pNext = nullptr, - .memory = handle, - .handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT_KHR, - }; - HANDLE win32_handle; - Check(dld->vkGetMemoryWin32HandleKHR(owner, &get_win32_handle_info, &win32_handle)); - return win32_handle; -} -#endif - -void DeviceMemory::SetObjectNameEXT(const char* name) const { - SetObjectName(dld, owner, handle, VK_OBJECT_TYPE_DEVICE_MEMORY, name); -} - void Fence::SetObjectNameEXT(const char* name) const { SetObjectName(dld, owner, handle, VK_OBJECT_TYPE_FENCE, name); } @@ -835,36 +771,6 @@ SwapchainKHR Device::CreateSwapchainKHR(const VkSwapchainCreateInfoKHR& ci) cons return SwapchainKHR(object, handle, *dld); } -DeviceMemory Device::TryAllocateMemory(const VkMemoryAllocateInfo& ai) const noexcept { - VkDeviceMemory memory; - if (dld->vkAllocateMemory(handle, &ai, nullptr, &memory) != VK_SUCCESS) { - return {}; - } - return DeviceMemory(memory, handle, *dld); -} - -DeviceMemory Device::AllocateMemory(const VkMemoryAllocateInfo& ai) const { - VkDeviceMemory memory; - Check(dld->vkAllocateMemory(handle, &ai, nullptr, &memory)); - return DeviceMemory(memory, handle, *dld); -} - -VkMemoryRequirements Device::GetBufferMemoryRequirements(VkBuffer buffer, - void* pnext) const noexcept { - const VkBufferMemoryRequirementsInfo2 info{ - .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_REQUIREMENTS_INFO_2, - .pNext = nullptr, - .buffer = buffer, - }; - VkMemoryRequirements2 requirements{ - .sType = VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2, - .pNext = pnext, - .memoryRequirements{}, - }; - dld->vkGetBufferMemoryRequirements2(handle, &info, &requirements); - return requirements.memoryRequirements; -} - VkMemoryRequirements Device::GetDeviceBufferMemoryRequirements( const VkBufferCreateInfo& ci) const noexcept { const VkDeviceBufferMemoryRequirements info{ @@ -881,12 +787,6 @@ VkMemoryRequirements Device::GetDeviceBufferMemoryRequirements( return requirements.memoryRequirements; } -VkMemoryRequirements Device::GetImageMemoryRequirements(VkImage image) const noexcept { - VkMemoryRequirements requirements; - dld->vkGetImageMemoryRequirements(handle, image, &requirements); - return requirements; -} - std::vector Device::GetPipelineExecutablePropertiesKHR( VkPipeline pipeline) const { const VkPipelineInfoKHR info{ @@ -938,14 +838,6 @@ void PhysicalDevice::GetProperties2(VkPhysicalDeviceProperties2& properties) con dld->vkGetPhysicalDeviceProperties2(physical_device, &properties); } -VkPhysicalDeviceFeatures PhysicalDevice::GetFeatures() const noexcept { - VkPhysicalDeviceFeatures2 features2; - features2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2; - features2.pNext = nullptr; - dld->vkGetPhysicalDeviceFeatures2(physical_device, &features2); - return features2.features; -} - void PhysicalDevice::GetFeatures2(VkPhysicalDeviceFeatures2& features) const noexcept { dld->vkGetPhysicalDeviceFeatures2(physical_device, &features); } diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index 9c60f03fd2..48b4b42221 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -169,10 +169,8 @@ struct InstanceDispatch { PFN_vkEnumerateInstanceLayerProperties vkEnumerateInstanceLayerProperties{}; PFN_vkCreateDebugUtilsMessengerEXT vkCreateDebugUtilsMessengerEXT{}; - PFN_vkCreateDebugReportCallbackEXT vkCreateDebugReportCallbackEXT{}; PFN_vkCreateDevice vkCreateDevice{}; PFN_vkDestroyDebugUtilsMessengerEXT vkDestroyDebugUtilsMessengerEXT{}; - PFN_vkDestroyDebugReportCallbackEXT vkDestroyDebugReportCallbackEXT{}; PFN_vkDestroyDevice vkDestroyDevice{}; PFN_vkDestroySurfaceKHR vkDestroySurfaceKHR{}; PFN_vkEnumerateDeviceExtensionProperties vkEnumerateDeviceExtensionProperties{}; @@ -181,7 +179,6 @@ struct InstanceDispatch { PFN_vkGetPhysicalDeviceFeatures2 vkGetPhysicalDeviceFeatures2{}; PFN_vkGetPhysicalDeviceFormatProperties vkGetPhysicalDeviceFormatProperties{}; PFN_vkGetPhysicalDeviceFormatProperties2 vkGetPhysicalDeviceFormatProperties2{}; - PFN_vkGetPhysicalDeviceMemoryProperties vkGetPhysicalDeviceMemoryProperties{}; PFN_vkGetPhysicalDeviceMemoryProperties2 vkGetPhysicalDeviceMemoryProperties2{}; PFN_vkGetPhysicalDeviceProperties vkGetPhysicalDeviceProperties{}; PFN_vkGetPhysicalDeviceProperties2 vkGetPhysicalDeviceProperties2{}; @@ -200,12 +197,8 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkAcquireNextImageKHR vkAcquireNextImageKHR{}; PFN_vkAllocateCommandBuffers vkAllocateCommandBuffers{}; PFN_vkAllocateDescriptorSets vkAllocateDescriptorSets{}; - PFN_vkAllocateMemory vkAllocateMemory{}; PFN_vkBeginCommandBuffer vkBeginCommandBuffer{}; - PFN_vkBindBufferMemory vkBindBufferMemory{}; - PFN_vkBindImageMemory vkBindImageMemory{}; PFN_vkCmdBeginConditionalRenderingEXT vkCmdBeginConditionalRenderingEXT{}; - PFN_vkCmdBeginDebugUtilsLabelEXT vkCmdBeginDebugUtilsLabelEXT{}; PFN_vkCmdBeginQuery vkCmdBeginQuery{}; PFN_vkCmdBeginRenderPass vkCmdBeginRenderPass{}; PFN_vkCmdBeginTransformFeedbackEXT vkCmdBeginTransformFeedbackEXT{}; @@ -234,7 +227,6 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdDrawIndexedIndirectCount vkCmdDrawIndexedIndirectCount{}; PFN_vkCmdDrawIndirectByteCountEXT vkCmdDrawIndirectByteCountEXT{}; PFN_vkCmdEndConditionalRenderingEXT vkCmdEndConditionalRenderingEXT{}; - PFN_vkCmdEndDebugUtilsLabelEXT vkCmdEndDebugUtilsLabelEXT{}; PFN_vkCmdEndQuery vkCmdEndQuery{}; PFN_vkCmdEndRenderPass vkCmdEndRenderPass{}; PFN_vkCmdEndTransformFeedbackEXT vkCmdEndTransformFeedbackEXT{}; @@ -260,13 +252,11 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdSetConservativeRasterizationModeEXT vkCmdSetConservativeRasterizationModeEXT{}; PFN_vkCmdSetLineRasterizationModeEXT vkCmdSetLineRasterizationModeEXT{}; PFN_vkCmdSetLineStippleEnableEXT vkCmdSetLineStippleEnableEXT{}; - PFN_vkCmdSetLineStippleEXT vkCmdSetLineStippleEXT{}; PFN_vkCmdSetDepthBiasEnableEXT vkCmdSetDepthBiasEnableEXT{}; PFN_vkCmdSetLogicOpEnableEXT vkCmdSetLogicOpEnableEXT{}; PFN_vkCmdSetDepthClampEnableEXT vkCmdSetDepthClampEnableEXT{}; PFN_vkCmdSetEvent2 vkCmdSetEvent2{}; PFN_vkCmdSetFrontFaceEXT vkCmdSetFrontFaceEXT{}; - PFN_vkCmdSetPatchControlPointsEXT vkCmdSetPatchControlPointsEXT{}; PFN_vkCmdSetLogicOpEXT vkCmdSetLogicOpEXT{}; PFN_vkCmdSetLineWidth vkCmdSetLineWidth{}; PFN_vkCmdSetPrimitiveTopologyEXT vkCmdSetPrimitiveTopologyEXT{}; @@ -294,7 +284,6 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCreateFence vkCreateFence{}; PFN_vkCreateFramebuffer vkCreateFramebuffer{}; PFN_vkCreateGraphicsPipelines vkCreateGraphicsPipelines{}; - PFN_vkCreateImage vkCreateImage{}; PFN_vkCreateImageView vkCreateImageView{}; PFN_vkCreatePipelineCache vkCreatePipelineCache{}; PFN_vkCreatePipelineLayout vkCreatePipelineLayout{}; @@ -328,30 +317,21 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkEndCommandBuffer vkEndCommandBuffer{}; PFN_vkFreeCommandBuffers vkFreeCommandBuffers{}; PFN_vkFreeDescriptorSets vkFreeDescriptorSets{}; - PFN_vkFreeMemory vkFreeMemory{}; PFN_vkGetBufferMemoryRequirements2 vkGetBufferMemoryRequirements2{}; PFN_vkGetDeviceBufferMemoryRequirements vkGetDeviceBufferMemoryRequirements{}; PFN_vkGetDeviceQueue vkGetDeviceQueue{}; PFN_vkGetEventStatus vkGetEventStatus{}; PFN_vkGetFenceStatus vkGetFenceStatus{}; - PFN_vkGetImageMemoryRequirements vkGetImageMemoryRequirements{}; PFN_vkGetPipelineCacheData vkGetPipelineCacheData{}; - PFN_vkGetMemoryFdKHR vkGetMemoryFdKHR{}; -#ifdef _WIN32 - PFN_vkGetMemoryWin32HandleKHR vkGetMemoryWin32HandleKHR{}; -#endif PFN_vkGetPipelineExecutablePropertiesKHR vkGetPipelineExecutablePropertiesKHR{}; PFN_vkGetPipelineExecutableStatisticsKHR vkGetPipelineExecutableStatisticsKHR{}; PFN_vkGetQueryPoolResults vkGetQueryPoolResults{}; PFN_vkGetSemaphoreCounterValue vkGetSemaphoreCounterValue{}; - PFN_vkMapMemory vkMapMemory{}; PFN_vkQueueBindSparse vkQueueBindSparse{}; PFN_vkQueueSubmit2 vkQueueSubmit2{}; PFN_vkResetFences vkResetFences{}; PFN_vkResetQueryPool vkResetQueryPool{}; PFN_vkSetDebugUtilsObjectNameEXT vkSetDebugUtilsObjectNameEXT{}; - PFN_vkSetDebugUtilsObjectTagEXT vkSetDebugUtilsObjectTagEXT{}; - PFN_vkUnmapMemory vkUnmapMemory{}; PFN_vkUpdateDescriptorSetWithTemplate vkUpdateDescriptorSetWithTemplate{}; PFN_vkGetBufferDeviceAddress vkGetBufferDeviceAddress{}; PFN_vkGetDescriptorSetLayoutSizeEXT vkGetDescriptorSetLayoutSizeEXT{}; @@ -381,7 +361,6 @@ void Destroy(VkDevice, VkCommandPool, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkDescriptorPool, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkDescriptorSetLayout, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkDescriptorUpdateTemplate, const DeviceDispatch&) noexcept; -void Destroy(VkDevice, VkDeviceMemory, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkEvent, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkFence, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkFramebuffer, const DeviceDispatch&) noexcept; @@ -397,7 +376,6 @@ void Destroy(VkDevice, VkSwapchainKHR, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkSemaphore, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkShaderModule, const DeviceDispatch&) noexcept; void Destroy(VkInstance, VkDebugUtilsMessengerEXT, const InstanceDispatch&) noexcept; -void Destroy(VkInstance, VkDebugReportCallbackEXT, const InstanceDispatch&) noexcept; void Destroy(VkInstance, VkSurfaceKHR, const InstanceDispatch&) noexcept; VkResult Free(VkDevice, VkDescriptorPool, Span, const DeviceDispatch&) noexcept; @@ -635,7 +613,6 @@ private: }; using DebugUtilsMessenger = Handle; -using DebugReportCallback = Handle; using DescriptorSetLayout = Handle; using DescriptorUpdateTemplate = Handle; using Pipeline = Handle; @@ -668,11 +645,6 @@ public: [[nodiscard]] DebugUtilsMessenger CreateDebugUtilsMessenger( const VkDebugUtilsMessengerCreateInfoEXT& create_info) const; - /// Creates a debug report callback. - /// @throw Exception on creation failure. - [[nodiscard]] DebugReportCallback CreateDebugReportCallback( - const VkDebugReportCallbackCreateInfoEXT& create_info) const; - /// Returns dispatch table. const InstanceDispatch& Dispatch() const noexcept { return *dld; @@ -880,30 +852,6 @@ public: void SetObjectNameEXT(const char* name) const; }; -class DeviceMemory : public Handle { - using Handle::Handle; - -public: - int GetMemoryFdKHR() const; - -#ifdef _WIN32 - HANDLE GetMemoryWin32HandleKHR() const; -#endif - - /// Set object name. - void SetObjectNameEXT(const char* name) const; - - u8* Map(VkDeviceSize offset, VkDeviceSize size) const { - void* data; - Check(dld->vkMapMemory(owner, handle, offset, size, 0, &data)); - return static_cast(data); - } - - void Unmap() const noexcept { - dld->vkUnmapMemory(owner, handle); - } -}; - class Fence : public Handle { using Handle::Handle; @@ -1088,17 +1036,8 @@ public: [[nodiscard]] SwapchainKHR CreateSwapchainKHR(const VkSwapchainCreateInfoKHR& ci) const; - [[nodiscard]] DeviceMemory TryAllocateMemory(const VkMemoryAllocateInfo& ai) const noexcept; - - [[nodiscard]] DeviceMemory AllocateMemory(const VkMemoryAllocateInfo& ai) const; - - VkMemoryRequirements GetBufferMemoryRequirements(VkBuffer buffer, - void* pnext = nullptr) const noexcept; - VkMemoryRequirements GetDeviceBufferMemoryRequirements(const VkBufferCreateInfo& ci) const noexcept; - VkMemoryRequirements GetImageMemoryRequirements(VkImage image) const noexcept; - std::vector GetPipelineExecutablePropertiesKHR( VkPipeline pipeline) const; @@ -1179,8 +1118,6 @@ public: void GetProperties2(VkPhysicalDeviceProperties2&) const noexcept; - VkPhysicalDeviceFeatures GetFeatures() const noexcept; - void GetFeatures2(VkPhysicalDeviceFeatures2&) const noexcept; VkFormatProperties GetFormatProperties(VkFormat) const noexcept; @@ -1566,11 +1503,6 @@ public: dld->vkCmdSetLineStippleEnableEXT(handle, enable ? VK_TRUE : VK_FALSE); } - void SetLineStippleEXT(u32 factor, u16 pattern) const noexcept - { - dld->vkCmdSetLineStippleEXT(handle, factor, pattern); - } - void SetDepthBiasEnableEXT(bool enable) const noexcept { dld->vkCmdSetDepthBiasEnableEXT(handle, enable ? VK_TRUE : VK_FALSE); } @@ -1599,10 +1531,6 @@ public: dld->vkCmdSetLogicOpEXT(handle, logic_op); } - void SetPatchControlPointsEXT(uint32_t patch_control_points) const noexcept { - dld->vkCmdSetPatchControlPointsEXT(handle, patch_control_points); - } - void SetColorWriteMaskEXT(u32 first, Span masks) const noexcept { dld->vkCmdSetColorWriteMaskEXT(handle, first, masks.size(), masks.data()); } @@ -1673,20 +1601,6 @@ public: dld->vkCmdEndConditionalRenderingEXT(handle); } - void BeginDebugUtilsLabelEXT(const char* label, std::span color) const noexcept { - const VkDebugUtilsLabelEXT label_info{ - .sType = VK_STRUCTURE_TYPE_DEBUG_UTILS_LABEL_EXT, - .pNext = nullptr, - .pLabelName = label, - .color{color[0], color[1], color[2], color[3]}, - }; - dld->vkCmdBeginDebugUtilsLabelEXT(handle, &label_info); - } - - void EndDebugUtilsLabelEXT() const noexcept { - dld->vkCmdEndDebugUtilsLabelEXT(handle); - } - private: VkCommandBuffer handle; const DeviceDispatch* dld; From 98ad2d5e4c20e246056281a3288ced8e97ab253a Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 24 Sep 2026 15:31:51 -0400 Subject: [PATCH 53/69] Quick tests on payloads --- src/shader_recompiler/ir_opt/texture_pass.cpp | 8 ++++---- src/video_core/gpu.cpp | 6 +++++- 2 files changed, 9 insertions(+), 5 deletions(-) diff --git a/src/shader_recompiler/ir_opt/texture_pass.cpp b/src/shader_recompiler/ir_opt/texture_pass.cpp index fe34a67555..82e1c1c39e 100644 --- a/src/shader_recompiler/ir_opt/texture_pass.cpp +++ b/src/shader_recompiler/ir_opt/texture_pass.cpp @@ -33,7 +33,8 @@ using TextureInstVector = boost::container::small_vector; constexpr u32 DESCRIPTOR_SIZE = 8; constexpr u32 DESCRIPTOR_SIZE_SHIFT = u32(std::countr_zero(DESCRIPTOR_SIZE)); -constexpr u32 DESCRIPTOR_MAX_COUNT = 1024; +constexpr u32 DESCRIPTOR_MAX_COUNT = 512; +constexpr u32 DESCRIPTOR_CBUF_BYTES = 16 * 1024; u32 DynamicDescriptorSizeShift(const IR::U32& dynamic_offset) { const IR::Inst* const inst = dynamic_offset.InstRecursive(); @@ -48,11 +49,10 @@ u32 DynamicDescriptorSizeShift(const IR::U32& dynamic_offset) { u32 DynamicDescriptorCount(u32 base_offset, u32 size_shift, u32 max_descriptors) { auto const descriptor_limit = (std::max)(1U, max_descriptors); - auto const max_cbuf_bytes = 16 * descriptor_limit; - if (size_shift >= 31 || base_offset >= max_cbuf_bytes) + if (size_shift >= 31 || base_offset >= DESCRIPTOR_CBUF_BYTES) return 1; auto const stride = 1U << size_shift; - auto const available = max_cbuf_bytes - base_offset; + auto const available = DESCRIPTOR_CBUF_BYTES - base_offset; if (available < DESCRIPTOR_SIZE) return 1; auto const available_count = 1U + (available - DESCRIPTOR_SIZE) / stride; diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 9901cf4262..eb834308e0 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -152,6 +152,10 @@ struct GPU::Impl { /// Tick pending requests within the GPU. void TickWork() { + if (last_sync_fence.load(std::memory_order_relaxed) == + current_sync_fence.load(std::memory_order_relaxed)) { + return; + } std::unique_lock lck{sync_request_mutex}; while (!sync_requests.empty()) { auto request = std::move(sync_requests.front()); @@ -328,7 +332,7 @@ struct GPU::Impl { std::list> sync_requests; std::atomic current_sync_fence{}; - u64 last_sync_fence{}; + std::atomic last_sync_fence{}; std::mutex sync_request_mutex; std::condition_variable sync_request_cv; From 108c991431a14b584b5e496be995938b934216da Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 24 Sep 2026 17:57:47 -0400 Subject: [PATCH 54/69] Experiment on descriptors payloads --- src/shader_recompiler/ir_opt/texture_pass.cpp | 2 +- .../renderer_opengl/gl_texture_cache.h | 1 - .../renderer_vulkan/vk_descriptor_buffer.h | 6 +-- .../renderer_vulkan/vk_texture_cache.h | 1 - .../texture_cache/descriptor_table.h | 54 +++++++++++-------- src/video_core/texture_cache/texture_cache.h | 40 +++++--------- .../texture_cache/texture_cache_base.h | 5 -- 7 files changed, 47 insertions(+), 62 deletions(-) diff --git a/src/shader_recompiler/ir_opt/texture_pass.cpp b/src/shader_recompiler/ir_opt/texture_pass.cpp index 82e1c1c39e..0228ce117d 100644 --- a/src/shader_recompiler/ir_opt/texture_pass.cpp +++ b/src/shader_recompiler/ir_opt/texture_pass.cpp @@ -33,7 +33,7 @@ using TextureInstVector = boost::container::small_vector; constexpr u32 DESCRIPTOR_SIZE = 8; constexpr u32 DESCRIPTOR_SIZE_SHIFT = u32(std::countr_zero(DESCRIPTOR_SIZE)); -constexpr u32 DESCRIPTOR_MAX_COUNT = 512; +constexpr u32 DESCRIPTOR_MAX_COUNT = 128; constexpr u32 DESCRIPTOR_CBUF_BYTES = 16 * 1024; u32 DynamicDescriptorSizeShift(const IR::U32& dynamic_offset) { diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index 9f3b244a8d..7c7ee14580 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -364,7 +364,6 @@ private: }; struct TextureCacheParams { - static constexpr bool ENABLE_VALIDATION = true; static constexpr bool FRAMEBUFFER_BLITS = true; static constexpr bool HAS_EMULATED_COPIES = true; static constexpr bool HAS_DEVICE_MEMORY_INFO = true; diff --git a/src/video_core/renderer_vulkan/vk_descriptor_buffer.h b/src/video_core/renderer_vulkan/vk_descriptor_buffer.h index 6f96bcc326..4cde5abcd4 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_buffer.h +++ b/src/video_core/renderer_vulkan/vk_descriptor_buffer.h @@ -17,9 +17,9 @@ class Device; class Scheduler; class DescriptorBufferRing final { - static constexpr size_t FRAMES_IN_FLIGHT = 8; - static constexpr VkDeviceSize TILER_FRAME_SIZE = 2 * 1024 * 1024; - static constexpr VkDeviceSize DESKTOP_FRAME_SIZE = 4 * 1024 * 1024; + static constexpr size_t FRAMES_IN_FLIGHT = 3; + static constexpr VkDeviceSize TILER_FRAME_SIZE = 5 * 1024 * 1024; + static constexpr VkDeviceSize DESKTOP_FRAME_SIZE = 8 * 1024 * 1024; public: explicit DescriptorBufferRing(const Device& device_, MemoryAllocator& memory_allocator); diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 67cf67872b..06ce5a5c32 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -552,7 +552,6 @@ private: }; struct TextureCacheParams { - static constexpr bool ENABLE_VALIDATION = true; static constexpr bool FRAMEBUFFER_BLITS = false; static constexpr bool HAS_EMULATED_COPIES = false; static constexpr bool HAS_DEVICE_MEMORY_INFO = true; diff --git a/src/video_core/texture_cache/descriptor_table.h b/src/video_core/texture_cache/descriptor_table.h index e40c128ab5..ec310803f4 100644 --- a/src/video_core/texture_cache/descriptor_table.h +++ b/src/video_core/texture_cache/descriptor_table.h @@ -7,12 +7,16 @@ #pragma once #include +#include +#include +#include #include #include "common/alignment.h" #include "common/common_types.h" #include "common/div_ceil.h" #include "common/assert.h" +#include "common/slot_vector.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" @@ -21,6 +25,12 @@ namespace VideoCommon { template class DescriptorTable { public: + struct Entry { + T descriptor; + Common::SlotId id; + u32 generation; + }; + [[nodiscard]] bool Synchronize(GPUVAddr gpu_addr, u32 limit) noexcept { bool ret = !(current_gpu_addr == gpu_addr && current_limit == limit); if (ret) { @@ -30,44 +40,42 @@ public: } void Invalidate() noexcept { - std::ranges::fill(read_descriptors, 0); + ++generation; } - [[nodiscard]] std::pair Read(Tegra::MemoryManager const& gpu_memory, u32 index) noexcept { + [[nodiscard]] std::pair Read(Tegra::MemoryManager const& gpu_memory, u32 index) noexcept { DEBUG_ASSERT(index <= current_limit); const GPUVAddr gpu_addr = current_gpu_addr + index * sizeof(T); - std::pair result; - gpu_memory.ReadBlockUnsafe(gpu_addr, std::addressof(result.first), sizeof(T)); - if ((read_descriptors[index / 64] & (1ULL << (index % 64))) != 0) { - result.second = result.first != descriptors[index]; - } else { - read_descriptors[index / 64] |= 1ULL << (index % 64); - result.second = true; + T value{}; + if (!aligned) { + gpu_memory.ReadBlockUnsafe(gpu_addr, std::addressof(value), sizeof(T)); + } else if (const u8* const ptr = gpu_memory.GetPointer(gpu_addr)) { + std::memcpy(std::addressof(value), ptr, sizeof(T)); } - if (result.second) { - descriptors[index] = result.first; + Entry& entry = entries[index]; + const bool is_new = entry.generation != generation || entry.descriptor != value; + if (is_new) { + entry.descriptor = value; + entry.generation = generation; } - return result; + return {entry, is_new}; } void Refresh(GPUVAddr gpu_addr, u32 limit) noexcept { current_gpu_addr = gpu_addr; current_limit = limit; - // Mario Brothership reallocates a lot of times, so use aggressive pre-alloc sizes - // std::vector by default uses quadratic growth, but that isn't even enough to satisfy brothership - const size_t num_descriptors = ((limit + 0x80000) & (~0x7ffff)) + 1; - size_t old_size = read_descriptors.size(); - read_descriptors.resize(Common::DivCeil(num_descriptors, 64U)); - old_size = (std::min)(old_size, read_descriptors.size()); - std::fill(read_descriptors.begin(), read_descriptors.begin() + old_size, 0); - // - descriptors.resize(num_descriptors); + aligned = gpu_addr % sizeof(T) == 0; + ++generation; + if (entries.size() <= limit) { + entries.resize(std::bit_ceil(size_t{limit} + 1)); + } } - std::vector read_descriptors; - std::vector descriptors; + std::vector entries; GPUVAddr current_gpu_addr{}; u32 current_limit{}; + u32 generation{1}; + bool aligned{}; }; } // namespace VideoCommon diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 4092036b9e..2b25184d9b 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -270,14 +270,11 @@ SamplerId TextureCache

::GetSamplerId(u32 index, bool compute) { LOG_DEBUG(HW_GPU, "Invalid sampler index={}", index); return NULL_SAMPLER_ID; } - auto const map_index = index | (compute ? Common::SlotId::TAGGED_VALUE : 0); - auto const [descriptor, is_new] = table.Read(*gpu_memory, index); + auto const [entry, is_new] = table.Read(*gpu_memory, index); if (is_new) { - auto const id = FindSampler(descriptor, compute); - channel_state->sampler_ids.insert_or_assign(map_index, id); - return id; + entry.id = FindSampler(entry.descriptor, compute); } - return channel_state->sampler_ids.find(map_index)->second; + return entry.id; } template @@ -500,26 +497,19 @@ ImageViewId TextureCache

::VisitImageView(u32 index, bool compute) { LOG_DEBUG(HW_GPU, "Invalid image view index={}", index); return NULL_IMAGE_VIEW_ID; } - auto const map_index = index | (compute ? Common::SlotId::TAGGED_VALUE : 0); - // Is new (on the tegra engine side)? - auto const [descriptor, is_new] = table.Read(*gpu_memory, index); + auto const [entry, is_new] = table.Read(*gpu_memory, index); if (is_new) { - if (IsValidEntry(*gpu_memory, descriptor)) { - // Is new (registered view) on the texture cache side? - const auto [pair, is_new_tc] = channel_state->image_views.try_emplace(descriptor); + entry.id = NULL_IMAGE_VIEW_ID; + if (IsValidEntry(*gpu_memory, entry.descriptor)) { + const auto [pair, is_new_tc] = channel_state->image_views.try_emplace(entry.descriptor); if (is_new_tc) - pair->second = CreateImageView(descriptor); - PrepareImageView(pair->second, false, false); - channel_state->image_view_ids.insert_or_assign(map_index, pair->second); - return pair->second; + pair->second = CreateImageView(entry.descriptor); + entry.id = pair->second; } - channel_state->image_view_ids.insert_or_assign(map_index, NULL_IMAGE_VIEW_ID); - return NULL_IMAGE_VIEW_ID; } - auto const it = channel_state->image_view_ids.find(map_index); - if (it->second != NULL_IMAGE_VIEW_ID) - PrepareImageView(it->second, false, false); - return it->second; + if (entry.id != NULL_IMAGE_VIEW_ID) + PrepareImageView(entry.id, false, false); + return entry.id; } template @@ -1296,9 +1286,6 @@ void TextureCache

::InvalidateScale(Image& image) { for (size_t c : active_channel_ids) { auto& channel_info = channel_storage[c]; - if constexpr (ENABLE_VALIDATION) - for (auto& e : channel_info.image_view_ids) - e.second = CORRUPT_ID; channel_info.graphics_image_table.Invalidate(); channel_info.compute_image_table.Invalidate(); } @@ -2295,9 +2282,6 @@ void TextureCache

::DeleteImage(ImageId image_id, bool immediate_delete) { } for (size_t c : active_channel_ids) { auto& channel_info = channel_storage[c]; - if constexpr (ENABLE_VALIDATION) - for (auto& e : channel_info.image_view_ids) - e.second = CORRUPT_ID; channel_info.graphics_image_table.Invalidate(); channel_info.compute_image_table.Invalidate(); } diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index 7a46134490..b51f03e532 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -89,9 +89,6 @@ public: std::unordered_map image_views; std::unordered_map samplers; - ::Common::unordered_map sampler_ids; - ::Common::unordered_map image_view_ids; - TextureCacheGPUMap* gpu_page_table = nullptr; TextureCacheGPUMap* sparse_page_table = nullptr; }; @@ -101,8 +98,6 @@ class TextureCache : public VideoCommon::ChannelSetupCaches Date: Thu, 24 Sep 2026 19:13:52 -0400 Subject: [PATCH 55/69] Another try on sparse binding --- src/video_core/buffer_cache/buffer_base.h | 11 ++++++ src/video_core/buffer_cache/buffer_cache.h | 35 +++++++++++++------ .../buffer_cache/buffer_cache_base.h | 3 +- .../renderer_vulkan/vk_buffer_cache.cpp | 1 + .../renderer_vulkan/vk_buffer_cache.h | 6 +--- .../renderer_vulkan/vk_multi_range_buffer.cpp | 11 +----- .../renderer_vulkan/vk_multi_range_buffer.h | 8 ++--- .../renderer_vulkan/vk_query_cache.cpp | 1 + 8 files changed, 44 insertions(+), 32 deletions(-) diff --git a/src/video_core/buffer_cache/buffer_base.h b/src/video_core/buffer_cache/buffer_base.h index d7eb21612f..6b2a37ea8c 100644 --- a/src/video_core/buffer_cache/buffer_base.h +++ b/src/video_core/buffer_cache/buffer_base.h @@ -129,13 +129,24 @@ public: write_tick = write_tick_; } + u64 ContentSerial() const noexcept { + return content_serial; + } + + void MarkContentModified() noexcept { + content_serial = ++next_content_serial; + } + private: + static inline u64 next_content_serial = 0; + VAddr cpu_addr = 0; BufferFlagBits flags{}; int stream_score = 0; size_t lru_id = SIZE_MAX; size_t size_bytes = 0; u64 write_tick = 0; + u64 content_serial = ++next_content_serial; }; } // namespace VideoCommon diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index eb214af50e..89d0003207 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -266,6 +266,7 @@ bool BufferCache

::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am const auto& copy = copies[0]; src_buffer.MarkUsage(copy.src_offset, copy.size); dest_buffer.MarkUsage(copy.dst_offset, copy.size); + dest_buffer.MarkContentModified(); runtime.CopyBuffer(dest_buffer, src_buffer, copies, true); if (has_new_downloads) { memory_tracker.MarkRegionAsGpuModified(*cpu_dest_address, amount); @@ -297,6 +298,7 @@ bool BufferCache

::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) { const u32 offset = dest_buffer.Offset(*cpu_dst_address); runtime.ClearBuffer(dest_buffer, offset, size, value); dest_buffer.MarkUsage(offset, size); + dest_buffer.MarkContentModified(); return true; } @@ -773,6 +775,7 @@ void BufferCache

::BindHostIndexBuffer() { if (draw_state.inline_index_draw_indexes.empty()) { SynchronizeBuffer(buffer, channel_state->index_buffer.device_addr, size); } else { + buffer.MarkContentModified(); if constexpr (USE_MEMORY_MAPS_FOR_UPLOADS) { auto upload_staging = runtime.UploadStagingBuffer(size); std::array copies{{BufferCopy{.src_offset = upload_staging.offset, .dst_offset = 0, .size = size}}}; @@ -1029,11 +1032,13 @@ void BufferCache

::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 template void BufferCache

::ResolveMultiRangeStorage(Binding& binding, - std::vector& pool) { + std::vector& pool, + bool is_written) { binding.segment_first = 0; binding.segment_count = 0; if constexpr (requires { runtime.BindMultiRangeStorageBuffer(u64{}, bool{}); }) { - if (binding.gpu_addr == 0 || binding.size == 0) { + if (binding.gpu_addr == 0 || binding.size == 0 || + (is_written && !runtime.PrefersSparseSources())) { return; } const VirtualSegments* found = @@ -1041,7 +1046,7 @@ void BufferCache

::ResolveMultiRangeStorage(Binding& binding, if (!found || found->size() < 2) { return; } - const VirtualSegments segments = *found; + const VirtualSegments& segments = *found; const u32 first = static_cast(pool.size()); const bool prefer_sparse = runtime.PrefersSparseSources(); for (const VirtualSegment& segment : segments) { @@ -1078,9 +1083,7 @@ bool BufferCache

::BindMultiRangeStorage(const Binding& binding, bool is_writt const MultiRangeSegment& segment = pool[binding.segment_first + index]; Buffer& buffer = slot_buffers[segment.buffer_id]; TouchBuffer(buffer, segment.buffer_id); - if (SynchronizeBuffer(buffer, segment.device_addr, segment.size)) { - runtime.InvalidateMultiRange(key); - } + SynchronizeBuffer(buffer, segment.device_addr, segment.size); const u32 offset = buffer.Offset(segment.device_addr); buffer.MarkUsage(offset, segment.size); if (is_written) { @@ -1456,9 +1459,12 @@ void BufferCache

::UpdateStorageBuffers(size_t stage) { ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) { // Resolve buffer Binding& binding = channel_state->storage_buffers[stage][index]; - const BufferId buffer_id = FindBuffer(binding.device_addr, binding.size, false); - binding.buffer_id = buffer_id; - ResolveMultiRangeStorage(binding, graphics_segments); + const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0; + ResolveMultiRangeStorage(binding, graphics_segments, is_written); + binding.buffer_id = NULL_BUFFER_ID; + if (binding.segment_count == 0 || is_written) { + binding.buffer_id = FindBuffer(binding.device_addr, binding.size, false); + } }); } @@ -1521,8 +1527,12 @@ void BufferCache

::UpdateComputeStorageBuffers() { ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) { // Resolve buffer Binding& binding = channel_state->compute_storage_buffers[index]; - binding.buffer_id = FindBuffer(binding.device_addr, binding.size, false); - ResolveMultiRangeStorage(binding, compute_segments); + const bool is_written = ((channel_state->written_compute_storage_buffers >> index) & 1) != 0; + ResolveMultiRangeStorage(binding, compute_segments, is_written); + binding.buffer_id = NULL_BUFFER_ID; + if (binding.segment_count == 0 || is_written) { + binding.buffer_id = FindBuffer(binding.device_addr, binding.size, false); + } }); } @@ -1539,6 +1549,7 @@ void BufferCache

::MarkWrittenBuffer(BufferId buffer_id, DAddr device_addr, u3 if constexpr (!IS_OPENGL) { Buffer& buffer = slot_buffers[buffer_id]; buffer.setWriteTick(runtime.CurrentTick()); + buffer.MarkContentModified(); } memory_tracker.MarkRegionAsGpuModified(device_addr, size); gpu_modified_ranges.Add(device_addr, size); @@ -1784,6 +1795,7 @@ bool BufferCache

::SynchronizeBuffer(Buffer& buffer, DAddr device_addr, u32 si template void BufferCache

::UploadMemory(Buffer& buffer, u64 total_size_bytes, u64 largest_copy, std::span copies) { + buffer.MarkContentModified(); if constexpr (USE_MEMORY_MAPS_FOR_UPLOADS) { MappedUploadMemory(buffer, total_size_bytes, copies); } else { @@ -1869,6 +1881,7 @@ void BufferCache

::InlineMemoryImplementation(DAddr dest_address, size_t copy_ BufferId buffer_id = FindBuffer(dest_address, static_cast(copy_size), false); auto& buffer = slot_buffers[buffer_id]; SynchronizeBuffer(buffer, dest_address, static_cast(copy_size)); + buffer.MarkContentModified(); if constexpr (USE_MEMORY_MAPS_FOR_UPLOADS) { auto upload_staging = runtime.UploadStagingBuffer(copy_size); diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h index b4af2820d5..80a0b2c7eb 100644 --- a/src/video_core/buffer_cache/buffer_cache_base.h +++ b/src/video_core/buffer_cache/buffer_cache_base.h @@ -235,7 +235,8 @@ public: - void ResolveMultiRangeStorage(Binding& binding, std::vector& pool); + void ResolveMultiRangeStorage(Binding& binding, std::vector& pool, + bool is_written); void UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t size); diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index dec26f328d..d3d3d7a437 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -428,6 +428,7 @@ u32 BufferCacheRuntime::GetStorageBufferAlignment() const { } void BufferCacheRuntime::TickFrame(Common::SlotVector& slot_buffers) noexcept { + multi_range_buffers.DrainRetired(scheduler); for (auto it = slot_buffers.begin(); it != slot_buffers.end(); it++) { if (scheduler.IsFree(it->LastUsageTick())) { it->ResetUsageTracking(); diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h index 582f665b4e..f83696f0de 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.h +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h @@ -192,7 +192,7 @@ public: .memory_offset = location.offset, .offset = offset, .size = size, - .write_tick = buffer.getWriteTick(), + .content_serial = buffer.ContentSerial(), .memory_type = location.memory_type, }); multi_range_total += size; @@ -200,10 +200,6 @@ public: bool BindMultiRangeStorageBuffer(u64 key, bool is_written); - void InvalidateMultiRange(u64 key) { - multi_range_buffers.Invalidate(key); - } - void OnBufferDeleted(const Buffer& buffer) { multi_range_buffers.DropOwner(scheduler, buffer.Handle()); } diff --git a/src/video_core/renderer_vulkan/vk_multi_range_buffer.cpp b/src/video_core/renderer_vulkan/vk_multi_range_buffer.cpp index 6d1ae5fc53..68dcab9250 100644 --- a/src/video_core/renderer_vulkan/vk_multi_range_buffer.cpp +++ b/src/video_core/renderer_vulkan/vk_multi_range_buffer.cpp @@ -87,7 +87,7 @@ u64 MultiRangeBufferCache::HashSources(std::span sources u64 MultiRangeBufferCache::HashContent(std::span sources) const { u64 hash = 0xcbf29ce484222325ULL; for (const MultiRangeSource& source : sources) { - hash ^= source.write_tick; + hash ^= source.content_serial; hash *= 0x100000001b3ULL; } return hash; @@ -219,9 +219,6 @@ MultiRangeRef MultiRangeBufferCache::Get(const Device& device, Scheduler& schedu if (sources.empty() || total == 0) { return MultiRangeRef{}; } - if (!retired.empty()) { - DrainRetired(scheduler); - } const u64 geometry = HashSources(sources); const u64 content = HashContent(sources); const auto it = entries.find(key); @@ -335,10 +332,4 @@ void MultiRangeBufferCache::DropOwner(Scheduler& scheduler, VkBuffer owner) { } } -void MultiRangeBufferCache::Invalidate(u64 key) { - if (auto const it = entries.find(key); it != entries.end()) { - it->second.dirty = true; - } -} - } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_multi_range_buffer.h b/src/video_core/renderer_vulkan/vk_multi_range_buffer.h index f40d3aa1f8..d9c065df3f 100644 --- a/src/video_core/renderer_vulkan/vk_multi_range_buffer.h +++ b/src/video_core/renderer_vulkan/vk_multi_range_buffer.h @@ -27,7 +27,7 @@ struct MultiRangeSource { VkDeviceSize memory_offset{}; VkDeviceSize offset{}; VkDeviceSize size{}; - u64 write_tick{}; + u64 content_serial{}; u32 memory_type{}; }; @@ -55,10 +55,10 @@ public: void MarkGathered(u64 key); - void Invalidate(u64 key); - void DropOwner(Scheduler& scheduler, VkBuffer owner); + void DrainRetired(Scheduler& scheduler); + VkDeviceSize block_size{DEFAULT_BLOCK_SIZE}; bool use_sparse{}; @@ -94,8 +94,6 @@ private: void RetireEntry(Scheduler& scheduler, Entry& entry); - void DrainRetired(Scheduler& scheduler); - ::Common::unordered_map entries; boost::container::static_vector retired; u32 sparse_memory_type_bits{}; diff --git a/src/video_core/renderer_vulkan/vk_query_cache.cpp b/src/video_core/renderer_vulkan/vk_query_cache.cpp index 7bdb5738c8..e14629dfc8 100644 --- a/src/video_core/renderer_vulkan/vk_query_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_query_cache.cpp @@ -1654,6 +1654,7 @@ void QueryCacheRuntime::SyncValues(std::span values, VkBuffer ba const auto post_op = VideoCommon::ObtainBufferOperation::DoNothing; const auto [buffer, offset] = impl->buffer_cache.ObtainCPUBuffer( pair.first, static_cast(pair.second - pair.first), sync_info, post_op); + buffer->MarkContentModified(); impl->buffers_to_upload_to.emplace_back(buffer->Handle(), offset); } }); From 19298b713b9eef4e2264b90107f8dd01440d50fe Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 24 Sep 2026 23:28:37 -0400 Subject: [PATCH 56/69] Removal of WFI + address device creation crash on A6xx --- src/android/app/src/main/jni/native.cpp | 101 +++++++++--------- src/video_core/engines/sw_blitter/blitter.cpp | 3 +- .../renderer_vulkan/vk_rasterizer.cpp | 41 +------ .../renderer_vulkan/vk_rasterizer.h | 1 - .../vulkan_common/vulkan_wrapper.cpp | 31 ------ src/video_core/vulkan_common/vulkan_wrapper.h | 28 ----- 6 files changed, 52 insertions(+), 153 deletions(-) diff --git a/src/android/app/src/main/jni/native.cpp b/src/android/app/src/main/jni/native.cpp index 746b00ad11..05b2f081eb 100644 --- a/src/android/app/src/main/jni/native.cpp +++ b/src/android/app/src/main/jni/native.cpp @@ -855,36 +855,53 @@ jboolean JNICALL Java_org_yuzu_yuzu_1emu_utils_GpuDriverHelper_supportsCustomDri #endif } +#ifdef ARCHITECTURE_arm64 +namespace { + +struct SystemDriverInfo { + VkPhysicalDeviceProperties properties{}; + VkPhysicalDeviceDriverProperties driver{}; +}; + +SystemDriverInfo QuerySystemDriverInfo(JNIEnv* env, jstring j_hook_lib_dir) { + const std::string hook_lib_dir = Common::Android::GetJString(env, j_hook_lib_dir); + const Common::DynamicLibrary library{adrenotools_open_libvulkan( + RTLD_NOW, 0, nullptr, hook_lib_dir.c_str(), nullptr, nullptr, nullptr, nullptr)}; + Vulkan::vk::InstanceDispatch dld; + const Vulkan::vk::Instance instance = Vulkan::CreateInstance(library, dld, VK_API_VERSION_1_1); + const std::vector devices = instance.EnumeratePhysicalDevices(); + if (devices.empty()) { + throw Vulkan::vk::Exception(VK_ERROR_INITIALIZATION_FAILED); + } + SystemDriverInfo info{}; + info.driver.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES; + VkPhysicalDeviceProperties2 properties2{ + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2, + .pNext = &info.driver, + .properties = {}, + }; + Vulkan::vk::PhysicalDevice(devices[0], dld).GetProperties2(properties2); + info.properties = properties2.properties; + return info; +} + +} +#endif + jobjectArray Java_org_yuzu_yuzu_1emu_utils_GpuDriverHelper_getSystemDriverInfo( JNIEnv* env, jobject j_obj, jobject j_surf, jstring j_hook_lib_dir) { + std::string version_string{"1.1.0"}; + std::string driver_name{"generic"}; #ifdef ARCHITECTURE_arm64 - const char* file_redirect_dir_{}; - int featureFlags{}; - std::string hook_lib_dir = Common::Android::GetJString(env, j_hook_lib_dir); - auto handle = adrenotools_open_libvulkan(RTLD_NOW, featureFlags, nullptr, hook_lib_dir.c_str(), - nullptr, nullptr, file_redirect_dir_, nullptr); - auto driver_library = std::make_shared(handle); - InputCommon::InputSubsystem input_subsystem; - auto window = - std::make_unique(ANativeWindow_fromSurface(env, j_surf), driver_library); - - Vulkan::vk::InstanceDispatch dld; - Vulkan::vk::Instance vk_instance = Vulkan::CreateInstance( - *driver_library, dld, VK_API_VERSION_1_1, Core::Frontend::WindowSystemType::Android); - - auto surface = Vulkan::CreateSurface(vk_instance, window->GetWindowInfo()); - - auto device = Vulkan::CreateDevice(vk_instance, dld, *surface); - - auto driver_version = device.GetDriverVersion(); - auto version_string = - fmt::format("{}.{}.{}", VK_API_VERSION_MAJOR(driver_version), - VK_API_VERSION_MINOR(driver_version), VK_API_VERSION_PATCH(driver_version)); - auto driver_name = device.GetDriverName(); -#else - auto driver_version = "1.0.0"; - auto version_string = "1.1.0"; //Assume lowest Vulkan level - auto driver_name = "generic"; + try { + const SystemDriverInfo info = QuerySystemDriverInfo(env, j_hook_lib_dir); + const u32 driver_version = info.properties.driverVersion; + version_string = + fmt::format("{}.{}.{}", VK_API_VERSION_MAJOR(driver_version), + VK_API_VERSION_MINOR(driver_version), VK_API_VERSION_PATCH(driver_version)); + driver_name = Vulkan::vk::GetDriverName(info.driver); + } catch (...) { + } #endif jobjectArray j_driver_info = env->NewObjectArray(2, Common::Android::GetStringClass(), Common::Android::ToJString(env, version_string)); env->SetObjectArrayElement(j_driver_info, 1, Common::Android::ToJString(env, driver_name)); @@ -893,32 +910,12 @@ jobjectArray Java_org_yuzu_yuzu_1emu_utils_GpuDriverHelper_getSystemDriverInfo( jstring Java_org_yuzu_yuzu_1emu_utils_GpuDriverHelper_getGpuModel(JNIEnv *env, jobject j_obj, jobject j_surf, jstring j_hook_lib_dir) { #ifdef ARCHITECTURE_arm64 - const char* file_redirect_dir_{}; - int featureFlags{}; - std::string hook_lib_dir = Common::Android::GetJString(env, j_hook_lib_dir); - auto handle = adrenotools_open_libvulkan(RTLD_NOW, featureFlags, nullptr, hook_lib_dir.c_str(), - nullptr, nullptr, file_redirect_dir_, nullptr); - auto driver_library = std::make_shared(handle); - InputCommon::InputSubsystem input_subsystem; - auto window = - std::make_unique(ANativeWindow_fromSurface(env, j_surf), driver_library); - - Vulkan::vk::InstanceDispatch dld; - Vulkan::vk::Instance vk_instance = Vulkan::CreateInstance( - *driver_library, dld, VK_API_VERSION_1_1, Core::Frontend::WindowSystemType::Android); - - auto surface = Vulkan::CreateSurface(vk_instance, window->GetWindowInfo()); - - auto device = Vulkan::CreateDevice(vk_instance, dld, *surface); - - const std::string model_name{device.GetModelName()}; - - window.release(); - - return Common::Android::ToJString(env, model_name); -#else - return Common::Android::ToJString(env, "no-info"); + try { + return Common::Android::ToJString(env, QuerySystemDriverInfo(env, j_hook_lib_dir).properties.deviceName); + } catch (...) { + } #endif + return Common::Android::ToJString(env, "no-info"); } jboolean Java_org_yuzu_yuzu_1emu_NativeLibrary_reloadKeys(JNIEnv* env, jclass clazz) { diff --git a/src/video_core/engines/sw_blitter/blitter.cpp b/src/video_core/engines/sw_blitter/blitter.cpp index 8bcc2f7a76..45df323e52 100644 --- a/src/video_core/engines/sw_blitter/blitter.cpp +++ b/src/video_core/engines/sw_blitter/blitter.cpp @@ -92,7 +92,8 @@ void Bilinear(std::span input, std::span output, size_t src_widt const auto read_src = [&](f32 in_x, f32 in_y) { const size_t read_from = - ((static_cast(in_x) * src_width + static_cast(in_y)) >> 32) * + ((std::min)(static_cast(in_y), src_height - 1) * src_width + + (std::min)(static_cast(in_x), src_width - 1)) * ir_components; return std::span(&input[read_from], ir_components); }; diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index c32cbd2a0d..6a0aef3447 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -226,8 +226,7 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra descriptor_buffer_ring, render_pass_cache, buffer_cache, texture_cache, gpu.ShaderNotify()), accelerate_dma(buffer_cache, texture_cache, scheduler), - fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler), - wfi_event(device.GetLogical().CreateEvent()) { + fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler) { scheduler.SetQueryCache(query_cache); } @@ -881,45 +880,7 @@ void RasterizerVulkan::FlushAndInvalidateRegion(DAddr addr, u64 size, } void RasterizerVulkan::WaitForIdle() { - // Everything but wait pixel operations. This intentionally includes FRAGMENT_SHADER_BIT because - // fragment shaders can still write storage buffers. - VkPipelineStageFlags2 flags = - VK_PIPELINE_STAGE_2_DRAW_INDIRECT_BIT | VK_PIPELINE_STAGE_2_VERTEX_INPUT_BIT | - VK_PIPELINE_STAGE_2_VERTEX_SHADER_BIT | - VK_PIPELINE_STAGE_2_TESSELLATION_CONTROL_SHADER_BIT | - VK_PIPELINE_STAGE_2_TESSELLATION_EVALUATION_SHADER_BIT | - VK_PIPELINE_STAGE_2_GEOMETRY_SHADER_BIT | VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | - VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_2_TRANSFER_BIT; - if (device.IsExtTransformFeedbackSupported()) { - flags |= VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT; - } - query_cache.NotifyWFI(); - - scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([event = *wfi_event, flags](vk::CommandBuffer cmdbuf) { - const VkMemoryBarrier2 barrier{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = flags, - .srcAccessMask = VK_ACCESS_2_NONE, - .dstStageMask = VK_PIPELINE_STAGE_2_TOP_OF_PIPE_BIT, - .dstAccessMask = VK_ACCESS_2_NONE, - }; - const VkDependencyInfo dependency_info{ - .sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO, - .pNext = nullptr, - .dependencyFlags = 0, - .memoryBarrierCount = 1, - .pMemoryBarriers = &barrier, - .bufferMemoryBarrierCount = 0, - .pBufferMemoryBarriers = nullptr, - .imageMemoryBarrierCount = 0, - .pImageMemoryBarriers = nullptr, - }; - cmdbuf.SetEvent(event, dependency_info); - cmdbuf.WaitEvents(event, dependency_info); - }); fence_manager.SignalOrdering(); } diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index ea98948df7..e47104783b 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -221,7 +221,6 @@ private: AccelerateDMA accelerate_dma; FenceManager fence_manager; - vk::Event wfi_event; boost::container::static_vector image_view_indices; std::array image_view_ids; diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp index 25c2218c53..44e519a3fa 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.cpp +++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp @@ -125,13 +125,11 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCmdSetDepthBias); X(vkCmdSetDepthBias2EXT); X(vkCmdSetDepthBounds); - X(vkCmdSetEvent2); X(vkCmdSetScissor); X(vkCmdSetStencilCompareMask); X(vkCmdSetStencilReference); X(vkCmdSetStencilWriteMask); X(vkCmdSetViewport); - X(vkCmdWaitEvents2); X(vkCmdBindVertexBuffers2EXT); X(vkCmdSetCullModeEXT); X(vkCmdSetDepthBoundsTestEnableEXT); @@ -168,7 +166,6 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCreateDescriptorPool); X(vkCreateDescriptorSetLayout); X(vkCreateDescriptorUpdateTemplate); - X(vkCreateEvent); X(vkCreateFence); X(vkCreateFramebuffer); X(vkCreateGraphicsPipelines); @@ -187,7 +184,6 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkDestroyDescriptorPool); X(vkDestroyDescriptorSetLayout); X(vkDestroyDescriptorUpdateTemplate); - X(vkDestroyEvent); X(vkDestroyFence); X(vkDestroyFramebuffer); X(vkDestroyImage); @@ -208,7 +204,6 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkGetBufferMemoryRequirements2); X(vkGetDeviceBufferMemoryRequirements); X(vkGetDeviceQueue); - X(vkGetEventStatus); X(vkGetFenceStatus); X(vkGetPipelineCacheData); X(vkGetQueryPoolResults); @@ -261,12 +256,6 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { if (!dld.vkQueueSubmit2) { Proc(dld.vkQueueSubmit2, dld, "vkQueueSubmit2KHR", device); } - if (!dld.vkCmdSetEvent2) { - Proc(dld.vkCmdSetEvent2, dld, "vkCmdSetEvent2KHR", device); - } - if (!dld.vkCmdWaitEvents2) { - Proc(dld.vkCmdWaitEvents2, dld, "vkCmdWaitEvents2KHR", device); - } if (!dld.vkCreateRenderPass2) { Proc(dld.vkCreateRenderPass2, dld, "vkCreateRenderPass2KHR", device); @@ -361,10 +350,6 @@ void Destroy(VkDevice device, VkDescriptorUpdateTemplate handle, dld.vkDestroyDescriptorUpdateTemplate(device, handle, nullptr); } -void Destroy(VkDevice device, VkEvent handle, const DeviceDispatch& dld) noexcept { - dld.vkDestroyEvent(device, handle, nullptr); -} - void Destroy(VkDevice device, VkFence handle, const DeviceDispatch& dld) noexcept { dld.vkDestroyFence(device, handle, nullptr); } @@ -590,10 +575,6 @@ std::vector SwapchainKHR::GetImages() const { return images; } -void Event::SetObjectNameEXT(const char* name) const { - SetObjectName(dld, owner, handle, VK_OBJECT_TYPE_EVENT, name); -} - void ShaderModule::SetObjectNameEXT(const char* name) const { SetObjectName(dld, owner, handle, VK_OBJECT_TYPE_SHADER_MODULE, name); } @@ -753,18 +734,6 @@ ShaderModule Device::CreateShaderModule(const VkShaderModuleCreateInfo& ci) cons return ShaderModule(object, handle, *dld); } -Event Device::CreateEvent() const { - static constexpr VkEventCreateInfo ci{ - .sType = VK_STRUCTURE_TYPE_EVENT_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - }; - - VkEvent object; - Check(dld->vkCreateEvent(handle, &ci, nullptr, &object)); - return Event(object, handle, *dld); -} - SwapchainKHR Device::CreateSwapchainKHR(const VkSwapchainCreateInfoKHR& ci) const { VkSwapchainKHR object; Check(dld->vkCreateSwapchainKHR(handle, &ci, nullptr, &object)); diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index 48b4b42221..7898886921 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -255,7 +255,6 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdSetDepthBiasEnableEXT vkCmdSetDepthBiasEnableEXT{}; PFN_vkCmdSetLogicOpEnableEXT vkCmdSetLogicOpEnableEXT{}; PFN_vkCmdSetDepthClampEnableEXT vkCmdSetDepthClampEnableEXT{}; - PFN_vkCmdSetEvent2 vkCmdSetEvent2{}; PFN_vkCmdSetFrontFaceEXT vkCmdSetFrontFaceEXT{}; PFN_vkCmdSetLogicOpEXT vkCmdSetLogicOpEXT{}; PFN_vkCmdSetLineWidth vkCmdSetLineWidth{}; @@ -272,7 +271,6 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdSetColorWriteEnableEXT vkCmdSetColorWriteEnableEXT{}; PFN_vkCmdSetColorBlendEnableEXT vkCmdSetColorBlendEnableEXT{}; PFN_vkCmdSetColorBlendEquationEXT vkCmdSetColorBlendEquationEXT{}; - PFN_vkCmdWaitEvents2 vkCmdWaitEvents2{}; PFN_vkCreateBuffer vkCreateBuffer{}; PFN_vkCreateBufferView vkCreateBufferView{}; PFN_vkCreateCommandPool vkCreateCommandPool{}; @@ -280,7 +278,6 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCreateDescriptorPool vkCreateDescriptorPool{}; PFN_vkCreateDescriptorSetLayout vkCreateDescriptorSetLayout{}; PFN_vkCreateDescriptorUpdateTemplate vkCreateDescriptorUpdateTemplate{}; - PFN_vkCreateEvent vkCreateEvent{}; PFN_vkCreateFence vkCreateFence{}; PFN_vkCreateFramebuffer vkCreateFramebuffer{}; PFN_vkCreateGraphicsPipelines vkCreateGraphicsPipelines{}; @@ -299,7 +296,6 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkDestroyDescriptorPool vkDestroyDescriptorPool{}; PFN_vkDestroyDescriptorSetLayout vkDestroyDescriptorSetLayout{}; PFN_vkDestroyDescriptorUpdateTemplate vkDestroyDescriptorUpdateTemplate{}; - PFN_vkDestroyEvent vkDestroyEvent{}; PFN_vkDestroyFence vkDestroyFence{}; PFN_vkDestroyFramebuffer vkDestroyFramebuffer{}; PFN_vkDestroyImage vkDestroyImage{}; @@ -320,7 +316,6 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkGetBufferMemoryRequirements2 vkGetBufferMemoryRequirements2{}; PFN_vkGetDeviceBufferMemoryRequirements vkGetDeviceBufferMemoryRequirements{}; PFN_vkGetDeviceQueue vkGetDeviceQueue{}; - PFN_vkGetEventStatus vkGetEventStatus{}; PFN_vkGetFenceStatus vkGetFenceStatus{}; PFN_vkGetPipelineCacheData vkGetPipelineCacheData{}; PFN_vkGetPipelineExecutablePropertiesKHR vkGetPipelineExecutablePropertiesKHR{}; @@ -361,7 +356,6 @@ void Destroy(VkDevice, VkCommandPool, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkDescriptorPool, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkDescriptorSetLayout, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkDescriptorUpdateTemplate, const DeviceDispatch&) noexcept; -void Destroy(VkDevice, VkEvent, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkFence, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkFramebuffer, const DeviceDispatch&) noexcept; void Destroy(VkDevice, VkImage, const DeviceDispatch&) noexcept; @@ -908,18 +902,6 @@ public: std::vector GetImages() const; }; -class Event : public Handle { - using Handle::Handle; - -public: - /// Set object name. - void SetObjectNameEXT(const char* name) const; - - VkResult GetStatus() const noexcept { - return dld->vkGetEventStatus(owner, handle); - } -}; - class ShaderModule : public Handle { using Handle::Handle; @@ -1032,8 +1014,6 @@ public: [[nodiscard]] ShaderModule CreateShaderModule(const VkShaderModuleCreateInfo& ci) const; - [[nodiscard]] Event CreateEvent() const; - [[nodiscard]] SwapchainKHR CreateSwapchainKHR(const VkSwapchainCreateInfoKHR& ci) const; VkMemoryRequirements GetDeviceBufferMemoryRequirements(const VkBufferCreateInfo& ci) const noexcept; @@ -1445,14 +1425,6 @@ public: dld->vkCmdSetDepthBounds(handle, min_depth_bounds, max_depth_bounds); } - void SetEvent(VkEvent event, const VkDependencyInfo& dependency_info) const noexcept { - dld->vkCmdSetEvent2(handle, event, &dependency_info); - } - - void WaitEvents(Span events, const VkDependencyInfo& dependency_info) const noexcept { - dld->vkCmdWaitEvents2(handle, events.size(), events.data(), &dependency_info); - } - void BindVertexBuffers2EXT(u32 first_binding, u32 binding_count, const VkBuffer* buffers, const VkDeviceSize* offsets, const VkDeviceSize* sizes, const VkDeviceSize* strides) const noexcept { From 3f3240caed66983a6759e6301c0ff7aced7e6de0 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Thu, 24 Sep 2026 23:46:01 -0400 Subject: [PATCH 57/69] Fix android build --- src/android/app/src/main/jni/native.cpp | 62 +++++++++---------- src/video_core/engines/sw_blitter/blitter.cpp | 3 + 2 files changed, 32 insertions(+), 33 deletions(-) diff --git a/src/android/app/src/main/jni/native.cpp b/src/android/app/src/main/jni/native.cpp index 05b2f081eb..bd4d843473 100644 --- a/src/android/app/src/main/jni/native.cpp +++ b/src/android/app/src/main/jni/native.cpp @@ -738,6 +738,35 @@ const char* fallback_cpu_detection() { return s_result.c_str(); } +#ifdef ARCHITECTURE_arm64 +struct SystemDriverInfo { + VkPhysicalDeviceProperties properties{}; + VkPhysicalDeviceDriverProperties driver{}; +}; + +SystemDriverInfo QuerySystemDriverInfo(JNIEnv* env, jstring j_hook_lib_dir) { + const std::string hook_lib_dir = Common::Android::GetJString(env, j_hook_lib_dir); + const Common::DynamicLibrary library{adrenotools_open_libvulkan( + RTLD_NOW, 0, nullptr, hook_lib_dir.c_str(), nullptr, nullptr, nullptr, nullptr)}; + Vulkan::vk::InstanceDispatch dld; + const Vulkan::vk::Instance instance = Vulkan::CreateInstance(library, dld, VK_API_VERSION_1_1); + const std::vector devices = instance.EnumeratePhysicalDevices(); + if (devices.empty()) { + throw Vulkan::vk::Exception(VK_ERROR_INITIALIZATION_FAILED); + } + SystemDriverInfo info{}; + info.driver.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES; + VkPhysicalDeviceProperties2 properties2{ + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2, + .pNext = &info.driver, + .properties = {}, + }; + Vulkan::vk::PhysicalDevice(devices[0], dld).GetProperties2(properties2); + info.properties = properties2.properties; + return info; +} +#endif + } // namespace extern "C" { @@ -855,39 +884,6 @@ jboolean JNICALL Java_org_yuzu_yuzu_1emu_utils_GpuDriverHelper_supportsCustomDri #endif } -#ifdef ARCHITECTURE_arm64 -namespace { - -struct SystemDriverInfo { - VkPhysicalDeviceProperties properties{}; - VkPhysicalDeviceDriverProperties driver{}; -}; - -SystemDriverInfo QuerySystemDriverInfo(JNIEnv* env, jstring j_hook_lib_dir) { - const std::string hook_lib_dir = Common::Android::GetJString(env, j_hook_lib_dir); - const Common::DynamicLibrary library{adrenotools_open_libvulkan( - RTLD_NOW, 0, nullptr, hook_lib_dir.c_str(), nullptr, nullptr, nullptr, nullptr)}; - Vulkan::vk::InstanceDispatch dld; - const Vulkan::vk::Instance instance = Vulkan::CreateInstance(library, dld, VK_API_VERSION_1_1); - const std::vector devices = instance.EnumeratePhysicalDevices(); - if (devices.empty()) { - throw Vulkan::vk::Exception(VK_ERROR_INITIALIZATION_FAILED); - } - SystemDriverInfo info{}; - info.driver.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES; - VkPhysicalDeviceProperties2 properties2{ - .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2, - .pNext = &info.driver, - .properties = {}, - }; - Vulkan::vk::PhysicalDevice(devices[0], dld).GetProperties2(properties2); - info.properties = properties2.properties; - return info; -} - -} -#endif - jobjectArray Java_org_yuzu_yuzu_1emu_utils_GpuDriverHelper_getSystemDriverInfo( JNIEnv* env, jobject j_obj, jobject j_surf, jstring j_hook_lib_dir) { std::string version_string{"1.1.0"}; diff --git a/src/video_core/engines/sw_blitter/blitter.cpp b/src/video_core/engines/sw_blitter/blitter.cpp index 45df323e52..a360ffb125 100644 --- a/src/video_core/engines/sw_blitter/blitter.cpp +++ b/src/video_core/engines/sw_blitter/blitter.cpp @@ -1,3 +1,6 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + // SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project // SPDX-License-Identifier: GPL-3.0-or-later From 7a007fbf487c98f1a5bec97c922a51fff68afaa9 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Fri, 25 Sep 2026 01:51:43 -0400 Subject: [PATCH 58/69] Another intent to fix qcom driver bugs with render --- src/video_core/renderer_vulkan/vk_texture_cache.cpp | 12 +++++++----- 1 file changed, 7 insertions(+), 5 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 32d4c5a52d..279b43b5ff 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -2028,6 +2028,9 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu } current_image = &Image::original_image; storage_image_views.resize(info.resources.levels); + if (original_image) { + runtime->TransitionImageLayout(*this); + } } Image::Image(const VideoCommon::NullImageParams& params) : VideoCommon::ImageBase{params} {} @@ -3238,9 +3241,9 @@ void TextureCacheRuntime::TransitionImageLayout(Image& image) { VkImageMemoryBarrier2 barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .srcStageMask = VK_PIPELINE_STAGE_2_NONE, .srcAccessMask = VK_ACCESS_2_NONE, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, + .dstStageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED, .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -3255,9 +3258,8 @@ void TextureCacheRuntime::TransitionImageLayout(Image& image) { .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }; - scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([barrier](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(barrier); + scheduler.RecordWithUploadBuffer([barrier](vk::CommandBuffer, vk::CommandBuffer upload_cmdbuf) { + upload_cmdbuf.PipelineBarrier(barrier); }); } } From a5c5966cd8d7683cafae390662937e5ddb2aa557 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Fri, 25 Sep 2026 02:25:13 -0400 Subject: [PATCH 59/69] Keep adjusting barriers --- src/video_core/renderer_vulkan/vk_compute_pass.cpp | 3 +-- src/video_core/renderer_vulkan/vk_texture_cache.cpp | 9 ++------- 2 files changed, 3 insertions(+), 9 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index ca5a267c6a..083b3e900c 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -644,8 +644,7 @@ void RecordUnswizzleEndBarrier(Scheduler& scheduler, VkImage vk_image, .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT | - VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 279b43b5ff..50b3bb7caf 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -692,9 +692,6 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im static constexpr VkAccessFlags2 WRITE_ACCESS_FLAGS = VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; - static constexpr VkAccessFlags2 READ_ACCESS_FLAGS = VK_ACCESS_2_SHADER_READ_BIT | - VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT; // Compute exact mip/layer range being written to RangedBarrierRange range; @@ -725,10 +722,8 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | - VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT, - .dstAccessMask = WRITE_ACCESS_FLAGS | READ_ACCESS_FLAGS, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, From ab11282def5f26dbc2f98ff37865cfa605ac8d92 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Fri, 25 Sep 2026 02:56:40 -0400 Subject: [PATCH 60/69] Keep adjusting the barriers until qcom is finally fixed --- src/video_core/renderer_vulkan/blit_image.cpp | 62 +++---------------- .../renderer_vulkan/vk_rasterizer.cpp | 5 +- .../renderer_vulkan/vk_texture_cache.cpp | 17 +---- 3 files changed, 15 insertions(+), 69 deletions(-) diff --git a/src/video_core/renderer_vulkan/blit_image.cpp b/src/video_core/renderer_vulkan/blit_image.cpp index 928e7bfb9c..b462572a5b 100644 --- a/src/video_core/renderer_vulkan/blit_image.cpp +++ b/src/video_core/renderer_vulkan/blit_image.cpp @@ -452,34 +452,6 @@ VkExtent2D GetConversionExtent(const ImageView& src_image_view) { }; } -void TransitionImageLayout(vk::CommandBuffer& cmdbuf, VkImage image, VkImageLayout target_layout, - VkImageLayout source_layout = VK_IMAGE_LAYOUT_GENERAL) { - constexpr VkAccessFlags2 flags{VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_SHADER_READ_BIT}; - const VkImageMemoryBarrier2 barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .srcAccessMask = flags, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = flags, - .oldLayout = source_layout, - .newLayout = target_layout, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .image = image, - .subresourceRange{ - .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, - .baseMipLevel = 0, - .levelCount = 1, - .baseArrayLayer = 0, - .layerCount = 1, - }, - }; - cmdbuf.PipelineBarrier(barrier); -} - void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view) { const VkImage image = image_view.ImageHandle(); const VkImageSubresourceRange subresource_range = SubresourceRangeFromView(image_view); @@ -562,25 +534,6 @@ void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view) }, }); } - -void BeginRenderPass(vk::CommandBuffer& cmdbuf, const Framebuffer* framebuffer) { - const VkRenderPass render_pass = framebuffer->RenderPass(); - const VkFramebuffer framebuffer_handle = framebuffer->Handle(); - const VkExtent2D render_area = framebuffer->RenderArea(); - const VkRenderPassBeginInfo renderpass_bi{ - .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO, - .pNext = nullptr, - .renderPass = render_pass, - .framebuffer = framebuffer_handle, - .renderArea{ - .offset{}, - .extent = render_area, - }, - .clearValueCount = 0, - .pClearValues = nullptr, - }; - cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE); -} } // Anonymous namespace BlitImageHelper::BlitImageHelper(const Device& device_, Scheduler& scheduler_, @@ -686,11 +639,14 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, VkImageView }; const VkPipelineLayout layout = *one_texture_pipeline_layout; const VkPipeline pipeline = FindOrEmplaceColorPipeline(key); - scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([this, dst_framebuffer, src_image_view, src_image, src_sampler, dst_region, - src_region, src_size, pipeline, layout](vk::CommandBuffer cmdbuf) { - TransitionImageLayout(cmdbuf, src_image, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL); - BeginRenderPass(cmdbuf, dst_framebuffer); + const auto attachments = + std::span(dst_framebuffer->Images()).first(dst_framebuffer->NumImages()); + if (std::ranges::find(attachments, src_image) != attachments.end()) { + scheduler.RequestOutsideRenderPassOperationContext(); + } + scheduler.RequestRenderpass(dst_framebuffer); + scheduler.Record([this, src_image_view, src_sampler, dst_region, src_region, src_size, + pipeline, layout](vk::CommandBuffer cmdbuf) { const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit(); UpdateOneTextureDescriptorSet(device, descriptor_set, src_sampler, src_image_view); cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline); @@ -698,8 +654,8 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, VkImageView nullptr); BindBlitState(cmdbuf, layout, dst_region, src_region, src_size); cmdbuf.Draw(3, 1, 0, 0); - cmdbuf.EndRenderPass(); }); + scheduler.InvalidateState(); } void BlitImageHelper::BlitImpl(const Framebuffer* dst_framebuffer, diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 6a0aef3447..eb0e52c04c 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -421,8 +421,9 @@ void RasterizerVulkan::DrawTexture() { .y = ScaleSrc(draw_texture_state.src_y1)}}; Extent3D src_size = {static_cast(ScaleSrc(texture.size.width)), static_cast(ScaleSrc(texture.size.height)), texture.size.depth}; - blit_image.BlitColor(framebuffer, texture.RenderTarget(), texture.ImageHandle(), - sampler->Handle(), dst_region, src_region, src_size); + blit_image.BlitColor(framebuffer, texture.Handle(Shader::TextureType::Color2D), + texture.ImageHandle(), sampler->Handle(), dst_region, src_region, + src_size); } void RasterizerVulkan::Clear(u32 layer_count) { diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 50b3bb7caf..71eb640975 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -1719,10 +1719,7 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = 0, - .dstStageMask = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | - VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, .dstAccessMask = 0, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1736,16 +1733,8 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | - VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT | - VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, From bcb3fbcf129479846892e0b469bfc4156f851ff3 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Fri, 25 Sep 2026 03:42:09 -0400 Subject: [PATCH 61/69] Remove the binding/rt image cache --- .../texture_cache/descriptor_table.h | 6 +- src/video_core/texture_cache/texture_cache.h | 93 +++---------------- .../texture_cache/texture_cache_base.h | 6 -- 3 files changed, 13 insertions(+), 92 deletions(-) diff --git a/src/video_core/texture_cache/descriptor_table.h b/src/video_core/texture_cache/descriptor_table.h index ec310803f4..37022842fe 100644 --- a/src/video_core/texture_cache/descriptor_table.h +++ b/src/video_core/texture_cache/descriptor_table.h @@ -31,12 +31,10 @@ public: u32 generation; }; - [[nodiscard]] bool Synchronize(GPUVAddr gpu_addr, u32 limit) noexcept { - bool ret = !(current_gpu_addr == gpu_addr && current_limit == limit); - if (ret) { + void Synchronize(GPUVAddr gpu_addr, u32 limit) noexcept { + if (current_gpu_addr != gpu_addr || current_limit != limit) { Refresh(gpu_addr, limit); } - return ret; } void Invalidate() noexcept { diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 2b25184d9b..46cead4c35 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -196,65 +196,16 @@ void TextureCache

::FillImageViews(std::span views, bool compu template void TextureCache

::CheckFeedbackLoop(std::span views) { - if (!Settings::values.barrier_feedback_loops.GetValue()) { + if (!rt_depth_image_id || !Settings::values.barrier_feedback_loops.GetValue()) { return; } - - if (render_targets_serial == last_feedback_loop_serial && - texture_bindings_serial == last_feedback_texture_serial) { - if (last_feedback_loop_result) { + const ImageViewId depth_view_id = render_targets.depth_buffer_id; + for (const auto& view : views) { + if (view.id && view.id != depth_view_id && + slot_image_views[view.id].image_id == rt_depth_image_id) { runtime.BarrierFeedbackLoop(); + return; } - return; - } - - if (rt_active_mask == 0) { - last_feedback_loop_serial = render_targets_serial; - last_feedback_texture_serial = texture_bindings_serial; - last_feedback_loop_result = false; - return; - } - const u32 depth_bit = 1u << NUM_RT; - const bool depth_active = (rt_active_mask & depth_bit) != 0; - - const bool requires_barrier = [&] { - for (const auto& view : views) { - if (!view.id) { - continue; - } - - { - bool is_continue = false; - for (size_t i = 0; i < 8; ++i) - is_continue |= (rt_active_mask & (1u << i)) && view.id == render_targets.color_buffer_ids[i]; - if (is_continue) - continue; - } - - if (depth_active && view.id == render_targets.depth_buffer_id) - continue; - - const ImageId view_image_id = slot_image_views[view.id].image_id; - { - bool is_continue = false; - for (size_t i = 0; i < 8; ++i) - is_continue |= (rt_active_mask & (1u << i)) && view_image_id == rt_image_id[i]; - if (is_continue) - continue; - } - if (depth_active && view_image_id == rt_depth_image_id) { - return true; - } - } - - return false; - }(); - - last_feedback_loop_serial = render_targets_serial; - last_feedback_texture_serial = texture_bindings_serial; - last_feedback_loop_result = requires_barrier; - if (requires_barrier) { - runtime.BarrierFeedbackLoop(); } } @@ -293,26 +244,14 @@ void TextureCache

::SynchronizeDescriptors(bool compute) { const bool linked_tsc = kepler_compute->launch_description.linked_tsc; const u32 tic_limit = kepler_compute->regs.tic.limit; const u32 tsc_limit = linked_tsc ? tic_limit : kepler_compute->regs.tsc.limit; - bool bindings_changed = false; - if (channel_state->compute_sampler_table.Synchronize(kepler_compute->regs.tsc.Address(), tsc_limit)) - bindings_changed = true; - if (channel_state->compute_image_table.Synchronize(kepler_compute->regs.tic.Address(), tic_limit)) - bindings_changed = true; - if (bindings_changed) { - ++texture_bindings_serial; - } + channel_state->compute_sampler_table.Synchronize(kepler_compute->regs.tsc.Address(), tsc_limit); + channel_state->compute_image_table.Synchronize(kepler_compute->regs.tic.Address(), tic_limit); } else { const bool linked_tsc = maxwell3d->regs.sampler_binding == Tegra::Engines::Maxwell3D::Regs::SamplerBinding::ViaHeaderBinding; const u32 tic_limit = maxwell3d->regs.tex_header.limit; const u32 tsc_limit = linked_tsc ? tic_limit : maxwell3d->regs.tex_sampler.limit; - bool bindings_changed = false; - if (channel_state->graphics_sampler_table.Synchronize(maxwell3d->regs.tex_sampler.Address(), tsc_limit)) - bindings_changed = true; - if (channel_state->graphics_image_table.Synchronize(maxwell3d->regs.tex_header.Address(), tic_limit)) - bindings_changed = true; - if (bindings_changed) { - ++texture_bindings_serial; - } + channel_state->graphics_sampler_table.Synchronize(maxwell3d->regs.tex_sampler.Address(), tsc_limit); + channel_state->graphics_image_table.Synchronize(maxwell3d->regs.tex_header.Address(), tic_limit); } } @@ -442,19 +381,9 @@ void TextureCache

::UpdateRenderTargets(bool is_clear) { PrepareImageView(depth_buffer_id, true, is_clear && IsFullClear(depth_buffer_id)); - rt_active_mask = 0; - rt_image_id = {}; - for (size_t i = 0; i < rt_image_id.size(); ++i) { - if (ImageViewId const view = render_targets.color_buffer_ids[i]; view) { - rt_active_mask |= 1u << i; - rt_image_id[i] = slot_image_views[view].image_id; - } - } + rt_depth_image_id = {}; if (depth_buffer_id) { - rt_active_mask |= (1u << NUM_RT); rt_depth_image_id = slot_image_views[depth_buffer_id].image_id; - } else { - rt_depth_image_id = ImageId{}; } for (size_t index = 0; index < NUM_RT; ++index) { diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index b51f03e532..0243a68a2d 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -418,13 +418,7 @@ private: RenderTargets render_targets; u64 render_targets_serial = 0; - u32 rt_active_mask = 0; - std::array rt_image_id{}; ImageId rt_depth_image_id{}; - u64 texture_bindings_serial = 0; - u64 last_feedback_loop_serial = 0; - u64 last_feedback_texture_serial = 0; - bool last_feedback_loop_result = false; FramebufferId last_framebuffer_id{}; u64 last_framebuffer_serial = 0; From bc121efc1900d29d469f210ae1ba5dbf2f2a43b1 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Fri, 25 Sep 2026 04:32:48 -0400 Subject: [PATCH 62/69] Corrections on the barriers for compute dispatch --- .../renderer_vulkan/vk_rasterizer.cpp | 30 +++++++++++++------ 1 file changed, 21 insertions(+), 9 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index eb0e52c04c..4ba8c2abbc 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -625,6 +625,23 @@ void RasterizerVulkan::DispatchCompute() { return; } + static constexpr VkMemoryBarrier2 READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = + VK_PIPELINE_STAGE_2_DRAW_INDIRECT_BIT | VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT, + }; + static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_ALL_GRAPHICS_BIT | VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, + }; const auto& qmd{kepler_compute->launch_description}; auto indirect_address = kepler_compute->GetIndirectComputeAddress(); if (indirect_address) { @@ -639,7 +656,9 @@ void RasterizerVulkan::DispatchCompute() { if (!pipeline->IsBound()) { return; } + cmdbuf.PipelineBarrier(READ_BARRIER); cmdbuf.DispatchIndirect(indirect_buffer, indirect_offset); + cmdbuf.PipelineBarrier(WRITE_BARRIER); }); return; } @@ -649,20 +668,13 @@ void RasterizerVulkan::DispatchCompute() { return; } scheduler.RequestOutsideRenderPassOperationContext(); - static constexpr VkMemoryBarrier2 READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT, - }; - scheduler.Record([](vk::CommandBuffer cmdbuf) { cmdbuf.PipelineBarrier(READ_BARRIER); }); scheduler.Record([pipeline, dim](vk::CommandBuffer cmdbuf) { if (!pipeline->IsBound()) { return; } + cmdbuf.PipelineBarrier(READ_BARRIER); cmdbuf.Dispatch(dim[0], dim[1], dim[2]); + cmdbuf.PipelineBarrier(WRITE_BARRIER); }); // Log compute dispatch From d2d2220b24491a3e16d3a80ea647a83303741f1a Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Fri, 25 Sep 2026 18:05:12 -0400 Subject: [PATCH 63/69] Address some regressions with TFB + SSBO's due to WFI removal --- src/video_core/buffer_cache/buffer_base.h | 20 ++++++++++++ src/video_core/buffer_cache/buffer_cache.h | 31 +++++++++++++++++++ .../buffer_cache/buffer_cache_base.h | 15 +++++++++ .../renderer_vulkan/vk_buffer_cache.cpp | 12 +++++++ .../renderer_vulkan/vk_buffer_cache.h | 6 ++++ .../renderer_vulkan/vk_graphics_pipeline.cpp | 4 +++ .../renderer_vulkan/vk_rasterizer.cpp | 1 + .../renderer_vulkan/vk_scheduler.cpp | 28 ++++++++--------- src/video_core/renderer_vulkan/vk_scheduler.h | 23 ++++++++++++++ 9 files changed, 125 insertions(+), 15 deletions(-) diff --git a/src/video_core/buffer_cache/buffer_base.h b/src/video_core/buffer_cache/buffer_base.h index 6b2a37ea8c..b8d0d1896f 100644 --- a/src/video_core/buffer_cache/buffer_base.h +++ b/src/video_core/buffer_cache/buffer_base.h @@ -137,6 +137,22 @@ public: content_serial = ++next_content_serial; } + [[nodiscard]] bool HasDrawHazard(u64 pass, u64 wfi, DAddr addr, u64 size) const noexcept { + return draw_write_pass == pass && draw_write_wfi < wfi && addr < draw_write_end && + addr + size > draw_write_begin; + } + + void MarkDrawWrite(u64 pass, u64 wfi, DAddr addr, u64 size) noexcept { + if (draw_write_pass != pass) { + draw_write_pass = pass; + draw_write_begin = addr; + draw_write_end = addr + size; + } + draw_write_wfi = wfi; + draw_write_begin = (std::min)(draw_write_begin, addr); + draw_write_end = (std::max)(draw_write_end, addr + size); + } + private: static inline u64 next_content_serial = 0; @@ -147,6 +163,10 @@ private: size_t size_bytes = 0; u64 write_tick = 0; u64 content_serial = ++next_content_serial; + u64 draw_write_pass = 0; + u64 draw_write_wfi = 0; + DAddr draw_write_begin = 0; + DAddr draw_write_end = 0; }; } // namespace VideoCommon diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 89d0003207..d4c8a61eb7 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -366,6 +366,13 @@ void BufferCache

::DisableGraphicsUniformBuffer(size_t stage, u32 index) { template void BufferCache

::UpdateGraphicsBuffers(bool is_indexed) { + if constexpr (!IS_OPENGL) { + draw_writes.clear(); + draw_pass = runtime.RenderPassSerial(); + draw_wfi = runtime.WaitForIdleSerial(); + draw_hazard = false; + recording_draw = true; + } ReclaimInline(); do { channel_state->has_deleted_buffers = false; @@ -373,6 +380,24 @@ void BufferCache

::UpdateGraphicsBuffers(bool is_indexed) { } while (channel_state->has_deleted_buffers); } +template +bool BufferCache

::TakeDrawHazard() noexcept { + return std::exchange(draw_hazard, false); +} + +template +void BufferCache

::CommitDrawWrites() { + recording_draw = false; + if (draw_writes.empty()) { + return; + } + const u64 pass = runtime.RenderPassSerial(); + for (const DrawWrite& write : draw_writes) { + slot_buffers[write.buffer_id].MarkDrawWrite(pass, draw_wfi, write.device_addr, write.size); + } + runtime.MarkRenderPassWrites(); +} + template void BufferCache

::UpdateComputeBuffers() { ReclaimInline(); @@ -1550,6 +1575,9 @@ void BufferCache

::MarkWrittenBuffer(BufferId buffer_id, DAddr device_addr, u3 Buffer& buffer = slot_buffers[buffer_id]; buffer.setWriteTick(runtime.CurrentTick()); buffer.MarkContentModified(); + if (recording_draw) { + draw_writes.push_back({buffer_id, device_addr, size}); + } } memory_tracker.MarkRegionAsGpuModified(device_addr, size); gpu_modified_ranges.Add(device_addr, size); @@ -1770,6 +1798,9 @@ void BufferCache

::TouchBuffer(Buffer& buffer, BufferId buffer_id) noexcept { template bool BufferCache

::SynchronizeBuffer(Buffer& buffer, DAddr device_addr, u32 size) { + if constexpr (!IS_OPENGL) { + draw_hazard |= buffer.HasDrawHazard(draw_pass, draw_wfi, device_addr, size); + } upload_copies.clear(); u64 total_size_bytes = 0; u64 largest_copy = 0; diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h index 80a0b2c7eb..6730291831 100644 --- a/src/video_core/buffer_cache/buffer_cache_base.h +++ b/src/video_core/buffer_cache/buffer_cache_base.h @@ -260,6 +260,10 @@ public: void UpdateComputeBuffers(); + [[nodiscard]] bool TakeDrawHazard() noexcept; + + void CommitDrawWrites(); + void BindHostGeometryBuffers(bool is_indexed); void BindHostStageBuffers(size_t stage); @@ -520,6 +524,17 @@ private: boost::container::small_vector upload_copies; + struct DrawWrite { + BufferId buffer_id; + DAddr device_addr; + u32 size; + }; + boost::container::small_vector draw_writes; + u64 draw_pass = 0; + u64 draw_wfi = 0; + bool draw_hazard = false; + bool recording_draw = false; + MemoryTracker memory_tracker; Common::RangeSet uncommitted_gpu_modified_ranges; Common::RangeSet gpu_modified_ranges; diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index d3d3d7a437..57494d9649 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -440,6 +440,18 @@ u64 BufferCacheRuntime::CurrentTick() { return scheduler.GetMasterSemaphore().CurrentTick(); } +u64 BufferCacheRuntime::RenderPassSerial() const noexcept { + return scheduler.ActiveRenderPassSerial(); +} + +u64 BufferCacheRuntime::WaitForIdleSerial() const noexcept { + return scheduler.WaitForIdleSerial(); +} + +void BufferCacheRuntime::MarkRenderPassWrites() noexcept { + scheduler.MarkRenderPassWrites(); +} + u64 BufferCacheRuntime::KnownGpuTick() { return scheduler.GetMasterSemaphore().KnownGpuTick(); } diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h index f83696f0de..1a8629aee8 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.h +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h @@ -112,6 +112,12 @@ public: u64 CurrentTick(); + u64 RenderPassSerial() const noexcept; + + u64 WaitForIdleSerial() const noexcept; + + void MarkRenderPassWrites() noexcept; + u64 KnownGpuTick(); void Wait(u64 buffer_tick); diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index 101fab301e..767099eea2 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -595,7 +595,11 @@ bool GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling, } } + if (buffer_cache.TakeDrawHazard()) { + scheduler.RequestOutsideRenderPassOperationContext(); + } scheduler.RequestRenderpass(texture_cache.GetFramebuffer()); + buffer_cache.CommitDrawWrites(); if (!is_built.load(std::memory_order::relaxed)) { // Wait for the pipeline to be built scheduler.Record([this](vk::CommandBuffer) { diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 4ba8c2abbc..2562ff1ad3 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -894,6 +894,7 @@ void RasterizerVulkan::FlushAndInvalidateRegion(DAddr addr, u64 size, void RasterizerVulkan::WaitForIdle() { query_cache.NotifyWFI(); + scheduler.NotifyWaitForIdle(); fence_manager.SignalOrdering(); } diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 33b03ea8f2..4cde9c8651 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -98,6 +98,7 @@ void Scheduler::BeginRenderPassImpl(const Framebuffer* framebuffer, VkRenderPass state.renderpass = renderpass; state.framebuffer = framebuffer_handle; state.render_area = render_area; + ++renderpass_serial; if (GPU::Logging::IsActive() && Settings::values.gpu_log_vulkan_calls.GetValue()) { const std::string render_pass_info = @@ -417,8 +418,17 @@ void Scheduler::EndRenderPass() Record([num_images = num_renderpass_images, images = renderpass_images, ranges = renderpass_image_ranges, - has_transform_feedback = device.IsExtTransformFeedbackSupported()]( + num_memory_barriers = + static_cast(std::exchange(renderpass_writes, false))]( vk::CommandBuffer cmdbuf) { + static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_ALL_GRAPHICS_BIT, + .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, + }; std::array barriers; for (size_t i = 0; i < num_images; ++i) { const VkImageSubresourceRange& range = ranges[i]; @@ -459,20 +469,8 @@ void Scheduler::EndRenderPass() }; } cmdbuf.EndRenderPass(); - cmdbuf.PipelineBarrier(0, {}, {}, vk::Span(barriers.data(), num_images)); - if (has_transform_feedback) { - static constexpr VkMemoryBarrier2 XFB_OUTPUT_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT, - .srcAccessMask = VK_ACCESS_2_TRANSFORM_FEEDBACK_WRITE_BIT_EXT, - .dstStageMask = VK_PIPELINE_STAGE_2_VERTEX_INPUT_BIT - | VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .dstAccessMask = VK_ACCESS_2_VERTEX_ATTRIBUTE_READ_BIT - | VK_ACCESS_2_TRANSFER_READ_BIT, - }; - cmdbuf.PipelineBarrier(XFB_OUTPUT_BARRIER); - } + cmdbuf.PipelineBarrier(0, vk::Span(&WRITE_BARRIER, num_memory_barriers), {}, + vk::Span(barriers.data(), num_images)); }); state.renderpass = VkRenderPass{}; diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index e82b0414fc..587ed27f5b 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -9,6 +9,7 @@ #include #include #include +#include #include #include #include @@ -77,6 +78,25 @@ public: return state.renderpass != VK_NULL_HANDLE; } + u64 ActiveRenderPassSerial() const noexcept { + if (state.renderpass) { + return renderpass_serial; + } + return (std::numeric_limits::max)(); + } + + u64 WaitForIdleSerial() const noexcept { + return wfi_serial; + } + + void NotifyWaitForIdle() noexcept { + ++wfi_serial; + } + + void MarkRenderPassWrites() noexcept { + renderpass_writes = true; + } + /// Update the pipeline to the current execution context. bool UpdateGraphicsPipeline(GraphicsPipeline* pipeline); @@ -311,6 +331,9 @@ private: State state; + u64 renderpass_serial = 0; + u64 wfi_serial = 0; + bool renderpass_writes = false; u32 num_renderpass_images = 0; std::array renderpass_images{}; std::array renderpass_image_ranges{}; From 63ce0a05273d93f2bd84c568a5ad991519d4c173 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Fri, 25 Sep 2026 18:20:25 -0400 Subject: [PATCH 64/69] Fix build --- src/video_core/buffer_cache/buffer_cache.h | 19 +++++++++++-------- 1 file changed, 11 insertions(+), 8 deletions(-) diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index d4c8a61eb7..971f40922d 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -387,15 +387,18 @@ bool BufferCache

::TakeDrawHazard() noexcept { template void BufferCache

::CommitDrawWrites() { - recording_draw = false; - if (draw_writes.empty()) { - return; + if constexpr (!IS_OPENGL) { + recording_draw = false; + if (draw_writes.empty()) { + return; + } + const u64 pass = runtime.RenderPassSerial(); + for (const DrawWrite& write : draw_writes) { + slot_buffers[write.buffer_id].MarkDrawWrite(pass, draw_wfi, write.device_addr, + write.size); + } + runtime.MarkRenderPassWrites(); } - const u64 pass = runtime.RenderPassSerial(); - for (const DrawWrite& write : draw_writes) { - slot_buffers[write.buffer_id].MarkDrawWrite(pass, draw_wfi, write.device_addr, write.size); - } - runtime.MarkRenderPassWrites(); } template From ccbaa75e5eb55480d4bebf7fb98c33d1cb69c489 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Fri, 25 Sep 2026 19:03:42 -0400 Subject: [PATCH 65/69] Adjust cuts on depth draws --- src/video_core/renderer_vulkan/vk_rasterizer.cpp | 3 +++ src/video_core/renderer_vulkan/vk_scheduler.cpp | 2 ++ src/video_core/renderer_vulkan/vk_scheduler.h | 9 +++++++++ src/video_core/renderer_vulkan/vk_texture_cache.cpp | 4 +++- 4 files changed, 17 insertions(+), 1 deletion(-) diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 2562ff1ad3..c77d032f94 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -264,6 +264,8 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, bool skip_empty, Func&& draw pipeline->SetEngine(maxwell3d, gpu_memory); if (!pipeline->Configure(is_indexed)) return; + scheduler.MarkDepthWrites(maxwell3d->regs.depth_test_enable != 0 && + maxwell3d->regs.depth_write_enabled != 0); UpdateDynamicStates(pipeline->HasDynamicVertexInput()); @@ -583,6 +585,7 @@ void RasterizerVulkan::Clear(u32 layer_count) { if (aspect_flags == 0) { return; } + scheduler.MarkDepthWrites(!can_defer_clear && (aspect_flags & VK_IMAGE_ASPECT_DEPTH_BIT) != 0); if (use_stencil && framebuffer->HasAspectStencilBit() && regs.stencil_front_mask != 0xFF && regs.stencil_front_mask != 0) { diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 4cde9c8651..457221b2b1 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -99,6 +99,7 @@ void Scheduler::BeginRenderPassImpl(const Framebuffer* framebuffer, VkRenderPass state.framebuffer = framebuffer_handle; state.render_area = render_area; ++renderpass_serial; + renderpass_depth_writes = false; if (GPU::Logging::IsActive() && Settings::values.gpu_log_vulkan_calls.GetValue()) { const std::string render_pass_info = @@ -161,6 +162,7 @@ void Scheduler::RealizeDeferredClear() { dc.color_clear_mask, dc.depth_stencil, color_discard_mask, depth_stencil_discard); EndRenderPass(); BeginRenderPassImpl(dc.framebuffer, renderpass, clear_values.data(), count); + renderpass_depth_writes = dc.depth_stencil; } bool Scheduler::DeferColorClear(const Framebuffer* framebuffer, u32 rt_slot, diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index 587ed27f5b..c78faeecb5 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -97,6 +97,14 @@ public: renderpass_writes = true; } + void MarkDepthWrites(bool writes) noexcept { + renderpass_depth_writes |= writes; + } + + bool HasDepthWrites() const noexcept { + return renderpass_depth_writes; + } + /// Update the pipeline to the current execution context. bool UpdateGraphicsPipeline(GraphicsPipeline* pipeline); @@ -334,6 +342,7 @@ private: u64 renderpass_serial = 0; u64 wfi_serial = 0; bool renderpass_writes = false; + bool renderpass_depth_writes = false; u32 num_renderpass_images = 0; std::array renderpass_images{}; std::array renderpass_image_ranges{}; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 71eb640975..9882e1d34c 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -1210,7 +1210,9 @@ void TextureCacheRuntime::EraseResolveShadow(VkImage msaa_image) { } void TextureCacheRuntime::BarrierFeedbackLoop() { - scheduler.RequestOutsideRenderPassOperationContext(); + if (scheduler.HasDepthWrites()) { + scheduler.RequestOutsideRenderPassOperationContext(); + } } void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src, From 5dc91d7e255bc94463dab53a3ba36b70b6e6ba45 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Fri, 25 Sep 2026 23:48:21 -0400 Subject: [PATCH 66/69] Another take on the qcom/turnip missing handling for geometry stages --- .../backend/spirv/emit_spirv.cpp | 33 ++++- .../backend/spirv/emit_spirv_warp.cpp | 127 +++++++++++++----- .../backend/spirv/spirv_emit_context.h | 1 + .../ir_opt/constant_propagation_pass.cpp | 46 +++++-- src/shader_recompiler/profile.h | 3 + .../renderer_vulkan/vk_pipeline_cache.cpp | 15 ++- src/video_core/vulkan_common/vulkan_device.h | 5 + 7 files changed, 178 insertions(+), 52 deletions(-) diff --git a/src/shader_recompiler/backend/spirv/emit_spirv.cpp b/src/shader_recompiler/backend/spirv/emit_spirv.cpp index c221888c09..a903480c78 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv.cpp @@ -4,6 +4,7 @@ // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later +#include #include #include #include @@ -439,10 +440,13 @@ void SetupCapabilities(const Profile& profile, const Info& info, EmitContext& ct ctx.AddCapability(spv::Capability::DrawParameters); } if ((info.uses_subgroup_vote || info.uses_subgroup_invocation_id || - info.uses_subgroup_shuffles) && + info.uses_subgroup_shuffles || info.uses_subgroup_mask) && profile.support_vote && profile.SupportsSubgroupStage(ctx.stage)) { ctx.AddCapability(spv::Capability::GroupNonUniformBallot); ctx.AddCapability(spv::Capability::GroupNonUniformShuffle); + if (info.uses_subgroup_shuffles && profile.support_shuffle_relative) { + ctx.AddCapability(spv::Capability::GroupNonUniformShuffleRelative); + } if (!profile.warp_size_potentially_larger_than_guest) { // vote ops are only used when not taking the long path ctx.AddCapability(spv::Capability::GroupNonUniformVote); @@ -521,6 +525,29 @@ void PatchPhiNodes(IR::Program& program, EmitContext& ctx) { return { ctx.Def(phi->Arg(phi_arg)), parent }; }); } + +void RewriteOpcodes(std::vector& code, std::span> rewrites) { + if (rewrites.empty()) { + return; + } + size_t offset = 5; + while (offset + 2 < code.size()) { + const u32 word_count{code[offset] >> 16}; + const auto opcode{static_cast(code[offset] & 0xFFFFu)}; + if (word_count == 0) { + return; + } + if (opcode == spv::Op::OpGroupNonUniformShuffleXor || + opcode == spv::Op::OpGroupNonUniformQuadBroadcast) { + const auto it{std::ranges::find(rewrites, code[offset + 2], + &std::pair::first)}; + if (it != rewrites.end()) { + code[offset] = (code[offset] & 0xFFFF0000u) | static_cast(it->second); + } + } + offset += word_count; + } +} } // Anonymous namespace std::vector EmitSPIRV(const Profile& profile, const RuntimeInfo& runtime_info, IR::Program& program, Bindings& bindings) { @@ -535,7 +562,9 @@ std::vector EmitSPIRV(const Profile& profile, const RuntimeInfo& runtime_in SetupCapabilities(profile, program.info, ctx); SetupTransformFeedbackCapabilities(ctx, main); PatchPhiNodes(program, ctx); - return ctx.Assemble(); + std::vector code{ctx.Assemble()}; + RewriteOpcodes(code, ctx.opcode_rewrites); + return code; } Id EmitPhi(EmitContext& ctx, IR::Inst* inst) { diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp index ffd5d9c29f..8d83509961 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp @@ -77,20 +77,57 @@ Id GetMaxThreadId(EmitContext& ctx, Id thread_id, Id clamp, Id segmentation_mask return ComputeMaxThreadId(ctx, min_thread_id, clamp, not_seg_mask); } -Id SelectValue(EmitContext& ctx, Id in_range, Id value, Id src_thread_id) { - if (!StageSupportsSubgroups(ctx)) { - return value; +Id HostThreadId(EmitContext& ctx, Id thread_id) { + if (!ctx.profile.warp_size_potentially_larger_than_guest) { + return thread_id; } - return ctx.OpSelect( - ctx.U32[1], in_range, - ctx.OpGroupNonUniformShuffle(ctx.U32[1], SubgroupScope(ctx), value, src_thread_id), value); -} - -Id AddPartitionBase(EmitContext& ctx, Id thread_id) { const Id partition_idx{ctx.OpShiftRightLogical(ctx.U32[1], GetThreadId(ctx), ctx.Const(5u))}; const Id partition_base{ctx.OpShiftLeftLogical(ctx.U32[1], partition_idx, ctx.Const(5u))}; return ctx.OpIAdd(ctx.U32[1], thread_id, partition_base); } + +Id GuestLane(EmitContext& ctx, Id index) { + return ctx.OpBitwiseAnd(ctx.U32[1], index, ctx.Const(31U)); +} + +Id ShuffleAbsolute(EmitContext& ctx, Id value, Id src_thread_id) { + if (!ctx.profile.has_broken_spirv_subgroup_shuffle) { + return ctx.OpGroupNonUniformShuffle(ctx.U32[1], SubgroupScope(ctx), value, src_thread_id); + } + Id result{ctx.u32_zero_value}; + for (u32 lane = 0; lane < ctx.profile.max_subgroup_size; ++lane) { + const Id read{ + ctx.OpGroupNonUniformBroadcast(ctx.U32[1], SubgroupScope(ctx), value, ctx.Const(lane))}; + const Id matches{ctx.OpIEqual(ctx.U1, src_thread_id, ctx.Const(lane))}; + result = ctx.OpSelect(ctx.U32[1], matches, read, result); + } + return result; +} + +Id ShuffleRelative(EmitContext& ctx, Id value, Id delta, Id src_thread_id, spv::Op op) { + if (!ctx.profile.support_shuffle_relative) { + return ShuffleAbsolute(ctx, value, HostThreadId(ctx, src_thread_id)); + } + const Id result{ctx.OpGroupNonUniformShuffleXor(ctx.U32[1], SubgroupScope(ctx), value, delta)}; + ctx.opcode_rewrites.emplace_back(result.value, op); + return result; +} + +Id BroadcastLane(EmitContext& ctx, Id value, u32 lane) { + Id result{ + ctx.OpGroupNonUniformBroadcast(ctx.U32[1], SubgroupScope(ctx), value, ctx.Const(lane))}; + if (!ctx.profile.warp_size_potentially_larger_than_guest) { + return result; + } + const Id partition_idx{ctx.OpShiftRightLogical(ctx.U32[1], GetThreadId(ctx), ctx.Const(5u))}; + for (u32 base = 32; base < ctx.profile.max_subgroup_size; base += 32) { + const Id read{ctx.OpGroupNonUniformBroadcast(ctx.U32[1], SubgroupScope(ctx), value, + ctx.Const(base + lane))}; + const Id matches{ctx.OpIEqual(ctx.U1, partition_idx, ctx.Const(base >> 5))}; + result = ctx.OpSelect(ctx.U32[1], matches, read, result); + } + return result; +} } // Anonymous namespace Id EmitLaneId(EmitContext& ctx) { @@ -203,61 +240,75 @@ Id EmitShuffleIndex(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id cla const Id min_thread_id{ComputeMinThreadId(ctx, thread_id, segmentation_mask)}; const Id max_thread_id{ComputeMaxThreadId(ctx, min_thread_id, clamp, not_seg_mask)}; - const Id lhs{ctx.OpBitwiseAnd(ctx.U32[1], index, not_seg_mask)}; - Id src_thread_id{ctx.OpBitwiseOr(ctx.U32[1], lhs, min_thread_id)}; + const Id lhs{ctx.OpBitwiseAnd(ctx.U32[1], GuestLane(ctx, index), not_seg_mask)}; + const Id src_thread_id{ctx.OpBitwiseOr(ctx.U32[1], lhs, min_thread_id)}; const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)}; - if (ctx.profile.warp_size_potentially_larger_than_guest) { - src_thread_id = AddPartitionBase(ctx, src_thread_id); - } - SetInBoundsFlag(inst, in_range); - return SelectValue(ctx, in_range, value, src_thread_id); + if (!StageSupportsSubgroups(ctx)) { + return value; + } + const IR::Value lane{inst->Arg(1).Resolve()}; + const IR::Value segment{inst->Arg(3).Resolve()}; + if (lane.IsImmediate() && segment.IsImmediate() && segment.U32() == 0) { + return ctx.OpSelect(ctx.U32[1], in_range, BroadcastLane(ctx, value, lane.U32() & 31), + value); + } + const Id shuffled{ShuffleAbsolute(ctx, value, HostThreadId(ctx, src_thread_id))}; + return ctx.OpSelect(ctx.U32[1], in_range, shuffled, value); } Id EmitShuffleUp(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp, Id segmentation_mask) { + if (!StageSupportsSubgroups(ctx)) { + SetInBoundsFlag(inst, ctx.false_value); + return value; + } + const Id delta{GuestLane(ctx, index)}; const Id thread_id{EmitLaneId(ctx)}; const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)}; - Id src_thread_id{ctx.OpISub(ctx.U32[1], thread_id, index)}; + const Id src_thread_id{ctx.OpISub(ctx.U32[1], thread_id, delta)}; const Id in_range{ctx.OpSGreaterThanEqual(ctx.U1, src_thread_id, max_thread_id)}; - if (ctx.profile.warp_size_potentially_larger_than_guest) { - src_thread_id = AddPartitionBase(ctx, src_thread_id); - } - SetInBoundsFlag(inst, in_range); - return SelectValue(ctx, in_range, value, src_thread_id); + const Id shuffled{ShuffleRelative(ctx, value, delta, src_thread_id, + spv::Op::OpGroupNonUniformShuffleUp)}; + return ctx.OpSelect(ctx.U32[1], in_range, shuffled, value); } Id EmitShuffleDown(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp, Id segmentation_mask) { + if (!StageSupportsSubgroups(ctx)) { + SetInBoundsFlag(inst, ctx.false_value); + return value; + } + const Id delta{GuestLane(ctx, index)}; const Id thread_id{EmitLaneId(ctx)}; const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)}; - Id src_thread_id{ctx.OpIAdd(ctx.U32[1], thread_id, index)}; + const Id src_thread_id{ctx.OpIAdd(ctx.U32[1], thread_id, delta)}; const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)}; - if (ctx.profile.warp_size_potentially_larger_than_guest) { - src_thread_id = AddPartitionBase(ctx, src_thread_id); - } - SetInBoundsFlag(inst, in_range); - return SelectValue(ctx, in_range, value, src_thread_id); + const Id shuffled{ShuffleRelative(ctx, value, delta, src_thread_id, + spv::Op::OpGroupNonUniformShuffleDown)}; + return ctx.OpSelect(ctx.U32[1], in_range, shuffled, value); } Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp, Id segmentation_mask) { + if (!StageSupportsSubgroups(ctx)) { + SetInBoundsFlag(inst, ctx.false_value); + return value; + } + const Id mask{GuestLane(ctx, index)}; const Id thread_id{EmitLaneId(ctx)}; const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)}; - Id src_thread_id{ctx.OpBitwiseXor(ctx.U32[1], thread_id, index)}; + const Id src_thread_id{ctx.OpBitwiseXor(ctx.U32[1], thread_id, mask)}; const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)}; - if (ctx.profile.warp_size_potentially_larger_than_guest) { - src_thread_id = AddPartitionBase(ctx, src_thread_id); - } - SetInBoundsFlag(inst, in_range); - return SelectValue(ctx, in_range, value, src_thread_id); + const Id shuffled{ctx.OpGroupNonUniformShuffleXor(ctx.U32[1], SubgroupScope(ctx), value, mask)}; + return ctx.OpSelect(ctx.U32[1], in_range, shuffled, value); } Id EmitQuadBroadcast(EmitContext& ctx, Id value, Id lane) { @@ -267,10 +318,16 @@ Id EmitQuadBroadcast(EmitContext& ctx, Id value, Id lane) { const Id base{ctx.OpBitwiseAnd(ctx.U32[1], GetThreadId(ctx), ctx.Const(~3u))}; const Id local_lane{ctx.OpBitwiseAnd(ctx.U32[1], lane, ctx.Const(3u))}; const Id src_thread_id{ctx.OpBitwiseOr(ctx.U32[1], base, local_lane)}; - return ctx.OpGroupNonUniformShuffle(ctx.U32[1], SubgroupScope(ctx), value, src_thread_id); + return ShuffleAbsolute(ctx, value, src_thread_id); } Id EmitQuadSwap(EmitContext& ctx, Id value, Id direction) { + if (ctx.profile.support_quad_shuffles) { + const Id result{ + ctx.OpGroupNonUniformQuadBroadcast(ctx.U32[1], SubgroupScope(ctx), value, direction)}; + ctx.opcode_rewrites.emplace_back(result.value, spv::Op::OpGroupNonUniformQuadSwap); + return result; + } const Id xor_mask{ctx.OpIAdd(ctx.U32[1], direction, ctx.Const(1u))}; return ctx.OpGroupNonUniformShuffleXor(ctx.U32[1], SubgroupScope(ctx), value, xor_mask); } diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.h b/src/shader_recompiler/backend/spirv/spirv_emit_context.h index 7edc104708..a5821e9e93 100644 --- a/src/shader_recompiler/backend/spirv/spirv_emit_context.h +++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.h @@ -361,6 +361,7 @@ public: Id frag_depth{}; std::vector interfaces; + std::vector> opcode_rewrites; Id load_const_func_u8{}; Id load_const_func_u16{}; diff --git a/src/shader_recompiler/ir_opt/constant_propagation_pass.cpp b/src/shader_recompiler/ir_opt/constant_propagation_pass.cpp index 57c08588fb..9424de188d 100644 --- a/src/shader_recompiler/ir_opt/constant_propagation_pass.cpp +++ b/src/shader_recompiler/ir_opt/constant_propagation_pass.cpp @@ -651,6 +651,29 @@ IR::Value GetThroughCast(IR::Value value, IR::Opcode expected_cast) { return value; } +u32 QuadButterflyMask(const IR::Inst& inst) { + if (inst.GetOpcode() == IR::Opcode::QuadSwap) { + const IR::Value direction{inst.Arg(1)}; + if (!direction.IsImmediate()) { + return 0; + } + return direction.U32() + 1; + } + if (inst.GetOpcode() != IR::Opcode::ShuffleButterfly) { + return 0; + } + const IR::Value index{inst.Arg(1)}; + const IR::Value clamp{inst.Arg(2)}; + const IR::Value segmentation_mask{inst.Arg(3)}; + if (!index.IsImmediate() || !clamp.IsImmediate() || !segmentation_mask.IsImmediate()) { + return 0; + } + if (clamp.U32() != 3 || segmentation_mask.U32() != 28) { + return 0; + } + return index.U32(); +} + void FoldFSwizzleAdd(IR::Block& block, IR::Inst& inst) { const IR::Value swizzle{inst.Arg(2)}; if (!swizzle.IsImmediate()) { @@ -666,7 +689,8 @@ void FoldFSwizzleAdd(IR::Block& block, IR::Inst& inst) { return; } IR::Inst* const inst2{value_1.InstRecursive()}; - if (inst2->GetOpcode() != IR::Opcode::ShuffleButterfly) { + const u32 lane_mask{QuadButterflyMask(*inst2)}; + if (lane_mask == 0) { return; } const IR::Value value_3{GetThroughCast(inst2->Arg(0).Resolve(), IR::Opcode::BitCastU32F32)}; @@ -678,24 +702,15 @@ void FoldFSwizzleAdd(IR::Block& block, IR::Inst& inst) { return; } } - const IR::Value index{inst2->Arg(1)}; - const IR::Value clamp{inst2->Arg(2)}; - const IR::Value segmentation_mask{inst2->Arg(3)}; - if (!index.IsImmediate() || !clamp.IsImmediate() || !segmentation_mask.IsImmediate()) { - return; - } - if (clamp.U32() != 3 || segmentation_mask.U32() != 28) { - return; - } if (swizzle_value == 0x99) { // DPdxFine - if (index.U32() == 1) { + if (lane_mask == 1) { IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; inst.ReplaceUsesWith(ir.DPdxFine(IR::F32{inst.Arg(1)})); } } else if (swizzle_value == 0xA5) { // DPdyFine - if (index.U32() == 2) { + if (lane_mask == 2) { IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; inst.ReplaceUsesWith(ir.DPdyFine(IR::F32{inst.Arg(1)})); } @@ -709,6 +724,13 @@ bool FindGradient3DDerivatives(std::array& results, IR::Value coor const auto check_through_shuffle = [](IR::Value input, IR::Value& result) { const IR::Value value_1{GetThroughCast(input.Resolve(), IR::Opcode::BitCastF32U32)}; IR::Inst* const inst2{value_1.InstRecursive()}; + if (inst2->GetOpcode() == IR::Opcode::QuadBroadcast) { + if (!inst2->Arg(1).Resolve().IsImmediate()) { + return false; + } + result = GetThroughCast(inst2->Arg(0).Resolve(), IR::Opcode::BitCastU32F32); + return true; + } if (inst2->GetOpcode() != IR::Opcode::ShuffleIndex) { return false; } diff --git a/src/shader_recompiler/profile.h b/src/shader_recompiler/profile.h index d2ac480ab8..8e9c83f888 100644 --- a/src/shader_recompiler/profile.h +++ b/src/shader_recompiler/profile.h @@ -40,6 +40,7 @@ struct Profile { bool support_shader_quad_control{}; bool support_quad_shuffles{}; bool support_vote{}; + bool support_shuffle_relative{}; u32 supported_subgroup_stages{0x7F}; bool support_viewport_index_layer_non_geometry{}; bool support_viewport_mask{}; @@ -102,6 +103,8 @@ struct Profile { bool ignore_nan_fp_comparisons{}; /// Some drivers have broken support for OpVectorExtractDynamic on subgroup mask inputs bool has_broken_spirv_subgroup_mask_vector_extract_dynamic{}; + bool has_broken_spirv_subgroup_shuffle{}; + u32 max_subgroup_size{}; u32 gl_max_compute_smem_size{}; diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index 642b9fa223..7c530e1cfb 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -362,7 +362,7 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, const auto subgroup_stage_bit{[subgroup_stages](VkShaderStageFlags flag, Shader::Stage stage) { return (subgroup_stages & flag) != 0 ? (1u << static_cast(stage)) : 0u; }}; - const u32 supported_subgroup_stages{ + u32 supported_subgroup_stages{ subgroup_stage_bit(VK_SHADER_STAGE_VERTEX_BIT, Shader::Stage::VertexA) | subgroup_stage_bit(VK_SHADER_STAGE_VERTEX_BIT, Shader::Stage::VertexB) | subgroup_stage_bit(VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT, @@ -372,6 +372,9 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, subgroup_stage_bit(VK_SHADER_STAGE_GEOMETRY_BIT, Shader::Stage::Geometry) | subgroup_stage_bit(VK_SHADER_STAGE_FRAGMENT_BIT, Shader::Stage::Fragment) | subgroup_stage_bit(VK_SHADER_STAGE_COMPUTE_BIT, Shader::Stage::Compute)}; + if (driver_id == VK_DRIVER_ID_MESA_TURNIP) { + supported_subgroup_stages &= ~(1u << static_cast(Shader::Stage::Geometry)); + } profile = Shader::Profile{ .supported_spirv = device.SupportedSpirvVersion(), .unified_descriptor_binding = true, @@ -409,6 +412,8 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, .support_shader_quad_control = device.IsKhrShaderQuadControlSupported(), .support_quad_shuffles = device.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_QUAD_BIT), .support_vote = device.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_VOTE_BIT), + .support_shuffle_relative = + device.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT), .supported_subgroup_stages = supported_subgroup_stages, .support_viewport_index_layer_non_geometry = device.IsExtShaderViewportIndexLayerSupported(), @@ -444,13 +449,17 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, driver_id == VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA, .has_broken_spirv_clamp = driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS, - .has_broken_spirv_position_input = driver_id == false, + .has_broken_spirv_position_input = driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY, .has_broken_unsigned_image_offsets = false, .has_broken_signed_operations = false, .has_broken_fp16_float_controls = driver_id == VK_DRIVER_ID_NVIDIA_PROPRIETARY, .has_broken_fp32_denorm_flush = driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY, .ignore_nan_fp_comparisons = false, - .has_broken_spirv_subgroup_mask_vector_extract_dynamic = false, + .has_broken_spirv_subgroup_mask_vector_extract_dynamic = + driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY && + device.GetDriverVersion() < VK_MAKE_VERSION(512, 672, 0), + .has_broken_spirv_subgroup_shuffle = driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY, + .max_subgroup_size = device.GetMaxSubgroupSize(), .has_broken_robust = device.IsNvidia() && device.GetNvidiaArch() <= NvidiaArchitecture::Arch_Pascal, .min_ssbo_alignment = device.GetStorageBufferAlignment(), diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index fba45d2385..3e3726651d 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -485,6 +485,11 @@ FN_MAX_LIMIT_LIST return properties.subgroup_properties.supportedStages; } + u32 GetMaxSubgroupSize() const { + return (std::max)(properties.subgroup_properties.subgroupSize, + properties.subgroup_size_control.maxSubgroupSize); + } + /// Returns the maximum number of push descriptors. u32 MaxPushDescriptors() const { return properties.push_descriptor.maxPushDescriptors; From d832f5e156a7108ddb7f6a4b57186146564ab286 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sat, 26 Sep 2026 00:42:58 -0400 Subject: [PATCH 67/69] Re-do geometry compaction pass --- src/shader_recompiler/CMakeLists.txt | 1 + .../frontend/maxwell/translate_program.cpp | 1 + src/shader_recompiler/host_translate_info.h | 1 + .../ir_opt/geometry_compaction_pass.cpp | 129 ++++++++++++++++++ src/shader_recompiler/ir_opt/passes.h | 1 + .../renderer_vulkan/vk_pipeline_cache.cpp | 1 + 6 files changed, 134 insertions(+) create mode 100644 src/shader_recompiler/ir_opt/geometry_compaction_pass.cpp diff --git a/src/shader_recompiler/CMakeLists.txt b/src/shader_recompiler/CMakeLists.txt index 957ed5fb55..cc9ce95d92 100644 --- a/src/shader_recompiler/CMakeLists.txt +++ b/src/shader_recompiler/CMakeLists.txt @@ -172,6 +172,7 @@ add_library(shader_recompiler STATIC ir_opt/constant_propagation_pass.cpp ir_opt/dead_code_elimination_pass.cpp ir_opt/dual_vertex_pass.cpp + ir_opt/geometry_compaction_pass.cpp ir_opt/global_memory_to_storage_buffer_pass.cpp ir_opt/identity_removal_pass.cpp ir_opt/layer_pass.cpp diff --git a/src/shader_recompiler/frontend/maxwell/translate_program.cpp b/src/shader_recompiler/frontend/maxwell/translate_program.cpp index ebc5a825dd..e7d7e5a6b3 100644 --- a/src/shader_recompiler/frontend/maxwell/translate_program.cpp +++ b/src/shader_recompiler/frontend/maxwell/translate_program.cpp @@ -299,6 +299,7 @@ IR::Program TranslateProgram(ObjectPool& inst_pool, ObjectPool + +#include "shader_recompiler/frontend/ir/ir_emitter.h" +#include "shader_recompiler/host_translate_info.h" +#include "shader_recompiler/ir_opt/passes.h" + +namespace Shader::Optimization { +namespace { +constexpr int MAX_BALLOT_DEPTH = 8; + +struct SlotAtomic { + IR::Block* block; + IR::Inst* atomic; + IR::Inst* ballot; +}; + +IR::Inst* FindBallot(const IR::Value& value, int depth) { + if (depth > MAX_BALLOT_DEPTH || value.IsImmediate()) { + return nullptr; + } + IR::Inst* const inst{value.InstRecursive()}; + if (inst->GetOpcode() == IR::Opcode::SubgroupBallot) { + return inst; + } + if (inst->GetOpcode() == IR::Opcode::Phi) { + return nullptr; + } + for (size_t index = 0; index < inst->NumArgs(); ++index) { + if (IR::Inst* const ballot{FindBallot(inst->Arg(index), depth + 1)}) { + return ballot; + } + } + return nullptr; +} + +bool IsShuffle(IR::Opcode opcode) { + switch (opcode) { + case IR::Opcode::ShuffleIndex: + case IR::Opcode::ShuffleUp: + case IR::Opcode::ShuffleDown: + case IR::Opcode::ShuffleButterfly: + return true; + default: + return false; + } +} + +IR::U32 PrimitiveSlot(IR::IREmitter& ir, u32 invocations, const IR::U32& amount) { + IR::U32 key{ir.GetAttributeU32(IR::Attribute::PrimitiveId)}; + if (invocations > 1) { + key = ir.IAdd(ir.IMul(key, ir.Imm32(invocations)), ir.InvocationId()); + } + return ir.IMul(key, amount); +} + +bool MergesAtomic(const IR::Inst& phi, const IR::Block& block, const IR::Inst& atomic) { + for (size_t index = 0; index < phi.NumArgs(); ++index) { + const IR::Value arg{phi.Arg(index)}; + if (phi.PhiBlock(index) == &block && !arg.IsImmediate() && + arg.InstRecursive() == &atomic) { + return true; + } + } + return false; +} + +void RewriteMergedSlots(IR::Block& block, IR::Inst& atomic, u32 invocations) { + for (IR::Block* const successor : block.ImmSuccessors()) { + for (IR::Inst& phi : successor->Instructions()) { + if (phi.GetOpcode() != IR::Opcode::Phi || !MergesAtomic(phi, block, atomic)) { + continue; + } + for (size_t index = 0; index < phi.NumArgs(); ++index) { + const IR::Value arg{phi.Arg(index)}; + if (arg.IsImmediate() || !IsShuffle(arg.InstRecursive()->GetOpcode())) { + continue; + } + IR::IREmitter ir{*phi.PhiBlock(index)}; + const IR::U32 amount{arg.InstRecursive()->Arg(0)}; + phi.SetArg(index, PrimitiveSlot(ir, invocations, amount)); + } + } + } +} + +void RewriteAtomic(IR::Block& block, IR::Inst& atomic, u32 invocations) { + const auto insert_point{IR::Block::InstructionList::s_iterator_to(atomic)}; + IR::IREmitter ir{block, insert_point}; + const IR::U32 amount{atomic.Arg(2)}; + const IR::U32 slot{PrimitiveSlot(ir, invocations, amount)}; + block.PrependNewInst(insert_point, IR::Opcode::StorageAtomicUMax32, + {atomic.Arg(0), atomic.Arg(1), ir.IAdd(slot, amount)}); + atomic.ReplaceUsesWith(slot); +} + +void NeutralizePredicate(IR::Inst& ballot) { + const IR::Value pred{ballot.Arg(0)}; + if (!pred.IsImmediate()) { + pred.InstRecursive()->ReplaceUsesWith(IR::Value{true}); + } +} +} + +void GeometryCompactionPass(IR::Program& program, const HostTranslateInfo& host_info) { + if (program.stage != Stage::Geometry || !host_info.single_lane_geometry_subgroups) { + return; + } + boost::container::small_vector slot_atomics; + for (IR::Block* const block : program.post_order_blocks) { + for (IR::Inst& inst : block->Instructions()) { + if (inst.GetOpcode() != IR::Opcode::StorageAtomicIAdd32) { + continue; + } + if (IR::Inst* const ballot{FindBallot(inst.Arg(2), 0)}) { + slot_atomics.push_back({block, &inst, ballot}); + } + } + } + for (const SlotAtomic& slot_atomic : slot_atomics) { + RewriteMergedSlots(*slot_atomic.block, *slot_atomic.atomic, program.invocations); + RewriteAtomic(*slot_atomic.block, *slot_atomic.atomic, program.invocations); + NeutralizePredicate(*slot_atomic.ballot); + } +} + +} diff --git a/src/shader_recompiler/ir_opt/passes.h b/src/shader_recompiler/ir_opt/passes.h index 1e637cb23c..0c2b44e022 100644 --- a/src/shader_recompiler/ir_opt/passes.h +++ b/src/shader_recompiler/ir_opt/passes.h @@ -16,6 +16,7 @@ void CollectShaderInfoPass(Environment& env, IR::Program& program); void ConditionalBarrierPass(IR::Program& program); void ConstantPropagationPass(Environment& env, IR::Program& program); void DeadCodeEliminationPass(IR::Program& program); +void GeometryCompactionPass(IR::Program& program, const HostTranslateInfo& host_info); void GlobalMemoryToStorageBufferPass(IR::Program& program, const HostTranslateInfo& host_info); void IdentityRemovalPass(IR::Program& program); void LowerFp64ToFp32(IR::Program& program); diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index 7c530e1cfb..d04d36046e 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -488,6 +488,7 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, .support_viewport_index_layer = device.IsExtShaderViewportIndexLayerSupported(), .support_geometry_shader_passthrough = device.IsNvGeometryShaderPassthroughSupported(), .support_conditional_barrier = device.SupportsConditionalBarriers(), + .single_lane_geometry_subgroups = !profile.SupportsSubgroupStage(Shader::Stage::Geometry), }; host_info.ApplyDescriptorLimitPolicy(); From 9d4355064b7dbc8f72eddee2444a412cccef73e3 Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sat, 26 Sep 2026 02:31:24 -0400 Subject: [PATCH 68/69] Another try to make barriers more explicit --- .../renderer_vulkan/vk_buffer_cache.cpp | 66 +++++-------------- .../renderer_vulkan/vk_rasterizer.cpp | 26 ++++---- .../renderer_vulkan/vk_scheduler.cpp | 43 +++++++++--- src/video_core/renderer_vulkan/vk_scheduler.h | 14 ++++ src/video_core/vulkan_common/vulkan_wrapper.h | 25 +++++++ 5 files changed, 103 insertions(+), 71 deletions(-) diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 57494d9649..30baf44e9d 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -36,6 +36,24 @@ struct VertexBindings { u32 count; }; +constexpr VkMemoryBarrier2 READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = vk::PIPELINE_STAGE_BUFFER_INPUTS | VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = vk::ACCESS_TRANSFER, +}; + +constexpr VkMemoryBarrier2 WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_BUFFER_INPUTS, + .dstAccessMask = vk::ACCESS_BUFFER_INPUTS, +}; + VkBufferCopy MakeBufferCopy(const VideoCommon::BufferCopy& copy) { return VkBufferCopy{ .srcOffset = copy.src_offset, @@ -482,22 +500,6 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer, if (dst_buffer == VK_NULL_HANDLE || src_buffer == VK_NULL_HANDLE) { return; } - static constexpr VkMemoryBarrier2 READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, - }; - static constexpr VkMemoryBarrier2 WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, - }; // Measuring a popular game, this number never exceeds the specified size once data is warmed up boost::container::small_vector vk_copies(copies.size()); @@ -523,14 +525,6 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer, } void BufferCacheRuntime::PreCopyBarrier() { - static constexpr VkMemoryBarrier2 READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, - }; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([](vk::CommandBuffer cmdbuf) { cmdbuf.PipelineBarrier(READ_BARRIER); @@ -538,14 +532,6 @@ void BufferCacheRuntime::PreCopyBarrier() { } void BufferCacheRuntime::PostCopyBarrier() { - static constexpr VkMemoryBarrier2 WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, - }; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([](vk::CommandBuffer cmdbuf) { cmdbuf.PipelineBarrier(WRITE_BARRIER); @@ -556,22 +542,6 @@ void BufferCacheRuntime::ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t si if (dest_buffer == VK_NULL_HANDLE) { return; } - static constexpr VkMemoryBarrier2 READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, - }; - static constexpr VkMemoryBarrier2 WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, - }; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([dest_buffer, offset, size, value](vk::CommandBuffer cmdbuf) { diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index c77d032f94..488fe8019c 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -631,19 +631,19 @@ void RasterizerVulkan::DispatchCompute() { static constexpr VkMemoryBarrier2 READ_BARRIER{ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, - .dstStageMask = - VK_PIPELINE_STAGE_2_DRAW_INDIRECT_BIT | VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT, + .srcStageMask = vk::PIPELINE_STAGE_BUFFER_INPUTS, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .dstAccessMask = vk::ACCESS_SHADER_RESOURCES, }; - static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + static constexpr VkMemoryBarrier2 INDIRECT_READ_BARRIER{ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .srcStageMask = vk::PIPELINE_STAGE_BUFFER_INPUTS, .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_ALL_GRAPHICS_BIT | VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = + VK_PIPELINE_STAGE_2_DRAW_INDIRECT_BIT | VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .dstAccessMask = VK_ACCESS_2_INDIRECT_COMMAND_READ_BIT | vk::ACCESS_SHADER_RESOURCES, }; const auto& qmd{kepler_compute->launch_description}; auto indirect_address = kepler_compute->GetIndirectComputeAddress(); @@ -653,15 +653,14 @@ void RasterizerVulkan::DispatchCompute() { const auto post_op = VideoCommon::ObtainBufferOperation::DiscardWrite; const auto [buffer, offset] = buffer_cache.ObtainBuffer(*indirect_address, 12, sync_info, post_op); - scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.RequestComputeDispatchContext(); scheduler.Record([pipeline, indirect_buffer = buffer->Handle(), indirect_offset = offset](vk::CommandBuffer cmdbuf) { if (!pipeline->IsBound()) { return; } - cmdbuf.PipelineBarrier(READ_BARRIER); + cmdbuf.PipelineBarrier(INDIRECT_READ_BARRIER); cmdbuf.DispatchIndirect(indirect_buffer, indirect_offset); - cmdbuf.PipelineBarrier(WRITE_BARRIER); }); return; } @@ -670,14 +669,13 @@ void RasterizerVulkan::DispatchCompute() { if (dim[0] > max_dim[0] || dim[1] > max_dim[1] || dim[2] > max_dim[2]) { return; } - scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.RequestComputeDispatchContext(); scheduler.Record([pipeline, dim](vk::CommandBuffer cmdbuf) { if (!pipeline->IsBound()) { return; } cmdbuf.PipelineBarrier(READ_BARRIER); cmdbuf.Dispatch(dim[0], dim[1], dim[2]); - cmdbuf.PipelineBarrier(WRITE_BARRIER); }); // Log compute dispatch diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 457221b2b1..ed7cc1ed19 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -47,6 +47,12 @@ Scheduler::Scheduler(const Device& device_, StateTracker& state_tracker_) : device{device_}, state_tracker{state_tracker_}, master_semaphore{std::make_unique(device)}, command_pool{std::make_unique(*master_semaphore, device)} { + if (device.IsExtTransformFeedbackSupported()) { + renderpass_write_barrier.srcStageMask |= VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT; + renderpass_write_barrier.srcAccessMask |= VK_ACCESS_2_TRANSFORM_FEEDBACK_WRITE_BIT_EXT; + renderpass_write_barrier.dstStageMask |= VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT; + renderpass_write_barrier.dstAccessMask |= VK_ACCESS_2_TRANSFORM_FEEDBACK_WRITE_BIT_EXT; + } AcquireNewChunk(); AllocateWorkerCommandBuffer(); @@ -95,6 +101,7 @@ void Scheduler::BeginRenderPassImpl(const Framebuffer* framebuffer, VkRenderPass const VkClearValue* clear_values, u32 clear_value_count) { const VkFramebuffer framebuffer_handle = framebuffer->Handle(); const VkExtent2D render_area = framebuffer->RenderArea(); + PublishComputeWrites(); state.renderpass = renderpass; state.framebuffer = framebuffer_handle; state.render_area = render_area; @@ -222,6 +229,31 @@ void Scheduler::RequestRenderpass(const Framebuffer* framebuffer) { void Scheduler::RequestOutsideRenderPassOperationContext() { EndRenderPass(); + PublishComputeWrites(); +} + +void Scheduler::RequestComputeDispatchContext() { + EndRenderPass(); + compute_writes = true; +} + +void Scheduler::PublishComputeWrites() { + if (!std::exchange(compute_writes, false)) { + return; + } + Record([](vk::CommandBuffer cmdbuf) { + static constexpr VkMemoryBarrier2 COMPUTE_WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_BUFFER_INPUTS | vk::PIPELINE_STAGE_ATTACHMENTS | + VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = + vk::ACCESS_BUFFER_INPUTS | vk::ACCESS_ATTACHMENTS | vk::ACCESS_TRANSFER, + }; + cmdbuf.PipelineBarrier(COMPUTE_WRITE_BARRIER); + }); } bool Scheduler::UpdateGraphicsPipeline(GraphicsPipeline* pipeline) { @@ -420,17 +452,10 @@ void Scheduler::EndRenderPass() Record([num_images = num_renderpass_images, images = renderpass_images, ranges = renderpass_image_ranges, + write_barrier = &renderpass_write_barrier, num_memory_barriers = static_cast(std::exchange(renderpass_writes, false))]( vk::CommandBuffer cmdbuf) { - static constexpr VkMemoryBarrier2 WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_ALL_GRAPHICS_BIT, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, - }; std::array barriers; for (size_t i = 0; i < num_images; ++i) { const VkImageSubresourceRange& range = ranges[i]; @@ -471,7 +496,7 @@ void Scheduler::EndRenderPass() }; } cmdbuf.EndRenderPass(); - cmdbuf.PipelineBarrier(0, vk::Span(&WRITE_BARRIER, num_memory_barriers), {}, + cmdbuf.PipelineBarrier(0, vk::Span(write_barrier, num_memory_barriers), {}, vk::Span(barriers.data(), num_images)); }); diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index c78faeecb5..53be37928b 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -73,6 +73,8 @@ public: /// of a renderpass. void RequestOutsideRenderPassOperationContext(); + void RequestComputeDispatchContext(); + /// Returns true when a render pass is currently active in the scheduler state. bool IsRenderPassActive() const { return state.renderpass != VK_NULL_HANDLE; @@ -319,6 +321,8 @@ private: void EndRenderPass(); + void PublishComputeWrites(); + void AcquireNewChunk(); const Device& device; @@ -343,6 +347,16 @@ private: u64 wfi_serial = 0; bool renderpass_writes = false; bool renderpass_depth_writes = false; + bool compute_writes = false; + VkMemoryBarrier2 renderpass_write_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_SHADERS, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_BUFFER_INPUTS | vk::PIPELINE_STAGE_ATTACHMENTS | + VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = vk::ACCESS_BUFFER_INPUTS | vk::ACCESS_ATTACHMENTS | vk::ACCESS_TRANSFER, + }; u32 num_renderpass_images = 0; std::array renderpass_images{}; std::array renderpass_image_ranges{}; diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index 7898886921..72403aff8b 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -158,6 +158,31 @@ inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER_ inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_HOST = VK_PIPELINE_STAGE_2_HOST_BIT; +inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_GRAPHICS_SHADERS = + VK_PIPELINE_STAGE_2_PRE_RASTERIZATION_SHADERS_BIT | VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT; + +inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_BUFFER_INPUTS = + VK_PIPELINE_STAGE_2_DRAW_INDIRECT_BIT | VK_PIPELINE_STAGE_2_VERTEX_INPUT_BIT | + PIPELINE_STAGE_GRAPHICS_SHADERS | VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT; + +inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_ATTACHMENTS = + VK_PIPELINE_STAGE_2_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | + VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT; + +inline constexpr VkAccessFlags2 ACCESS_SHADER_RESOURCES = + VK_ACCESS_2_UNIFORM_READ_BIT | VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT; + +inline constexpr VkAccessFlags2 ACCESS_BUFFER_INPUTS = + VK_ACCESS_2_INDIRECT_COMMAND_READ_BIT | VK_ACCESS_2_INDEX_READ_BIT | + VK_ACCESS_2_VERTEX_ATTRIBUTE_READ_BIT | ACCESS_SHADER_RESOURCES; + +inline constexpr VkAccessFlags2 ACCESS_ATTACHMENTS = + VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; + +inline constexpr VkAccessFlags2 ACCESS_TRANSFER = + VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT; + /// Table holding Vulkan instance function pointers. struct InstanceDispatch { From 49c3a79fa8e6508551141e53b20d373937cbf3cf Mon Sep 17 00:00:00 2001 From: CamilleLaVey Date: Sat, 26 Sep 2026 04:13:18 -0400 Subject: [PATCH 69/69] Keep adjusting memory barriers to be more explicit --- src/video_core/renderer_vulkan/blit_image.cpp | 40 +--- src/video_core/renderer_vulkan/blit_image.h | 4 - .../renderer_vulkan/present/util.cpp | 25 +-- .../renderer_vulkan/vk_buffer_cache.cpp | 56 +++-- .../renderer_vulkan/vk_buffer_cache.h | 3 + .../renderer_vulkan/vk_compute_pass.cpp | 61 +++--- .../renderer_vulkan/vk_present_manager.cpp | 14 +- .../renderer_vulkan/vk_query_cache.cpp | 34 ++- .../renderer_vulkan/vk_rasterizer.cpp | 26 +-- .../renderer_vulkan/vk_scheduler.cpp | 47 ++--- src/video_core/renderer_vulkan/vk_scheduler.h | 12 +- .../renderer_vulkan/vk_texture_cache.cpp | 194 +++++++----------- src/video_core/vulkan_common/vulkan_device.h | 26 +++ src/video_core/vulkan_common/vulkan_wrapper.h | 30 ++- 14 files changed, 246 insertions(+), 326 deletions(-) diff --git a/src/video_core/renderer_vulkan/blit_image.cpp b/src/video_core/renderer_vulkan/blit_image.cpp index b462572a5b..dad11ae613 100644 --- a/src/video_core/renderer_vulkan/blit_image.cpp +++ b/src/video_core/renderer_vulkan/blit_image.cpp @@ -460,16 +460,8 @@ void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view) const VkImageMemoryBarrier2 barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | - VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | - VK_PIPELINE_STAGE_2_TRANSFER_BIT | - VK_PIPELINE_STAGE_2_EARLY_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT, - .srcAccessMask = VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_SHADER_WRITE_BIT | - VK_ACCESS_2_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT, @@ -950,8 +942,8 @@ void BlitImageHelper::CopyMSAAImpl(VkRenderPass renderpass, VkPipeline pipeline, VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = aspect_info.pre_src_stages, - .srcAccessMask = aspect_info.pre_src_access, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = aspect_info.pre_dst_stages, .dstAccessMask = aspect_info.pre_src_dst_access, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -964,8 +956,8 @@ void BlitImageHelper::CopyMSAAImpl(VkRenderPass renderpass, VkPipeline pipeline, VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = aspect_info.pre_src_stages, - .srcAccessMask = aspect_info.pre_src_access, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = aspect_info.pre_dst_stages, .dstAccessMask = aspect_info.pre_dst_dst_access, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1017,8 +1009,8 @@ void BlitImageHelper::CopyMSAAImpl(VkRenderPass renderpass, VkPipeline pipeline, .pNext = nullptr, .srcStageMask = aspect_info.post_src_stages, .srcAccessMask = aspect_info.post_src_access, - .dstStageMask = aspect_info.post_dst_stages, - .dstAccessMask = aspect_info.post_dst_access, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1075,21 +1067,13 @@ void BlitImageHelper::CopyMSAA(RenderPassCache& render_pass_cache, VkImage dst_i .src_view_aspect = VK_IMAGE_ASPECT_COLOR_BIT, .attachment_aspect = VK_IMAGE_ASPECT_COLOR_BIT, .barrier_aspect = VK_IMAGE_ASPECT_COLOR_BIT, - .pre_src_access = VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_2_SHADER_WRITE_BIT | - VK_ACCESS_2_TRANSFER_WRITE_BIT, .pre_src_dst_access = VK_ACCESS_2_SHADER_READ_BIT, .pre_dst_dst_access = VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT, - .pre_src_stages = VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | - VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_2_TRANSFER_BIT, .pre_dst_stages = VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT, .post_src_access = VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT, - .post_dst_access = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_TRANSFER_READ_BIT, .post_src_stages = VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT, - .post_dst_stages = VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | - VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_2_TRANSFER_BIT, }; const VkFormat src_vk_format = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, true, src_format).format; @@ -1584,21 +1568,13 @@ void BlitImageHelper::CopyMSAADepth(RenderPassCache& render_pass_cache, VkImage .src_view_aspect = VK_IMAGE_ASPECT_DEPTH_BIT, .attachment_aspect = attachment_aspect, .barrier_aspect = attachment_aspect, - .pre_src_access = - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, .pre_src_dst_access = VK_ACCESS_2_SHADER_READ_BIT, .pre_dst_dst_access = VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, - .pre_src_stages = VK_PIPELINE_STAGE_2_EARLY_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_2_TRANSFER_BIT, .pre_dst_stages = VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_2_EARLY_FRAGMENT_TESTS_BIT, .post_src_access = VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, - .post_dst_access = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_TRANSFER_READ_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT, .post_src_stages = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT, - .post_dst_stages = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, }; const VkFormat src_vk_format = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, true, src_format).format; diff --git a/src/video_core/renderer_vulkan/blit_image.h b/src/video_core/renderer_vulkan/blit_image.h index 895d0f8de4..bf9bdfdc41 100644 --- a/src/video_core/renderer_vulkan/blit_image.h +++ b/src/video_core/renderer_vulkan/blit_image.h @@ -140,15 +140,11 @@ private: VkImageAspectFlags src_view_aspect; VkImageAspectFlags attachment_aspect; VkImageAspectFlags barrier_aspect; - VkAccessFlags2 pre_src_access; VkAccessFlags2 pre_src_dst_access; VkAccessFlags2 pre_dst_dst_access; - VkPipelineStageFlags2 pre_src_stages; VkPipelineStageFlags2 pre_dst_stages; VkAccessFlags2 post_src_access; - VkAccessFlags2 post_dst_access; VkPipelineStageFlags2 post_src_stages; - VkPipelineStageFlags2 post_dst_stages; }; void BlitImpl(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, diff --git a/src/video_core/renderer_vulkan/present/util.cpp b/src/video_core/renderer_vulkan/present/util.cpp index 80d5e19578..7c9e0aa76d 100644 --- a/src/video_core/renderer_vulkan/present/util.cpp +++ b/src/video_core/renderer_vulkan/present/util.cpp @@ -49,16 +49,13 @@ vk::Image CreateWrappedImage(MemoryAllocator& allocator, VkExtent2D dimensions, void TransitionImageLayout(vk::CommandBuffer& cmdbuf, VkImage image, VkImageLayout target_layout, VkImageLayout source_layout) { - constexpr VkAccessFlags2 flags{VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_SHADER_READ_BIT}; const VkImageMemoryBarrier2 barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .srcAccessMask = flags, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = flags, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = source_layout, .newLayout = target_layout, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -121,8 +118,8 @@ void DownloadColorImage(vk::CommandBuffer& cmdbuf, VkImage image, VkBuffer buffe const VkImageMemoryBarrier2 read_barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -143,8 +140,8 @@ void DownloadColorImage(vk::CommandBuffer& cmdbuf, VkImage image, VkBuffer buffe .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_NONE, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -162,9 +159,9 @@ void DownloadColorImage(vk::CommandBuffer& cmdbuf, VkImage image, VkBuffer buffe .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_HOST, + .dstAccessMask = VK_ACCESS_2_HOST_READ_BIT, }; const VkBufferImageCopy copy{ .bufferOffset = 0, diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 30baf44e9d..7b9287d797 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -36,24 +36,6 @@ struct VertexBindings { u32 count; }; -constexpr VkMemoryBarrier2 READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_BUFFER_INPUTS | VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .dstAccessMask = vk::ACCESS_TRANSFER, -}; - -constexpr VkMemoryBarrier2 WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_BUFFER_INPUTS, - .dstAccessMask = vk::ACCESS_BUFFER_INPUTS, -}; - VkBufferCopy MakeBufferCopy(const VideoCommon::BufferCopy& copy) { return VkBufferCopy{ .srcOffset = copy.src_offset, @@ -392,7 +374,23 @@ BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& m staging_pool{staging_pool_}, guest_descriptor_queue{guest_descriptor_queue_}, quad_index_pass(device, scheduler, descriptor_pool, staging_pool, compute_pass_descriptor_queue), - multi_range_buffers(device_) { + multi_range_buffers(device_), + read_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = device_.GetBufferUserStages(), + .srcAccessMask = vk::ACCESS_BUFFER_WRITES, + .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .dstAccessMask = vk::ACCESS_TRANSFER, + }, + write_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = device_.GetBufferConsumerStages(), + .dstAccessMask = device_.GetBufferConsumerAccess(), + } { const VkDriverIdKHR driver_id = device.GetDriverID(); limit_dynamic_storage_buffers = driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY || driver_id == VK_DRIVER_ID_ARM_PROPRIETARY; @@ -513,28 +511,28 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer, } scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([src_buffer, dst_buffer, vk_copies, barrier](vk::CommandBuffer cmdbuf) { + scheduler.Record([this, src_buffer, dst_buffer, vk_copies, barrier](vk::CommandBuffer cmdbuf) { if (barrier) { - cmdbuf.PipelineBarrier(READ_BARRIER); + cmdbuf.PipelineBarrier(read_barrier); } cmdbuf.CopyBuffer(src_buffer, dst_buffer, VideoCommon::FixSmallVectorADL(vk_copies)); if (barrier) { - cmdbuf.PipelineBarrier(WRITE_BARRIER); + cmdbuf.PipelineBarrier(write_barrier); } }); } void BufferCacheRuntime::PreCopyBarrier() { scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(READ_BARRIER); + scheduler.Record([this](vk::CommandBuffer cmdbuf) { + cmdbuf.PipelineBarrier(read_barrier); }); } void BufferCacheRuntime::PostCopyBarrier() { scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(WRITE_BARRIER); + scheduler.Record([this](vk::CommandBuffer cmdbuf) { + cmdbuf.PipelineBarrier(write_barrier); }); } @@ -544,10 +542,10 @@ void BufferCacheRuntime::ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t si } scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([dest_buffer, offset, size, value](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(READ_BARRIER); + scheduler.Record([this, dest_buffer, offset, size, value](vk::CommandBuffer cmdbuf) { + cmdbuf.PipelineBarrier(read_barrier); cmdbuf.FillBuffer(dest_buffer, offset, size, value); - cmdbuf.PipelineBarrier(WRITE_BARRIER); + cmdbuf.PipelineBarrier(write_barrier); }); } diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h index 1a8629aee8..2f40cc2ac1 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.h +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h @@ -270,6 +270,9 @@ private: boost::container::small_vector multi_range_sources; VkDeviceSize multi_range_total{}; + VkMemoryBarrier2 read_barrier; + VkMemoryBarrier2 write_barrier; + bool limit_dynamic_storage_buffers = false; u32 max_dynamic_storage_buffers = (std::numeric_limits::max)(); }; diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 083b3e900c..2fcba26ce4 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -328,7 +328,7 @@ std::pair Uint8Pass::Assemble(u32 num_vertices, VkBuffer .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, .dstStageMask = VK_PIPELINE_STAGE_2_VERTEX_INPUT_BIT, - .dstAccessMask = VK_ACCESS_2_VERTEX_ATTRIBUTE_READ_BIT, + .dstAccessMask = VK_ACCESS_2_INDEX_READ_BIT, }; const VkDescriptorSet set = descriptor_allocator.Commit(); device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); @@ -487,13 +487,13 @@ void ConditionalRenderingResolvePass::Resolve(VkBuffer dst_buffer, VkBuffer src_ scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([this, descriptor_data, compare_to_zero](vk::CommandBuffer cmdbuf) { - static constexpr VkMemoryBarrier2 read_barrier{ + const VkMemoryBarrier2 read_barrier{ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT | VK_ACCESS_2_SHADER_WRITE_BIT, + .srcStageMask = device.GetBufferUserStages(), + .srcAccessMask = vk::ACCESS_BUFFER_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, - .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT, + .dstAccessMask = vk::ACCESS_SHADER_RESOURCES, }; static constexpr VkMemoryBarrier2 write_barrier{ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, @@ -536,14 +536,22 @@ QueriesPrefixScanPass::QueriesPrefixScanPass( void QueriesPrefixScanPass::Run(VkBuffer accumulation_buffer, VkBuffer dst_buffer, VkBuffer src_buffer, size_t number_of_sums, size_t min_accumulation_limit, size_t max_accumulation_limit) { - constexpr VkAccessFlags2 BASE_DST_ACCESS = VK_ACCESS_2_SHADER_READ_BIT | - VK_ACCESS_2_TRANSFER_READ_BIT | - VK_ACCESS_2_VERTEX_ATTRIBUTE_READ_BIT | - VK_ACCESS_2_INDIRECT_COMMAND_READ_BIT | - VK_ACCESS_2_INDEX_READ_BIT | - VK_ACCESS_2_UNIFORM_READ_BIT; - const VkAccessFlags2 conditional_access = - device.IsExtConditionalRendering() ? VK_ACCESS_2_CONDITIONAL_RENDERING_READ_BIT_EXT : 0; + const VkMemoryBarrier2 read_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = device.GetBufferUserStages(), + .srcAccessMask = vk::ACCESS_BUFFER_WRITES, + .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .dstAccessMask = vk::ACCESS_SHADER_RESOURCES, + }; + const VkMemoryBarrier2 write_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, + .dstStageMask = device.GetBufferConsumerStages(), + .dstAccessMask = device.GetBufferConsumerAccess(), + }; size_t current_runs = number_of_sums; size_t offset = 0; while (current_runs != 0) { @@ -560,23 +568,8 @@ void QueriesPrefixScanPass::Run(VkBuffer accumulation_buffer, VkBuffer dst_buffe scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([this, descriptor_data, min_accumulation_limit, max_accumulation_limit, - runs_to_do, used_offset, conditional_access](vk::CommandBuffer cmdbuf) { - static constexpr VkMemoryBarrier2 read_barrier{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, - .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT, - }; - const VkMemoryBarrier2 write_barrier{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, - .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = BASE_DST_ACCESS | conditional_access, - }; + runs_to_do, used_offset, read_barrier, + write_barrier](vk::CommandBuffer cmdbuf) { const QueriesPrefixScanPushConstants uniforms{ .min_accumulation_base = static_cast(min_accumulation_limit), .max_accumulation_base = static_cast(max_accumulation_limit), @@ -604,9 +597,9 @@ void RecordUnswizzleBeginBarrier(Scheduler& scheduler, VkPipeline vk_pipeline, V VkImageLayout old_layout = VK_IMAGE_LAYOUT_UNDEFINED; VkPipelineStageFlags2 src_stage = VK_PIPELINE_STAGE_2_TOP_OF_PIPE_BIT; if (is_initialized) { - src_access = VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT; + src_access = vk::ACCESS_IMAGE_WRITES; old_layout = VK_IMAGE_LAYOUT_GENERAL; - src_stage = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER; + src_stage = vk::PIPELINE_STAGE_IMAGE_USERS; } scheduler.Record([vk_pipeline, vk_image, aspect_mask, src_access, old_layout, src_stage](vk::CommandBuffer cmdbuf) { @@ -643,8 +636,8 @@ void RecordUnswizzleEndBarrier(Scheduler& scheduler, VkImage vk_image, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, diff --git a/src/video_core/renderer_vulkan/vk_present_manager.cpp b/src/video_core/renderer_vulkan/vk_present_manager.cpp index 4db267f724..87eb109877 100644 --- a/src/video_core/renderer_vulkan/vk_present_manager.cpp +++ b/src/video_core/renderer_vulkan/vk_present_manager.cpp @@ -458,7 +458,7 @@ void PresentManager::CopyToSwapchainImpl(Frame* frame) { VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, + .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_NONE, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, @@ -478,8 +478,8 @@ void PresentManager::CopyToSwapchainImpl(Frame* frame) { VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -502,8 +502,8 @@ void PresentManager::CopyToSwapchainImpl(Frame* frame) { .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_ALL_GRAPHICS_BIT, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT, + .dstStageMask = VK_PIPELINE_STAGE_2_NONE, + .dstAccessMask = VK_ACCESS_2_NONE, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -522,8 +522,8 @@ void PresentManager::CopyToSwapchainImpl(Frame* frame) { .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_ALL_GRAPHICS_BIT, - .dstAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, diff --git a/src/video_core/renderer_vulkan/vk_query_cache.cpp b/src/video_core/renderer_vulkan/vk_query_cache.cpp index e14629dfc8..af8a0502b8 100644 --- a/src/video_core/renderer_vulkan/vk_query_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_query_cache.cpp @@ -869,7 +869,7 @@ public: .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .dstStageMask = vk::PIPELINE_STAGE_HOST, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstAccessMask = VK_ACCESS_2_HOST_READ_BIT, }; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([](vk::CommandBuffer cmdbuf) { @@ -1040,7 +1040,7 @@ private: .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT, + .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, }; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([dst_buffer = current_bank->GetBuffer(), @@ -1577,32 +1577,24 @@ VideoCommon::StreamerInterface* QueryCacheRuntime::GetStreamerInterface(QueryTyp } void QueryCacheRuntime::Barriers(bool is_prebarrier) { - static constexpr VkMemoryBarrier2 READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, - }; - static constexpr VkMemoryBarrier2 WRITE_BARRIER{ + VkMemoryBarrier2 barrier{ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER_HOST, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = impl->device.GetBufferConsumerStages(), + .dstAccessMask = impl->device.GetBufferConsumerAccess(), }; - impl->scheduler.RequestOutsideRenderPassOperationContext(); if (is_prebarrier) { - impl->scheduler.Record([](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(READ_BARRIER); - }); - } else { - impl->scheduler.Record([](vk::CommandBuffer cmdbuf) { - cmdbuf.PipelineBarrier(WRITE_BARRIER); - }); + barrier.srcStageMask = impl->device.GetBufferUserStages(); + barrier.srcAccessMask = vk::ACCESS_BUFFER_WRITES; + barrier.dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT; + barrier.dstAccessMask = vk::ACCESS_TRANSFER; } + impl->scheduler.RequestOutsideRenderPassOperationContext(); + impl->scheduler.Record([barrier](vk::CommandBuffer cmdbuf) { + cmdbuf.PipelineBarrier(barrier); + }); } template diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 488fe8019c..a1dc2259a5 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -628,23 +628,14 @@ void RasterizerVulkan::DispatchCompute() { return; } - static constexpr VkMemoryBarrier2 READ_BARRIER{ + const VkMemoryBarrier2 read_barrier{ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_BUFFER_INPUTS, + .srcStageMask = device.GetBufferUserStages(), .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, .dstAccessMask = vk::ACCESS_SHADER_RESOURCES, }; - static constexpr VkMemoryBarrier2 INDIRECT_READ_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_BUFFER_INPUTS, - .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, - .dstStageMask = - VK_PIPELINE_STAGE_2_DRAW_INDIRECT_BIT | VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, - .dstAccessMask = VK_ACCESS_2_INDIRECT_COMMAND_READ_BIT | vk::ACCESS_SHADER_RESOURCES, - }; const auto& qmd{kepler_compute->launch_description}; auto indirect_address = kepler_compute->GetIndirectComputeAddress(); if (indirect_address) { @@ -654,12 +645,15 @@ void RasterizerVulkan::DispatchCompute() { const auto [buffer, offset] = buffer_cache.ObtainBuffer(*indirect_address, 12, sync_info, post_op); scheduler.RequestComputeDispatchContext(); - scheduler.Record([pipeline, indirect_buffer = buffer->Handle(), - indirect_offset = offset](vk::CommandBuffer cmdbuf) { + VkMemoryBarrier2 indirect_read_barrier = read_barrier; + indirect_read_barrier.dstStageMask |= VK_PIPELINE_STAGE_2_DRAW_INDIRECT_BIT; + indirect_read_barrier.dstAccessMask |= VK_ACCESS_2_INDIRECT_COMMAND_READ_BIT; + scheduler.Record([pipeline, indirect_buffer = buffer->Handle(), indirect_offset = offset, + indirect_read_barrier](vk::CommandBuffer cmdbuf) { if (!pipeline->IsBound()) { return; } - cmdbuf.PipelineBarrier(INDIRECT_READ_BARRIER); + cmdbuf.PipelineBarrier(indirect_read_barrier); cmdbuf.DispatchIndirect(indirect_buffer, indirect_offset); }); return; @@ -670,11 +664,11 @@ void RasterizerVulkan::DispatchCompute() { return; } scheduler.RequestComputeDispatchContext(); - scheduler.Record([pipeline, dim](vk::CommandBuffer cmdbuf) { + scheduler.Record([pipeline, dim, read_barrier](vk::CommandBuffer cmdbuf) { if (!pipeline->IsBound()) { return; } - cmdbuf.PipelineBarrier(READ_BARRIER); + cmdbuf.PipelineBarrier(read_barrier); cmdbuf.Dispatch(dim[0], dim[1], dim[2]); }); diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index ed7cc1ed19..42ecc9c3c6 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -47,11 +47,22 @@ Scheduler::Scheduler(const Device& device_, StateTracker& state_tracker_) : device{device_}, state_tracker{state_tracker_}, master_semaphore{std::make_unique(device)}, command_pool{std::make_unique(*master_semaphore, device)} { + compute_write_barrier = VkMemoryBarrier2{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, + .pNext = nullptr, + .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, + .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, + .dstStageMask = device.GetBufferConsumerStages() | vk::PIPELINE_STAGE_ATTACHMENTS, + .dstAccessMask = device.GetBufferConsumerAccess() | vk::ACCESS_ATTACHMENTS, + }; + renderpass_write_barrier = compute_write_barrier; + renderpass_write_barrier.srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_SHADERS; + upload_write_barrier = compute_write_barrier; + upload_write_barrier.srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT; + upload_write_barrier.srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT; if (device.IsExtTransformFeedbackSupported()) { renderpass_write_barrier.srcStageMask |= VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT; renderpass_write_barrier.srcAccessMask |= VK_ACCESS_2_TRANSFORM_FEEDBACK_WRITE_BIT_EXT; - renderpass_write_barrier.dstStageMask |= VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT; - renderpass_write_barrier.dstAccessMask |= VK_ACCESS_2_TRANSFORM_FEEDBACK_WRITE_BIT_EXT; } AcquireNewChunk(); @@ -241,18 +252,8 @@ void Scheduler::PublishComputeWrites() { if (!std::exchange(compute_writes, false)) { return; } - Record([](vk::CommandBuffer cmdbuf) { - static constexpr VkMemoryBarrier2 COMPUTE_WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, - .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_BUFFER_INPUTS | vk::PIPELINE_STAGE_ATTACHMENTS | - VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .dstAccessMask = - vk::ACCESS_BUFFER_INPUTS | vk::ACCESS_ATTACHMENTS | vk::ACCESS_TRANSFER, - }; - cmdbuf.PipelineBarrier(COMPUTE_WRITE_BARRIER); + Record([barrier = &compute_write_barrier](vk::CommandBuffer cmdbuf) { + cmdbuf.PipelineBarrier(*barrier); }); } @@ -379,15 +380,7 @@ u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_se const u64 signal_value = master_semaphore->NextTick(); RecordWithUploadBuffer([signal_semaphore, wait_semaphore, signal_value, this](vk::CommandBuffer cmdbuf, vk::CommandBuffer upload_cmdbuf) { - static constexpr VkMemoryBarrier2 WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, - }; - upload_cmdbuf.PipelineBarrier(WRITE_BARRIER); + upload_cmdbuf.PipelineBarrier(upload_write_barrier); upload_cmdbuf.End(); cmdbuf.End(); @@ -481,12 +474,8 @@ void Scheduler::EndRenderPass() | VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT, .srcAccessMask = src_access, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT - | VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT - | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT - | VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT - | VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index 53be37928b..a7109e4a48 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -348,15 +348,9 @@ private: bool renderpass_writes = false; bool renderpass_depth_writes = false; bool compute_writes = false; - VkMemoryBarrier2 renderpass_write_barrier{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_SHADERS, - .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_BUFFER_INPUTS | vk::PIPELINE_STAGE_ATTACHMENTS | - VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .dstAccessMask = vk::ACCESS_BUFFER_INPUTS | vk::ACCESS_ATTACHMENTS | vk::ACCESS_TRANSFER, - }; + VkMemoryBarrier2 renderpass_write_barrier{}; + VkMemoryBarrier2 compute_write_barrier{}; + VkMemoryBarrier2 upload_write_barrier{}; u32 num_renderpass_images = 0; std::array renderpass_images{}; std::array renderpass_image_ranges{}; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 9882e1d34c..04b4eabc07 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -689,10 +689,6 @@ struct RangedBarrierRange { void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage image, VkImageAspectFlags aspect_mask, bool is_initialized, std::span copies) { - static constexpr VkAccessFlags2 WRITE_ACCESS_FLAGS = - VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; - // Compute exact mip/layer range being written to RangedBarrierRange range; for (const auto& region : copies) { @@ -703,10 +699,8 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im const VkImageMemoryBarrier2 read_barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT, - .srcAccessMask = WRITE_ACCESS_FLAGS, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED, @@ -722,8 +716,8 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -938,8 +932,8 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -952,10 +946,8 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED, // Discard contents @@ -972,8 +964,8 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = 0, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT | VK_ACCESS_2_MEMORY_READ_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -986,8 +978,8 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT | VK_ACCESS_2_MEMORY_READ_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1259,27 +1251,17 @@ void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src, static constexpr VkMemoryBarrier2 READ_BARRIER{ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_BUFFER_USERS, + .srcAccessMask = vk::ACCESS_BUFFER_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, - }; - static constexpr VkMemoryBarrier2 WRITE_BARRIER{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, - .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstAccessMask = vk::ACCESS_TRANSFER, }; const std::array pre_barriers{ VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1296,8 +1278,8 @@ void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = 0, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = 0, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1310,10 +1292,8 @@ void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src, VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1330,13 +1310,8 @@ void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT | - VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1349,7 +1324,7 @@ void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src, cmdbuf.CopyImageToBuffer(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, copy_buffer, vk_in_copies); - cmdbuf.PipelineBarrier(0, WRITE_BARRIER, {}, middle_in_barrier); + cmdbuf.PipelineBarrier(0, {}, {}, middle_in_barrier); cmdbuf.PipelineBarrier(0, READ_BARRIER, {}, middle_out_barrier); cmdbuf.CopyBufferToImage(copy_buffer, dst_image, VK_IMAGE_LAYOUT_GENERAL, vk_out_copies); @@ -1435,10 +1410,8 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1457,10 +1430,8 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1482,11 +1453,8 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1679,13 +1647,8 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | - VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1698,13 +1661,8 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | - VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT | - VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_2_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1721,8 +1679,8 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = 0, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .dstAccessMask = 0, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1735,8 +1693,8 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1768,17 +1726,6 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src, static_cast(copy.src_subresource.base_layer + copy.src_subresource.num_layers) <= shadow->layers) { const VkImageAspectFlags aspect_mask = shadow->aspect_mask; - VkPipelineStageFlags2 attachment_stage = VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT; - VkAccessFlags2 attachment_write = VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT; - VkAccessFlags2 attachment_read_write = - VK_ACCESS_2_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT; - if ((aspect_mask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) != 0) { - attachment_stage = VK_PIPELINE_STAGE_2_EARLY_FRAGMENT_TESTS_BIT | - VK_PIPELINE_STAGE_2_LATE_FRAGMENT_TESTS_BIT; - attachment_write = VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; - attachment_read_write = VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT | - VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; - } const VkImage shadow_image = *shadow->image; const VkImage dst_image = dst.Handle(); const VkImageCopy region{ @@ -1799,15 +1746,14 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src, .extent = {copy.extent.width, copy.extent.height, 1}, }; scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([shadow_image, dst_image, region, aspect_mask, attachment_stage, - attachment_write, - attachment_read_write](vk::CommandBuffer cmdbuf) { + scheduler.Record([shadow_image, dst_image, region, + aspect_mask](vk::CommandBuffer cmdbuf) { const std::array pre_barriers{ VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = attachment_stage | VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = attachment_write, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1821,9 +1767,8 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src, VkImageMemoryBarrier2{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = attachment_stage | VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT | attachment_write | - VK_ACCESS_2_TRANSFER_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -1841,8 +1786,8 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = 0, - .dstStageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, - .dstAccessMask = 0, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -1856,9 +1801,8 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, - .dstStageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, - .dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | attachment_read_write | - VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -2243,12 +2187,15 @@ void Image::DownloadMemory(std::span buffers_span, std::span o scheduler->RequestOutsideRenderPassOperationContext(); scheduler->Record([buffers = std::move(buffers_vector), image = temp_vk_image, - aspect_mask_ = aspect_mask, vk_copies](vk::CommandBuffer cmdbuf) { + aspect_mask_ = aspect_mask, vk_copies, + consumer_stages = runtime->device.GetBufferConsumerStages(), + consumer_access = runtime->device.GetBufferConsumerAccess()]( + vk::CommandBuffer cmdbuf) { const VkImageMemoryBarrier2 read_barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -2275,17 +2222,17 @@ void Image::DownloadMemory(std::span buffers_span, std::span o .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = consumer_stages, + .dstAccessMask = consumer_access, }; const VkImageMemoryBarrier2 image_write_barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = 0, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -2318,12 +2265,15 @@ void Image::DownloadMemory(std::span buffers_span, std::span o } scheduler->RequestOutsideRenderPassOperationContext(); scheduler->Record([buffers = std::move(buffers_vector), image = *original_image, - aspect_mask_ = aspect_mask, vk_copies](vk::CommandBuffer cmdbuf) { + aspect_mask_ = aspect_mask, vk_copies, + consumer_stages = runtime->device.GetBufferConsumerStages(), + consumer_access = runtime->device.GetBufferConsumerAccess()]( + vk::CommandBuffer cmdbuf) { const VkImageMemoryBarrier2 read_barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, - .srcStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .srcStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .srcAccessMask = vk::ACCESS_IMAGE_WRITES, .dstStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -2350,17 +2300,17 @@ void Image::DownloadMemory(std::span buffers_span, std::span o .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, - .srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, + .srcAccessMask = VK_ACCESS_2_TRANSFER_WRITE_BIT, + .dstStageMask = consumer_stages, + .dstAccessMask = consumer_access, }; const VkImageMemoryBarrier2 image_write_barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_TRANSFER_BIT, .srcAccessMask = 0, - .dstStageMask = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, - .dstAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -3229,8 +3179,8 @@ void TextureCacheRuntime::TransitionImageLayout(Image& image) { .pNext = nullptr, .srcStageMask = VK_PIPELINE_STAGE_2_NONE, .srcAccessMask = VK_ACCESS_2_NONE, - .dstStageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, - .dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT, + .dstStageMask = vk::PIPELINE_STAGE_IMAGE_USERS, + .dstAccessMask = vk::ACCESS_IMAGE_USERS, .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 3e3726651d..12d2267876 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -869,6 +869,32 @@ FN_MAX_LIMIT_LIST return extensions.conditional_rendering; } + VkPipelineStageFlags2 GetBufferUserStages() const { + VkPipelineStageFlags2 stages = vk::PIPELINE_STAGE_BUFFER_USERS; + if (IsExtConditionalRendering()) { + stages |= VK_PIPELINE_STAGE_2_CONDITIONAL_RENDERING_BIT_EXT; + } + if (IsExtTransformFeedbackSupported()) { + stages |= VK_PIPELINE_STAGE_2_TRANSFORM_FEEDBACK_BIT_EXT; + } + return stages; + } + + VkPipelineStageFlags2 GetBufferConsumerStages() const { + return GetBufferUserStages() | VK_PIPELINE_STAGE_2_HOST_BIT; + } + + VkAccessFlags2 GetBufferConsumerAccess() const { + VkAccessFlags2 access = vk::ACCESS_BUFFER_CONSUMERS; + if (IsExtConditionalRendering()) { + access |= VK_ACCESS_2_CONDITIONAL_RENDERING_READ_BIT_EXT; + } + if (IsExtTransformFeedbackSupported()) { + access |= VK_ACCESS_2_TRANSFORM_FEEDBACK_WRITE_BIT_EXT; + } + return access; + } + bool IsExtAstcDecodeModeSupported() const { return extensions.astc_decode_mode; } diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index 72403aff8b..824b567e7f 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -147,15 +147,6 @@ inline VkResult Filter(VkResult result) { return result; } -inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_GRAPHICS_COMPUTE = - VK_PIPELINE_STAGE_2_ALL_GRAPHICS_BIT | VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT; - -inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER = - PIPELINE_STAGE_GRAPHICS_COMPUTE | VK_PIPELINE_STAGE_2_TRANSFER_BIT; - -inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER_HOST = - PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER | VK_PIPELINE_STAGE_2_HOST_BIT; - inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_HOST = VK_PIPELINE_STAGE_2_HOST_BIT; inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_GRAPHICS_SHADERS = @@ -183,6 +174,27 @@ inline constexpr VkAccessFlags2 ACCESS_ATTACHMENTS = inline constexpr VkAccessFlags2 ACCESS_TRANSFER = VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT; +inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_BUFFER_USERS = + PIPELINE_STAGE_BUFFER_INPUTS | VK_PIPELINE_STAGE_2_TRANSFER_BIT; + +inline constexpr VkAccessFlags2 ACCESS_BUFFER_WRITES = + VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT; + +inline constexpr VkAccessFlags2 ACCESS_BUFFER_CONSUMERS = + ACCESS_BUFFER_INPUTS | ACCESS_TRANSFER | VK_ACCESS_2_HOST_READ_BIT; + +inline constexpr VkPipelineStageFlags2 PIPELINE_STAGE_IMAGE_USERS = + PIPELINE_STAGE_GRAPHICS_SHADERS | VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | + PIPELINE_STAGE_ATTACHMENTS | VK_PIPELINE_STAGE_2_TRANSFER_BIT; + +inline constexpr VkAccessFlags2 ACCESS_IMAGE_WRITES = + VK_ACCESS_2_SHADER_WRITE_BIT | VK_ACCESS_2_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT; + +inline constexpr VkAccessFlags2 ACCESS_IMAGE_USERS = + VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT | ACCESS_ATTACHMENTS | + ACCESS_TRANSFER; + /// Table holding Vulkan instance function pointers. struct InstanceDispatch {