[TEST] Line/ conservative rasterization adjustment

src/android/app/build.gradle.kts

@@ -65,7 +65,7 @@ android {
 
     defaultConfig {
         applicationId = "dev.eden.eden_emulator"
-        minSdk = 24
+        minSdk = 33
         targetSdk = 36
         versionName = getGitVersion()
         versionCode = autoVersion

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt

@@ -594,7 +594,7 @@ abstract class SettingsItem(
                     IntSetting.ANDROID_PIPELINE_WORKERS,
                     titleId = R.string.pipeline_worker_cores,
                     descriptionId = R.string.pipeline_worker_cores_description,
-                    min = 4,
+                    min = 2,
                     max = 8,
                     units = "cores"
                 )

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/InputDialogFragment.kt

@@ -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: 2024 yuzu Emulator Project
@@ -169,7 +169,7 @@ class InputDialogFragment : DialogFragment() {
         NativeInput.onGamePadButtonEvent(
             controllerData.getGUID(),
             controllerData.getPort(),
-            InputHandler.getButtonIdFromEvent(event),
+            event.keyCode,
             action
         )
         onInputReceived(event.device)

src/android/app/src/main/java/org/yuzu/yuzu_emu/utils/InputHandler.kt

@@ -49,12 +49,6 @@ object InputHandler {
         MotionEvent.AXIS_RTRIGGER
     )
 
-    // Currently, Android doesn't support Joy-Con D-pad buttons. We fall back to the scan code
-    private const val LINUX_BUTTON_DPAD_UP = 0x220
-    private const val LINUX_BUTTON_DPAD_DOWN = 0x221
-    private const val LINUX_BUTTON_DPAD_LEFT = 0x222
-    private const val LINUX_BUTTON_DPAD_RIGHT = 0x223
-
     fun isPhysicalGameController(device: InputDevice?): Boolean {
         device ?: return false
 
@@ -93,25 +87,12 @@ object InputHandler {
         NativeInput.onGamePadButtonEvent(
             controllerData.getGUID(),
             controllerData.getPort(),
-            getButtonIdFromEvent(event),
+            event.keyCode,
             action
         )
         return true
     }
 
-    fun getButtonIdFromEvent(event: KeyEvent): Int {
-        if (event.keyCode == 0) {
-            return when (event.scanCode) {
-                LINUX_BUTTON_DPAD_UP -> KeyEvent.KEYCODE_DPAD_UP
-                LINUX_BUTTON_DPAD_DOWN -> KeyEvent.KEYCODE_DPAD_DOWN
-                LINUX_BUTTON_DPAD_LEFT -> KeyEvent.KEYCODE_DPAD_LEFT
-                LINUX_BUTTON_DPAD_RIGHT -> KeyEvent.KEYCODE_DPAD_RIGHT
-                else -> return 0
-            }
-        }
-        return event.keyCode
-    }
-
     fun dispatchGenericMotionEvent(event: MotionEvent): Boolean {
         val controllerData =
             androidControllers[event.device.controllerNumber] ?: return false

src/android/app/src/main/jni/game_metadata.cpp

@@ -20,96 +20,132 @@ struct RomMetadata {
     std::vector<u8> icon;
     bool isHomebrew;
 };
-static ankerl::unordered_dense::map<std::string, RomMetadata> m_rom_metadata_cache;
 
-static RomMetadata CacheRomMetadata(const std::string& path) {
-    auto& instance = EmulationSession::GetInstance();
-    const auto file = Core::GetGameFileFromPath(instance.System().GetFilesystem(), path);
-    if (auto loader = Loader::GetLoader(instance.System(), file, 0, 0); loader) {
-        RomMetadata entry;
-        loader->ReadTitle(entry.title);
-        loader->ReadProgramId(entry.programId);
-        loader->ReadIcon(entry.icon);
+ankerl::unordered_dense::map<std::string, RomMetadata> m_rom_metadata_cache;
 
-        const FileSys::PatchManager pm{
-            entry.programId,
-            instance.System().GetFileSystemController(),
-            instance.System().GetContentProvider()
-        };
-        const auto control = pm.GetControlMetadata();
+RomMetadata CacheRomMetadata(const std::string& path) {
+    const auto file =
+        Core::GetGameFileFromPath(EmulationSession::GetInstance().System().GetFilesystem(), path);
+    auto loader = Loader::GetLoader(EmulationSession::GetInstance().System(), file, 0, 0);
 
-        if (control.first != nullptr) {
-            entry.developer = control.first->GetDeveloperName();
-            entry.version = control.first->GetVersionString();
+    RomMetadata entry;
+    loader->ReadTitle(entry.title);
+    loader->ReadProgramId(entry.programId);
+    loader->ReadIcon(entry.icon);
+
+    const FileSys::PatchManager pm{
+        entry.programId, EmulationSession::GetInstance().System().GetFileSystemController(),
+        EmulationSession::GetInstance().System().GetContentProvider()};
+    const auto control = pm.GetControlMetadata();
+
+    if (control.first != nullptr) {
+        entry.developer = control.first->GetDeveloperName();
+        entry.version = control.first->GetVersionString();
+    } else {
+        FileSys::NACP nacp;
+        if (loader->ReadControlData(nacp) == Loader::ResultStatus::Success) {
+            entry.developer = nacp.GetDeveloperName();
         } else {
-            FileSys::NACP nacp{};
-            entry.developer = loader->ReadControlData(nacp) == Loader::ResultStatus::Success
-                ? nacp.GetDeveloperName()
-                : "";
-            entry.version = "1.0.0";
+            entry.developer = "";
         }
-        if (loader->GetFileType() == Loader::FileType::NRO) {
-            auto loader_nro = reinterpret_cast<Loader::AppLoader_NRO*>(loader.get());
-            entry.isHomebrew = loader_nro->IsHomebrew();
-        } else {
-            entry.isHomebrew = false;
-        }
-        m_rom_metadata_cache[path] = entry;
-        return entry;
+
+        entry.version = "1.0.0";
     }
-    return {};
+
+    if (loader->GetFileType() == Loader::FileType::NRO) {
+        auto loader_nro = reinterpret_cast<Loader::AppLoader_NRO*>(loader.get());
+        entry.isHomebrew = loader_nro->IsHomebrew();
+    } else {
+        entry.isHomebrew = false;
+    }
+
+    m_rom_metadata_cache[path] = entry;
+
+    return entry;
 }
 
-static RomMetadata GetRomMetadata(const std::string& path, bool reload = false) {
-    if (reload)
+RomMetadata GetRomMetadata(const std::string& path, bool reload = false) {
+    if (reload) {
         return CacheRomMetadata(path);
-    if (auto it = m_rom_metadata_cache.find(path); it != m_rom_metadata_cache.end())
-        return it->second;
+    }
+
+    if (auto search = m_rom_metadata_cache.find(path); search != m_rom_metadata_cache.end()) {
+        return search->second;
+    }
+
     return CacheRomMetadata(path);
 }
 
 extern "C" {
 
-jboolean Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getIsValid(JNIEnv* env, jobject obj, jstring jpath) {
-    if (auto const file = EmulationSession::GetInstance().System().GetFilesystem()->OpenFile(Common::Android::GetJString(env, jpath), FileSys::OpenMode::Read); file) {
-        if (auto loader = Loader::GetLoader(EmulationSession::GetInstance().System(), file); loader) {
-            auto const file_type = loader->GetFileType();
-            if (file_type == Loader::FileType::Unknown || file_type == Loader::FileType::Error)
-                return false;
-            if ((file_type == Loader::FileType::NSP || file_type == Loader::FileType::XCI) && !Loader::IsBootableGameContainer(file, file_type))
-                return false;
-            u64 program_id = 0;
-            return loader->ReadProgramId(program_id) == Loader::ResultStatus::Success;
-        }
+jboolean Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getIsValid(JNIEnv* env, jobject obj,
+                                                               jstring jpath) {
+    const auto file = EmulationSession::GetInstance().System().GetFilesystem()->OpenFile(
+        Common::Android::GetJString(env, jpath), FileSys::OpenMode::Read);
+    if (!file) {
+        return false;
     }
-    return false;
+
+    auto loader = Loader::GetLoader(EmulationSession::GetInstance().System(), file);
+    if (!loader) {
+        return false;
+    }
+
+    const auto file_type = loader->GetFileType();
+    if (file_type == Loader::FileType::Unknown || file_type == Loader::FileType::Error) {
+        return false;
+    }
+
+    if ((file_type == Loader::FileType::NSP || file_type == Loader::FileType::XCI) &&
+        !Loader::IsBootableGameContainer(file, file_type)) {
+        return false;
+    }
+
+    u64 program_id = 0;
+    Loader::ResultStatus res = loader->ReadProgramId(program_id);
+    if (res != Loader::ResultStatus::Success) {
+        return false;
+    }
+    return true;
 }
 
-jstring Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getTitle(JNIEnv* env, jobject obj, jstring jpath) {
-    return Common::Android::ToJString(env, GetRomMetadata(Common::Android::GetJString(env, jpath)).title);
+jstring Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getTitle(JNIEnv* env, jobject obj,
+                                                            jstring jpath) {
+    return Common::Android::ToJString(
+        env, GetRomMetadata(Common::Android::GetJString(env, jpath)).title);
 }
 
-jstring Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getProgramId(JNIEnv* env, jobject obj, jstring jpath) {
-    return Common::Android::ToJString(env, std::to_string(GetRomMetadata(Common::Android::GetJString(env, jpath)).programId));
+jstring Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getProgramId(JNIEnv* env, jobject obj,
+                                                                jstring jpath) {
+    return Common::Android::ToJString(
+        env, std::to_string(GetRomMetadata(Common::Android::GetJString(env, jpath)).programId));
 }
 
-jstring Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getDeveloper(JNIEnv* env, jobject obj, jstring jpath) {
-    return Common::Android::ToJString(env, GetRomMetadata(Common::Android::GetJString(env, jpath)).developer);
+jstring Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getDeveloper(JNIEnv* env, jobject obj,
+                                                                jstring jpath) {
+    return Common::Android::ToJString(
+        env, GetRomMetadata(Common::Android::GetJString(env, jpath)).developer);
 }
 
-jstring Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getVersion(JNIEnv* env, jobject obj, jstring jpath, jboolean jreload) {
-    return Common::Android::ToJString(env, GetRomMetadata(Common::Android::GetJString(env, jpath), jreload).version);
+jstring Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getVersion(JNIEnv* env, jobject obj,
+                                                              jstring jpath, jboolean jreload) {
+    return Common::Android::ToJString(
+        env, GetRomMetadata(Common::Android::GetJString(env, jpath), jreload).version);
 }
 
-jbyteArray Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getIcon(JNIEnv* env, jobject obj, jstring jpath) {
+jbyteArray Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getIcon(JNIEnv* env, jobject obj,
+                                                              jstring jpath) {
     auto icon_data = GetRomMetadata(Common::Android::GetJString(env, jpath)).icon;
-    jbyteArray icon = env->NewByteArray(jsize(icon_data.size()));
-    env->SetByteArrayRegion(icon, 0, env->GetArrayLength(icon), reinterpret_cast<jbyte*>(icon_data.data()));
+    jbyteArray icon = env->NewByteArray(static_cast<jsize>(icon_data.size()));
+    env->SetByteArrayRegion(icon, 0, env->GetArrayLength(icon),
+                            reinterpret_cast<jbyte*>(icon_data.data()));
     return icon;
 }
 
-jboolean Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getIsHomebrew(JNIEnv* env, jobject obj, jstring jpath) {
-    return jboolean(GetRomMetadata(Common::Android::GetJString(env, jpath)).isHomebrew);
+jboolean Java_org_yuzu_yuzu_1emu_utils_GameMetadata_getIsHomebrew(JNIEnv* env, jobject obj,
+                                                                  jstring jpath) {
+    return static_cast<jboolean>(
+        GetRomMetadata(Common::Android::GetJString(env, jpath)).isHomebrew);
 }
 
 void Java_org_yuzu_yuzu_1emu_utils_GameMetadata_resetMetadata(JNIEnv* env, jobject obj) {

src/audio_core/renderer/command/data_source/decode.cpp

@@ -18,15 +18,20 @@
 namespace AudioCore::Renderer {
 
 constexpr u32 TempBufferSize = 0x3F00;
+constexpr std::array<u8, 3> PitchBySrcQuality = {4, 8, 4};
 
-/// @brief Decode PCM data. Only s16 or f32 is supported.
-/// @tparam T         - Type to decode. Only s16 and f32 are supported.
-/// @param memory     - Core memory for reading samples.
-/// @param out_buffer - Output mix buffer to receive the samples.
-/// @param req        - Information for how to decode.
-/// @return Number of samples decoded.
+/**
+ * Decode PCM data. Only s16 or f32 is supported.
+ *
+ * @tparam T         - Type to decode. Only s16 and f32 are supported.
+ * @param memory     - Core memory for reading samples.
+ * @param out_buffer - Output mix buffer to receive the samples.
+ * @param req        - Information for how to decode.
+ * @return Number of samples decoded.
+ */
 template <typename T>
-static u32 DecodePcm(Core::Memory::Memory& memory, std::span<s16> out_buffer, const DecodeArg& req) {
+static u32 DecodePcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
+                     const DecodeArg& req) {
     constexpr s32 min{(std::numeric_limits<s16>::min)()};
     constexpr s32 max{(std::numeric_limits<s16>::max)()};
 
@@ -89,12 +94,16 @@ static u32 DecodePcm(Core::Memory::Memory& memory, std::span<s16> out_buffer, co
     return samples_to_decode;
 }
 
-/// @brief Decode ADPCM data.
-/// @param memory     - Core memory for reading samples.
-/// @param out_buffer - Output mix buffer to receive the samples.
-/// @param req        - Information for how to decode.
-/// @return Number of samples decoded.
-static u32 DecodeAdpcm(Core::Memory::Memory& memory, std::span<s16> out_buffer, const DecodeArg& req) {
+/**
+ * Decode ADPCM data.
+ *
+ * @param memory     - Core memory for reading samples.
+ * @param out_buffer - Output mix buffer to receive the samples.
+ * @param req        - Information for how to decode.
+ * @return Number of samples decoded.
+ */
+static u32 DecodeAdpcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
+                       const DecodeArg& req) {
     constexpr u32 SamplesPerFrame{14};
     constexpr u32 NibblesPerFrame{16};
 
@@ -106,7 +115,8 @@ static u32 DecodeAdpcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
         return 0;
     }
 
-    auto end{(req.end_offset % SamplesPerFrame) + NibblesPerFrame * (req.end_offset / SamplesPerFrame)};
+    auto end{(req.end_offset % SamplesPerFrame) +
+             NibblesPerFrame * (req.end_offset / SamplesPerFrame)};
     if (req.end_offset % SamplesPerFrame) {
         end += 3;
     } else {
@@ -123,49 +133,52 @@ static u32 DecodeAdpcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
         return 0;
     }
 
-    auto samples_to_read = samples_to_process;
-    auto samples_remaining_in_frame = start_pos % SamplesPerFrame;
-    auto position_in_frame = (start_pos / SamplesPerFrame) * NibblesPerFrame + samples_remaining_in_frame;
+    auto samples_to_read{samples_to_process};
+    auto samples_remaining_in_frame{start_pos % SamplesPerFrame};
+    auto position_in_frame{(start_pos / SamplesPerFrame) * NibblesPerFrame +
+                           samples_remaining_in_frame};
+
     if (samples_remaining_in_frame) {
         position_in_frame += 2;
     }
 
     const auto size{(std::max)((samples_to_process / 8U) * SamplesPerFrame, 8U)};
-    Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::UnsafeRead> wavebuffer(memory, req.buffer + position_in_frame / 2, size);
+    Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::UnsafeRead> wavebuffer(
+        memory, req.buffer + position_in_frame / 2, size);
 
-    auto context = req.adpcm_context;
-    auto header = context->header;
-    u8 scale = u8(header & 0xfU);
-    u8 coeff_index = u8((header >> 4U) & 0x7u);
-    s32 coeff0 = req.coefficients[coeff_index * 2 + 0];
-    s32 coeff1 = req.coefficients[coeff_index * 2 + 1];
+    auto context{req.adpcm_context};
+    auto header{context->header};
+    u8 coeff_index{static_cast<u8>((header >> 4U) & 0xFU)};
+    u8 scale{static_cast<u8>(header & 0xFU)};
+    s32 coeff0{req.coefficients[coeff_index * 2 + 0]};
+    s32 coeff1{req.coefficients[coeff_index * 2 + 1]};
 
-    auto yn0 = context->yn0;
-    auto yn1 = context->yn1;
-    auto const get_step = [](u32 index) {
-        // Emulates the following table
-        // 0, 1, 2, 3, 4, 5, 6, 7, -8, -7, -6, -5, -4, -3, -2, -1,
-        constexpr u64 steps_table = 0x1234567876543210ull;
-        constexpr u64 steps_sign = 0b1111111100000000ull;
-        auto const r = s32((steps_table >> (index * 4)) & 0xf);
-        return ((steps_sign >> index) & 1) == 0 ? -r : r;
+    auto yn0{context->yn0};
+    auto yn1{context->yn1};
+
+    static constexpr std::array<s32, 16> Steps{
+        0, 1, 2, 3, 4, 5, 6, 7, -8, -7, -6, -5, -4, -3, -2, -1,
     };
-    auto const decode_sample = [&](const s32 code) -> s16 {
+
+    const auto decode_sample = [&](const s32 code) -> s16 {
         const auto xn = code * (1 << scale);
         const auto prediction = coeff0 * yn0 + coeff1 * yn1;
         const auto sample = ((xn << 11) + 0x400 + prediction) >> 11;
         const auto saturated = std::clamp<s32>(sample, -0x8000, 0x7FFF);
         yn1 = yn0;
-        return yn0 = s16(saturated);
+        yn0 = static_cast<s16>(saturated);
+        return yn0;
     };
 
-    u32 read_index = 0;
-    for (u32 write_index = 0; samples_to_read > 0 && write_index < out_buffer.size(); ) {
+    u32 read_index{0};
+    u32 write_index{0};
+
+    while (samples_to_read > 0) {
         // Are we at a new frame?
         if ((position_in_frame % NibblesPerFrame) == 0) {
             header = wavebuffer[read_index++];
-            scale = header & 0xFu;
-            coeff_index = (header >> 4) & 0x7u;
+            coeff_index = (header >> 4) & 0xF;
+            scale = header & 0xF;
             coeff0 = req.coefficients[coeff_index * 2 + 0];
             coeff1 = req.coefficients[coeff_index * 2 + 1];
             position_in_frame += 2;
@@ -174,12 +187,14 @@ static u32 DecodeAdpcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
             if (samples_to_read >= SamplesPerFrame) {
                 // Can grab all samples until the next header
                 for (u32 i = 0; i < SamplesPerFrame / 2; i++) {
-                    auto code0 = get_step((wavebuffer[read_index + i] >> 4) & 0xF);
-                    auto code1 = get_step(wavebuffer[read_index + i] & 0xF);
+                    auto code0{Steps[(wavebuffer[read_index] >> 4) & 0xF]};
+                    auto code1{Steps[wavebuffer[read_index] & 0xF]};
+                    read_index++;
+
                     out_buffer[write_index++] = decode_sample(code0);
                     out_buffer[write_index++] = decode_sample(code1);
                 }
-                read_index += SamplesPerFrame / 2;
+
                 position_in_frame += SamplesPerFrame;
                 samples_to_read -= SamplesPerFrame;
                 continue;
@@ -187,14 +202,15 @@ static u32 DecodeAdpcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
         }
 
         // Decode a single sample
-        auto code = wavebuffer[read_index];
+        auto code{wavebuffer[read_index]};
         if (position_in_frame & 1) {
             code &= 0xF;
             read_index++;
         } else {
             code >>= 4;
         }
-        out_buffer[write_index++] = decode_sample(get_step(code));
+
+        out_buffer[write_index++] = decode_sample(Steps[code]);
 
         position_in_frame++;
         samples_to_read--;
@@ -203,21 +219,27 @@ static u32 DecodeAdpcm(Core::Memory::Memory& memory, std::span<s16> out_buffer,
     context->header = header;
     context->yn0 = yn0;
     context->yn1 = yn1;
+
     return samples_to_process;
 }
 
-/// @brief Decode implementation.
-/// Decode wavebuffers according to the given args.
-///
-/// @param memory - Core memory to read data from.
-/// @param args   - The wavebuffer data, and information for how to decode it.
+/**
+ * Decode implementation.
+ * Decode wavebuffers according to the given args.
+ *
+ * @param memory - Core memory to read data from.
+ * @param args   - The wavebuffer data, and information for how to decode it.
+ */
 void DecodeFromWaveBuffers(Core::Memory::Memory& memory, const DecodeFromWaveBuffersArgs& args) {
-    constexpr auto EndWaveBuffer = [](auto& voice_state, auto& wavebuffer, auto& index, auto& played_samples, auto& consumed) -> void {
+    static constexpr auto EndWaveBuffer = [](auto& voice_state, auto& wavebuffer, auto& index,
+                                             auto& played_samples, auto& consumed) -> void {
         voice_state.wave_buffer_valid[index] = false;
         voice_state.loop_count = 0;
+
         if (wavebuffer.stream_ended) {
             played_samples = 0;
         }
+
         index = (index + 1) % MaxWaveBuffers;
         consumed++;
     };
@@ -233,9 +255,8 @@ void DecodeFromWaveBuffers(Core::Memory::Memory& memory, const DecodeFromWaveBuf
         return;
     }
 
-    // 0 -> 4, 1 -> 8, 2 -> 4
-    auto pitch = u32((0x040804ul >> (u32(args.src_quality) * 8)) & 0xfful);
-    if (u32(pitch + size_required.to_int_floor()) > TempBufferSize) {
+    auto pitch{PitchBySrcQuality[static_cast<u32>(args.src_quality)]};
+    if (static_cast<u32>(pitch + size_required.to_int_floor()) > TempBufferSize) {
         return;
     }
 
@@ -251,7 +272,7 @@ void DecodeFromWaveBuffers(Core::Memory::Memory& memory, const DecodeFromWaveBuf
     bool is_buffer_starved{false};
     u32 offset{voice_state.offset};
 
-    auto output_buffer = args.output;
+    auto output_buffer{args.output};
     std::array<s16, TempBufferSize> temp_buffer{};
 
     while (remaining_sample_count > 0) {
@@ -273,8 +294,8 @@ void DecodeFromWaveBuffers(Core::Memory::Memory& memory, const DecodeFromWaveBuf
             if (wavebuffer_index >= MaxWaveBuffers) {
                 LOG_ERROR(Service_Audio, "Invalid wavebuffer index! {}", wavebuffer_index);
                 wavebuffer_index = 0;
-                wavebuffers_consumed = MaxWaveBuffers;
                 voice_state.wave_buffer_valid.fill(false);
+                wavebuffers_consumed = MaxWaveBuffers;
             }
 
             if (!voice_state.wave_buffer_valid[wavebuffer_index]) {
@@ -282,9 +303,12 @@ void DecodeFromWaveBuffers(Core::Memory::Memory& memory, const DecodeFromWaveBuf
                 break;
             }
 
-            auto& wavebuffer = args.wave_buffers[wavebuffer_index];
-            if (offset == 0 && args.sample_format == SampleFormat::Adpcm && wavebuffer.context != 0) {
-                memory.ReadBlockUnsafe(wavebuffer.context, &voice_state.adpcm_context, wavebuffer.context_size);
+            auto& wavebuffer{args.wave_buffers[wavebuffer_index]};
+
+            if (offset == 0 && args.sample_format == SampleFormat::Adpcm &&
+                wavebuffer.context != 0) {
+                memory.ReadBlockUnsafe(wavebuffer.context, &voice_state.adpcm_context,
+                                       wavebuffer.context_size);
             }
 
             auto start_offset{wavebuffer.start_offset};
@@ -327,7 +351,9 @@ void DecodeFromWaveBuffers(Core::Memory::Memory& memory, const DecodeFromWaveBuf
             case SampleFormat::Adpcm: {
                 decode_arg.adpcm_context = &voice_state.adpcm_context;
                 memory.ReadBlockUnsafe(args.data_address, &decode_arg.coefficients, args.data_size);
-                samples_decoded = DecodeAdpcm( memory, {&temp_buffer[temp_buffer_pos], TempBufferSize - temp_buffer_pos}, decode_arg);
+                samples_decoded = DecodeAdpcm(
+                    memory, {&temp_buffer[temp_buffer_pos], TempBufferSize - temp_buffer_pos},
+                    decode_arg);
             } break;
 
             default:

src/audio_core/renderer/command/resample/resample.cpp

@@ -1,6 +1,3 @@
-// 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-2.0-or-later
 
@@ -8,13 +5,21 @@
 
 namespace AudioCore::Renderer {
 
-static void ResampleLowQuality(std::span<s32> output, std::span<const s16> input, const Common::FixedPoint<49, 15>& sample_rate_ratio, Common::FixedPoint<49, 15>& fraction, const u32 samples_to_write) {
-    u32 read_index{0};
-    for (u32 i = 0; i < samples_to_write; i++) {
-        output[i] = input[read_index + (fraction >= 0.5f)];
-        fraction += sample_rate_ratio;
-        read_index += u32(fraction.to_int_floor());
-        fraction.clear_int();
+static void ResampleLowQuality(std::span<s32> output, std::span<const s16> input,
+                               const Common::FixedPoint<49, 15>& sample_rate_ratio,
+                               Common::FixedPoint<49, 15>& fraction, const u32 samples_to_write) {
+    if (sample_rate_ratio == 1.0f) {
+        for (u32 i = 0; i < samples_to_write; i++) {
+            output[i] = input[i];
+        }
+    } else {
+        u32 read_index{0};
+        for (u32 i = 0; i < samples_to_write; i++) {
+            output[i] = input[read_index + (fraction >= 0.5f)];
+            fraction += sample_rate_ratio;
+            read_index += static_cast<u32>(fraction.to_int_floor());
+            fraction.clear_int();
+        }
     }
 }
 
@@ -290,7 +295,7 @@ static void ResampleNormalQuality(std::span<s32> output, std::span<const s16> in
     auto lut{get_lut()};
     u32 read_index{0};
     for (u32 i = 0; i < samples_to_write; i++) {
-        const auto lut_index = ((fraction.get_frac() >> 8) << 2) & 511;
+        const auto lut_index{(fraction.get_frac() >> 8) * 4};
         const Common::FixedPoint<56, 8> sample0{input[read_index + 0] * lut[lut_index + 0]};
         const Common::FixedPoint<56, 8> sample1{input[read_index + 1] * lut[lut_index + 1]};
         const Common::FixedPoint<56, 8> sample2{input[read_index + 2] * lut[lut_index + 2]};
@@ -840,7 +845,7 @@ static void ResampleHighQuality(std::span<s32> output, std::span<const s16> inpu
     auto lut{get_lut()};
     u32 read_index{0};
     for (u32 i = 0; i < samples_to_write; i++) {
-        const auto lut_index = ((fraction.get_frac() >> 8) << 3) & 1023;
+        const auto lut_index{(fraction.get_frac() >> 8) * 8};
         const Common::FixedPoint<56, 8> sample0{input[read_index + 0] * lut[lut_index + 0]};
         const Common::FixedPoint<56, 8> sample1{input[read_index + 1] * lut[lut_index + 1]};
         const Common::FixedPoint<56, 8> sample2{input[read_index + 2] * lut[lut_index + 2]};

src/common/thread.cpp

@@ -1,5 +1,6 @@
 // SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: 2013 Dolphin Emulator Project
 // SPDX-FileCopyrightText: 2014 Citra Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
@@ -39,6 +40,110 @@
 #include <unistd.h>
 #endif
 
+#ifdef __ANDROID__
+#include <sys/resource.h>
+#include <algorithm>
+#include <fstream>
+#include <utility>
+#include <vector>
+
+namespace {
+[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_URGENT_AUDIO = -19;
+[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_AUDIO = -16;
+[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_URGENT_DISPLAY = -8;
+[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_DISPLAY = -4;
+[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_FOREGROUND = -2;
+[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_MORE_FAVORABLE = -1;
+[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_DEFAULT = 0;
+[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_LESS_FAVORABLE = 1;
+[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_BACKGROUND = 10;
+[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_LOWEST = 19;
+
+constexpr size_t ANDROID_MINIMUM_PERFORMANCE_CORES = 4;
+
+cpu_set_t ComputePerformanceCoreMask() {
+    cpu_set_t mask;
+    CPU_ZERO(&mask);
+
+    cpu_set_t allowed;
+    CPU_ZERO(&allowed);
+    if (sched_getaffinity(gettid(), sizeof(allowed), &allowed) != 0) {
+        return mask;
+    }
+
+    std::vector<std::pair<long, int>> cores;
+    const int total = static_cast<int>(std::thread::hardware_concurrency());
+    for (int cpu = 0; cpu < total; ++cpu) {
+        if (!CPU_ISSET(cpu, &allowed)) {
+            continue;
+        }
+        long max_frequency = 0;
+        std::ifstream file("/sys/devices/system/cpu/cpu" + std::to_string(cpu) +
+                           "/cpufreq/cpuinfo_max_freq");
+        if (!file || !(file >> max_frequency) || max_frequency <= 0) {
+            CPU_ZERO(&mask);
+            return mask;
+        }
+        cores.emplace_back(max_frequency, cpu);
+    }
+    if (cores.empty()) {
+        return mask;
+    }
+
+    std::sort(cores.begin(), cores.end(),
+              [](const auto& lhs, const auto& rhs) { return lhs.first > rhs.first; });
+
+    size_t taken = 0;
+    long cluster_frequency = cores.front().first;
+    for (const auto& [frequency, cpu] : cores) {
+        if (frequency != cluster_frequency) {
+            if (taken >= ANDROID_MINIMUM_PERFORMANCE_CORES) {
+                break;
+            }
+            cluster_frequency = frequency;
+        }
+        CPU_SET(cpu, &mask);
+        ++taken;
+    }
+    return mask;
+}
+
+const cpu_set_t& PerformanceCoreMask() {
+    static const cpu_set_t mask = ComputePerformanceCoreMask();
+    return mask;
+}
+
+cpu_set_t ComputeEfficiencyCoreMask() {
+    cpu_set_t mask;
+    CPU_ZERO(&mask);
+
+    const cpu_set_t& performance = PerformanceCoreMask();
+    if (CPU_COUNT(&performance) == 0) {
+        return mask;
+    }
+
+    cpu_set_t allowed;
+    CPU_ZERO(&allowed);
+    if (sched_getaffinity(gettid(), sizeof(allowed), &allowed) != 0) {
+        return mask;
+    }
+
+    const int total = static_cast<int>(std::thread::hardware_concurrency());
+    for (int cpu = 0; cpu < total; ++cpu) {
+        if (CPU_ISSET(cpu, &allowed) && !CPU_ISSET(cpu, &performance)) {
+            CPU_SET(cpu, &mask);
+        }
+    }
+    return mask;
+}
+
+const cpu_set_t& EfficiencyCoreMask() {
+    static const cpu_set_t mask = ComputeEfficiencyCoreMask();
+    return mask;
+}
+} // Anonymous namespace
+#endif
+
 #include "common/cpu_features.h"
 #ifdef ARCHITECTURE_x86_64
 #ifdef _MSC_VER
@@ -78,6 +183,21 @@ void SetCurrentThreadPriority(ThreadPriority new_priority) {
         }
     }();
     set_thread_priority(find_thread(NULL), priority);
+#elif defined(__ANDROID__)
+    const int nice_value = [&]() {
+        switch (new_priority) {
+        case ThreadPriority::Low: return ANDROID_THREAD_PRIORITY_BACKGROUND;
+        case ThreadPriority::Normal: return ANDROID_THREAD_PRIORITY_DEFAULT;
+        case ThreadPriority::High: return ANDROID_THREAD_PRIORITY_DISPLAY;
+        case ThreadPriority::VeryHigh: return ANDROID_THREAD_PRIORITY_URGENT_DISPLAY;
+        case ThreadPriority::Critical: return ANDROID_THREAD_PRIORITY_AUDIO;
+        default: return ANDROID_THREAD_PRIORITY_DEFAULT;
+        }
+    }();
+    if (setpriority(PRIO_PROCESS, static_cast<id_t>(gettid()), nice_value) != 0) {
+        LOG_DEBUG(Common, "Could not set thread nice value to {}: {}", nice_value,
+                  GetLastErrorMsg());
+    }
 #else
     pthread_t this_thread = pthread_self();
     const auto scheduling_type = SCHED_OTHER;
@@ -132,29 +252,28 @@ void SetCurrentThreadName(const char* name) {
 #endif
 }
 
-void PinCurrentThreadToPerformanceCore(size_t core_id) {
-    ASSERT(core_id < 4);
-    // If we set a flag for a CPU that doesn't exist, the thread may not be allowed to
-    // run in ANY processor!
-    auto const total_cores = std::thread::hardware_concurrency();
-    if (core_id < total_cores) {
+void SetCurrentThreadToPerformanceCores() {
 #if defined(__ANDROID__)
-        cpu_set_t set;
-        CPU_ZERO(&set);
-        CPU_SET(core_id, &set);
-        sched_setaffinity(pthread_self(), sizeof(set), &set);
-#elif defined(__linux__) || defined(__FreeBSD__)
-        cpu_set_t set;
-        CPU_ZERO(&set);
-        CPU_SET(core_id, &set);
-        pthread_setaffinity_np(pthread_self(), sizeof(set), &set);
-#elif defined(_WIN32)
-        DWORD set = 1UL << core_id;
-        SetThreadAffinityMask(GetCurrentThread(), set);
-#else
-        // No pin functionality implemented
-#endif
+    const cpu_set_t& mask = PerformanceCoreMask();
+    if (CPU_COUNT(&mask) == 0) {
+        return;
     }
+    if (sched_setaffinity(gettid(), sizeof(mask), &mask) != 0) {
+        LOG_DEBUG(Common, "Could not restrict thread to performance cores: {}", GetLastErrorMsg());
+    }
+#endif
+}
+
+void SetCurrentThreadToEfficiencyCores() {
+#if defined(__ANDROID__)
+    const cpu_set_t& mask = EfficiencyCoreMask();
+    if (CPU_COUNT(&mask) == 0) {
+        return;
+    }
+    if (sched_setaffinity(gettid(), sizeof(mask), &mask) != 0) {
+        LOG_DEBUG(Common, "Could not restrict thread to efficiency cores: {}", GetLastErrorMsg());
+    }
+#endif
 }
 
 #ifdef ARCHITECTURE_x86_64

src/common/thread.h

@@ -99,8 +99,14 @@ enum class ThreadPriority : u32 {
     Critical = 4,
 };
 
+enum class ThreadPlacement : u32 {
+    Default = 0,
+    Background = 1,
+};
+
 void SetCurrentThreadPriority(ThreadPriority new_priority);
 void SetCurrentThreadName(const char* name);
-void PinCurrentThreadToPerformanceCore(size_t core_id);
+void SetCurrentThreadToPerformanceCores();
+void SetCurrentThreadToEfficiencyCores();
 
 } // namespace Common

src/common/thread_worker.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
@@ -37,10 +37,15 @@ class StatefulThreadWorker {
     using StateMaker = std::conditional_t<with_state, std::function<StateType()>, DummyCallable>;
 
 public:
-    explicit StatefulThreadWorker(size_t num_workers, std::string name, StateMaker func = {})
+    explicit StatefulThreadWorker(size_t num_workers, std::string name, StateMaker func = {},
+                                  ThreadPlacement placement = ThreadPlacement::Default)
         : workers_queued{num_workers}, thread_name{std::move(name)} {
-        const auto lambda = [this, func](std::stop_token stop_token) {
+        const auto lambda = [this, func, placement](std::stop_token stop_token) {
             Common::SetCurrentThreadName(thread_name.c_str());
+            if (placement == ThreadPlacement::Background) {
+                Common::SetCurrentThreadPriority(ThreadPriority::Low);
+                Common::SetCurrentThreadToEfficiencyCores();
+            }
             {
                 [[maybe_unused]] std::conditional_t<with_state, StateType, int> state{func()};
                 while (!stop_token.stop_requested()) {

src/common/uuid.cpp

@@ -208,12 +208,4 @@ UUID UUID::MakeRandomRFC4122V4() {
     return uuid;
 }
 
-UUID UUID::MakeRFC4122V5(std::span<u8, 20> sha1) {
-    UUID uuid{};
-    std::memcpy(&uuid.uuid, sha1.data(), sizeof(UUID));
-    uuid.uuid[8] = 0x80 | (uuid.uuid[8] & 0x3F);
-    uuid.uuid[6] = 0x50 | (uuid.uuid[6] & 0xF);
-    return uuid;
-}
-
 } // namespace Common

src/common/uuid.h

@@ -1,6 +1,3 @@
-// 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-2.0-or-later
 
@@ -8,7 +5,6 @@
 
 #include <array>
 #include <functional>
-#include <span>
 #include <string>
 
 #include "common/common_types.h"
@@ -90,20 +86,28 @@ struct UUID {
         };
     }
 
-    /// @brief Creates a random UUID.
-    /// @returns A random UUID.
-    [[nodiscard]] static UUID MakeRandom();
+    /**
+     * Creates a random UUID.
+     *
+     * @returns A random UUID.
+     */
+    static UUID MakeRandom();
 
-    /// @brief Creates a random UUID with a seed.
-    /// @param seed A seed to initialize the Mersenne-Twister RNG
-    /// @returns A random UUID.
-    [[nodiscard]] static UUID MakeRandomWithSeed(u32 seed);
+    /**
+     * Creates a random UUID with a seed.
+     *
+     * @param seed A seed to initialize the Mersenne-Twister RNG
+     *
+     * @returns A random UUID.
+     */
+    static UUID MakeRandomWithSeed(u32 seed);
 
-    /// @brief Creates a random UUID. The generated UUID is RFC 4122 Version 4 compliant.
-    /// @returns A random UUID that is RFC 4122 Version 4 compliant.
-    [[nodiscard]] static UUID MakeRandomRFC4122V4();
-
-    [[nodiscard]] static UUID MakeRFC4122V5(std::span<u8, 20> sha1);
+    /**
+     * Creates a random UUID. The generated UUID is RFC 4122 Version 4 compliant.
+     *
+     * @returns A random UUID that is RFC 4122 Version 4 compliant.
+     */
+    static UUID MakeRandomRFC4122V4();
 
     friend constexpr bool operator==(const UUID& lhs, const UUID& rhs) = default;
 };

src/core/cpu_manager.cpp

@@ -174,12 +174,7 @@ void CpuManager::RunThread(std::stop_token token, std::size_t core) {
     std::string name = is_multicore ? ("CPUCore_" + std::to_string(core)) : std::string{"CPUThread"};
     Common::SetCurrentThreadName(name.c_str());
     Common::SetCurrentThreadPriority(Common::ThreadPriority::Critical);
-#ifdef __ANDROID__
-    // Aimed specifically for Snapdragon 8 Elite devices
-    // This kills performance on desktop, but boosts perf for UMA devices
-    // like the S8E. Mediatek and Mali likely won't suffer.
-    Common::PinCurrentThreadToPerformanceCore(core);
-#endif
+    Common::SetCurrentThreadToPerformanceCores();
     auto& data = core_data[core];
     data.host_context = Common::Fiber::ThreadToFiber();
 

src/core/file_sys/control_metadata.h

@@ -274,22 +274,23 @@ public:
     explicit NACP(VirtualFile file);
     ~NACP();
 
-    [[nodiscard]] const LanguageEntry& GetLanguageEntry() const;
-    [[nodiscard]] std::string GetApplicationName() const;
-    [[nodiscard]] std::string GetDeveloperName() const;
-    [[nodiscard]] u64 GetTitleId() const;
-    [[nodiscard]] u64 GetDLCBaseTitleId() const;
-    [[nodiscard]] std::string GetVersionString() const;
-    [[nodiscard]] u64 GetDefaultNormalSaveSize() const;
-    [[nodiscard]] u64 GetDefaultJournalSaveSize() const;
-    [[nodiscard]] u32 GetSupportedLanguages() const;
-    [[nodiscard]] std::vector<std::string> GetApplicationNames() const;
-    [[nodiscard]] std::vector<u8> GetRawBytes() const;
-    [[nodiscard]] bool GetUserAccountSwitchLock() const;
-    [[nodiscard]] u64 GetDeviceSaveDataSize() const;
-    [[nodiscard]] u32 GetParentalControlFlag() const;
-    [[nodiscard]] const std::array<u8, 0x20>& GetRatingAge() const;
+    const LanguageEntry& GetLanguageEntry() const;
+    std::string GetApplicationName() const;
+    std::string GetDeveloperName() const;
+    u64 GetTitleId() const;
+    u64 GetDLCBaseTitleId() const;
+    std::string GetVersionString() const;
+    u64 GetDefaultNormalSaveSize() const;
+    u64 GetDefaultJournalSaveSize() const;
+    u32 GetSupportedLanguages() const;
+    std::vector<std::string> GetApplicationNames() const;
+    std::vector<u8> GetRawBytes() const;
+    bool GetUserAccountSwitchLock() const;
+    u64 GetDeviceSaveDataSize() const;
+    u32 GetParentalControlFlag() const;
+    const std::array<u8, 0x20>& GetRatingAge() const;
 
+private:
     RawNACP raw{};
     std::vector<LanguageEntry> language_entries;
 };

src/core/file_sys/patch_manager.cpp

@@ -1156,14 +1156,4 @@ PatchManager::Metadata PatchManager::ParseControlNCA(const NCA& nca) const {
 
     return {std::move(nacp), icon_file};
 }
-
-[[nodiscard]] PatchManager::Metadata PatchManager::GetMetadataFromBaseOrUpdate(Core::System& system, u64 application_id) noexcept {
-    const FileSys::PatchManager pm{application_id, system.GetFileSystemController(), system.GetContentProvider()};
-    auto metadata = pm.GetControlMetadata();
-    if (metadata.first != nullptr)
-        return metadata;
-    const FileSys::PatchManager pm_update{FileSys::GetUpdateTitleID(application_id), system.GetFileSystemController(), system.GetContentProvider()};
-    return pm_update.GetControlMetadata();
-}
-
 } // namespace FileSys

src/core/file_sys/patch_manager.h

@@ -105,9 +105,6 @@ public:
     // Version of GetControlMetadata that takes an arbitrary NCA
     [[nodiscard]] Metadata ParseControlNCA(const NCA& nca) const;
 
-    /// @brief Gets NACP metadata (accounting for any patches or updates)
-    [[nodiscard]] static PatchManager::Metadata GetMetadataFromBaseOrUpdate(Core::System& system, u64 application_id) noexcept;
-
 private:
     [[nodiscard]] std::vector<VirtualFile> CollectPatches(const std::vector<VirtualDir>& patch_dirs,
                                                           const std::string& build_id) const;

src/core/hle/service/am/service/application_functions.cpp

@@ -4,8 +4,6 @@
 // SPDX-FileCopyrightText: Copyright 2024 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-#include <openssl/evp.h>
-
 #include "common/settings.h"
 #include "common/uuid.h"
 #include "core/file_sys/control_metadata.h"
@@ -156,7 +154,20 @@ Result IApplicationFunctions::GetDesiredLanguage(Out<u64> out_language_code) {
     // Default to 0 (all languages supported)
     u32 supported_languages = 0;
 
-    const auto res = FileSys::PatchManager::GetMetadataFromBaseOrUpdate(system, m_applet->program_id);
+    const auto res = [this] {
+        const FileSys::PatchManager pm{m_applet->program_id, system.GetFileSystemController(),
+                                       system.GetContentProvider()};
+        auto metadata = pm.GetControlMetadata();
+        if (metadata.first != nullptr) {
+            return metadata;
+        }
+
+        const FileSys::PatchManager pm_update{FileSys::GetUpdateTitleID(m_applet->program_id),
+                                              system.GetFileSystemController(),
+                                              system.GetContentProvider()};
+        return pm_update.GetControlMetadata();
+    }();
+
     if (res.first != nullptr) {
         supported_languages = res.first->GetSupportedLanguages();
     }
@@ -194,7 +205,20 @@ Result IApplicationFunctions::SetTerminateResult(Result terminate_result) {
 Result IApplicationFunctions::GetDisplayVersion(Out<DisplayVersion> out_display_version) {
     LOG_DEBUG(Service_AM, "called");
 
-    const auto res = FileSys::PatchManager::GetMetadataFromBaseOrUpdate(system, m_applet->program_id);
+    const auto res = [this] {
+        const FileSys::PatchManager pm{m_applet->program_id, system.GetFileSystemController(),
+                                       system.GetContentProvider()};
+        auto metadata = pm.GetControlMetadata();
+        if (metadata.first != nullptr) {
+            return metadata;
+        }
+
+        const FileSys::PatchManager pm_update{FileSys::GetUpdateTitleID(m_applet->program_id),
+                                              system.GetFileSystemController(),
+                                              system.GetContentProvider()};
+        return pm_update.GetControlMetadata();
+    }();
+
     if (res.first != nullptr) {
         const auto& version = res.first->GetVersionString();
         std::memcpy(out_display_version->string.data(), version.data(),
@@ -323,21 +347,8 @@ Result IApplicationFunctions::NotifyRunning(Out<bool> out_became_running) {
 }
 
 Result IApplicationFunctions::GetPseudoDeviceId(Out<Common::UUID> out_pseudo_device_id) {
-    LOG_WARNING(Service_AM, "(stubbed)");
-
-    // This should be hashed with the device specific hash
-    // for now this will do
-    const auto res = FileSys::PatchManager::GetMetadataFromBaseOrUpdate(system, m_applet->program_id);
-    u8 hash[EVP_MAX_MD_SIZE];
-    unsigned int hash_len = 0;
-    auto const seed = res.first->raw.seed_for_pseudo_device_id;
-    EVP_MD_CTX *ctx = EVP_MD_CTX_new();
-    auto const algorithm = EVP_sha1();
-    EVP_DigestInit_ex(ctx, algorithm, nullptr);
-    EVP_DigestUpdate(ctx, &seed, sizeof(seed));
-    EVP_DigestFinal_ex(ctx, hash, &hash_len);
-    EVP_MD_CTX_free(ctx);
-    *out_pseudo_device_id = Common::UUID::MakeRFC4122V5(std::span<u8, 20>{hash, std::size(hash)});
+    LOG_WARNING(Service_AM, "(STUBBED) called");
+    *out_pseudo_device_id = {};
     R_SUCCEED();
 }
 

src/core/hle/service/am/service/library_applet_self_accessor.cpp

@@ -252,7 +252,19 @@ Result ILibraryAppletSelfAccessor::GetMainAppletApplicationDesiredLanguage(
     // Default to 0 (all languages supported)
     u32 supported_languages = 0;
 
-    const auto res = FileSys::PatchManager::GetMetadataFromBaseOrUpdate(system, identity.application_id);
+    const auto res = [this, identity] {
+        const FileSys::PatchManager pm{identity.application_id, system.GetFileSystemController(),
+                                       system.GetContentProvider()};
+        auto metadata = pm.GetControlMetadata();
+        if (metadata.first != nullptr) {
+            return metadata;
+        }
+
+        const FileSys::PatchManager pm_update{FileSys::GetUpdateTitleID(identity.application_id),
+                                              system.GetFileSystemController(),
+                                              system.GetContentProvider()};
+        return pm_update.GetControlMetadata();
+    }();
 
     if (res.first != nullptr) {
         supported_languages = res.first->GetSupportedLanguages();

src/hid_core/frontend/emulated_console.cpp

@@ -100,7 +100,6 @@ void EmulatedConsole::ReloadInput() {
     motion.gyro = emulated_motion.GetGyroscope();
     motion.rotation = emulated_motion.GetRotations();
     motion.orientation = emulated_motion.GetOrientation();
-    motion.quaternion = emulated_motion.GetQuaternion();
     motion.is_at_rest = !emulated_motion.IsMoving(motion_sensitivity);
 
     // Unique index for identifying touch device source

src/video_core/CMakeLists.txt

@@ -33,7 +33,7 @@ add_library(video_core STATIC
     control/channel_state_cache.h
     control/scheduler.cpp
     control/scheduler.h
-    delayed_destruction_ring.h
+    deferred_destruction_queue.h
     dirty_flags.cpp
     dirty_flags.h
     dma_pusher.cpp
@@ -158,8 +158,6 @@ add_library(video_core STATIC
     renderer_vulkan/vk_compute_pass.h
     renderer_vulkan/vk_compute_pipeline.cpp
     renderer_vulkan/vk_compute_pipeline.h
-    renderer_vulkan/vk_descriptor_buffer.cpp
-    renderer_vulkan/vk_descriptor_buffer.h
     renderer_vulkan/vk_descriptor_pool.cpp
     renderer_vulkan/vk_descriptor_pool.h
     renderer_vulkan/vk_fence_manager.cpp

src/video_core/buffer_cache/buffer_cache.h

@@ -31,44 +31,77 @@ BufferCache<P>::BufferCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, R
     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;
+        memory_budget = FALLBACK_MEMORY_BUDGET;
         return;
     }
 
-    const s64 device_local_memory = static_cast<s64>(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<u64>(
-        (std::max)((std::min)(device_local_memory - min_vacancy_expected, min_spacing_expected),
-                 DEFAULT_EXPECTED_MEMORY));
-    critical_memory = static_cast<u64>(
-        (std::max)((std::min)(device_local_memory - min_vacancy_critical, min_spacing_critical),
-                 DEFAULT_CRITICAL_MEMORY));
+    memory_budget = runtime.GetDeviceLocalMemory();
 }
 
 template <class P>
 BufferCache<P>::~BufferCache() = default;
 
 template <class P>
-void BufferCache<P>::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 auto clean_up = [this, &num_iterations](BufferId buffer_id) {
-        if (num_iterations == 0) {
+u64 BufferCache<P>::DeviceUsage(bool force_refresh) {
+    if (!runtime.CanReportAllocationUsage()) {
+        return total_used_memory;
+    }
+    if (force_refresh || usage_refresh_countdown == 0) {
+        cached_device_usage = runtime.GetDeviceAllocationUsage();
+        usage_refresh_countdown = USAGE_REFRESH_INTERVAL;
+    } else {
+        --usage_refresh_countdown;
+    }
+    return cached_device_usage;
+}
+
+template <class P>
+u64 BufferCache<P>::ReclaimMemory(u64 target_bytes, bool allow_download) {
+    if (target_bytes == 0 || in_reclaim) {
+        return 0;
+    }
+    in_reclaim = true;
+    sentenced_buffers.Reclaim(runtime.CompletedSyncPoint());
+    u64 freed = 0;
+    const auto clean_up = [&](BufferId buffer_id) {
+        if (freed >= target_bytes) {
             return true;
         }
-        --num_iterations;
         auto& buffer = slot_buffers[buffer_id];
+        if (!allow_download && IsRegionGpuModified(buffer.CpuAddr(), buffer.SizeBytes())) {
+            return false;
+        }
+        const u64 buffer_bytes = Common::AlignUp(buffer.SizeBytes(), 1024);
         DownloadBufferMemory(buffer);
         DeleteBuffer(buffer_id);
+        freed += buffer_bytes;
         return false;
     };
-    lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, clean_up);
+    const u64 cold_tick =
+        frame_tick > RECLAIM_GUARD_FRAMES ? frame_tick - RECLAIM_GUARD_FRAMES : 0;
+    lru_cache.ForEachItemBelow(cold_tick, clean_up);
+    if (freed < target_bytes) {
+        lru_cache.ForEachItemBelow(frame_tick > 0 ? frame_tick - 1 : 0, clean_up);
+    }
+    sentenced_buffers.Reclaim(runtime.CompletedSyncPoint());
+    in_reclaim = false;
+    usage_refresh_countdown = 0;
+    reclaim_stalled = freed == 0;
+    return freed;
+}
+
+template <class P>
+void BufferCache<P>::EnsureHeadroom(bool allow_download) {
+    if (reclaim_stalled) {
+        return;
+    }
+    const u64 limit = memory_budget > RECLAIM_HEADROOM ? memory_budget - RECLAIM_HEADROOM : 0;
+    const u64 usage = DeviceUsage(false);
+    if (usage <= limit) {
+        return;
+    }
+    const u64 target = (limit / 100) * RECLAIM_TARGET_PERCENT;
+    ReclaimMemory((std::min)(usage - target, total_used_memory), allow_download);
 }
 
 template <class P>
@@ -96,15 +129,11 @@ void BufferCache<P>::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();
-    }
-    if (total_used_memory >= minimum_memory) {
-        RunGarbageCollector();
-    }
+    usage_refresh_countdown = 0;
+    reclaim_stalled = false;
+    EnsureHeadroom(true);
     ++frame_tick;
-    delayed_destruction_ring.Tick();
+    sentenced_buffers.Reclaim(runtime.CompletedSyncPoint());
 
     for (auto& buffer : async_buffers_death_ring) {
         runtime.FreeDeferredStagingBuffer(buffer);
@@ -1576,6 +1605,7 @@ void BufferCache<P>::JoinOverlap(BufferId new_buffer_id, BufferId overlap_id,
 
 template <class P>
 BufferId BufferCache<P>::CreateBuffer(DAddr device_addr, u32 wanted_size) {
+    EnsureHeadroom(false);
     DAddr device_addr_end = Common::AlignUp(device_addr + wanted_size, CACHING_PAGESIZE);
     device_addr = Common::AlignDown(device_addr, CACHING_PAGESIZE);
     wanted_size = static_cast<u32>(device_addr_end - device_addr);
@@ -1613,7 +1643,7 @@ void BufferCache<P>::ChangeRegister(BufferId buffer_id) {
         total_used_memory += Common::AlignUp(size, 1024);
         buffer.setLRUID(lru_cache.Insert(buffer_id, frame_tick));
     } else {
-        total_used_memory -= Common::AlignUp(size, 1024);
+        total_used_memory -= std::min<u64>(total_used_memory, Common::AlignUp(size, 1024));
         lru_cache.Free(buffer.getLRUID());
     }
     const DAddr device_addr_begin = buffer.CpuAddr();
@@ -1872,7 +1902,7 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
 #ifdef YUZU_LEGACY
     if (!do_not_mark || !immediately_free)
 #endif
-        delayed_destruction_ring.Push(std::move(slot_buffers[buffer_id]));
+        sentenced_buffers.Push(std::move(slot_buffers[buffer_id]), runtime.CurrentSyncPoint());
 
     slot_buffers.erase(buffer_id);
 

src/video_core/buffer_cache/buffer_cache_base.h

@@ -9,6 +9,7 @@
 #include <algorithm>
 #include <array>
 #include <bit>
+#include <deque>
 #include <functional>
 #include <memory>
 #include <mutex>
@@ -30,7 +31,7 @@
 #include "common/slot_vector.h"
 #include "video_core/buffer_cache/buffer_base.h"
 #include "video_core/control/channel_state_cache.h"
-#include "video_core/delayed_destruction_ring.h"
+#include "video_core/deferred_destruction_queue.h"
 #include "video_core/dirty_flags.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/kepler_compute.h"
@@ -182,13 +183,15 @@ class BufferCache : public VideoCommon::ChannelSetupCaches<BufferCacheChannelInf
     static constexpr bool USE_MEMORY_MAPS_FOR_UPLOADS = P::USE_MEMORY_MAPS_FOR_UPLOADS;
 
 #ifdef YUZU_LEGACY
-    static constexpr s64 TARGET_THRESHOLD = 3_GiB;
+    static constexpr u64 RECLAIM_HEADROOM = 384_MiB;
 #else
-    static constexpr s64 TARGET_THRESHOLD = 4_GiB;
+    static constexpr u64 RECLAIM_HEADROOM = 512_MiB;
 #endif
 
-    static constexpr s64 DEFAULT_EXPECTED_MEMORY = 512_MiB;
-    static constexpr s64 DEFAULT_CRITICAL_MEMORY = 1_GiB;
+    static constexpr u64 FALLBACK_MEMORY_BUDGET = 2_GiB;
+    static constexpr u32 USAGE_REFRESH_INTERVAL = 16;
+    static constexpr u64 RECLAIM_GUARD_FRAMES = 8;
+    static constexpr u64 RECLAIM_TARGET_PERCENT = 88;
 
     // Debug Flags.
 
@@ -215,6 +218,8 @@ public:
 
     void TickFrame();
 
+    u64 ReclaimMemory(u64 target_bytes, bool allow_download);
+
     void WriteMemory(DAddr device_addr, u64 size);
 
     void CachedWriteMemory(DAddr device_addr, u64 size);
@@ -358,7 +363,9 @@ private:
                ((device_addr + size) & ~Core::DEVICE_PAGEMASK);
     }
 
-    void RunGarbageCollector();
+    u64 DeviceUsage(bool force_refresh);
+
+    void EnsureHeadroom(bool allow_download);
 
     void BindHostIndexBuffer();
 
@@ -475,12 +482,7 @@ private:
     Tegra::MaxwellDeviceMemoryManager& device_memory;
 
     Common::SlotVector<Buffer> slot_buffers;
-#ifdef YUZU_LEGACY
-    static constexpr size_t TICKS_TO_DESTROY = 6;
-#else
-    static constexpr size_t TICKS_TO_DESTROY = 8;
-#endif
-    DelayedDestructionRing<Buffer, TICKS_TO_DESTROY> delayed_destruction_ring;
+    DeferredDestructionQueue<Buffer> sentenced_buffers;
 
     const Tegra::Engines::Maxwell3D::DrawManager::IndirectParams* current_draw_indirect{};
 
@@ -515,8 +517,11 @@ private:
     Common::LeastRecentlyUsedCache<LRUItemParams> lru_cache;
     u64 frame_tick = 0;
     u64 total_used_memory = 0;
-    u64 minimum_memory = 0;
-    u64 critical_memory = 0;
+    u64 memory_budget = 0;
+    u64 cached_device_usage = 0;
+    u32 usage_refresh_countdown = 0;
+    bool in_reclaim = false;
+    bool reclaim_stalled = false;
     BufferId inline_buffer_id;
 #ifdef YUZU_LEGACY
     bool immediately_free = false;

src/video_core/deferred_destruction_queue.h

@@ -0,0 +1,56 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include <cstddef>
+#include <utility>
+
+#include <boost/container/deque.hpp>
+#include <boost/container/options.hpp>
+
+#include "common/common_types.h"
+
+namespace VideoCommon {
+
+template <typename T>
+class DeferredDestructionQueue {
+public:
+    void Push(T&& object, u64 sync_point) {
+        entries.emplace_back(std::move(object), sync_point);
+    }
+
+    void Reclaim(u64 completed_sync_point) {
+        while (!entries.empty() && entries.front().sync_point <= completed_sync_point) {
+            entries.pop_front();
+        }
+    }
+
+    void Clear() {
+        entries.clear();
+    }
+
+    [[nodiscard]] size_t Size() const noexcept {
+        return entries.size();
+    }
+
+    [[nodiscard]] bool Empty() const noexcept {
+        return entries.empty();
+    }
+
+private:
+    struct Entry {
+        Entry(T&& object_, u64 sync_point_) noexcept
+            : object{std::move(object_)}, sync_point{sync_point_} {}
+
+        T object;
+        u64 sync_point;
+    };
+
+    using EntryDequeOptions =
+        boost::container::deque_options<boost::container::block_size<8u>>::type;
+
+    boost::container::deque<Entry, void, EntryDequeOptions> entries;
+};
+
+} // namespace VideoCommon

src/video_core/delayed_destruction_ring.h

@@ -1,34 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2025 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 <array>
-#include <cstddef>
-#include <utility>
-#include <vector>
-
-namespace VideoCommon {
-
-/// Container to push objects to be destroyed a few ticks in the future
-template <typename T, size_t TICKS_TO_DESTROY>
-class DelayedDestructionRing {
-public:
-    void Tick() {
-        index = (index + 1) % TICKS_TO_DESTROY;
-        elements[index].clear();
-    }
-
-    void Push(T&& object) {
-        elements[index].push_back(std::move(object));
-    }
-
-private:
-    size_t index = 0;
-    std::array<std::vector<T>, TICKS_TO_DESTROY> elements;
-};
-
-} // namespace VideoCommon

src/video_core/fence_manager.h

@@ -18,7 +18,7 @@
 #include "common/common_types.h"
 #include "common/settings.h"
 #include "common/thread.h"
-#include "video_core/delayed_destruction_ring.h"
+#include "video_core/deferred_destruction_queue.h"
 #include "video_core/gpu.h"
 #include "video_core/host1x/host1x.h"
 #include "video_core/host1x/syncpoint_manager.h"
@@ -50,7 +50,8 @@ public:
     /// Notify the fence manager about a new frame
     void TickFrame() {
         std::unique_lock lock(ring_guard);
-        delayed_destruction_ring.Tick();
+        ++retire_tick;
+        sentenced_fences.Reclaim(retire_tick > RETIRE_DELAY ? retire_tick - RETIRE_DELAY : 0);
     }
 
     // Unlike other fences, this one doesn't
@@ -186,7 +187,7 @@ private:
             }
             {
                 std::unique_lock lock(ring_guard);
-                delayed_destruction_ring.Push(std::move(current_fence));
+                sentenced_fences.Push(std::move(current_fence), retire_tick);
             }
             fences.pop();
         }
@@ -219,7 +220,7 @@ private:
             }
             {
                 std::unique_lock lock(ring_guard);
-                delayed_destruction_ring.Push(std::move(current_fence));
+                sentenced_fences.Push(std::move(current_fence), retire_tick);
             }
         }
     }
@@ -264,7 +265,9 @@ private:
 
     std::jthread fence_thread;
 
-    DelayedDestructionRing<TFence, 8> delayed_destruction_ring;
+    static constexpr u64 RETIRE_DELAY = 8;
+    u64 retire_tick = 1;
+    DeferredDestructionQueue<TFence> sentenced_fences;
 };
 
 } // namespace VideoCommon

src/video_core/gpu_thread.cpp

@@ -30,6 +30,7 @@ void ThreadManager::StartThread(VideoCore::RendererBase& renderer, Core::Fronten
     thread = std::jthread([&](std::stop_token stop_token) {
         Common::SetCurrentThreadName("GPU");
         Common::SetCurrentThreadPriority(Common::ThreadPriority::Critical);
+        Common::SetCurrentThreadToPerformanceCores();
         system.RegisterHostThread();
 
         auto current_context = context.Acquire();

src/video_core/host_shaders/CMakeLists.txt

@@ -32,6 +32,7 @@ set(SHADER_FILES
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_msaa_to_non_msaa.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_non_msaa_to_msaa.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_non_msaa_to_msaa.frag
+    ${CMAKE_CURRENT_SOURCE_DIR}/convert_non_msaa_to_msaa_depth.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_s8d24_to_abgr8.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/full_screen_triangle.vert
     ${CMAKE_CURRENT_SOURCE_DIR}/fxaa.frag

src/video_core/host_shaders/astc_decoder.comp

@@ -77,8 +77,14 @@ uvec4 local_buff;
 uvec4 color_endpoint_data;
 int color_bitsread = 0;
 
-#define MAX_WEIGHT_VALUES 64
-uint result_vector[MAX_WEIGHT_VALUES];
+// 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 DIVCEIL(number, divisor) (number + divisor - 1) / divisor
+#define ARRAY_NUM_ELEMENTS 144
+#define VECTOR_ARRAY_SIZE DIVCEIL(ARRAY_NUM_ELEMENTS * 2, 4)
+uint result_vector[ARRAY_NUM_ELEMENTS * 2];
 
 int result_index = 0;
 uint result_vector_max_index;
@@ -486,7 +492,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits, o
         A = ReplicateBitTo9((bitval & 1));
         switch (encoding) {
         case JUST_BITS:
-            color_values[out_index++] = FastReplicateTo8(bitval, bitlen);
+            color_values[++out_index] = FastReplicateTo8(bitval, bitlen);
             break;
         case TRIT: {
             D = QuintTritValue(val);
@@ -565,7 +571,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits, o
             uint T = (D * C) + B;
             T ^= A;
             T = (A & 0x80) | (T >> 2);
-            color_values[out_index++] = T;
+            color_values[++out_index] = T;
         }
     }
 }
@@ -747,12 +753,12 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, ui
 #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 / 4][i % 4] = 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 / 4][i % 4] = int(color_values[++colvals_index]);                      \
     }
 
     switch (color_endpoint_mode) {
@@ -1219,10 +1225,6 @@ void DecompressBlock(ivec3 coord) {
         FillError(coord);
         return;
     }
-    if (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;
@@ -1237,10 +1239,6 @@ void DecompressBlock(ivec3 coord) {
     const uint base_mode = base_cem & 3;
     const uint max_weight = DecodeMaxWeight(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) {
@@ -1255,7 +1253,6 @@ void DecompressBlock(ivec3 coord) {
             extra_cem_bits += 8;
             break;
         default:
-            FillError(coord);
             return;
         }
     }
@@ -1265,7 +1262,6 @@ void DecompressBlock(ivec3 coord) {
     if (remaining_bits > 128) {
         // Bad data, more remaining bits than 4 bytes
         // return early
-        FillError(coord);
         return;
     }
     // Read color data...

src/video_core/host_shaders/convert_non_msaa_to_msaa_depth.frag

@@ -0,0 +1,19 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#version 450 core
+
+layout(binding = 0) uniform sampler2D img_in;
+
+layout(push_constant) uniform PushConstants {
+    ivec2 dst_offset;
+    ivec2 src_offset;
+    ivec2 scale;
+};
+
+void main() {
+    const ivec2 msaa_coord = ivec2(gl_FragCoord.xy) - dst_offset;
+    const ivec2 sample_offset = ivec2(gl_SampleID % scale.x, gl_SampleID / scale.x);
+    const ivec2 coord = msaa_coord * scale + sample_offset + src_offset;
+    gl_FragDepth = texelFetch(img_in, coord, 0).r;
+}

src/video_core/macro.cpp

@@ -417,6 +417,14 @@ void HLE_TransformFeedbackSetup::Execute(Core::System& system, Engines::Maxwell3
     default: return std::monostate{};
     }
 }
+[[nodiscard]] inline bool CanBeHLEProgram(u64 hash) noexcept {
+    switch (hash) {
+#define HLE_MACRO_ELEM(HASH, TY, VAL) case HASH: return true;
+    HLE_MACRO_LIST
+#undef HLE_MACRO_ELEM
+    default: return false;
+    }
+}
 
 void MacroInterpreterImpl::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, std::span<const u32> params, u32 method) {
     Reset();
@@ -1337,92 +1345,80 @@ static void Dump(u64 hash, std::span<const u32> code, bool decompiled = false) {
     macro_file.write(reinterpret_cast<const char*>(code.data()), code.size_bytes());
 }
 
-void MacroEngine::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, u32 method,
-                          std::span<const u32> parameters) {
-    const auto execute_variant = [&system, &maxwell3d, &parameters,
-                                  method](AnyCachedMacro& cached) {
-        if (std::holds_alternative<MacroInterpreterImpl>(cached) ||
-            std::holds_alternative<std::unique_ptr<DynamicCachedMacro>>(cached) ||
-            Settings::values.disable_macro_hle) {
-            maxwell3d.RefreshParameters();
-        }
-
-        if (auto program = std::get_if<HLE_DrawArraysIndirect>(&cached))
-            return program->Execute(system, maxwell3d, parameters, method);
-        if (auto program = std::get_if<HLE_DrawIndexedIndirect>(&cached))
-            return program->Execute(system, maxwell3d, parameters, method);
-        if (auto program = std::get_if<HLE_MultiDrawIndexedIndirectCount>(&cached))
-            return program->Execute(system, maxwell3d, parameters, method);
-        if (auto program = std::get_if<HLE_MultiLayerClear>(&cached))
-            return program->Execute(system, maxwell3d, parameters, method);
-        if (auto program = std::get_if<HLE_C713C83D8F63CCF3>(&cached))
-            return program->Execute(system, maxwell3d, parameters, method);
-        if (auto program = std::get_if<HLE_D7333D26E0A93EDE>(&cached))
-            return program->Execute(system, maxwell3d, parameters, method);
-        if (auto program = std::get_if<HLE_BindShader>(&cached))
-            return program->Execute(system, maxwell3d, parameters, method);
-        if (auto program = std::get_if<HLE_SetRasterBoundingBox>(&cached))
-            return program->Execute(system, maxwell3d, parameters, method);
-        if (auto program = std::get_if<HLE_ClearConstBuffer>(&cached))
-            return program->Execute(system, maxwell3d, parameters, method);
-        if (auto program = std::get_if<HLE_ClearMemory>(&cached))
-            return program->Execute(system, maxwell3d, parameters, method);
-        if (auto program = std::get_if<HLE_TransformFeedbackSetup>(&cached))
-            return program->Execute(system, maxwell3d, parameters, method);
-        if (auto program = std::get_if<HLE_DrawIndirectByteCount>(&cached))
-            return program->Execute(system, maxwell3d, parameters, method);
-        if (auto program = std::get_if<MacroInterpreterImpl>(&cached))
-            return program->Execute(system, maxwell3d, parameters, method);
-        if (auto program = std::get_if<std::unique_ptr<DynamicCachedMacro>>(&cached))
-            return program->get()->Execute(system, maxwell3d, parameters, method);
-
-        UNREACHABLE();
+void MacroEngine::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, u32 method, std::span<const u32> parameters) {
+    auto const execute_variant = [&system, &maxwell3d, &parameters, method](AnyCachedMacro& acm) {
+        if (auto a = std::get_if<HLE_DrawArraysIndirect>(&acm))
+            return a->Execute(system, maxwell3d, parameters, method);
+        if (auto a = std::get_if<HLE_DrawIndexedIndirect>(&acm))
+            return a->Execute(system, maxwell3d, parameters, method);
+        if (auto a = std::get_if<HLE_MultiDrawIndexedIndirectCount>(&acm))
+            return a->Execute(system, maxwell3d, parameters, method);
+        if (auto a = std::get_if<HLE_MultiLayerClear>(&acm))
+            return a->Execute(system, maxwell3d, parameters, method);
+        if (auto a = std::get_if<HLE_C713C83D8F63CCF3>(&acm))
+            return a->Execute(system, maxwell3d, parameters, method);
+        if (auto a = std::get_if<HLE_D7333D26E0A93EDE>(&acm))
+            return a->Execute(system, maxwell3d, parameters, method);
+        if (auto a = std::get_if<HLE_BindShader>(&acm))
+            return a->Execute(system, maxwell3d, parameters, method);
+        if (auto a = std::get_if<HLE_SetRasterBoundingBox>(&acm))
+            return a->Execute(system, maxwell3d, parameters, method);
+        if (auto a = std::get_if<HLE_ClearConstBuffer>(&acm))
+            return a->Execute(system, maxwell3d, parameters, method);
+        if (auto a = std::get_if<HLE_ClearMemory>(&acm))
+            return a->Execute(system, maxwell3d, parameters, method);
+        if (auto a = std::get_if<HLE_TransformFeedbackSetup>(&acm))
+            return a->Execute(system, maxwell3d, parameters, method);
+        if (auto a = std::get_if<HLE_DrawIndirectByteCount>(&acm))
+            return a->Execute(system, maxwell3d, parameters, method);
+        if (auto a = std::get_if<MacroInterpreterImpl>(&acm))
+            return a->Execute(system, maxwell3d, parameters, method);
+        if (auto a = std::get_if<std::unique_ptr<DynamicCachedMacro>>(&acm))
+            return a->get()->Execute(system, maxwell3d, parameters, method);
     };
     if (auto const it = macro_cache.find(method); it != macro_cache.end()) {
-        execute_variant(it->second.program);
-        return;
-    }
-
-    // Macro not compiled, check if it's uploaded and if so, compile it
-    std::span<const u32> code;
-    auto macro_code = uploaded_macro_code.find(method);
-    if (macro_code == uploaded_macro_code.end()) {
+        auto& ci = it->second;
+        if (!CanBeHLEProgram(ci.hash) || Settings::values.disable_macro_hle)
+            maxwell3d.RefreshParameters(); //LLE must reload parameters
+        execute_variant(ci.program);
+    } else {
+        // Macro not compiled, check if it's uploaded and if so, compile it
         std::optional<u32> mid_method;
-        for (const auto& [method_base, uploaded_code] : uploaded_macro_code) {
-            if (method >= method_base && (method - method_base) < uploaded_code.size()) {
-                mid_method = method_base;
-                break;
+        const auto macro_code = uploaded_macro_code.find(method);
+        if (macro_code == uploaded_macro_code.end()) {
+            for (const auto& [method_base, code] : uploaded_macro_code) {
+                if (method >= method_base && (method - method_base) < code.size()) {
+                    mid_method = method_base;
+                    break;
+                }
+            }
+            if (!mid_method.has_value()) {
+                ASSERT_MSG(false, "Macro 0x{0:x} was not uploaded", method);
+                return;
             }
         }
-        if (!mid_method) {
-            ASSERT_MSG(false, "Macro 0x{0:x} was not uploaded", method);
-            return;
+        auto& ci = macro_cache[method];
+        if (mid_method) {
+            const auto& macro_cached = uploaded_macro_code[mid_method.value()];
+            const auto rebased_method = method - mid_method.value();
+            auto& code = uploaded_macro_code[method];
+            code.resize(macro_cached.size() - rebased_method);
+            std::memcpy(code.data(), macro_cached.data() + rebased_method, code.size() * sizeof(u32));
+            ci.hash = Common::HashValue(code);
+            ci.program = Compile(system, maxwell3d, code);
+        } else {
+            ci.program = Compile(system, maxwell3d, macro_code->second);
+            ci.hash = Common::HashValue(macro_code->second);
+        }
+        if (CanBeHLEProgram(ci.hash) && !Settings::values.disable_macro_hle) {
+            ci.program = GetHLEProgram(ci.hash);
+        } else {
+            maxwell3d.RefreshParameters();
+        }
+        execute_variant(ci.program);
+        if (Settings::values.dump_macros) {
+            Dump(ci.hash, macro_code->second, !std::holds_alternative<std::monostate>(ci.program));
         }
-
-        const auto source = uploaded_macro_code.find(*mid_method);
-        ASSERT(source != uploaded_macro_code.end());
-        const auto rebased_method = method - *mid_method;
-        std::vector<u32> rebased_code(source->second.begin() + rebased_method,
-                                      source->second.end());
-        const auto [it, inserted] = uploaded_macro_code.emplace(method, std::move(rebased_code));
-        ASSERT(inserted);
-        code = it->second;
-    } else {
-        code = macro_code->second;
-    }
-
-    auto& ci = macro_cache[method];
-    ci.hash = Common::HashRange(code.begin(), code.end());
-    if (!Settings::values.disable_macro_hle) {
-        ci.program = GetHLEProgram(ci.hash);
-    }
-    if (std::holds_alternative<std::monostate>(ci.program)) {
-        ci.program = Compile(system, maxwell3d, code);
-    }
-
-    execute_variant(ci.program);
-    if (Settings::values.dump_macros) {
-        Dump(ci.hash, code, !std::holds_alternative<std::monostate>(ci.program));
     }
 }
 

src/video_core/renderer_opengl/gl_buffer_cache.h

@@ -93,7 +93,17 @@ public:
     void PostCopyBarrier();
     void Finish();
 
-    void TickFrame(Common::SlotVector<Buffer>&) noexcept {}
+    void TickFrame(Common::SlotVector<Buffer>&) noexcept {
+        ++sync_point;
+    }
+
+    u64 CurrentSyncPoint() const noexcept {
+        return sync_point;
+    }
+
+    u64 CompletedSyncPoint() const noexcept {
+        return sync_point > SYNC_POINT_DELAY ? sync_point - SYNC_POINT_DELAY : 0;
+    }
 
     void ClearBuffer(Buffer& dest_buffer, u32 offset, size_t size, u32 value);
 
@@ -128,6 +138,14 @@ public:
 
     u64 GetDeviceMemoryUsage() const;
 
+    u64 GetDeviceAllocationUsage() const {
+        return GetDeviceMemoryUsage();
+    }
+
+    bool CanReportAllocationUsage() const {
+        return device.CanReportMemoryUsage();
+    }
+
     void BindFastUniformBuffer(size_t stage, u32 binding_index, u32 size) {
         const GLuint handle = fast_uniforms[stage][binding_index].handle;
         const GLsizeiptr gl_size = static_cast<GLsizeiptr>(size);
@@ -213,9 +231,13 @@ private:
         GL_FRAGMENT_PROGRAM_PARAMETER_BUFFER_NV,
     };
 
+    static constexpr u64 SYNC_POINT_DELAY = 8;
+
     const Device& device;
     StagingBufferPool& staging_buffer_pool;
 
+    u64 sync_point = 1;
+
     bool has_fast_buffer_sub_data = false;
     bool use_assembly_shaders = false;
     bool has_unified_vertex_buffers = false;

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -87,6 +87,14 @@ public:
 
     u64 GetDeviceMemoryUsage() const;
 
+    u64 GetDeviceAllocationUsage() const {
+        return GetDeviceMemoryUsage();
+    }
+
+    bool CanReportAllocationUsage() const {
+        return device.CanReportMemoryUsage();
+    }
+
     bool CanReportMemoryUsage() const {
         return device.CanReportMemoryUsage();
     }
@@ -139,7 +147,19 @@ public:
 
     bool HasNativeASTC() const noexcept;
 
-    void TickFrame() {}
+    void TickFrame() {
+        ++sync_point;
+    }
+
+    u64 CurrentSyncPoint() const noexcept {
+        return sync_point;
+    }
+
+    u64 CompletedSyncPoint() const noexcept {
+        return sync_point > SYNC_POINT_DELAY ? sync_point - SYNC_POINT_DELAY : 0;
+    }
+
+    void WaitSyncPoint(u64) {}
 
     StateTracker& GetStateTracker() {
         return state_tracker;
@@ -174,6 +194,9 @@ private:
     std::array<OGLFramebuffer, 4> rescale_read_fbos;
     const Settings::ResolutionScalingInfo& resolution;
     u64 device_access_memory;
+
+    static constexpr u64 SYNC_POINT_DELAY = 8;
+    u64 sync_point = 1;
 };
 
 class Image : public VideoCommon::ImageBase {
@@ -370,6 +393,7 @@ struct TextureCacheParams {
     static constexpr bool HAS_EMULATED_COPIES = true;
     static constexpr bool HAS_DEVICE_MEMORY_INFO = true;
     static constexpr bool IMPLEMENTS_ASYNC_DOWNLOADS = true;
+    static constexpr bool HAS_TIMELINE_SYNC_POINTS = false;
 
     using Runtime = OpenGL::TextureCacheRuntime;
     using Image = OpenGL::Image;

src/video_core/renderer_vulkan/blit_image.cpp

@@ -21,6 +21,7 @@
 #include "video_core/host_shaders/convert_depth_to_float_frag_spv.h"
 #include "video_core/host_shaders/convert_float_to_depth_frag_spv.h"
 #include "video_core/host_shaders/convert_msaa_to_non_msaa_frag_spv.h"
+#include "video_core/host_shaders/convert_non_msaa_to_msaa_depth_frag_spv.h"
 #include "video_core/host_shaders/convert_non_msaa_to_msaa_frag_spv.h"
 #include "video_core/host_shaders/convert_s8d24_to_abgr8_frag_spv.h"
 #include "video_core/host_shaders/full_screen_triangle_vert_spv.h"
@@ -519,7 +520,8 @@ void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view)
 }
 
 [[nodiscard]] vk::ImageView MakeMSAACopyView(const vk::Device& device, VkImage image,
-                                             VkFormat format, u32 base_level) {
+                                             VkFormat format, u32 base_level,
+                                             VkImageAspectFlags aspect_mask) {
     return device.CreateImageView(VkImageViewCreateInfo{
         .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
         .pNext = nullptr,
@@ -534,7 +536,7 @@ void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view)
             .a = VK_COMPONENT_SWIZZLE_IDENTITY,
         },
         .subresourceRange{
-            .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+            .aspectMask = aspect_mask,
             .baseMipLevel = base_level,
             .levelCount = 1,
             .baseArrayLayer = 0,
@@ -610,6 +612,8 @@ BlitImageHelper::BlitImageHelper(const Device& device_, Scheduler& scheduler_,
       convert_s8d24_to_abgr8_frag(BuildShader(device, CONVERT_S8D24_TO_ABGR8_FRAG_SPV)),
       convert_msaa_to_non_msaa_frag(BuildShader(device, CONVERT_MSAA_TO_NON_MSAA_FRAG_SPV)),
       convert_non_msaa_to_msaa_frag(BuildShader(device, CONVERT_NON_MSAA_TO_MSAA_FRAG_SPV)),
+      convert_non_msaa_to_msaa_depth_frag(
+          BuildShader(device, CONVERT_NON_MSAA_TO_MSAA_DEPTH_FRAG_SPV)),
       linear_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_LINEAR>)),
       nearest_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_NEAREST>)) {}
 
@@ -895,16 +899,34 @@ void BlitImageHelper::CopyMSAA(RenderPassCache& render_pass_cache, VkImage dst_i
     const s32 scale_y = 1 << samples_y;
     const VkSampleCountFlagBits samples =
         msaa_to_non_msaa ? VK_SAMPLE_COUNT_1_BIT : SampleCountFlag(num_samples);
+    const auto dst_surface_type = VideoCore::Surface::GetFormatType(dst_format);
+    const bool is_depth = dst_surface_type == VideoCore::Surface::SurfaceType::Depth ||
+                          dst_surface_type == VideoCore::Surface::SurfaceType::DepthStencil;
+    const bool has_stencil = dst_surface_type == VideoCore::Surface::SurfaceType::DepthStencil;
+    const VkImageAspectFlags view_aspect =
+        is_depth ? VK_IMAGE_ASPECT_DEPTH_BIT : VK_IMAGE_ASPECT_COLOR_BIT;
+    VkImageAspectFlags barrier_aspect = VK_IMAGE_ASPECT_COLOR_BIT;
+    if (is_depth) {
+        barrier_aspect = VK_IMAGE_ASPECT_DEPTH_BIT;
+        if (has_stencil) {
+            barrier_aspect |= VK_IMAGE_ASPECT_STENCIL_BIT;
+        }
+    }
     RenderPassKey renderpass_key{};
     renderpass_key.color_formats.fill(VideoCore::Surface::PixelFormat::Invalid);
-    renderpass_key.color_formats[0] = dst_format;
-    renderpass_key.depth_format = VideoCore::Surface::PixelFormat::Invalid;
+    if (is_depth) {
+        renderpass_key.depth_format = dst_format;
+    } else {
+        renderpass_key.color_formats[0] = dst_format;
+        renderpass_key.depth_format = VideoCore::Surface::PixelFormat::Invalid;
+    }
     renderpass_key.samples = samples;
     const VkRenderPass renderpass = render_pass_cache.Get(renderpass_key);
     const MSAACopyPipelineKey key{
         .renderpass = renderpass,
         .samples = samples,
         .msaa_to_non_msaa = msaa_to_non_msaa,
+        .is_depth = is_depth,
     };
     const VkPipeline pipeline = FindOrEmplaceMSAACopyPipeline(key);
     const VkPipelineLayout layout = *msaa_copy_pipeline_layout;
@@ -920,10 +942,10 @@ void BlitImageHelper::CopyMSAA(RenderPassCache& render_pass_cache, VkImage dst_i
         ASSERT(copy.dst_subresource.num_layers == 1);
         vk::ImageView src_view =
             MakeMSAACopyView(device.GetLogical(), src_image, src_vk_format,
-                             static_cast<u32>(copy.src_subresource.base_level));
+                             static_cast<u32>(copy.src_subresource.base_level), view_aspect);
         vk::ImageView dst_view =
             MakeMSAACopyView(device.GetLogical(), dst_image, dst_vk_format,
-                             static_cast<u32>(copy.dst_subresource.base_level));
+                             static_cast<u32>(copy.dst_subresource.base_level), view_aspect);
         const VkOffset2D dst_offset{copy.dst_offset.x, copy.dst_offset.y};
         const VkExtent2D dst_extent{copy.extent.width, copy.extent.height};
         const VkRect2D render_area{
@@ -949,50 +971,64 @@ void BlitImageHelper::CopyMSAA(RenderPassCache& render_pass_cache, VkImage dst_i
         scheduler.RequestOutsideRenderPassOperationContext();
         scheduler.Record([this, pipeline, layout, sampler, renderpass,
                           framebuffer_handle = *framebuffer, src_view_handle = *src_view,
-                          src = src_image, dst = dst_image, render_area,
+                          src = src_image, dst = dst_image, render_area, is_depth, barrier_aspect,
                           push_constants](vk::CommandBuffer cmdbuf) {
-            constexpr VkImageSubresourceRange color_range{
-                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+            const VkImageSubresourceRange src_range{
+                .aspectMask = barrier_aspect,
                 .baseMipLevel = 0,
                 .levelCount = VK_REMAINING_MIP_LEVELS,
                 .baseArrayLayer = 0,
                 .layerCount = VK_REMAINING_ARRAY_LAYERS,
             };
+            const VkImageSubresourceRange dst_range = src_range;
+            const VkAccessFlags attachment_read =
+                is_depth ? VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT
+                         : VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
+            const VkAccessFlags attachment_write =
+                is_depth ? VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT
+                         : VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
+            const VkPipelineStageFlags depth_stage =
+                VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
+                VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
+            const VkPipelineStageFlags attachment_stage =
+                is_depth ? depth_stage
+                         : static_cast<VkPipelineStageFlags>(
+                               VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT);
             const std::array pre_barriers{
                 VkImageMemoryBarrier{
                     .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                     .pNext = nullptr,
-                    .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
-                                     VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
+                    .srcAccessMask = attachment_write | VK_ACCESS_SHADER_WRITE_BIT |
+                                     VK_ACCESS_TRANSFER_WRITE_BIT,
                     .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
                     .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
                     .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                     .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                     .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                     .image = src,
-                    .subresourceRange = color_range,
+                    .subresourceRange = src_range,
                 },
                 VkImageMemoryBarrier{
                     .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                     .pNext = nullptr,
-                    .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
-                                     VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
-                    .dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
-                                     VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
+                    .srcAccessMask = attachment_write | VK_ACCESS_SHADER_WRITE_BIT |
+                                     VK_ACCESS_TRANSFER_WRITE_BIT,
+                    .dstAccessMask = attachment_read | attachment_write,
                     .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
                     .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                     .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                     .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                     .image = dst,
-                    .subresourceRange = color_range,
+                    .subresourceRange = dst_range,
                 },
             };
             cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
+                                       VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
+                                       VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
                                        VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
                                        VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
                                        VK_PIPELINE_STAGE_TRANSFER_BIT,
-                                   VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                                       VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
+                                   VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | attachment_stage,
                                    0, nullptr, nullptr, pre_barriers);
             const VkRenderPassBeginInfo renderpass_bi{
                 .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
@@ -1025,16 +1061,16 @@ void BlitImageHelper::CopyMSAA(RenderPassCache& render_pass_cache, VkImage dst_i
             const VkImageMemoryBarrier post_barrier{
                 .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                 .pNext = nullptr,
-                .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
+                .srcAccessMask = attachment_write,
                 .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT,
                 .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
                 .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                 .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                 .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                 .image = dst,
-                .subresourceRange = color_range,
+                .subresourceRange = dst_range,
             };
-            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
+            cmdbuf.PipelineBarrier(attachment_stage,
                                    VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
                                        VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
                                        VK_PIPELINE_STAGE_TRANSFER_BIT,
@@ -1423,9 +1459,36 @@ VkPipeline BlitImageHelper::FindOrEmplaceMSAACopyPipeline(const MSAACopyPipeline
         return *msaa_copy_pipelines[std::distance(msaa_copy_keys.begin(), it)];
     }
     msaa_copy_keys.push_back(key);
-    const std::array stages = MakeStages(*clear_color_vert, key.msaa_to_non_msaa
-                                                                ? *convert_msaa_to_non_msaa_frag
-                                                                : *convert_non_msaa_to_msaa_frag);
+    const VkShaderModule frag_module =
+        key.msaa_to_non_msaa
+            ? *convert_msaa_to_non_msaa_frag
+            : (key.is_depth ? *convert_non_msaa_to_msaa_depth_frag
+                            : *convert_non_msaa_to_msaa_frag);
+    const std::array stages = MakeStages(*clear_color_vert, frag_module);
+    const VkPipelineDepthStencilStateCreateInfo depth_stencil_ci{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .depthTestEnable = VK_TRUE,
+        .depthWriteEnable = VK_TRUE,
+        .depthCompareOp = VK_COMPARE_OP_ALWAYS,
+        .depthBoundsTestEnable = VK_FALSE,
+        .stencilTestEnable = VK_FALSE,
+        .front = {},
+        .back = {},
+        .minDepthBounds = 0.0f,
+        .maxDepthBounds = 0.0f,
+    };
+    static constexpr VkPipelineColorBlendStateCreateInfo no_color_blend_ci{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .logicOpEnable = VK_FALSE,
+        .logicOp = VK_LOGIC_OP_CLEAR,
+        .attachmentCount = 0,
+        .pAttachments = nullptr,
+        .blendConstants = {0.0f, 0.0f, 0.0f, 0.0f},
+    };
     const VkPipelineMultisampleStateCreateInfo multisample_ci{
         .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
         .pNext = nullptr,
@@ -1450,8 +1513,9 @@ VkPipeline BlitImageHelper::FindOrEmplaceMSAACopyPipeline(const MSAACopyPipeline
         .pViewportState = &PIPELINE_VIEWPORT_STATE_CREATE_INFO,
         .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
         .pMultisampleState = &multisample_ci,
-        .pDepthStencilState = nullptr,
-        .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO,
+        .pDepthStencilState = key.is_depth ? &depth_stencil_ci : nullptr,
+        .pColorBlendState = key.is_depth ? &no_color_blend_ci
+                                         : &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO,
         .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
         .layout = *msaa_copy_pipeline_layout,
         .renderPass = key.renderpass,

src/video_core/renderer_vulkan/blit_image.h

@@ -51,6 +51,7 @@ struct MSAACopyPipelineKey {
     VkRenderPass renderpass;
     VkSampleCountFlagBits samples;
     bool msaa_to_non_msaa;
+    bool is_depth;
 };
 
 struct BlitMSAAPipelineKey {
@@ -180,6 +181,7 @@ private:
     vk::ShaderModule convert_s8d24_to_abgr8_frag;
     vk::ShaderModule convert_msaa_to_non_msaa_frag;
     vk::ShaderModule convert_non_msaa_to_msaa_frag;
+    vk::ShaderModule convert_non_msaa_to_msaa_depth_frag;
     vk::Sampler linear_sampler;
     vk::Sampler nearest_sampler;
 

src/video_core/renderer_vulkan/fixed_pipeline_state.cpp

@@ -164,7 +164,9 @@ void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d, DynamicFe
     }
 
     provoking_vertex_last.Assign(use_last_provoking_vertex ? 1 : 0);
-    conservative_raster_enable.Assign(regs.conservative_raster_enable != 0 ? 1 : 0);
+    if (!features.has_dynamic_state3_conservative_raster_mode) {
+        conservative_raster_enable.Assign(regs.conservative_raster_enable != 0 ? 1 : 0);
+    }
     smooth_lines.Assign(regs.line_anti_alias_enable != 0 ? 1 : 0);
     alpha_to_coverage_enabled.Assign(regs.anti_alias_alpha_control.alpha_to_coverage != 0 ? 1 : 0);
     alpha_to_one_enabled.Assign(regs.anti_alias_alpha_control.alpha_to_one != 0 ? 1 : 0);
@@ -360,18 +362,35 @@ void FixedPipelineState::DynamicState::Refresh2(const Maxwell& regs,
     depth_bias_enable.Assign(enabled_lut[POLYGON_OFFSET_ENABLE_LUT[topology_index]] != 0 ? 1 : 0);
 }
 
+bool IsDepthClipEnabled(const Maxwell& regs) {
+    const auto clip = regs.viewport_clip_control.geometry_clip.Value();
+    return clip == Maxwell::ViewportClipControl::GeometryClip::Passthrough ||
+           clip == Maxwell::ViewportClipControl::GeometryClip::FrustumXYZ ||
+           clip == Maxwell::ViewportClipControl::GeometryClip::FrustumZ;
+}
+
+bool IsDepthClampEnabled(const Maxwell& regs, bool has_depth_clip_enable) {
+    if (!IsDepthClipEnabled(regs)) {
+        return true;
+    }
+    if (!has_depth_clip_enable) {
+        return false;
+    }
+    return regs.viewport_clip_control.pixel_min_z.Value() != 0 ||
+           regs.viewport_clip_control.pixel_max_z.Value() != 0;
+}
+
 void FixedPipelineState::DynamicState::Refresh3(const Maxwell& regs,
                                                 const DynamicFeatures& features) {
     if (!features.has_dynamic_state3_logic_op_enable) {
         logic_op_enable.Assign(regs.logic_op.enable != 0 ? 1 : 0);
     }
+    if (features.has_depth_clip_enable) {
+        depth_clip_disabled.Assign(IsDepthClipEnabled(regs) ? 0 : 1);
+    }
     if (!features.has_dynamic_state3_depth_clamp_enable) {
-        depth_clamp_disabled.Assign(regs.viewport_clip_control.geometry_clip ==
-                                        Maxwell::ViewportClipControl::GeometryClip::Passthrough ||
-                                    regs.viewport_clip_control.geometry_clip ==
-                                        Maxwell::ViewportClipControl::GeometryClip::FrustumXYZ ||
-                                    regs.viewport_clip_control.geometry_clip ==
-                                        Maxwell::ViewportClipControl::GeometryClip::FrustumZ);
+        depth_clamp_disabled.Assign(
+            IsDepthClampEnabled(regs, features.has_depth_clip_enable) ? 0 : 1);
     }
     if (!features.has_dynamic_state3_line_stipple_enable) {
         line_stipple_enable.Assign(regs.line_stipple_enable);

src/video_core/renderer_vulkan/fixed_pipeline_state.h

@@ -30,6 +30,8 @@ struct DynamicFeatures {
     bool has_extended_dynamic_state_3_blend;
     bool has_extended_dynamic_state_3_enables;
     bool has_dynamic_state3_depth_clamp_enable;
+    bool has_dynamic_state3_conservative_raster_mode;
+    bool has_depth_clip_enable;
     bool has_dynamic_state3_logic_op_enable;
     bool has_dynamic_state3_line_stipple_enable;
     bool has_dynamic_vertex_input;
@@ -165,6 +167,7 @@ struct FixedPipelineState {
             BitField<10, 1, u32> logic_op_enable;
             BitField<11, 1, u32> depth_clamp_disabled;
             BitField<12, 1, u32> line_stipple_enable;
+            BitField<13, 1, u32> depth_clip_disabled;
         };
         union {
             u32 raw2;
@@ -298,6 +301,9 @@ static_assert(std::has_unique_object_representations_v<FixedPipelineState>);
 static_assert(std::is_trivially_copyable_v<FixedPipelineState>);
 static_assert(std::is_trivially_constructible_v<FixedPipelineState>);
 
+bool IsDepthClipEnabled(const Maxwell& regs);
+bool IsDepthClampEnabled(const Maxwell& regs, bool has_depth_clip_enable);
+
 } // namespace Vulkan
 
 namespace std {

src/video_core/renderer_vulkan/pipeline_helper.h

@@ -47,93 +47,6 @@ using Shader::Backend::SPIRV::NUM_TEXTURE_AND_IMAGE_SCALING_WORDS;
     return std::nullopt;
 }
 
-[[nodiscard]] inline VkDeviceSize DescriptorSizeForType(const Device& device,
-                                                        VkDescriptorType type) {
-    const auto& props = device.DescriptorBufferProperties();
-    const bool robust = device.IsRobustBufferAccessEnabled();
-    switch (type) {
-    case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
-        return robust ? props.robustUniformBufferDescriptorSize : props.uniformBufferDescriptorSize;
-    case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
-        return robust ? props.robustStorageBufferDescriptorSize : props.storageBufferDescriptorSize;
-    case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
-        return robust ? props.robustUniformTexelBufferDescriptorSize
-                      : props.uniformTexelBufferDescriptorSize;
-    case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
-        return robust ? props.robustStorageTexelBufferDescriptorSize
-                      : props.storageTexelBufferDescriptorSize;
-    case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
-        return props.combinedImageSamplerDescriptorSize;
-    case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
-        return props.storageImageDescriptorSize;
-    default:
-        return 0;
-    }
-}
-
-struct DescriptorBufferBinding {
-    VkDescriptorType type;
-    u32 count;
-    VkDeviceSize offset;
-    VkDeviceSize stride;
-};
-
-struct DescriptorBufferLayout {
-    VkDeviceSize size{};
-    boost::container::small_vector<DescriptorBufferBinding, 32> bindings;
-
-    [[nodiscard]] bool Empty() const noexcept {
-        return bindings.empty();
-    }
-};
-
-inline void WriteDescriptorBuffer(const Device& device, const DescriptorBufferLayout& layout,
-                                  const DescriptorUpdateEntry* payload, u8* host) {
-    const vk::Device& dev = device.GetLogical();
-    for (const DescriptorBufferBinding& binding : layout.bindings) {
-        for (u32 index = 0; index < binding.count; ++index) {
-            const DescriptorUpdateEntry& entry = *(payload++);
-            const VkDescriptorAddressInfoEXT address_info{
-                .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_ADDRESS_INFO_EXT,
-                .pNext = nullptr,
-                .address = entry.address.address,
-                .range = entry.address.range,
-                .format = entry.address.format,
-            };
-            VkDescriptorGetInfoEXT get_info{
-                .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_GET_INFO_EXT,
-                .pNext = nullptr,
-                .type = binding.type,
-                .data{},
-            };
-            switch (binding.type) {
-            case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
-                get_info.data.pUniformBuffer = &address_info;
-                break;
-            case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
-                get_info.data.pStorageBuffer = &address_info;
-                break;
-            case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
-                get_info.data.pUniformTexelBuffer = &address_info;
-                break;
-            case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
-                get_info.data.pStorageTexelBuffer = &address_info;
-                break;
-            case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
-                get_info.data.pCombinedImageSampler = &entry.image;
-                break;
-            case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
-                get_info.data.pStorageImage = &entry.image;
-                break;
-            default:
-                continue;
-            }
-            dev.GetDescriptorEXT(get_info, binding.stride,
-                                 host + binding.offset + index * binding.stride);
-        }
-    }
-}
-
 [[nodiscard]] inline u32 NumDescriptorEntries(const Shader::Info& info) {
     return Shader::NumDescriptors(info.constant_buffer_descriptors) +
            Shader::NumDescriptors(info.storage_buffers_descriptors) +
@@ -148,70 +61,20 @@ public:
     DescriptorLayoutBuilder(const Device& device_) : device{&device_} {}
 
     bool CanUsePushDescriptor() const noexcept {
-        if (!device->IsKhrPushDescriptorSupported() ||
-            num_descriptors > device->MaxPushDescriptors()) {
-            return false;
-        }
-        return !device->IsExtDescriptorBufferSupported() ||
-               device->DescriptorBufferProperties().bufferlessPushDescriptors;
+        return device->IsKhrPushDescriptorSupported() &&
+               num_descriptors <= device->MaxPushDescriptors();
     }
 
-    bool CanUseDescriptorBuffer() const noexcept {
-        const auto& props = device->DescriptorBufferProperties();
-        if (!device->IsExtDescriptorBufferSupported() || bindings.empty() ||
-            !props.combinedImageSamplerDescriptorSingleArray) {
-            return false;
-        }
-        return !props.bufferlessPushDescriptors || !CanUsePushDescriptor();
-    }
-
-    DescriptorBufferLayout MakeDescriptorBufferLayout(VkDescriptorSetLayout layout) const {
-        DescriptorBufferLayout result;
-        if (!layout) {
-            return result;
-        }
-        const vk::Device& dev = device->GetLogical();
-        result.size = dev.GetDescriptorSetLayoutSizeEXT(layout);
-        result.bindings.reserve(bindings.size());
-        for (const VkDescriptorSetLayoutBinding& entry : bindings) {
-            result.bindings.push_back(DescriptorBufferBinding{
-                .type = entry.descriptorType,
-                .count = entry.descriptorCount,
-                .offset = dev.GetDescriptorSetLayoutBindingOffsetEXT(layout, entry.binding),
-                .stride = DescriptorSizeForType(*device, entry.descriptorType),
-            });
-        }
-        return result;
-    }
-
-    vk::DescriptorSetLayout CreateDescriptorSetLayout(bool use_push_descriptor,
-                                                      bool use_descriptor_buffer = false) const {
+    // TODO(crueter): utilize layout binding flags
+    vk::DescriptorSetLayout CreateDescriptorSetLayout(bool use_push_descriptor) const {
         if (bindings.empty()) {
             return nullptr;
         }
-        VkDescriptorSetLayoutCreateFlags flags = 0;
-        if (use_push_descriptor) {
-            flags |= VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR;
-        }
-        if (use_descriptor_buffer) {
-            flags |= VK_DESCRIPTOR_SET_LAYOUT_CREATE_DESCRIPTOR_BUFFER_BIT_EXT;
-        }
-        boost::container::small_vector<VkDescriptorBindingFlags, 32> binding_flags;
-        VkDescriptorSetLayoutBindingFlagsCreateInfo binding_flags_ci{};
-        const void* pnext = nullptr;
-        if (!use_push_descriptor && device->IsDescriptorBindingPartiallyBoundSupported()) {
-            binding_flags.assign(bindings.size(), VK_DESCRIPTOR_BINDING_PARTIALLY_BOUND_BIT);
-            binding_flags_ci = {
-                .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_BINDING_FLAGS_CREATE_INFO,
-                .pNext = nullptr,
-                .bindingCount = static_cast<u32>(binding_flags.size()),
-                .pBindingFlags = binding_flags.data(),
-            };
-            pnext = &binding_flags_ci;
-        }
+        const VkDescriptorSetLayoutCreateFlags flags =
+            use_push_descriptor ? VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR : 0;
         return device->GetLogical().CreateDescriptorSetLayout({
             .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
-            .pNext = pnext,
+            .pNext = nullptr,
             .flags = flags,
             .bindingCount = static_cast<u32>(bindings.size()),
             .pBindings = bindings.data(),

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -69,9 +69,6 @@ vk::Buffer CreateBuffer(const Device& device, const MemoryAllocator& memory_allo
     if (device.IsExtConditionalRendering()) {
         flags |= VK_BUFFER_USAGE_CONDITIONAL_RENDERING_BIT_EXT;
     }
-    if (device.IsBufferDeviceAddressSupported()) {
-        flags |= VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT;
-    }
     const VkBufferCreateInfo buffer_ci = {
         .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
         .pNext = nullptr,
@@ -94,9 +91,6 @@ Buffer::Buffer(BufferCacheRuntime& runtime, VideoCommon::NullBufferParams null_p
     device = &runtime.device;
     buffer = runtime.CreateNullBuffer();
     is_null = true;
-    if (device->IsBufferDeviceAddressSupported()) {
-        device_address = device->GetLogical().GetBufferDeviceAddress(*buffer);
-    }
 }
 
 Buffer::Buffer(BufferCacheRuntime& runtime, DAddr cpu_addr_, u64 size_bytes_)
@@ -106,9 +100,6 @@ Buffer::Buffer(BufferCacheRuntime& runtime, DAddr cpu_addr_, u64 size_bytes_)
     if (runtime.device.HasDebuggingToolAttached()) {
         buffer.SetObjectNameEXT(fmt::format("Buffer {:#x}", CpuAddr()).c_str());
     }
-    if (device->IsBufferDeviceAddressSupported()) {
-        device_address = device->GetLogical().GetBufferDeviceAddress(*buffer);
-    }
 }
 
 void Buffer::MarkUsage(u64 offset, u64 size) noexcept {
@@ -255,7 +246,6 @@ protected:
     StagingBufferPool& staging_pool;
 
     vk::Buffer buffer{};
-    MemoryCommit memory_commit{};
     VkIndexType index_type{};
     u32 num_indices = 0;
 };
@@ -373,10 +363,6 @@ StagingBufferRef BufferCacheRuntime::DownloadStagingBuffer(size_t size, bool def
     return staging_pool.Request(size, MemoryUsage::Download, deferred);
 }
 
-VkFormat BufferCacheRuntime::TexelBufferFormat(VideoCore::Surface::PixelFormat format) const {
-    return MaxwellToVK::SurfaceFormat(device, FormatType::Buffer, false, format).format;
-}
-
 void BufferCacheRuntime::FreeDeferredStagingBuffer(StagingBufferRef& ref) {
     staging_pool.FreeDeferred(ref);
 }
@@ -389,6 +375,10 @@ u64 BufferCacheRuntime::GetDeviceMemoryUsage() const {
     return device.GetDeviceMemoryUsage();
 }
 
+u64 BufferCacheRuntime::GetDeviceAllocationUsage() const {
+    return device.GetMemoryBudgetInfo().allocation_bytes;
+}
+
 bool BufferCacheRuntime::CanReportMemoryUsage() const {
     return device.CanReportMemoryUsage();
 }
@@ -417,6 +407,16 @@ u64 BufferCacheRuntime::KnownGpuTick() {
     return scheduler.GetMasterSemaphore().KnownGpuTick();
 }
 
+u64 BufferCacheRuntime::CurrentSyncPoint() const noexcept {
+    return scheduler.GetMasterSemaphore().CurrentTick();
+}
+
+u64 BufferCacheRuntime::CompletedSyncPoint() const {
+    auto& master_semaphore = scheduler.GetMasterSemaphore();
+    master_semaphore.Refresh();
+    return master_semaphore.KnownGpuTick();
+}
+
 void BufferCacheRuntime::Wait(u64 buffer_tick) {
     scheduler.Wait(buffer_tick);
 }
@@ -654,6 +654,7 @@ void BufferCacheRuntime::BindTransformFeedbackBuffer(u32 index, VkBuffer buffer,
         offset = 0;
         size = 0;
     }
+    scheduler.MarkTransformFeedbackUsed();
     scheduler.Record([index, buffer, offset, size](vk::CommandBuffer cmdbuf) {
         const VkDeviceSize vk_offset = offset;
         const VkDeviceSize vk_size = size;
@@ -666,19 +667,26 @@ void BufferCacheRuntime::BindTransformFeedbackBuffers(VideoCommon::HostBindings<
         // Already logged in the rasterizer
         return;
     }
-    boost::container::static_vector<VkBuffer, VideoCommon::NUM_VERTEX_BUFFERS> buffer_handles(bindings.buffers.size());
-    for (u32 i = 0; i < bindings.buffers.size(); ++i) {
+    const u32 count = std::min<u32>(static_cast<u32>(bindings.buffers.size()),
+                                    VideoCommon::NUM_TRANSFORM_FEEDBACK_BUFFERS);
+    std::array<VkBuffer, VideoCommon::NUM_TRANSFORM_FEEDBACK_BUFFERS> handles{};
+    std::array<VkDeviceSize, VideoCommon::NUM_TRANSFORM_FEEDBACK_BUFFERS> offsets{};
+    std::array<VkDeviceSize, VideoCommon::NUM_TRANSFORM_FEEDBACK_BUFFERS> sizes{};
+    for (u32 i = 0; i < count; ++i) {
         auto handle = bindings.buffers[i]->Handle();
         if (handle == VK_NULL_HANDLE) {
             ReserveNullBuffer();
             handle = *null_buffer;
-            bindings.offsets[i] = 0;
-            bindings.sizes[i] = 0;
+        } else {
+            offsets[i] = bindings.offsets[i];
+            sizes[i] = bindings.sizes[i];
         }
-        buffer_handles[i] = handle;
+        handles[i] = handle;
     }
-    scheduler.Record([bindings_ = std::move(bindings), buffer_handles_ = std::move(buffer_handles)](vk::CommandBuffer cmdbuf) {
-        cmdbuf.BindTransformFeedbackBuffersEXT(0, u32(buffer_handles_.size()), buffer_handles_.data(), bindings_.offsets.data(), bindings_.sizes.data());
+    scheduler.MarkTransformFeedbackUsed();
+    scheduler.Record([count, handles, offsets, sizes](vk::CommandBuffer cmdbuf) {
+        cmdbuf.BindTransformFeedbackBuffersEXT(0, count, handles.data(), offsets.data(),
+                                               sizes.data());
     });
 }
 
@@ -703,9 +711,6 @@ vk::Buffer BufferCacheRuntime::CreateNullBuffer() {
     if (device.IsExtTransformFeedbackSupported()) {
         create_info.usage |= VK_BUFFER_USAGE_TRANSFORM_FEEDBACK_BUFFER_BIT_EXT;
     }
-    if (device.IsBufferDeviceAddressSupported()) {
-        create_info.usage |= VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT;
-    }
     vk::Buffer ret = memory_allocator.CreateBuffer(create_info, MemoryUsage::DeviceLocal);
     if (device.HasDebuggingToolAttached()) {
         ret.SetObjectNameEXT("Null buffer");

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -39,10 +39,6 @@ public:
         return *buffer;
     }
 
-    [[nodiscard]] VkDeviceAddress DeviceAddress() const noexcept {
-        return device_address;
-    }
-
     [[nodiscard]] bool IsRegionUsed(u64 offset, u64 size) const noexcept {
         return tracker.IsUsed(offset, size);
     }
@@ -74,7 +70,6 @@ private:
     vk::Buffer buffer;
     std::vector<BufferView> views;
     VideoCommon::UsageTracker tracker;
-    VkDeviceAddress device_address{};
     u64 last_usage_tick{};
     bool is_null{};
 };
@@ -105,10 +100,20 @@ public:
 
     void Finish();
 
+    u64 CurrentSyncPoint() const noexcept;
+
+    u64 CompletedSyncPoint() const;
+
     u64 GetDeviceLocalMemory() const;
 
     u64 GetDeviceMemoryUsage() const;
 
+    u64 GetDeviceAllocationUsage() const;
+
+    bool CanReportAllocationUsage() const noexcept {
+        return true;
+    }
+
     bool CanReportMemoryUsage() const;
 
     u32 GetUniformBufferAlignment() const;
@@ -150,25 +155,22 @@ public:
                                           [[maybe_unused]] u32 binding_index,
                                           u32 size) {
         const StagingBufferRef ref = staging_pool.Request(size, MemoryUsage::Upload);
-        guest_descriptor_queue.AddBuffer(ref.buffer, ref.device_address,
-                                         static_cast<u32>(ref.offset), size);
+        BindBuffer(ref.buffer, static_cast<u32>(ref.offset), size);
         return ref.mapped_span;
     }
 
-    void BindUniformBuffer(const Buffer& buffer, u32 offset, u32 size) {
+    void BindUniformBuffer(VkBuffer buffer, u32 offset, u32 size) {
         BindBuffer(buffer, offset, size);
     }
 
-    void BindStorageBuffer(const Buffer& buffer, u32 offset, u32 size,
+    void BindStorageBuffer(VkBuffer buffer, u32 offset, u32 size,
                            [[maybe_unused]] bool is_written) {
         BindBuffer(buffer, offset, size);
     }
 
     void BindTextureBuffer(Buffer& buffer, u32 offset, u32 size,
                            VideoCore::Surface::PixelFormat format) {
-        guest_descriptor_queue.AddTexelBuffer(buffer.View(offset, size, format),
-                                              buffer.DeviceAddress(), offset, size,
-                                              TexelBufferFormat(format));
+        guest_descriptor_queue.AddTexelBuffer(buffer.View(offset, size, format));
     }
 
     bool ShouldLimitDynamicStorageBuffers() const {
@@ -180,17 +182,14 @@ public:
     }
 
 private:
-    void BindBuffer(const Buffer& buffer, u32 offset, u32 size) {
-        const VkBuffer handle = buffer.Handle();
-        if (handle == VK_NULL_HANDLE) {
-            guest_descriptor_queue.AddBuffer(handle, 0, 0, VK_WHOLE_SIZE);
+    void BindBuffer(VkBuffer buffer, u32 offset, u32 size) {
+        if (buffer == VK_NULL_HANDLE) {
+            guest_descriptor_queue.AddBuffer(buffer, 0, VK_WHOLE_SIZE);
         } else {
-            guest_descriptor_queue.AddBuffer(handle, buffer.DeviceAddress(), offset, size);
+            guest_descriptor_queue.AddBuffer(buffer, offset, size);
         }
     }
 
-    VkFormat TexelBufferFormat(VideoCore::Surface::PixelFormat format) const;
-
     void ReserveNullBuffer();
     vk::Buffer CreateNullBuffer();
 

src/video_core/renderer_vulkan/vk_compute_pipeline.cpp

@@ -34,14 +34,12 @@ using Tegra::Texture::TexturePair;
 ComputePipeline::ComputePipeline(const Device& device_, Scheduler& scheduler, vk::PipelineCache& pipeline_cache_,
                                  DescriptorPool& descriptor_pool,
                                  GuestDescriptorQueue& guest_descriptor_queue_,
-                                 DescriptorBufferRing& descriptor_buffer_ring_,
                                  Common::ThreadWorker* thread_worker,
                                  PipelineStatistics* pipeline_statistics,
                                  VideoCore::ShaderNotify* shader_notify, const Shader::Info& info_,
                                  vk::ShaderModule spv_module_, u64 shader_hash_)
     : device{device_},
-      pipeline_cache(pipeline_cache_), guest_descriptor_queue{guest_descriptor_queue_},
-      descriptor_buffer_ring{descriptor_buffer_ring_}, info{info_},
+      pipeline_cache(pipeline_cache_), guest_descriptor_queue{guest_descriptor_queue_}, info{info_},
       shader_hash{shader_hash_}, spv_module(std::move(spv_module_)) {
     if (shader_notify) {
         shader_notify->MarkShaderBuilding();
@@ -50,36 +48,18 @@ ComputePipeline::ComputePipeline(const Device& device_, Scheduler& scheduler, vk
                 uniform_buffer_sizes.begin());
     num_descriptor_entries = NumDescriptorEntries(info);
 
-    DescriptorLayoutBuilder builder{device};
-    builder.Add(info, VK_SHADER_STAGE_COMPUTE_BIT);
+    auto func{[this, &scheduler, &descriptor_pool, shader_notify, pipeline_statistics] {
+        DescriptorLayoutBuilder builder{device};
+        builder.Add(info, VK_SHADER_STAGE_COMPUTE_BIT);
 
-    uses_push_descriptor = builder.CanUsePushDescriptor();
-    uses_descriptor_buffer = builder.CanUseDescriptorBuffer() && descriptor_buffer_ring.IsValid();
-    descriptor_set_layout =
-        builder.CreateDescriptorSetLayout(uses_push_descriptor, uses_descriptor_buffer);
-    if (uses_descriptor_buffer) {
-        descriptor_buffer_layout = builder.MakeDescriptorBufferLayout(*descriptor_set_layout);
-        if (!descriptor_buffer_ring.CanAllocate(descriptor_buffer_layout.size)) {
-            LOG_DEBUG(Render_Vulkan,
-                        "Compute shader {:016X} needs {} descriptor bytes per dispatch, falling "
-                        "back to sets",
-                        shader_hash, descriptor_buffer_layout.size);
-            uses_descriptor_buffer = false;
-            descriptor_buffer_layout = {};
-            descriptor_set_layout = builder.CreateDescriptorSetLayout(false);
-        }
-    }
-    pipeline_layout = builder.CreatePipelineLayout(*descriptor_set_layout);
-    if (!uses_descriptor_buffer) {
+        uses_push_descriptor = builder.CanUsePushDescriptor();
+        descriptor_set_layout = builder.CreateDescriptorSetLayout(uses_push_descriptor);
+        pipeline_layout = builder.CreatePipelineLayout(*descriptor_set_layout);
         descriptor_update_template =
             builder.CreateTemplate(*descriptor_set_layout, *pipeline_layout, uses_push_descriptor);
         if (!uses_push_descriptor) {
-            descriptor_allocator =
-                descriptor_pool.Allocator(device, scheduler, *descriptor_set_layout, info);
+            descriptor_allocator = descriptor_pool.Allocator(device, scheduler, *descriptor_set_layout, info);
         }
-    }
-
-    auto func{[this, shader_notify, pipeline_statistics] {
         const VkPipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci{
             .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_REQUIRED_SUBGROUP_SIZE_CREATE_INFO_EXT,
             .pNext = nullptr,
@@ -89,9 +69,6 @@ ComputePipeline::ComputePipeline(const Device& device_, Scheduler& scheduler, vk
         if (device.IsKhrPipelineExecutablePropertiesEnabled() && Settings::values.renderer_debug.GetValue()) {
             flags |= VK_PIPELINE_CREATE_CAPTURE_STATISTICS_BIT_KHR;
         }
-        if (uses_descriptor_buffer) {
-            flags |= VK_PIPELINE_CREATE_DESCRIPTOR_BUFFER_BIT_EXT;
-        }
         const VkComputePipelineCreateInfo compute_ci{
             .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
             .pNext = nullptr,
@@ -148,10 +125,10 @@ ComputePipeline::ComputePipeline(const Device& device_, Scheduler& scheduler, vk
     }
 }
 
-bool ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
+void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
                                 Tegra::MemoryManager& gpu_memory, Scheduler& scheduler,
                                 BufferCache& buffer_cache, TextureCache& texture_cache) {
-    guest_descriptor_queue.Acquire(scheduler, num_descriptor_entries, uses_descriptor_buffer);
+    guest_descriptor_queue.Acquire(scheduler, num_descriptor_entries);
 
     buffer_cache.SetComputeUniformBufferState(info.constant_buffer_mask, &uniform_buffer_sizes);
     buffer_cache.UnbindComputeStorageBuffers();
@@ -272,33 +249,10 @@ bool ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
         GPU::Logging::GPULogger::GetInstance().LogPipelineBind(true, "compute pipeline");
     }
 
-    const DescriptorUpdateEntry* const descriptor_data{guest_descriptor_queue.UpdateData()};
-    VkDeviceSize descriptor_buffer_offset{};
-    u32 descriptor_buffer_chunk{};
-    if (uses_descriptor_buffer) {
-        const DescriptorBufferRing::Allocation alloc{
-            descriptor_buffer_ring.Allocate(scheduler, descriptor_buffer_layout.size)};
-        if (!alloc.host) {
-            LOG_DEBUG(Render_Vulkan, "Failed to reserve descriptor memory, skipping dispatch");
-            return false;
-        }
-        WriteDescriptorBuffer(device, descriptor_buffer_layout, descriptor_data, alloc.host);
-        descriptor_buffer_offset = alloc.offset;
-        descriptor_buffer_chunk = alloc.chunk;
-    }
-
-    const bool bind_descriptor_buffer{
-        uses_descriptor_buffer && scheduler.UpdateDescriptorBufferChunk(descriptor_buffer_chunk)};
-
+    const void* const descriptor_data{guest_descriptor_queue.UpdateData()};
     const bool is_rescaling = !info.texture_descriptors.empty() || !info.image_descriptors.empty();
-    scheduler.Record([this, descriptor_data, is_rescaling, descriptor_buffer_offset,
-                      descriptor_buffer_chunk, bind_descriptor_buffer,
+    scheduler.Record([this, descriptor_data, is_rescaling,
                       rescaling_data = rescaling.Data()](vk::CommandBuffer cmdbuf) {
-        if (bind_descriptor_buffer) {
-            const VkDescriptorBufferBindingInfoEXT binding_info{
-                descriptor_buffer_ring.BindingInfo(descriptor_buffer_chunk)};
-            cmdbuf.BindDescriptorBuffersEXT(binding_info);
-        }
         if (!pipeline) {
             return;
         }
@@ -311,11 +265,7 @@ bool ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
                                  RESCALING_LAYOUT_WORDS_OFFSET, sizeof(rescaling_data),
                                  rescaling_data.data());
         }
-        if (uses_descriptor_buffer) {
-            const u32 buffer_index{};
-            cmdbuf.SetDescriptorBufferOffsetsEXT(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline_layout,
-                                                 0, buffer_index, descriptor_buffer_offset);
-        } else if (uses_push_descriptor) {
+        if (uses_push_descriptor) {
             cmdbuf.PushDescriptorSetWithTemplateKHR(*descriptor_update_template, *pipeline_layout,
                                                     0, descriptor_data);
         } else {
@@ -326,7 +276,6 @@ bool ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
                                       descriptor_set, nullptr);
         }
     });
-    return true;
 }
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_compute_pipeline.h

@@ -13,9 +13,7 @@
 #include "common/common_types.h"
 #include "common/thread_worker.h"
 #include "shader_recompiler/shader_info.h"
-#include "video_core/renderer_vulkan/pipeline_helper.h"
 #include "video_core/renderer_vulkan/vk_buffer_cache.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_texture_cache.h"
 #include "video_core/renderer_vulkan/vk_update_descriptor.h"
@@ -36,7 +34,6 @@ public:
     explicit ComputePipeline(const Device& device, Scheduler& scheduler, vk::PipelineCache& pipeline_cache,
                              DescriptorPool& descriptor_pool,
                              GuestDescriptorQueue& guest_descriptor_queue,
-                             DescriptorBufferRing& descriptor_buffer_ring,
                              Common::ThreadWorker* thread_worker,
                              PipelineStatistics* pipeline_statistics,
                              VideoCore::ShaderNotify* shader_notify, const Shader::Info& info,
@@ -48,9 +45,8 @@ public:
     ComputePipeline& operator=(const ComputePipeline&) = delete;
     ComputePipeline(const ComputePipeline&) = delete;
 
-    [[nodiscard]] bool Configure(Tegra::Engines::KeplerCompute& kepler_compute,
-                                 Tegra::MemoryManager& gpu_memory, Scheduler& scheduler,
-                                 BufferCache& buffer_cache, TextureCache& texture_cache);
+    void Configure(Tegra::Engines::KeplerCompute& kepler_compute, Tegra::MemoryManager& gpu_memory,
+                   Scheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache);
 
     bool IsBound() const noexcept {
         return static_cast<bool>(pipeline);
@@ -60,7 +56,6 @@ private:
     const Device& device;
     vk::PipelineCache& pipeline_cache;
     GuestDescriptorQueue& guest_descriptor_queue;
-    DescriptorBufferRing& descriptor_buffer_ring;
     Shader::Info info;
     u64 shader_hash{};
     u32 num_descriptor_entries{};
@@ -70,8 +65,6 @@ private:
     vk::ShaderModule spv_module;
     vk::DescriptorSetLayout descriptor_set_layout;
     bool uses_push_descriptor{false};
-    bool uses_descriptor_buffer{false};
-    DescriptorBufferLayout descriptor_buffer_layout;
     DescriptorAllocator descriptor_allocator;
     vk::PipelineLayout pipeline_layout;
     vk::DescriptorUpdateTemplate descriptor_update_template;

src/video_core/renderer_vulkan/vk_descriptor_buffer.cpp

@@ -1,141 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
-#include <algorithm>
-
-#include "common/alignment.h"
-#include "common/assert.h"
-#include "common/logging.h"
-#include "video_core/renderer_vulkan/vk_descriptor_buffer.h"
-#include "video_core/renderer_vulkan/vk_scheduler.h"
-#include "video_core/vulkan_common/vulkan_device.h"
-
-namespace Vulkan {
-
-DescriptorBufferRing::DescriptorBufferRing(const Device& device_,
-                                           MemoryAllocator& memory_allocator)
-    : device{device_} {
-    if (!device.IsExtDescriptorBufferSupported() || !device.IsBufferDeviceAddressSupported()) {
-        return;
-    }
-    const VkPhysicalDeviceDescriptorBufferPropertiesEXT& props{device.DescriptorBufferProperties()};
-    alignment = std::max<VkDeviceSize>(props.descriptorBufferOffsetAlignment, 1);
-
-    const VkDeviceSize max_bound{(std::min)({props.maxSamplerDescriptorBufferRange,
-                                             props.maxResourceDescriptorBufferRange,
-                                             props.samplerDescriptorBufferAddressSpaceSize,
-                                             props.resourceDescriptorBufferAddressSpaceSize,
-                                             props.descriptorBufferAddressSpaceSize})};
-    const VkDeviceSize frame_size{device.IsTiler() ? TILER_FRAME_SIZE : DESKTOP_FRAME_SIZE};
-    const VkDeviceSize chunk_size{
-        Common::AlignDown((std::min)(frame_size, max_bound), alignment)};
-    if (chunk_size <= alignment) {
-        LOG_DEBUG(Render_Vulkan, "Descriptor buffer binding limit of {} is unusable, disabling",
-                  max_bound);
-        return;
-    }
-    chunk_capacity = chunk_size - alignment;
-    chunks_per_frame = static_cast<size_t>(frame_size / chunk_size);
-
-    const VkBufferCreateInfo buffer_ci{
-        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .size = chunk_size,
-        .usage = VK_BUFFER_USAGE_RESOURCE_DESCRIPTOR_BUFFER_BIT_EXT |
-                 VK_BUFFER_USAGE_SAMPLER_DESCRIPTOR_BUFFER_BIT_EXT |
-                 VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT,
-        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
-        .queueFamilyIndexCount = 0,
-        .pQueueFamilyIndices = nullptr,
-    };
-    const size_t total_chunks{chunks_per_frame * FRAMES_IN_FLIGHT};
-    chunks.reserve(total_chunks);
-    chunk_addresses.reserve(total_chunks);
-    chunk_hosts.reserve(total_chunks);
-    for (size_t index = 0; index < total_chunks; ++index) {
-        vk::Buffer buffer{memory_allocator.CreateBuffer(buffer_ci, MemoryUsage::Upload)};
-        if (!buffer.IsHostVisible()) {
-            LOG_DEBUG(Render_Vulkan, "Descriptor buffer is not host visible, disabling");
-            chunks.clear();
-            return;
-        }
-        if (!buffer.IsHostCoherent()) {
-            LOG_DEBUG(Render_Vulkan, "Descriptor buffer is not host coherent, disabling");
-            chunks.clear();
-            return;
-        }
-        if (device.HasDebuggingToolAttached()) {
-            buffer.SetObjectNameEXT("Descriptor buffer");
-        }
-        const VkDeviceAddress raw_address{device.GetLogical().GetBufferDeviceAddress(*buffer)};
-        const VkDeviceAddress address{Common::AlignUp(raw_address, alignment)};
-        chunk_addresses.push_back(address);
-        chunk_hosts.push_back(buffer.Mapped().data() + (address - raw_address));
-        chunks.push_back(std::move(buffer));
-    }
-}
-
-DescriptorBufferRing::~DescriptorBufferRing() = default;
-
-void DescriptorBufferRing::TickFrame() {
-    if (++frame_index >= FRAMES_IN_FLIGHT) {
-        frame_index = 0;
-    }
-    chunk_cursor = 0;
-    cursor = 0;
-    ++generation;
-    frame_reused = true;
-}
-
-void DescriptorBufferRing::TouchFrame(Scheduler& scheduler) {
-    frame_ticks[frame_index] = scheduler.CurrentTick();
-}
-
-DescriptorBufferRing::Allocation DescriptorBufferRing::Allocate(Scheduler& scheduler,
-                                                                VkDeviceSize size) {
-    ASSERT(!chunks.empty());
-    if (!CanAllocate(size)) {
-        LOG_DEBUG(Render_Vulkan, "Descriptor set of {} bytes exceeds chunk capacity {}", size,
-                  chunk_capacity);
-        return Allocation{};
-    }
-    const VkDeviceSize needed{Common::AlignUp(size, alignment)};
-    if (frame_reused) {
-        frame_reused = false;
-        scheduler.Wait(frame_ticks[frame_index]);
-    }
-    if (cursor + needed > chunk_capacity) {
-        if (chunk_cursor + 1 < chunks_per_frame) {
-            ++chunk_cursor;
-        } else {
-            LOG_DEBUG(Render_Vulkan, "Descriptor buffer frame exhausted, stalling on the GPU");
-            scheduler.Finish();
-            chunk_cursor = 0;
-            ++generation;
-        }
-        cursor = 0;
-    }
-    const size_t chunk{frame_index * chunks_per_frame + chunk_cursor};
-    const VkDeviceSize offset{cursor};
-    cursor += needed;
-    frame_ticks[frame_index] = scheduler.CurrentTick();
-    return Allocation{
-        .host = chunk_hosts[chunk] + offset,
-        .offset = offset,
-        .chunk = static_cast<u32>(chunk),
-        .generation = generation,
-    };
-}
-
-VkDescriptorBufferBindingInfoEXT DescriptorBufferRing::BindingInfo(u32 chunk) const noexcept {
-    return VkDescriptorBufferBindingInfoEXT{
-        .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_BUFFER_BINDING_INFO_EXT,
-        .pNext = nullptr,
-        .address = chunk_addresses[chunk],
-        .usage = VK_BUFFER_USAGE_RESOURCE_DESCRIPTOR_BUFFER_BIT_EXT |
-                 VK_BUFFER_USAGE_SAMPLER_DESCRIPTOR_BUFFER_BIT_EXT,
-    };
-}
-
-} // namespace Vulkan

src/video_core/renderer_vulkan/vk_descriptor_buffer.h

@@ -1,71 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
-#pragma once
-
-#include <array>
-#include <vector>
-
-#include "common/alignment.h"
-#include "common/common_types.h"
-#include "video_core/vulkan_common/vulkan_memory_allocator.h"
-#include "video_core/vulkan_common/vulkan_wrapper.h"
-
-namespace Vulkan {
-
-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;
-
-public:
-    explicit DescriptorBufferRing(const Device& device_, MemoryAllocator& memory_allocator);
-    ~DescriptorBufferRing();
-
-    struct Allocation {
-        u8* host{};
-        VkDeviceSize offset{};
-        u32 chunk{};
-        u64 generation{};
-    };
-
-    [[nodiscard]] u64 CurrentGeneration() const noexcept {
-        return generation;
-    }
-
-    void TouchFrame(Scheduler& scheduler);
-
-    [[nodiscard]] bool CanAllocate(VkDeviceSize size) const noexcept {
-        return Common::AlignUp(size, alignment) <= chunk_capacity;
-    }
-
-    void TickFrame();
-
-    [[nodiscard]] Allocation Allocate(Scheduler& scheduler, VkDeviceSize size);
-
-    [[nodiscard]] VkDescriptorBufferBindingInfoEXT BindingInfo(u32 chunk) const noexcept;
-
-    [[nodiscard]] bool IsValid() const noexcept {
-        return !chunks.empty();
-    }
-
-private:
-    const Device& device;
-    std::vector<vk::Buffer> chunks;
-    std::vector<VkDeviceAddress> chunk_addresses;
-    std::vector<u8*> chunk_hosts;
-    VkDeviceSize alignment{1};
-    VkDeviceSize chunk_capacity{};
-    size_t chunks_per_frame{};
-    size_t frame_index{};
-    size_t chunk_cursor{};
-    VkDeviceSize cursor{};
-    u64 generation{1};
-    std::array<u64, FRAMES_IN_FLIGHT> frame_ticks{};
-    bool frame_reused{};
-};
-
-} // namespace Vulkan

src/video_core/renderer_vulkan/vk_fence_manager.cpp

@@ -37,7 +37,7 @@ void InnerFence::Wait() {
     if (is_stubbed) {
         return;
     }
-    scheduler.Wait(wait_tick);
+    scheduler.WaitSubmitted(wait_tick);
 }
 
 FenceManager::FenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_,

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -5,7 +5,6 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <algorithm>
-#include <cstring>
 #include <iostream>
 #include <span>
 
@@ -251,15 +250,13 @@ GraphicsPipeline::GraphicsPipeline(
     Scheduler& scheduler_, BufferCache& buffer_cache_, TextureCache& texture_cache_,
     vk::PipelineCache& pipeline_cache_, VideoCore::ShaderNotify* shader_notify,
     const Device& device_, DescriptorPool& descriptor_pool,
-    GuestDescriptorQueue& guest_descriptor_queue_, DescriptorBufferRing& descriptor_buffer_ring_,
-    Common::ThreadWorker* worker_thread,
+    GuestDescriptorQueue& guest_descriptor_queue_, Common::ThreadWorker* worker_thread,
     PipelineStatistics* pipeline_statistics, RenderPassCache& render_pass_cache,
     const GraphicsPipelineCacheKey& key_, std::array<vk::ShaderModule, NUM_STAGES> stages,
     const std::array<const Shader::Info*, NUM_STAGES>& infos)
     : key{key_}, device{device_}, texture_cache{texture_cache_}, buffer_cache{buffer_cache_},
       pipeline_cache(pipeline_cache_), scheduler{scheduler_},
-      guest_descriptor_queue{guest_descriptor_queue_},
-      descriptor_buffer_ring{descriptor_buffer_ring_}, spv_modules{std::move(stages)} {
+      guest_descriptor_queue{guest_descriptor_queue_}, spv_modules{std::move(stages)} {
     if (shader_notify) {
         shader_notify->MarkShaderBuilding();
     }
@@ -279,37 +276,20 @@ GraphicsPipeline::GraphicsPipeline(
         num_descriptor_entries += NumDescriptorEntries(*info);
     }
     fragment_has_color0_output = stage_infos[NUM_STAGES - 1].stores_frag_color[0];
+    auto func{[this, shader_notify, &render_pass_cache, &descriptor_pool, pipeline_statistics] {
+        DescriptorLayoutBuilder builder{MakeBuilder(device, stage_infos)};
+        uses_push_descriptor = builder.CanUsePushDescriptor();
+        descriptor_set_layout = builder.CreateDescriptorSetLayout(uses_push_descriptor);
 
-    DescriptorLayoutBuilder builder{MakeBuilder(device, stage_infos)};
-    uses_push_descriptor = builder.CanUsePushDescriptor();
-    uses_descriptor_buffer = builder.CanUseDescriptorBuffer() && descriptor_buffer_ring.IsValid();
-    descriptor_set_layout =
-        builder.CreateDescriptorSetLayout(uses_push_descriptor, uses_descriptor_buffer);
-    if (uses_descriptor_buffer) {
-        descriptor_buffer_layout = builder.MakeDescriptorBufferLayout(*descriptor_set_layout);
-        if (!descriptor_buffer_ring.CanAllocate(descriptor_buffer_layout.size)) {
-            LOG_WARNING(Render_Vulkan,
-                        "Graphics pipeline {:016X} needs {} descriptor bytes per draw, falling back "
-                        "to sets",
-                        key.Hash(), descriptor_buffer_layout.size);
-            uses_descriptor_buffer = false;
-            descriptor_buffer_layout = {};
-            descriptor_set_layout = builder.CreateDescriptorSetLayout(uses_push_descriptor);
+        if (!uses_push_descriptor) {
+            descriptor_allocator = descriptor_pool.Allocator(device, scheduler, *descriptor_set_layout, stage_infos);
         }
-    }
 
-    const VkDescriptorSetLayout set_layout{*descriptor_set_layout};
-    pipeline_layout = builder.CreatePipelineLayout(set_layout);
-    if (!uses_descriptor_buffer) {
+        const VkDescriptorSetLayout set_layout{*descriptor_set_layout};
+        pipeline_layout = builder.CreatePipelineLayout(set_layout);
         descriptor_update_template =
             builder.CreateTemplate(set_layout, *pipeline_layout, uses_push_descriptor);
-        if (!uses_push_descriptor) {
-            descriptor_allocator =
-                descriptor_pool.Allocator(device, scheduler, set_layout, stage_infos);
-        }
-    }
 
-    auto func{[this, shader_notify, &render_pass_cache, pipeline_statistics] {
         const VkRenderPass render_pass{render_pass_cache.Get(MakeRenderPassKey(key.state, device))};
         Validate();
         try {
@@ -516,7 +496,7 @@ bool GraphicsPipeline::ConfigureImpl(bool is_indexed) {
     buffer_cache.UpdateGraphicsBuffers(is_indexed);
     buffer_cache.BindHostGeometryBuffers(is_indexed);
 
-    guest_descriptor_queue.Acquire(scheduler, num_descriptor_entries, uses_descriptor_buffer);
+    guest_descriptor_queue.Acquire(scheduler, num_descriptor_entries);
 
     RescalingPushConstant rescaling;
     RenderAreaPushConstant render_area;
@@ -558,43 +538,13 @@ bool GraphicsPipeline::ConfigureImpl(bool is_indexed) {
     if (IsBuilt() && !pipeline) {
         return false;
     }
-    return ConfigureDraw(rescaling, render_area);
+    ConfigureDraw(rescaling, render_area);
+
+    return true;
 }
 
-bool GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling,
+void GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling,
                                      const RenderAreaPushConstant& render_area) {
-    const void* const descriptor_data{guest_descriptor_queue.UpdateData()};
-
-    VkDeviceSize descriptor_buffer_offset{};
-    u32 descriptor_buffer_chunk{};
-    if (descriptor_set_layout && uses_descriptor_buffer) {
-        const auto* const entries = static_cast<const DescriptorUpdateEntry*>(descriptor_data);
-        const bool reuse_allocation =
-            last_descriptor_buffer_generation == descriptor_buffer_ring.CurrentGeneration() &&
-            last_descriptor_payload.size() == num_descriptor_entries &&
-            std::memcmp(last_descriptor_payload.data(), entries,
-                        num_descriptor_entries * sizeof(DescriptorUpdateEntry)) == 0;
-        if (reuse_allocation) {
-            descriptor_buffer_offset = last_descriptor_buffer_offset;
-            descriptor_buffer_chunk = last_descriptor_buffer_chunk;
-            descriptor_buffer_ring.TouchFrame(scheduler);
-        } else {
-            const DescriptorBufferRing::Allocation alloc{
-                descriptor_buffer_ring.Allocate(scheduler, descriptor_buffer_layout.size)};
-            if (!alloc.host) {
-                LOG_DEBUG(Render_Vulkan, "Failed to reserve descriptor memory, skipping draw");
-                return false;
-            }
-            WriteDescriptorBuffer(device, descriptor_buffer_layout, entries, alloc.host);
-            descriptor_buffer_offset = alloc.offset;
-            descriptor_buffer_chunk = alloc.chunk;
-            last_descriptor_buffer_offset = alloc.offset;
-            last_descriptor_buffer_chunk = alloc.chunk;
-            last_descriptor_buffer_generation = alloc.generation;
-            last_descriptor_payload.assign(entries, entries + num_descriptor_entries);
-        }
-    }
-
     scheduler.RequestRenderpass(texture_cache.GetFramebuffer());
     if (!is_built.load(std::memory_order::relaxed)) {
         // Wait for the pipeline to be built
@@ -606,9 +556,6 @@ bool GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling,
     const bool is_rescaling{texture_cache.IsRescaling()};
     const bool update_rescaling{scheduler.UpdateRescaling(is_rescaling)};
     const bool bind_pipeline{scheduler.UpdateGraphicsPipeline(this)};
-    const bool bind_descriptor_buffer{
-        descriptor_set_layout && uses_descriptor_buffer &&
-        scheduler.UpdateDescriptorBufferChunk(descriptor_buffer_chunk)};
 
     // Log graphics pipeline binding
     if (bind_pipeline && GPU::Logging::IsActive() &&
@@ -617,27 +564,11 @@ bool GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling,
         GPU::Logging::GPULogger::GetInstance().LogPipelineBind(false, pipeline_info);
     }
 
-    bool update_descriptors = true;
-    if (descriptor_set_layout && !uses_push_descriptor && !uses_descriptor_buffer) {
-        const auto* const entries = static_cast<const DescriptorUpdateEntry*>(descriptor_data);
-        update_descriptors =
-            bind_pipeline || last_descriptor_payload.size() != num_descriptor_entries ||
-            std::memcmp(last_descriptor_payload.data(), entries,
-                        num_descriptor_entries * sizeof(DescriptorUpdateEntry)) != 0;
-        if (update_descriptors) {
-            last_descriptor_payload.assign(entries, entries + num_descriptor_entries);
-        }
-    }
-    scheduler.Record([this, descriptor_data, bind_pipeline, update_descriptors,
-                      descriptor_buffer_offset, descriptor_buffer_chunk, bind_descriptor_buffer,
-                      rescaling_data = rescaling.Data(), is_rescaling, update_rescaling,
+    const void* const descriptor_data{guest_descriptor_queue.UpdateData()};
+    scheduler.Record([this, descriptor_data, bind_pipeline, rescaling_data = rescaling.Data(),
+                      is_rescaling, update_rescaling,
                       uses_render_area = render_area.uses_render_area,
                       render_area_data = render_area.words](vk::CommandBuffer cmdbuf) {
-        if (bind_descriptor_buffer) {
-            const VkDescriptorBufferBindingInfoEXT binding_info{
-                descriptor_buffer_ring.BindingInfo(descriptor_buffer_chunk)};
-            cmdbuf.BindDescriptorBuffersEXT(binding_info);
-        }
         if (bind_pipeline) {
             if (!pipeline) {
                 return;
@@ -662,14 +593,10 @@ bool GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling,
         if (!descriptor_set_layout) {
             return;
         }
-        if (uses_descriptor_buffer) {
-            const u32 buffer_index{};
-            cmdbuf.SetDescriptorBufferOffsetsEXT(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline_layout,
-                                                 0, buffer_index, descriptor_buffer_offset);
-        } else if (uses_push_descriptor) {
+        if (uses_push_descriptor) {
             cmdbuf.PushDescriptorSetWithTemplateKHR(*descriptor_update_template, *pipeline_layout,
                                                     0, descriptor_data);
-        } else if (update_descriptors) {
+        } else {
             const VkDescriptorSet descriptor_set{descriptor_allocator.Commit()};
             const vk::Device& dev{device.GetLogical()};
             dev.UpdateDescriptorSet(descriptor_set, *descriptor_update_template, descriptor_data);
@@ -677,7 +604,6 @@ bool GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling,
                                       descriptor_set, nullptr);
         }
     });
-    return true;
 }
 
 void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
@@ -831,16 +757,13 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
         .lineWidth = 1.0f,
         // TODO(alekpop): Transfer from regs
     };
-    const bool smooth_lines_supported =
-        device.IsExtLineRasterizationSupported() && device.SupportsSmoothLines();
-    const bool stippled_lines_supported =
-        device.IsExtLineRasterizationSupported() && device.SupportsStippledRectangularLines();
+    const VkLineRasterizationModeEXT line_raster_mode =
+        device.GetLineRasterizationMode(key.state.smooth_lines != 0);
+    const bool stippled_lines_supported = device.SupportsStippleForMode(line_raster_mode);
     VkPipelineRasterizationLineStateCreateInfoEXT line_state{
         .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_LINE_STATE_CREATE_INFO_EXT,
         .pNext = nullptr,
-        .lineRasterizationMode = key.state.smooth_lines != 0 && smooth_lines_supported
-                                     ? VK_LINE_RASTERIZATION_MODE_RECTANGULAR_SMOOTH_EXT
-                                     : VK_LINE_RASTERIZATION_MODE_RECTANGULAR_EXT,
+        .lineRasterizationMode = line_raster_mode,
         .stippledLineEnable =
             (dynamic.line_stipple_enable && stippled_lines_supported) ? VK_TRUE : VK_FALSE,
         .lineStippleFactor = key.state.line_stipple_factor,
@@ -879,6 +802,16 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
     if (device.IsExtProvokingVertexSupported()) {
         provoking_vertex.pNext = std::exchange(rasterization_ci.pNext, &provoking_vertex);
     }
+    VkPipelineRasterizationDepthClipStateCreateInfoEXT depth_clip_state{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_DEPTH_CLIP_STATE_CREATE_INFO_EXT,
+        .pNext = nullptr,
+        .flags = 0,
+        .depthClipEnable = static_cast<VkBool32>(dynamic.depth_clip_disabled == 0 ? VK_TRUE
+                                                                                  : VK_FALSE),
+    };
+    if (device.IsExtDepthClipEnableSupported()) {
+        depth_clip_state.pNext = std::exchange(rasterization_ci.pNext, &depth_clip_state);
+    }
 
     const bool supports_alpha_output = fragment_has_color0_output;
     const bool alpha_to_one_supported = device.SupportsAlphaToOne();
@@ -1069,9 +1002,6 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
     if (device.IsKhrPipelineExecutablePropertiesEnabled() && Settings::values.renderer_debug.GetValue()) {
         flags |= VK_PIPELINE_CREATE_CAPTURE_STATISTICS_BIT_KHR;
     }
-    if (uses_descriptor_buffer) {
-        flags |= VK_PIPELINE_CREATE_DESCRIPTOR_BUFFER_BIT_EXT;
-    }
 
     pipeline = device.GetLogical().CreateGraphicsPipeline({
         .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,

src/video_core/renderer_vulkan/vk_graphics_pipeline.h

@@ -12,18 +12,14 @@
 #include <condition_variable>
 #include <mutex>
 #include <type_traits>
-#include <vector>
 
 #include "common/thread_worker.h"
 #include "shader_recompiler/shader_info.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_vulkan/fixed_pipeline_state.h"
-#include "video_core/renderer_vulkan/pipeline_helper.h"
 #include "video_core/renderer_vulkan/vk_buffer_cache.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_texture_cache.h"
-#include "video_core/renderer_vulkan/vk_update_descriptor.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
 namespace VideoCore {
@@ -80,8 +76,7 @@ public:
         Scheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache,
         vk::PipelineCache& pipeline_cache, VideoCore::ShaderNotify* shader_notify,
         const Device& device, DescriptorPool& descriptor_pool,
-        GuestDescriptorQueue& guest_descriptor_queue,
-        DescriptorBufferRing& descriptor_buffer_ring, Common::ThreadWorker* worker_thread,
+        GuestDescriptorQueue& guest_descriptor_queue, Common::ThreadWorker* worker_thread,
         PipelineStatistics* pipeline_statistics, RenderPassCache& render_pass_cache,
         const GraphicsPipelineCacheKey& key, std::array<vk::ShaderModule, NUM_STAGES> stages,
         const std::array<const Shader::Info*, NUM_STAGES>& infos);
@@ -137,7 +132,7 @@ private:
     template <typename Spec>
     bool ConfigureImpl(bool is_indexed);
 
-    bool ConfigureDraw(const RescalingPushConstant& rescaling,
+    void ConfigureDraw(const RescalingPushConstant& rescaling,
                        const RenderAreaPushConstant& render_are);
 
     void MakePipeline(VkRenderPass render_pass);
@@ -153,7 +148,6 @@ private:
     vk::PipelineCache& pipeline_cache;
     Scheduler& scheduler;
     GuestDescriptorQueue& guest_descriptor_queue;
-    DescriptorBufferRing& descriptor_buffer_ring;
 
     bool (*configure_func)(GraphicsPipeline*, bool){};
 
@@ -176,17 +170,10 @@ private:
     vk::DescriptorUpdateTemplate descriptor_update_template;
     vk::Pipeline pipeline;
 
-    DescriptorBufferLayout descriptor_buffer_layout;
-    std::vector<DescriptorUpdateEntry> last_descriptor_payload;
-    VkDeviceSize last_descriptor_buffer_offset{};
-    u32 last_descriptor_buffer_chunk{};
-    u64 last_descriptor_buffer_generation{};
-
     std::condition_variable build_condvar;
     std::mutex build_mutex;
     std::atomic_bool is_built{false};
     bool uses_push_descriptor{false};
-    bool uses_descriptor_buffer{false};
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -305,7 +305,7 @@ size_t GetTotalPipelineWorkers() {
         std::max<size_t>(static_cast<size_t>(std::thread::hardware_concurrency()), 2ULL) - 1ULL;
 #ifdef __ANDROID__
     const int configured = AndroidSettings::values.pipeline_worker_count.GetValue();
-    const int clamped = std::clamp(configured, 4, 8);
+    const int clamped = std::clamp(configured, 2, 8);
     const size_t desired = static_cast<size_t>(clamped);
     if (desired == 0) {
         return 1ULL;
@@ -340,19 +340,18 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
                              const Device& device_, Scheduler& scheduler_,
                              DescriptorPool& descriptor_pool_,
                              GuestDescriptorQueue& guest_descriptor_queue_,
-                             DescriptorBufferRing& descriptor_buffer_ring_,
                              RenderPassCache& render_pass_cache_, BufferCache& buffer_cache_,
                              TextureCache& texture_cache_, VideoCore::ShaderNotify& shader_notify_)
     : VideoCommon::ShaderCache{device_memory_}, device{device_}, scheduler{scheduler_},
       descriptor_pool{descriptor_pool_}, guest_descriptor_queue{guest_descriptor_queue_},
-      descriptor_buffer_ring{descriptor_buffer_ring_},
       render_pass_cache{render_pass_cache_}, buffer_cache{buffer_cache_},
       texture_cache{texture_cache_}, shader_notify{shader_notify_},
       use_asynchronous_shaders{Settings::values.use_asynchronous_shaders.GetValue()},
       use_vulkan_pipeline_cache{Settings::values.use_vulkan_driver_pipeline_cache.GetValue()},
       workers(device.HasBrokenParallelShaderCompiling() ? 1ULL : GetTotalPipelineWorkers(),
-              "VkPipelineBuilder"),
-      serialization_thread(1, "VkPipelineSerialization") {
+              "VkPipelineBuilder", {}, Common::ThreadPlacement::Background),
+      serialization_thread(1, "VkPipelineSerialization", {},
+                           Common::ThreadPlacement::Background) {
     const auto& float_control{device.FloatControlProperties()};
     const VkDriverId driver_id{device.GetDriverID()};
     const VkShaderStageFlags subgroup_stages{device.GetSubgroupSupportedStages()};
@@ -516,6 +515,11 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
     dynamic_features.has_dynamic_state3_depth_clamp_enable =
         dynamic_features.has_extended_dynamic_state_3_enables &&
         device.SupportsDynamicState3DepthClampEnable();
+    dynamic_features.has_dynamic_state3_conservative_raster_mode =
+        dynamic_features.has_extended_dynamic_state_3_enables &&
+        device.SupportsDynamicState3ConservativeRasterizationMode();
+    dynamic_features.has_depth_clip_enable = 
+        device.IsExtDepthClipEnableSupported();
     dynamic_features.has_dynamic_state3_logic_op_enable =
         dynamic_features.has_extended_dynamic_state_3_enables &&
         device.SupportsDynamicState3LogicOpEnable();
@@ -528,7 +532,8 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
         device.IsExtVertexInputDynamicStateSupported() &&
         Settings::values.vertex_input_dynamic_state.GetValue();
 
-    dynamic_features.has_provoking_vertex = device.IsExtProvokingVertexSupported();
+    dynamic_features.has_provoking_vertex =
+        device.IsExtProvokingVertexSupported();
     dynamic_features.has_provoking_vertex_first_mode =
         device.SupportsProvokingVertexFirstMode();
     dynamic_features.has_provoking_vertex_last_mode =
@@ -838,8 +843,8 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
     Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr};
     return std::make_unique<GraphicsPipeline>(
         scheduler, buffer_cache, texture_cache, vulkan_pipeline_cache, &shader_notify, device,
-        descriptor_pool, guest_descriptor_queue, descriptor_buffer_ring, thread_worker, statistics,
-        render_pass_cache, key, std::move(modules), infos);
+        descriptor_pool, guest_descriptor_queue, thread_worker, statistics, render_pass_cache, key,
+        std::move(modules), infos);
 
 } catch (const Shader::Exception& exception) {
     auto hash = key.Hash();
@@ -959,8 +964,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
     }
     Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr};
     return std::make_unique<ComputePipeline>(device, scheduler, vulkan_pipeline_cache, descriptor_pool,
-                                             guest_descriptor_queue, descriptor_buffer_ring,
-                                             thread_worker, statistics,
+                                             guest_descriptor_queue, thread_worker, statistics,
                                              &shader_notify, program.info, std::move(spv_module),
                                              key.unique_hash);
 

src/video_core/renderer_vulkan/vk_pipeline_cache.h

@@ -105,7 +105,6 @@ public:
     explicit PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, const Device& device,
                            Scheduler& scheduler, DescriptorPool& descriptor_pool,
                            GuestDescriptorQueue& guest_descriptor_queue,
-                           DescriptorBufferRing& descriptor_buffer_ring,
                            RenderPassCache& render_pass_cache, BufferCache& buffer_cache,
                            TextureCache& texture_cache, VideoCore::ShaderNotify& shader_notify_);
     ~PipelineCache();
@@ -148,7 +147,6 @@ private:
     Scheduler& scheduler;
     DescriptorPool& descriptor_pool;
     GuestDescriptorQueue& guest_descriptor_queue;
-    DescriptorBufferRing& descriptor_buffer_ring;
     RenderPassCache& render_pass_cache;
     BufferCache& buffer_cache;
     TextureCache& texture_cache;

src/video_core/renderer_vulkan/vk_present_manager.cpp

@@ -296,9 +296,6 @@ void PresentManager::RecreateSwapchain(Frame* frame) {
 }
 
 void PresentManager::SetImageCount() {
-    // We cannot have more than 7 images in flight at any given time.
-    // FRAMES_IN_FLIGHT is 8, and the cache TICKS_TO_DESTROY is 8.
-    // Mali drivers will give us 6.
     image_count = std::min<size_t>(swapchain.GetImageCount(), 7);
 }
 

src/video_core/renderer_vulkan/vk_query_cache.cpp

@@ -919,7 +919,7 @@ private:
             return;
         }
         has_flushed_end_pending = true;
-        // Refresh buffers state before beginning transform feedback so counters are up-to-date
+        scheduler.MarkTransformFeedbackUsed();
         UpdateBuffers();
         if (!has_started || buffers_count == 0) {
             // No counter buffers available: begin without counters

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -6,6 +6,7 @@
 
 #include <algorithm>
 #include <array>
+#include <limits>
 #include <memory>
 #include <mutex>
 
@@ -203,10 +204,8 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
     : 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),
-      guest_descriptor_queue(device, UpdateDescriptorQueue::GUEST_FRAME_PAYLOAD_SIZE,
-                             device.IsExtDescriptorBufferSupported()),
+      guest_descriptor_queue(device, UpdateDescriptorQueue::GUEST_FRAME_PAYLOAD_SIZE),
       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),
       texture_cache_runtime{
           device,     scheduler,         memory_allocator, staging_pool,
@@ -219,15 +218,28 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
                           staging_pool, compute_pass_descriptor_queue, descriptor_pool, texture_cache),
       query_cache(gpu, *this, device_memory, query_cache_runtime),
       pipeline_cache(device_memory, device, scheduler, descriptor_pool, guest_descriptor_queue,
-                     descriptor_buffer_ring, render_pass_cache, buffer_cache, texture_cache,
-                     gpu.ShaderNotify()),
+                     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()) {
     scheduler.SetQueryCache(query_cache);
+    memory_allocator.SetReclaimCallback([this](u64 bytes) -> u64 {
+        u64 freed = staging_pool.ReclaimMemory(bytes);
+        if (freed < bytes) {
+            freed += texture_cache.ReclaimMemory(bytes - freed, false);
+        }
+        if (freed < bytes) {
+            freed += buffer_cache.ReclaimMemory(bytes - freed, false);
+        }
+        auto& master_semaphore = scheduler.GetMasterSemaphore();
+        master_semaphore.Refresh();
+        vk::TickDeletionQueue(master_semaphore.KnownGpuTick());
+        return freed;
+    });
 }
 
 RasterizerVulkan::~RasterizerVulkan() {
+    memory_allocator.SetReclaimCallback(nullptr);
     scheduler.WaitWorker();
     scheduler.Finish();
 }
@@ -587,10 +599,7 @@ void RasterizerVulkan::DispatchCompute() {
         return;
     }
     std::scoped_lock lock{texture_cache.mutex, buffer_cache.mutex};
-    if (!pipeline->Configure(*kepler_compute, *gpu_memory, scheduler, buffer_cache,
-                             texture_cache)) {
-        return;
-    }
+    pipeline->Configure(*kepler_compute, *gpu_memory, scheduler, buffer_cache, texture_cache);
 
     const auto& qmd{kepler_compute->launch_description};
     auto indirect_address = kepler_compute->GetIndirectComputeAddress();
@@ -887,9 +896,11 @@ void RasterizerVulkan::FlushCommands() {
 
 void RasterizerVulkan::TickFrame() {
     draw_counter = 0;
+    auto& master_semaphore = scheduler.GetMasterSemaphore();
+    master_semaphore.Refresh();
+    vk::TickDeletionQueue(master_semaphore.KnownGpuTick());
     guest_descriptor_queue.TickFrame();
     compute_pass_descriptor_queue.TickFrame();
-    descriptor_buffer_ring.TickFrame();
     fence_manager.TickFrame();
     staging_pool.TickFrame();
     {
@@ -1459,7 +1470,10 @@ void RasterizerVulkan::UpdateLineWidth(Tegra::Engines::Maxwell3D::Regs& regs) {
     }
     const float width =
         regs.line_anti_alias_enable ? regs.line_width_smooth : regs.line_width_aliased;
-    scheduler.Record([width](vk::CommandBuffer cmdbuf) { cmdbuf.SetLineWidth(width); });
+    const float clamped_width = device.ClampLineWidth(width);
+    scheduler.Record([clamped_width](vk::CommandBuffer cmdbuf) {
+        cmdbuf.SetLineWidth(clamped_width);
+    });
 }
 
 void RasterizerVulkan::UpdateCullMode(Tegra::Engines::Maxwell3D::Regs& regs) {
@@ -1556,7 +1570,10 @@ void RasterizerVulkan::UpdateLineStippleEnable(Tegra::Engines::Maxwell3D::Regs&
         return;
     }
 
-    scheduler.Record([enable = regs.line_stipple_enable](vk::CommandBuffer cmdbuf) {
+    const VkLineRasterizationModeEXT mode =
+        device.GetLineRasterizationMode(regs.line_anti_alias_enable != 0);
+    const bool enable = regs.line_stipple_enable != 0 && device.SupportsStippleForMode(mode);
+    scheduler.Record([enable](vk::CommandBuffer cmdbuf) {
         cmdbuf.SetLineStippleEnableEXT(enable);
     });
 }
@@ -1570,28 +1587,24 @@ void RasterizerVulkan::UpdateLineRasterizationMode(Tegra::Engines::Maxwell3D::Re
     }
 
     if (!device.SupportsDynamicState3LineRasterizationMode()) {
-        static std::once_flag warn_missing_rect;
-        std::call_once(warn_missing_rect, [] {
+        static std::once_flag warn_missing_dynamic_state;
+        std::call_once(warn_missing_dynamic_state, [] {
             LOG_WARNING(Render_Vulkan,
-                        "Driver lacks rectangular line rasterization support; skipping dynamic "
-                        "line state updates");
+                        "Driver lacks dynamic line rasterization mode; the pipeline static value "
+                        "is used instead");
         });
         return;
     }
 
     const bool wants_smooth = regs.line_anti_alias_enable != 0;
-    VkLineRasterizationModeEXT mode = VK_LINE_RASTERIZATION_MODE_RECTANGULAR_EXT;
-    if (wants_smooth) {
-        if (device.SupportsSmoothLines()) {
-            mode = VK_LINE_RASTERIZATION_MODE_RECTANGULAR_SMOOTH_EXT;
-        } else {
-            static std::once_flag warn_missing_smooth;
-            std::call_once(warn_missing_smooth, [] {
-                LOG_WARNING(Render_Vulkan,
-                            "Line anti-aliasing requested but smoothLines feature unavailable; "
-                            "using rectangular rasterization");
-            });
-        }
+    const VkLineRasterizationModeEXT mode = device.GetLineRasterizationMode(wants_smooth);
+    if (wants_smooth && mode != VK_LINE_RASTERIZATION_MODE_RECTANGULAR_SMOOTH_EXT) {
+        static std::once_flag warn_missing_smooth;
+        std::call_once(warn_missing_smooth, [] {
+            LOG_WARNING(Render_Vulkan,
+                        "Line anti-aliasing requested but smoothLines feature unavailable; "
+                        "falling back to the closest supported mode");
+        });
     }
     scheduler.Record([mode](vk::CommandBuffer cmdbuf) {
         cmdbuf.SetLineRasterizationModeEXT(mode);
@@ -1651,12 +1664,7 @@ void RasterizerVulkan::UpdateDepthClampEnable(Tegra::Engines::Maxwell3D::Regs& r
     if (!device.SupportsDynamicState3DepthClampEnable()) {
         return;
     }
-    bool is_enabled = !(regs.viewport_clip_control.geometry_clip ==
-                            Maxwell::ViewportClipControl::GeometryClip::Passthrough ||
-                        regs.viewport_clip_control.geometry_clip ==
-                            Maxwell::ViewportClipControl::GeometryClip::FrustumXYZ ||
-                        regs.viewport_clip_control.geometry_clip ==
-                            Maxwell::ViewportClipControl::GeometryClip::FrustumZ);
+    const bool is_enabled = IsDepthClampEnabled(regs, device.IsExtDepthClipEnableSupported());
     scheduler.Record(
         [is_enabled](vk::CommandBuffer cmdbuf) { cmdbuf.SetDepthClampEnableEXT(is_enabled); });
 }

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_descriptor_buffer.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
 #include "video_core/renderer_vulkan/vk_fence_manager.h"
 #include "video_core/renderer_vulkan/vk_pipeline_cache.h"
@@ -180,7 +179,6 @@ 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);
@@ -208,7 +206,6 @@ private:
     DescriptorPool descriptor_pool;
     GuestDescriptorQueue guest_descriptor_queue;
     ComputePassDescriptorQueue compute_pass_descriptor_queue;
-    DescriptorBufferRing descriptor_buffer_ring;
     BlitImageHelper blit_image;
     RenderPassCache render_pass_cache;
 

src/video_core/renderer_vulkan/vk_resource_pool.cpp

@@ -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
 
@@ -15,7 +12,10 @@ ResourcePool::ResourcePool(MasterSemaphore& master_semaphore_, size_t grow_step_
     : master_semaphore{&master_semaphore_}, grow_step{grow_step_} {}
 
 size_t ResourcePool::CommitResource() {
-    const auto search = [this](size_t begin, size_t end, u64 gpu_tick) -> std::optional<size_t> {
+    // Refresh semaphore to query updated results
+    master_semaphore->Refresh();
+    const u64 gpu_tick = master_semaphore->KnownGpuTick();
+    const auto search = [this, gpu_tick](size_t begin, size_t end) -> std::optional<size_t> {
         for (size_t iterator = begin; iterator < end; ++iterator) {
             if (gpu_tick >= ticks[iterator]) {
                 ticks[iterator] = master_semaphore->CurrentTick();
@@ -24,17 +24,11 @@ size_t ResourcePool::CommitResource() {
         }
         return std::nullopt;
     };
-    const auto find_free = [&](u64 gpu_tick) -> std::optional<size_t> {
-        std::optional<size_t> result = search(hint_iterator, ticks.size(), gpu_tick);
-        if (!result) {
-            result = search(0, hint_iterator, gpu_tick);
-        }
-        return result;
-    };
-    std::optional<size_t> found = find_free(master_semaphore->KnownGpuTick());
+    // Try to find a free resource from the hinted position to the end.
+    std::optional<size_t> found = search(hint_iterator, ticks.size());
     if (!found) {
-        master_semaphore->Refresh();
-        found = find_free(master_semaphore->KnownGpuTick());
+        // Search from beginning to the hinted position.
+        found = search(0, hint_iterator);
         if (!found) {
             // Both searches failed, the pool is full; handle it.
             const size_t free_resource = ManageOverflow();

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -47,6 +47,7 @@ Scheduler::Scheduler(const Device& device_, StateTracker& state_tracker_)
       master_semaphore{std::make_unique<MasterSemaphore>(device)},
       command_pool{std::make_unique<CommandPool>(*master_semaphore, device)} {
 
+    vk::SetDeletionTimeline(master_semaphore->CurrentTick());
     AcquireNewChunk();
     AllocateWorkerCommandBuffer();
     worker_thread = std::jthread([this](std::stop_token token) { WorkerThread(token); });
@@ -247,15 +248,6 @@ bool Scheduler::UpdateRescaling(bool is_rescaling) {
     return true;
 }
 
-bool Scheduler::UpdateDescriptorBufferChunk(u32 descriptor_chunk) {
-    if (state.descriptor_buffer_bound && descriptor_chunk == state.descriptor_buffer_chunk) {
-        return false;
-    }
-    state.descriptor_buffer_bound = true;
-    state.descriptor_buffer_chunk = descriptor_chunk;
-    return true;
-}
-
 void Scheduler::WorkerThread(std::stop_token stop_token) {
     Common::SetCurrentThreadName("VulkanWorker");
 
@@ -331,6 +323,7 @@ u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_se
     InvalidateState();
 
     const u64 signal_value = master_semaphore->NextTick();
+    vk::SetDeletionTimeline(master_semaphore->CurrentTick());
     RecordWithUploadBuffer([signal_semaphore, wait_semaphore, signal_value,
                             this](vk::CommandBuffer cmdbuf, vk::CommandBuffer upload_cmdbuf) {
         static constexpr VkMemoryBarrier WRITE_BARRIER{
@@ -378,7 +371,6 @@ void Scheduler::AllocateNewContext() {
 void Scheduler::InvalidateState() {
     state.graphics_pipeline = nullptr;
     state.rescaling_defined = false;
-    state.descriptor_buffer_bound = false;
     state_tracker.InvalidateCommandBufferState();
 }
 
@@ -408,7 +400,7 @@ void Scheduler::EndRenderPass()
         Record([num_images = num_renderpass_images,
                        images = renderpass_images,
                        ranges = renderpass_image_ranges,
-                       has_transform_feedback = device.IsExtTransformFeedbackSupported()](
+                       has_transform_feedback = state.uses_transform_feedback](
                           vk::CommandBuffer cmdbuf) {
             std::array<VkImageMemoryBarrier, 9> barriers;
             for (size_t i = 0; i < num_images; ++i) {
@@ -463,6 +455,7 @@ void Scheduler::EndRenderPass()
         });
 
         state.renderpass = VkRenderPass{};
+        state.uses_transform_feedback = false;
         num_renderpass_images = 0;
     }
 

src/video_core/renderer_vulkan/vk_scheduler.h

@@ -74,15 +74,17 @@ public:
         return state.renderpass != VK_NULL_HANDLE;
     }
 
+    /// Flags that transform feedback writes have been recorded since the last render pass end.
+    void MarkTransformFeedbackUsed() noexcept {
+        state.uses_transform_feedback = true;
+    }
+
     /// Update the pipeline to the current execution context.
     bool UpdateGraphicsPipeline(GraphicsPipeline* pipeline);
 
     /// Update the rescaling state. Returns true if the state has to be updated.
     bool UpdateRescaling(bool is_rescaling);
 
-    /// Returns true when the descriptor buffer chunk has to be bound into the command buffer.
-    bool UpdateDescriptorBufferChunk(u32 descriptor_chunk);
-
     /// Invalidates current command buffer state except for render passes
     void InvalidateState();
 
@@ -134,33 +136,16 @@ public:
             }
             master_semaphore->Wait(tick);
         }
-        if (Settings::values.use_speed_limit.GetValue() && target_fps > 0.0) {
-            auto now = std::chrono::steady_clock::now();
-            if (last_target_fps != target_fps) {
-                frame_interval = std::chrono::duration_cast<std::chrono::steady_clock::duration>(std::chrono::duration<double>(1.0 / target_fps));
-                max_frame_count = static_cast<int>(0.1 * target_fps);
-                last_target_fps = target_fps;
-                frame_counter = 0;
-                start_time = now;
-            }
-            frame_counter++;
-            auto target_time = start_time + frame_interval * frame_counter;
-            if (target_time >= now) {
-                auto sleep_time = target_time - now;
-                if (sleep_time > std::chrono::milliseconds(15)) {
-                    std::this_thread::sleep_for(sleep_time - std::chrono::milliseconds(1));
-                }
-                while (std::chrono::steady_clock::now() < target_time) {
-                    std::this_thread::yield();
-                }
-            } else if (frame_counter > max_frame_count) {
-                frame_counter = 0;
-                start_time = now;
-            }
-        }
+        ApplyFramePacing(target_fps);
+    }
+
+    void WaitSubmitted(u64 tick, double target_fps = 0.0) {
+        if (tick > 0 && tick < master_semaphore->CurrentTick()) {
+            master_semaphore->Wait(tick);
+        }
+        ApplyFramePacing(target_fps);
     }
 
-    /// Returns the master timeline semaphore.
     [[nodiscard]] MasterSemaphore& GetMasterSemaphore() const noexcept {
         return *master_semaphore;
     }
@@ -168,6 +153,35 @@ public:
     std::mutex submit_mutex;
 
 private:
+    void ApplyFramePacing(double target_fps) {
+        if (!Settings::values.use_speed_limit.GetValue() || target_fps <= 0.0) {
+            return;
+        }
+        auto now = std::chrono::steady_clock::now();
+        if (last_target_fps != target_fps) {
+            frame_interval = std::chrono::duration_cast<std::chrono::steady_clock::duration>(
+                std::chrono::duration<double>(1.0 / target_fps));
+            max_frame_count = static_cast<int>(0.1 * target_fps);
+            last_target_fps = target_fps;
+            frame_counter = 0;
+            start_time = now;
+        }
+        frame_counter++;
+        auto target_time = start_time + frame_interval * frame_counter;
+        if (target_time >= now) {
+            auto sleep_time = target_time - now;
+            if (sleep_time > std::chrono::milliseconds(15)) {
+                std::this_thread::sleep_for(sleep_time - std::chrono::milliseconds(1));
+            }
+            while (std::chrono::steady_clock::now() < target_time) {
+                std::this_thread::yield();
+            }
+        } else if (frame_counter > max_frame_count) {
+            frame_counter = 0;
+            start_time = now;
+        }
+    }
+
     class Command {
     public:
         virtual ~Command() = default;
@@ -258,8 +272,7 @@ private:
         bool is_rescaling = false;
         bool rescaling_defined = false;
         bool needs_state_enable_refresh = false;
-        u32 descriptor_buffer_chunk = 0;
-        bool descriptor_buffer_bound = false;
+        bool uses_transform_feedback = false;
     };
 
     struct DeferredClear {

src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp

@@ -84,16 +84,10 @@ StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& mem
     if (device.IsExtTransformFeedbackSupported()) {
         stream_ci.usage |= VK_BUFFER_USAGE_TRANSFORM_FEEDBACK_BUFFER_BIT_EXT;
     }
-    if (device.IsBufferDeviceAddressSupported()) {
-        stream_ci.usage |= VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT;
-    }
     stream_buffer = memory_allocator.CreateBuffer(stream_ci, MemoryUsage::Stream);
     if (device.HasDebuggingToolAttached()) {
         stream_buffer.SetObjectNameEXT("Stream Buffer");
     }
-    if (device.IsBufferDeviceAddressSupported()) {
-        stream_buffer_address = device.GetLogical().GetBufferDeviceAddress(*stream_buffer);
-    }
     stream_pointer = stream_buffer.Mapped();
     ASSERT_MSG(!stream_pointer.empty(), "Stream buffer must be host visible!");
 }
@@ -155,7 +149,6 @@ StagingBufferRef StagingBufferPool::GetStreamBuffer(size_t size) {
     iterator = Common::AlignUp(iterator + size, MAX_ALIGNMENT);
     return StagingBufferRef{
         .buffer = *stream_buffer,
-        .device_address = stream_buffer_address,
         .offset = static_cast<VkDeviceSize>(offset),
         .mapped_span = stream_pointer.subspan(offset, size),
         .usage{},
@@ -219,22 +212,14 @@ StagingBufferRef StagingBufferPool::CreateStagingBuffer(size_t size, MemoryUsage
     if (device.IsExtTransformFeedbackSupported()) {
         buffer_ci.usage |= VK_BUFFER_USAGE_TRANSFORM_FEEDBACK_BUFFER_BIT_EXT;
     }
-    if (device.IsBufferDeviceAddressSupported()) {
-        buffer_ci.usage |= VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT;
-    }
     vk::Buffer buffer = memory_allocator.CreateBuffer(buffer_ci, usage);
     if (device.HasDebuggingToolAttached()) {
         ++buffer_index;
         buffer.SetObjectNameEXT(fmt::format("Staging Buffer {}", buffer_index).c_str());
     }
     const std::span<u8> mapped_span = buffer.Mapped();
-    const VkDeviceAddress buffer_address =
-        device.IsBufferDeviceAddressSupported()
-            ? device.GetLogical().GetBufferDeviceAddress(*buffer)
-            : VkDeviceAddress{};
     StagingBuffer& entry = GetCache(usage)[log2_size].entries.emplace_back(StagingBuffer{
         .buffer = std::move(buffer),
-        .device_address = buffer_address,
         .mapped_span = mapped_span,
         .usage = usage,
         .log2_level = log2_size,
@@ -267,25 +252,62 @@ void StagingBufferPool::ReleaseLevel(StagingBuffersCache& cache, size_t log2) {
     constexpr size_t deletions_per_tick = 16;
     auto& staging = cache[log2];
     auto& entries = staging.entries;
-    const size_t old_size = entries.size();
+    if (entries.empty()) {
+        staging.delete_index = 0;
+        staging.iterate_index = 0;
+        return;
+    }
 
     const auto is_deletable = [this](const StagingBuffer& entry) {
         return scheduler.IsFree(entry.tick);
     };
-    const size_t begin_offset = staging.delete_index;
-    const size_t end_offset = (std::min)(begin_offset + deletions_per_tick, old_size);
+    const size_t begin_offset = (std::min)(staging.delete_index, entries.size());
+    const size_t end_offset = (std::min)(begin_offset + deletions_per_tick, entries.size());
     const auto begin = entries.begin() + begin_offset;
     const auto end = entries.begin() + end_offset;
-    entries.erase(std::remove_if(begin, end, is_deletable), end);
+    const auto surviving_end = std::remove_if(begin, end, is_deletable);
+    const size_t removed = static_cast<size_t>(std::distance(surviving_end, end));
+    entries.erase(surviving_end, end);
 
-    const size_t new_size = entries.size();
-    staging.delete_index += deletions_per_tick;
-    if (staging.delete_index >= new_size) {
+    staging.delete_index = end_offset - removed;
+    if (staging.delete_index >= entries.size()) {
         staging.delete_index = 0;
     }
-    if (staging.iterate_index > new_size) {
+    if (staging.iterate_index > entries.size()) {
         staging.iterate_index = 0;
     }
 }
 
+u64 StagingBufferPool::ReclaimMemory(u64 target_bytes) {
+    u64 freed = 0;
+    const auto is_deletable = [this](const StagingBuffer& entry) {
+        return scheduler.IsFree(entry.tick);
+    };
+    const auto reclaim_cache = [&](StagingBuffersCache& cache) {
+        for (size_t level = NUM_LEVELS; level-- > 0 && freed < target_bytes;) {
+            auto& staging = cache[level];
+            auto& entries = staging.entries;
+            if (entries.empty()) {
+                continue;
+            }
+            const u64 entry_bytes = 1ULL << level;
+            auto it = entries.begin();
+            while (it != entries.end() && freed < target_bytes) {
+                if (is_deletable(*it)) {
+                    it = entries.erase(it);
+                    freed += entry_bytes;
+                } else {
+                    ++it;
+                }
+            }
+            staging.delete_index = 0;
+            staging.iterate_index = 0;
+        }
+    };
+    reclaim_cache(device_local_cache);
+    reclaim_cache(upload_cache);
+    reclaim_cache(download_cache);
+    return freed;
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_staging_buffer_pool.h

@@ -21,7 +21,6 @@ class Scheduler;
 
 struct StagingBufferRef {
     VkBuffer buffer;
-    VkDeviceAddress device_address;
     VkDeviceSize offset;
     std::span<u8> mapped_span;
     MemoryUsage usage;
@@ -46,6 +45,8 @@ public:
 
     void TickFrame();
 
+    u64 ReclaimMemory(u64 target_bytes);
+
 private:
     struct StreamBufferCommit {
         size_t upper_bound;
@@ -54,7 +55,6 @@ private:
 
     struct StagingBuffer {
         vk::Buffer buffer;
-        VkDeviceAddress device_address;
         std::span<u8> mapped_span;
         MemoryUsage usage;
         u32 log2_level;
@@ -65,7 +65,6 @@ private:
         StagingBufferRef Ref() const noexcept {
             return {
                 .buffer = *buffer,
-                .device_address = device_address,
                 .offset = 0,
                 .mapped_span = mapped_span,
                 .usage = usage,
@@ -109,7 +108,6 @@ private:
     Scheduler& scheduler;
 
     vk::Buffer stream_buffer;
-    VkDeviceAddress stream_buffer_address{};
     std::span<u8> stream_pointer;
     VkDeviceSize stream_buffer_size;
     VkDeviceSize region_size;

src/video_core/renderer_vulkan/vk_swapchain.cpp

@@ -9,10 +9,6 @@
 #include <limits>
 #include <vector>
 
-#ifdef __ANDROID__
-#include <android/api-level.h>
-#endif
-
 #include "common/logging.h"
 #include "common/settings.h"
 #include "common/settings_enums.h"
@@ -176,34 +172,26 @@ bool Swapchain::AcquireNextImage() {
         break;
     }
 
-    const auto wait_with_frame_pacing = [this] {
+#ifdef __ANDROID__
+    scheduler.WaitSubmitted(resource_ticks[image_index]);
+#else
     switch (Settings::values.frame_pacing_mode.GetValue()) {
     case Settings::FramePacingMode::Target_Auto:
-        scheduler.Wait(resource_ticks[image_index]);
+        scheduler.WaitSubmitted(resource_ticks[image_index]);
         break;
     case Settings::FramePacingMode::Target_30:
-        scheduler.Wait(resource_ticks[image_index], 30.0);
+        scheduler.WaitSubmitted(resource_ticks[image_index], 30.0);
         break;
     case Settings::FramePacingMode::Target_60:
-        scheduler.Wait(resource_ticks[image_index], 60.0);
+        scheduler.WaitSubmitted(resource_ticks[image_index], 60.0);
         break;
     case Settings::FramePacingMode::Target_90:
-        scheduler.Wait(resource_ticks[image_index], 90.0);
+        scheduler.WaitSubmitted(resource_ticks[image_index], 90.0);
         break;
     case Settings::FramePacingMode::Target_120:
-        scheduler.Wait(resource_ticks[image_index], 120.0);
+        scheduler.WaitSubmitted(resource_ticks[image_index], 120.0);
         break;
     }
-    };
-
-#ifdef __ANDROID__
-    if (android_get_device_api_level() >= 30) {
-        scheduler.Wait(resource_ticks[image_index]);
-    } else {
-        wait_with_frame_pacing();
-    }
-#else
-    wait_with_frame_pacing();
 #endif
 
     resource_ticks[image_index] = scheduler.CurrentTick();

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -207,7 +207,11 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
                 return device.IsFormatSupported(view_format, VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT,
                                                 FormatType::Optimal);
             });
-        if (has_storage_compatible_view) {
+        const bool storage_allowed_for_samples =
+            image_ci.samples == VK_SAMPLE_COUNT_1_BIT ||
+            (device.GetStorageImageSampleCounts() &
+             static_cast<VkSampleCountFlags>(image_ci.samples)) != 0;
+        if (has_storage_compatible_view && storage_allowed_for_samples) {
             image_ci.usage |= VK_IMAGE_USAGE_STORAGE_BIT;
         }
 
@@ -264,10 +268,6 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
     }
 }
 
-[[nodiscard]] bool IsLdrAstcFormat(VkFormat format) {
-    return format >= VK_FORMAT_ASTC_4x4_UNORM_BLOCK && format <= VK_FORMAT_ASTC_12x12_SRGB_BLOCK;
-}
-
 [[nodiscard]] VkImageAspectFlags ImageViewAspectMask(const VideoCommon::ImageViewInfo& info) {
     if (info.IsRenderTarget()) {
         return ImageAspectMask(info.format);
@@ -895,6 +895,15 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const
                        0, nullptr, nullptr, write_barriers);
     });
 }
+
+[[nodiscard]] bool CanBlitNatively(const Device& device, PixelFormat format) {
+    static constexpr auto OPTIMAL_FORMAT = FormatType::Optimal;
+    static constexpr VkFormatFeatureFlags BLIT_USAGE =
+        VK_FORMAT_FEATURE_BLIT_SRC_BIT | VK_FORMAT_FEATURE_BLIT_DST_BIT;
+    const VkFormat vk_format =
+        MaxwellToVK::SurfaceFormat(device, OPTIMAL_FORMAT, false, format).format;
+    return device.IsFormatSupported(vk_format, BLIT_USAGE, OPTIMAL_FORMAT);
+}
 } // Anonymous namespace
 
 TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& scheduler_,
@@ -1227,27 +1236,19 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst
         blit_image_helper.ResolveDepthStencil(dst_framebuffer, src, dst_region, src_region);
         return;
     }
-    if (aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
-        const auto format = src.format;
-        const auto can_blit_depth_stencil = [this, format] {
-            switch (format) {
-            case VideoCore::Surface::PixelFormat::D24_UNORM_S8_UINT:
-            case VideoCore::Surface::PixelFormat::S8_UINT_D24_UNORM:
-                return device.IsBlitDepth24Stencil8Supported();
-            case VideoCore::Surface::PixelFormat::D32_FLOAT_S8_UINT:
-                return device.IsBlitDepth32Stencil8Supported();
-            default:
-                UNREACHABLE();
-            }
-        }();
-        // Use shader-based depth/stencil blits if hardware doesn't support the format
-        // Note: MSAA resolves (MSAA->single) use vkCmdResolveImage which works fine
-        if (!can_blit_depth_stencil) {
-            UNIMPLEMENTED_IF(is_src_msaa || is_dst_msaa);
-            blit_image_helper.BlitDepthStencil(dst_framebuffer, src, dst_region, src_region,
-                                               filter, operation);
+    static constexpr VkImageAspectFlags DEPTH_STENCIL_ASPECTS =
+        VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT;
+    if ((aspect_mask & DEPTH_STENCIL_ASPECTS) != 0 && !CanBlitNatively(device, src.format)) {
+        if (aspect_mask != DEPTH_STENCIL_ASPECTS) {
+            UNIMPLEMENTED_MSG("Host cannot blit format {} and no helper path exists for aspect "
+                              "mask 0x{:x}",
+                              src.format, aspect_mask);
             return;
         }
+        UNIMPLEMENTED_IF(is_src_msaa || is_dst_msaa);
+        blit_image_helper.BlitDepthStencil(dst_framebuffer, src, dst_region, src_region, filter,
+                                           operation);
+        return;
     }
     ASSERT(!(is_dst_msaa && !is_src_msaa));
     ASSERT(operation == Fermi2D::Operation::SrcCopy);
@@ -1642,7 +1643,14 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src,
     const u32 num_samples = msaa_to_non_msaa ? src.info.num_samples : dst.info.num_samples;
     if (dst.AspectMask() != VK_IMAGE_ASPECT_COLOR_BIT ||
         VideoCore::Surface::IsPixelFormatInteger(dst.info.format)) {
-        UNIMPLEMENTED_MSG("Copying images with different samples is not supported.");
+        const u64 key{(static_cast<u64>(dst.AspectMask()) << 32) |
+                      static_cast<u64>(dst.info.format)};
+        if (unsupported_msaa_resolves.insert(key).second) {
+            LOG_WARNING(Render_Vulkan,
+                        "MSAA resolve unsupported: format={}, aspect={:#x}, samples {}->{}",
+                        dst.info.format, dst.AspectMask(), src.info.num_samples,
+                        dst.info.num_samples);
+        }
         return;
     }
     if (ENABLE_MSAA_RESOLVE_CONSUME && msaa_to_non_msaa && copies.size() == 1 &&
@@ -1752,6 +1760,20 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src,
                                src.info.format, num_samples, copies, msaa_to_non_msaa);
 }
 
+u64 TextureCacheRuntime::CurrentSyncPoint() const noexcept {
+    return scheduler.CurrentTick();
+}
+
+u64 TextureCacheRuntime::CompletedSyncPoint() const {
+    auto& master_semaphore = scheduler.GetMasterSemaphore();
+    master_semaphore.Refresh();
+    return master_semaphore.KnownGpuTick();
+}
+
+void TextureCacheRuntime::WaitSyncPoint(u64 sync_point) {
+    scheduler.Wait(sync_point);
+}
+
 u64 TextureCacheRuntime::GetDeviceLocalMemory() const {
     return device.GetDeviceLocalMemory();
 }
@@ -1760,6 +1782,10 @@ u64 TextureCacheRuntime::GetDeviceMemoryUsage() const {
     return device.GetDeviceMemoryUsage();
 }
 
+u64 TextureCacheRuntime::GetDeviceAllocationUsage() const {
+    return device.GetMemoryBudgetInfo().allocation_bytes;
+}
+
 bool TextureCacheRuntime::CanReportMemoryUsage() const {
     return device.CanReportMemoryUsage();
 }
@@ -1769,6 +1795,7 @@ std::optional<size_t> TextureCacheRuntime::GetSamplerHeapBudget() const {
 }
 
 void TextureCacheRuntime::TickFrame() {
+    device.TickAllocatorFrame();
     std::erase_if(pending_msaa_images, [this](const auto& pending) {
         return scheduler.IsFree(pending.first);
     });
@@ -1806,7 +1833,9 @@ 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)) {
+    if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported() &&
+        Settings::values.astc_recompression.GetValue() ==
+            Settings::AstcRecompression::Uncompressed) {
         const auto& device = runtime->device.GetLogical();
         const VkFormat storage_format = VK_FORMAT_A8B8G8R8_UNORM_PACK32;
         for (s32 level = 0; level < info.resources.levels; ++level) {
@@ -1882,9 +1911,11 @@ void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
         ScaleDown(true);
     }
 
-    const bool wants_msaa_upload = info.num_samples > 1
-        && (aspect_mask & VK_IMAGE_ASPECT_COLOR_BIT) != 0
+    const bool is_color_upload = (aspect_mask & VK_IMAGE_ASPECT_COLOR_BIT) != 0
         && !VideoCore::Surface::IsPixelFormatInteger(info.format);
+    const bool is_depth_upload = (aspect_mask & VK_IMAGE_ASPECT_DEPTH_BIT) != 0;
+    const bool wants_msaa_upload =
+        info.num_samples > 1 && (is_color_upload || is_depth_upload);
 
     if (wants_msaa_upload) {
         ImageInfo temp_info = info;
@@ -1923,10 +1954,10 @@ void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
             image_copies.push_back(image_copy);
         }
 
+        runtime->TransitionImageLayout(*this);
         runtime->blit_image_helper.CopyMSAA(runtime->render_pass_cache, Handle(), info.format,
                                             temp_vk_image, info.format, info.num_samples,
                                             image_copies, false);
-        initialized = true;
         runtime->pending_msaa_images.emplace_back(scheduler->CurrentTick(), std::move(temp_image));
 
         if (is_rescaled) {
@@ -1937,6 +1968,9 @@ void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
 
     if (info.num_samples > 1) {
         LOG_WARNING(Render_Vulkan, "MSAA upload not implemented for format {}", info.format);
+        if (runtime != nullptr) {
+            runtime->TransitionImageLayout(*this);
+        }
         if (is_rescaled) {
             ScaleUp();
         }
@@ -2388,18 +2422,9 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
         .pNext = nullptr,
         .usage = clamped_view_usage,
     };
-    const VkImageViewASTCDecodeModeEXT astc_decode_mode{
-        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_ASTC_DECODE_MODE_EXT,
-        .pNext = &image_view_usage,
-        .decodeMode = VK_FORMAT_R8G8B8A8_UNORM,
-    };
-    const void* view_next = &image_view_usage;
-    if (device->IsExtAstcDecodeModeSupported() && IsLdrAstcFormat(format_info.format)) {
-        view_next = &astc_decode_mode;
-    }
     const VkImageViewCreateInfo create_info{
         .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
-        .pNext = view_next,
+        .pNext = &image_view_usage,
         .flags = 0,
         .image = image.Handle(),
         .viewType = VkImageViewType{},
@@ -2522,11 +2547,12 @@ VkImageView ImageView::StorageView(Shader::TextureType texture_type,
                                    Shader::ImageFormat image_format) {
     if (image_handle) {
         if (image_format == Shader::ImageFormat::Typeless) {
-            if (!typeless_storage_view) {
+            auto& view{typeless_storage_views[static_cast<size_t>(texture_type)]};
+            if (!view) {
                 auto info = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, true, format);
-                typeless_storage_view = MakeView(info.format, VK_IMAGE_ASPECT_COLOR_BIT, texture_type);
+                view = MakeView(info.format, VK_IMAGE_ASPECT_COLOR_BIT, texture_type);
             }
-            return *typeless_storage_view;
+            return *view;
         }
         const bool is_signed = image_format == Shader::ImageFormat::R8_SINT
             || image_format == Shader::ImageFormat::R16_SINT;

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -60,10 +60,22 @@ public:
 
     void TickFrame();
 
+    u64 CurrentSyncPoint() const noexcept;
+
+    u64 CompletedSyncPoint() const;
+
+    void WaitSyncPoint(u64 sync_point);
+
     u64 GetDeviceLocalMemory() const;
 
     u64 GetDeviceMemoryUsage() const;
 
+    u64 GetDeviceAllocationUsage() const;
+
+    bool CanReportAllocationUsage() const noexcept {
+        return true;
+    }
+
     bool CanReportMemoryUsage() const;
 
     std::optional<size_t> GetSamplerHeapBudget() const;
@@ -156,6 +168,7 @@ public:
     std::array<vk::Buffer, indexing_slots> buffers{};
     std::vector<std::pair<u64, vk::Image>> pending_msaa_images;
     ankerl::unordered_dense::map<VkImage, ResolveShadow> resolve_shadows;
+    ankerl::unordered_dense::set<u64> unsupported_msaa_resolves;
 };
 
 class Framebuffer {
@@ -426,7 +439,7 @@ private:
 
     std::array<vk::ImageView, Shader::NUM_TEXTURE_TYPES> image_views;
     std::optional<StorageViews> storage_views;
-    vk::ImageView typeless_storage_view;
+    std::array<vk::ImageView, Shader::NUM_TEXTURE_TYPES> typeless_storage_views;
     vk::ImageView depth_view;
     vk::ImageView stencil_view;
     vk::ImageView color_view;
@@ -486,6 +499,7 @@ struct TextureCacheParams {
     static constexpr bool HAS_EMULATED_COPIES = false;
     static constexpr bool HAS_DEVICE_MEMORY_INFO = true;
     static constexpr bool IMPLEMENTS_ASYNC_DOWNLOADS = true;
+    static constexpr bool HAS_TIMELINE_SYNC_POINTS = true;
 
     using Runtime = Vulkan::TextureCacheRuntime;
     using Image = Vulkan::Image;

src/video_core/renderer_vulkan/vk_update_descriptor.cpp

@@ -16,10 +16,8 @@
 
 namespace Vulkan {
 
-UpdateDescriptorQueue::UpdateDescriptorQueue(const Device& device_, size_t frame_payload_size_,
-                                             bool supports_descriptor_buffer_)
+UpdateDescriptorQueue::UpdateDescriptorQueue(const Device& device_, size_t frame_payload_size_)
     : device{device_}, frame_payload_size{frame_payload_size_},
-      supports_descriptor_buffer{supports_descriptor_buffer_},
       payload(frame_payload_size_ * FRAMES_IN_FLIGHT)
 {
     payload_start = payload.data();
@@ -36,9 +34,7 @@ void UpdateDescriptorQueue::TickFrame() {
     payload_cursor = payload_start;
 }
 
-void UpdateDescriptorQueue::Acquire(Scheduler& scheduler, size_t required_entries,
-                                    bool use_descriptor_buffer_) {
-    use_descriptor_buffer = supports_descriptor_buffer && use_descriptor_buffer_;
+void UpdateDescriptorQueue::Acquire(Scheduler& scheduler, size_t required_entries) {
     static constexpr size_t DEFAULT_REQUIRED_ENTRIES = 0x400;
     const size_t reserve = required_entries > 0 ? required_entries : DEFAULT_REQUIRED_ENTRIES;
     ASSERT_MSG(reserve < frame_payload_size, "Descriptor reservation {} >= frame capacity {}",

src/video_core/renderer_vulkan/vk_update_descriptor.h

@@ -15,23 +15,15 @@ namespace Vulkan {
 class Device;
 class Scheduler;
 
-struct DescriptorAddress {
-    VkDeviceAddress address;
-    VkDeviceSize range;
-    VkFormat format;
-};
-
 union DescriptorUpdateEntry {
     DescriptorUpdateEntry() = default;
     DescriptorUpdateEntry(VkDescriptorImageInfo image_) : image{image_} {}
     DescriptorUpdateEntry(VkDescriptorBufferInfo buffer_) : buffer{buffer_} {}
     DescriptorUpdateEntry(VkBufferView texel_buffer_) : texel_buffer{texel_buffer_} {}
-    DescriptorUpdateEntry(DescriptorAddress address_) : address{address_} {}
     std::monostate empty{};
     VkDescriptorImageInfo image;
     VkDescriptorBufferInfo buffer;
     VkBufferView texel_buffer;
-    DescriptorAddress address;
 };
 
 class UpdateDescriptorQueue final {
@@ -43,17 +35,11 @@ public:
     static constexpr size_t GUEST_FRAME_PAYLOAD_SIZE = 0x80000;
     static constexpr size_t COMPUTE_FRAME_PAYLOAD_SIZE = 0x20000;
 
-    explicit UpdateDescriptorQueue(const Device& device_, size_t frame_payload_size_,
-                                   bool supports_descriptor_buffer_ = false);
+    explicit UpdateDescriptorQueue(const Device& device_, size_t frame_payload_size_);
     ~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);
+    void Acquire(Scheduler& scheduler, size_t required_entries = 0);
 
     const DescriptorUpdateEntry* UpdateData() const noexcept {
         return upload_start;
@@ -83,41 +69,13 @@ public:
         };
     }
 
-    void AddBuffer(VkBuffer buffer, VkDeviceAddress base_address, VkDeviceSize offset,
-                   VkDeviceSize size) {
-        if (!use_descriptor_buffer) {
-            AddBuffer(buffer, offset, size);
-            return;
-        }
-        *(payload_cursor++) = DescriptorAddress{
-            .address = base_address == 0 ? 0 : base_address + offset,
-            .range = base_address == 0 ? VK_WHOLE_SIZE : size,
-            .format = VK_FORMAT_UNDEFINED,
-        };
-    }
-
     void AddTexelBuffer(VkBufferView texel_buffer) {
         *(payload_cursor++) = texel_buffer;
     }
 
-    void AddTexelBuffer(VkBufferView texel_buffer, VkDeviceAddress base_address,
-                        VkDeviceSize offset, VkDeviceSize size, VkFormat format) {
-        if (!use_descriptor_buffer) {
-            AddTexelBuffer(texel_buffer);
-            return;
-        }
-        *(payload_cursor++) = DescriptorAddress{
-            .address = base_address == 0 ? 0 : base_address + offset,
-            .range = base_address == 0 ? VK_WHOLE_SIZE : size,
-            .format = format,
-        };
-    }
-
 private:
     const Device& device;
     const size_t frame_payload_size;
-    const bool supports_descriptor_buffer;
-    bool use_descriptor_buffer{false};
     size_t frame_index{0};
     DescriptorUpdateEntry* payload_cursor = nullptr;
     DescriptorUpdateEntry* payload_start = nullptr;

src/video_core/shader_environment.cpp

@@ -284,7 +284,24 @@ std::optional<u64> GenericEnvironment::TryFindSize() {
 Tegra::Texture::TICEntry GenericEnvironment::ReadTextureInfo(GPUVAddr tic_addr, u32 tic_limit,
                                                              bool via_header_index, u32 raw) {
     const auto handle{Tegra::Texture::TexturePair(raw, via_header_index)};
-    ASSERT(handle.first <= tic_limit);
+    if (handle.first > tic_limit) {
+        LOG_CRITICAL(Shader,
+                    "TIC index out of range: raw=0x{:08x} tic_index={} tsc_index={} tic_limit={} "
+                    "tic_addr=0x{:x} via_header_index={} stage={} program_base=0x{:x} "
+                    "start_address=0x{:x}",
+                    raw, handle.first, handle.second, tic_limit, tic_addr, via_header_index,
+                    static_cast<u32>(stage), program_base, start_address);
+        ASSERT(handle.first <= tic_limit);
+        Tegra::Texture::TICEntry fallback{};
+        fallback.format.Assign(Tegra::Texture::TextureFormat::A8B8G8R8);
+        fallback.r_type.Assign(Tegra::Texture::ComponentType::UNORM);
+        fallback.g_type.Assign(Tegra::Texture::ComponentType::UNORM);
+        fallback.b_type.Assign(Tegra::Texture::ComponentType::UNORM);
+        fallback.a_type.Assign(Tegra::Texture::ComponentType::UNORM);
+        fallback.texture_type.Assign(Tegra::Texture::TextureType::Texture2D);
+        fallback.normalized_coords.Assign(1);
+        return fallback;
+    }
     const GPUVAddr descriptor_addr{tic_addr + handle.first * sizeof(Tegra::Texture::TICEntry)};
     Tegra::Texture::TICEntry entry;
     gpu_memory->ReadBlock(descriptor_addr, &entry, sizeof(entry));

src/video_core/texture_cache/texture_cache.h

@@ -58,23 +58,9 @@ TextureCache<P>::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<s64>(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<u64>(
-            (std::max)((std::min)(device_local_memory - min_vacancy_expected, min_spacing_expected),
-                     DEFAULT_EXPECTED_MEMORY));
-        critical_memory = static_cast<u64>(
-            (std::max)((std::min)(device_local_memory - min_vacancy_critical, min_spacing_critical),
-                     DEFAULT_CRITICAL_MEMORY));
-        minimum_memory = static_cast<u64>((device_local_memory - mem_threshold) / 2);
+        memory_budget = runtime.GetDeviceLocalMemory();
     } else {
-        expected_memory = DEFAULT_EXPECTED_MEMORY + 512_MiB;
-        critical_memory = DEFAULT_CRITICAL_MEMORY + 1_GiB;
-        minimum_memory = 0;
+        memory_budget = FALLBACK_MEMORY_BUDGET;
     }
 
     const bool gpu_unswizzle_enabled = Settings::values.gpu_unswizzle_enabled.GetValue();
@@ -114,71 +100,154 @@ TextureCache<P>::TextureCache(Runtime& runtime_, Tegra::MaxwellDeviceMemoryManag
 }
 
 template <class P>
-void TextureCache<P>::RunGarbageCollector() {
-    bool high_priority_mode = false;
-    bool aggressive_mode = false;
-    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;
-        ticks_to_destroy = aggressive_mode ? 10ULL : high_priority_mode ? 25ULL : 50ULL;
-        num_iterations = aggressive_mode ? 40 : (high_priority_mode ? 20 : 10);
-    };
-    const auto Cleanup = [this, &num_iterations, &high_priority_mode, &aggressive_mode](ImageId image_id) {
-        if (num_iterations == 0) {
-            return true;
-        }
-        --num_iterations;
-        auto& image = slot_images[image_id];
-        if (True(image.flags & ImageFlagBits::IsDecoding)) {
-            return false;
-        }
-        const bool must_download = image.IsSafeDownload() && False(image.flags & ImageFlagBits::BadOverlap);
-        if ((!aggressive_mode && True(image.flags & ImageFlagBits::CostlyLoad)) || (!high_priority_mode && must_download)) {
-            return false;
-        }
-        if (must_download) {
-            auto map = runtime.DownloadStagingBuffer(image.unswizzled_size_bytes);
-            const auto copies = FixSmallVectorADL(FullDownloadCopies(image.info));
-            image.DownloadMemory(map, copies);
-            runtime.Finish();
-            SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span, swizzle_data_buffer);
-        }
-        if (True(image.flags & ImageFlagBits::Tracked)) {
-            UntrackImage(image, image_id);
-        }
-        UnregisterImage(image_id);
-        DeleteImage(image_id, image.scale_tick > frame_tick + 5);
-        if (aggressive_mode && total_used_memory < critical_memory) {
-            num_iterations >>= 2;
-            aggressive_mode = false;
-        } else if (high_priority_mode && total_used_memory < expected_memory) {
-            num_iterations >>= 1;
-            high_priority_mode = false;
-        }
-        return false;
-    };
-    Configure(false);
-    lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, Cleanup);
-    if (total_used_memory >= critical_memory) {
-        Configure(true);
-        lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, Cleanup);
+void TextureCache<P>::QueueEvictionDownload(Image& image) {
+    auto copies = FullDownloadCopies(image.info);
+    auto staging = runtime.DownloadStagingBuffer(image.unswizzled_size_bytes, true);
+    image.DownloadMemory(staging, FixSmallVectorADL(copies));
+    pending_eviction_downloads.push_back(PendingEvictionDownload{
+        .staging = staging,
+        .gpu_memory = gpu_memory,
+        .copies = std::move(copies),
+        .info = image.info,
+        .gpu_addr = image.gpu_addr,
+        .sync_point = runtime.CurrentSyncPoint(),
+    });
+}
+
+template <class P>
+void TextureCache<P>::TickEvictionDownloads(u64 completed_sync_point) {
+    while (!pending_eviction_downloads.empty() &&
+           pending_eviction_downloads.front().sync_point <= completed_sync_point) {
+        auto& entry = pending_eviction_downloads.front();
+        SwizzleImage(*entry.gpu_memory, entry.gpu_addr, entry.info, FixSmallVectorADL(entry.copies),
+                     entry.staging.mapped_span.subspan(entry.staging.offset), swizzle_data_buffer);
+        runtime.FreeDeferredStagingBuffer(entry.staging);
+        pending_eviction_downloads.pop_front();
     }
 }
 
+template <class P>
+void TextureCache<P>::FlushEvictionDownloads() {
+    if (pending_eviction_downloads.empty()) {
+        return;
+    }
+    const u64 last_sync_point = pending_eviction_downloads.back().sync_point;
+    runtime.WaitSyncPoint(last_sync_point);
+    TickEvictionDownloads(last_sync_point);
+}
+
+template <class P>
+u64 TextureCache<P>::ImageSizeBytes(const ImageBase& image) {
+    u64 tentative_size = (std::max)(image.guest_size_bytes, image.unswizzled_size_bytes);
+    if ((IsPixelFormatASTC(image.info.format) &&
+         True(image.flags & ImageFlagBits::AcceleratedUpload)) ||
+        True(image.flags & ImageFlagBits::Converted)) {
+        tentative_size = TranscodedAstcSize(tentative_size, image.info.format);
+    }
+    u64 size = Common::AlignUp(tentative_size, 1024);
+    if (image.HasScaled()) {
+        size += GetScaledImageSizeBytes(image);
+    }
+    return size;
+}
+
+template <class P>
+u64 TextureCache<P>::DeviceUsage(bool force_refresh) {
+    if (!runtime.CanReportAllocationUsage()) {
+        return total_used_memory;
+    }
+    if (force_refresh || usage_refresh_countdown == 0) {
+        cached_device_usage = runtime.GetDeviceAllocationUsage();
+        usage_refresh_countdown = USAGE_REFRESH_INTERVAL;
+    } else {
+        --usage_refresh_countdown;
+    }
+    return cached_device_usage;
+}
+
+template <class P>
+u64 TextureCache<P>::ReclaimMemory(u64 target_bytes, bool allow_download) {
+    if (target_bytes == 0 || in_reclaim) {
+        return 0;
+    }
+    in_reclaim = true;
+    const u64 drain_point = runtime.CompletedSyncPoint();
+    TickEvictionDownloads(drain_point);
+    sentenced_images.Reclaim(drain_point);
+    sentenced_image_view.Reclaim(drain_point);
+    sentenced_framebuffers.Reclaim(drain_point);
+    u64 freed = 0;
+    const auto evict = [&](ImageId image_id) {
+        if (freed >= target_bytes) {
+            return true;
+        }
+        auto& image = slot_images[image_id];
+        if (True(image.flags & ImageFlagBits::IsDecoding)) {
+            return false;
+        }
+        const bool must_download =
+            image.IsSafeDownload() && False(image.flags & ImageFlagBits::BadOverlap);
+        bool queued_download = false;
+        if (must_download) {
+            if constexpr (HAS_TIMELINE_SYNC_POINTS) {
+                if (!allow_download) {
+                    return false;
+                }
+                QueueEvictionDownload(image);
+                queued_download = true;
+            } else {
+                return false;
+            }
+        }
+        const u64 image_bytes = ImageSizeBytes(image);
+        if (True(image.flags & ImageFlagBits::Tracked)) {
+            UntrackImage(image, image_id);
+        }
+        UnregisterImage(image_id);
+        DeleteImage(image_id, !queued_download && image.scale_tick > frame_tick + 5);
+        freed += image_bytes;
+        return false;
+    };
+    const u64 cold_tick =
+        frame_tick > RECLAIM_GUARD_FRAMES ? frame_tick - RECLAIM_GUARD_FRAMES : 0;
+    lru_cache.ForEachItemBelow(cold_tick, evict);
+    if (freed < target_bytes) {
+        lru_cache.ForEachItemBelow(frame_tick > 0 ? frame_tick - 1 : 0, evict);
+    }
+    const u64 exit_point = runtime.CompletedSyncPoint();
+    sentenced_images.Reclaim(exit_point);
+    sentenced_image_view.Reclaim(exit_point);
+    sentenced_framebuffers.Reclaim(exit_point);
+    in_reclaim = false;
+    usage_refresh_countdown = 0;
+    reclaim_stalled = freed == 0;
+    return freed;
+}
+
+template <class P>
+void TextureCache<P>::EnsureHeadroom(bool allow_download) {
+    if (reclaim_stalled) {
+        return;
+    }
+    const u64 limit = memory_budget > RECLAIM_HEADROOM ? memory_budget - RECLAIM_HEADROOM : 0;
+    const u64 usage = DeviceUsage(false);
+    if (usage <= limit) {
+        return;
+    }
+    const u64 target = (limit / 100) * RECLAIM_TARGET_PERCENT;
+    ReclaimMemory((std::min)(usage - target, total_used_memory), allow_download);
+}
+
 template <class P>
 void TextureCache<P>::TickFrame() {
-    // If we can obtain the memory info, use it instead of the estimate.
-    if (runtime.CanReportMemoryUsage()) {
-        total_used_memory = runtime.GetDeviceMemoryUsage();
-    }
-    if (total_used_memory > minimum_memory) {
-        RunGarbageCollector();
-    }
-    sentenced_images.Tick();
-    sentenced_framebuffers.Tick();
-    sentenced_image_view.Tick();
+    usage_refresh_countdown = 0;
+    reclaim_stalled = false;
+    EnsureHeadroom(true);
+    const u64 completed_sync_point = runtime.CompletedSyncPoint();
+    TickEvictionDownloads(completed_sync_point);
+    sentenced_images.Reclaim(completed_sync_point);
+    sentenced_framebuffers.Reclaim(completed_sync_point);
+    sentenced_image_view.Reclaim(completed_sync_point);
     TickAsyncDecode();
     TickAsyncUnswizzle();
 
@@ -596,6 +665,7 @@ void TextureCache<P>::WriteMemory(DAddr cpu_addr, size_t size) {
 
 template <class P>
 void TextureCache<P>::DownloadMemory(DAddr cpu_addr, size_t size) {
+    FlushEvictionDownloads();
     boost::container::small_vector<ImageId, 16> images;
     ForEachImageInRegion(cpu_addr, size, [&images](ImageId image_id, ImageBase& image) {
         if (!image.IsSafeDownload()) {
@@ -894,6 +964,7 @@ void TextureCache<P>::CommitAsyncFlushes() {
 
 template <class P>
 void TextureCache<P>::PopAsyncFlushes() {
+    TickEvictionDownloads(runtime.CompletedSyncPoint());
     if (committed_downloads.empty()) {
         return;
     }
@@ -1294,8 +1365,9 @@ void TextureCache<P>::InvalidateScale(Image& image) {
     }
     RemoveImageViewReferences(image_view_ids);
     RemoveFramebuffers(image_view_ids);
+    const u64 sync_point = runtime.CurrentSyncPoint();
     for (const ImageViewId image_view_id : image_view_ids) {
-        sentenced_image_view.Push(std::move(slot_image_views[image_view_id]));
+        sentenced_image_view.Push(std::move(slot_image_views[image_view_id]), sync_point);
         slot_image_views.erase(image_view_id);
     }
     image.image_view_ids.clear();
@@ -1331,6 +1403,7 @@ void TextureCache<P>::QueueAsyncDecode(Image& image, ImageId image_id) {
     LOG_INFO(HW_GPU, "Queuing async texture decode");
 
     image.flags |= ImageFlagBits::IsDecoding;
+    runtime.TransitionImageLayout(image);
     auto decode = std::make_unique<AsyncDecodeContext>();
     auto* decode_ptr = decode.get();
     decode->image_id = image_id;
@@ -1363,6 +1436,7 @@ void TextureCache<P>::QueueAsyncUnswizzle(Image& image, ImageId image_id) {
     }
 
     image.flags |= ImageFlagBits::IsDecoding;
+    runtime.TransitionImageLayout(image);
 
     unswizzle_queue.push_back({
         .image_id = image_id,
@@ -1523,6 +1597,7 @@ ImageId TextureCache<P>::InsertImage(const ImageInfo& info, GPUVAddr gpu_addr,
 
 template <class P>
 ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, DAddr cpu_addr) {
+    EnsureHeadroom(false);
     ImageInfo new_info = info;
     const size_t size_bytes = CalculateGuestSizeInBytes(new_info);
     const bool broken_views = runtime.HasBrokenTextureViewFormats();
@@ -1631,7 +1706,28 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, DA
     for (const ImageId overlap_id : join_ignore_textures) {
         Image& overlap = slot_images[overlap_id];
         if (True(overlap.flags & ImageFlagBits::GpuModified)) {
-            UNIMPLEMENTED();
+            if (new_image.TryFindBase(overlap.gpu_addr) &&
+                (!can_rescale || ImageCanRescale(overlap))) {
+                if (can_rescale) {
+                    ScaleUp(overlap);
+                } else {
+                    ScaleDown(overlap);
+                }
+                join_copies_to_do.emplace_back(JoinCopy{false, overlap_id});
+                continue;
+            }
+            if (overlap.IsSafeDownload() && False(overlap.flags & ImageFlagBits::BadOverlap) &&
+                gpu_memory->GpuToCpuAddress(overlap.gpu_addr).has_value()) {
+                QueueEvictionDownload(overlap);
+            } else {
+                LOG_WARNING(HW_GPU,
+                            "Dropping GPU modified overlap, contents are not recoverable: "
+                            "gpu_addr=0x{:x} format={} size={}x{}x{} levels={} layers={}",
+                            overlap.gpu_addr, static_cast<int>(overlap.info.format),
+                            overlap.info.size.width, overlap.info.size.height,
+                            overlap.info.size.depth, overlap.info.resources.levels,
+                            overlap.info.resources.layers);
+            }
         }
         if (True(overlap.flags & ImageFlagBits::Tracked)) {
             UntrackImage(overlap, overlap_id);
@@ -2185,13 +2281,7 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {
     ASSERT_MSG(False(image.flags & ImageFlagBits::Registered),
                "Trying to register an already registered image");
     image.flags |= ImageFlagBits::Registered;
-    u64 tentative_size = (std::max)(image.guest_size_bytes, image.unswizzled_size_bytes);
-    if ((IsPixelFormatASTC(image.info.format) &&
-         True(image.flags & ImageFlagBits::AcceleratedUpload)) ||
-        True(image.flags & ImageFlagBits::Converted)) {
-        tentative_size = TranscodedAstcSize(tentative_size, image.info.format);
-    }
-    total_used_memory += Common::AlignUp(tentative_size, 1024);
+    total_used_memory += ImageSizeBytes(image);
     image.lru_index = lru_cache.Insert(image_id, frame_tick);
 
     ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, image_id](u64 page) {
@@ -2354,16 +2444,7 @@ void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) {
 template <class P>
 void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) {
     ImageBase& image = slot_images[image_id];
-    if (image.HasScaled()) {
-        total_used_memory -= GetScaledImageSizeBytes(image);
-    }
-    u64 tentative_size = (std::max)(image.guest_size_bytes, image.unswizzled_size_bytes);
-    if ((IsPixelFormatASTC(image.info.format) &&
-         True(image.flags & ImageFlagBits::AcceleratedUpload)) ||
-        True(image.flags & ImageFlagBits::Converted)) {
-        tentative_size = TranscodedAstcSize(tentative_size, image.info.format);
-    }
-    total_used_memory -= Common::AlignUp(tentative_size, 1024);
+    total_used_memory -= std::min<u64>(total_used_memory, ImageSizeBytes(image));
     const GPUVAddr gpu_addr = image.gpu_addr;
     const auto alloc_it = image_allocs_table.find(gpu_addr);
     if (alloc_it == image_allocs_table.end()) {
@@ -2417,14 +2498,15 @@ void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) {
         ASSERT_MSG(num_removed_overlaps == 1, "Invalid number of removed overlapps: {}",
                    num_removed_overlaps);
     }
+    const u64 sync_point = runtime.CurrentSyncPoint();
     for (const ImageViewId image_view_id : image_view_ids) {
         if (!immediate_delete) {
-            sentenced_image_view.Push(std::move(slot_image_views[image_view_id]));
+            sentenced_image_view.Push(std::move(slot_image_views[image_view_id]), sync_point);
         }
         slot_image_views.erase(image_view_id);
     }
     if (!immediate_delete) {
-        sentenced_images.Push(std::move(slot_images[image_id]));
+        sentenced_images.Push(std::move(slot_images[image_id]), sync_point);
     }
     slot_images.erase(image_id);
 
@@ -2470,7 +2552,8 @@ void TextureCache<P>::RemoveFramebuffers(std::span<const ImageViewId> removed_vi
                 last_framebuffer_id = {};
                 last_framebuffer_serial = 0;
             }
-            sentenced_framebuffers.Push(std::move(slot_framebuffers[framebuffer_id]));
+            sentenced_framebuffers.Push(std::move(slot_framebuffers[framebuffer_id]),
+                                        runtime.CurrentSyncPoint());
             it = framebuffers.erase(it);
         } else {
             ++it;

src/video_core/texture_cache/texture_cache_base.h

@@ -30,7 +30,7 @@
 #include "common/thread_worker.h"
 #include "video_core/compatible_formats.h"
 #include "video_core/control/channel_state_cache.h"
-#include "video_core/delayed_destruction_ring.h"
+#include "video_core/deferred_destruction_queue.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/surface.h"
 #include "video_core/texture_cache/descriptor_table.h"
@@ -108,18 +108,20 @@ class TextureCache : public VideoCommon::ChannelSetupCaches<TextureCacheChannelI
     static constexpr bool HAS_DEVICE_MEMORY_INFO = P::HAS_DEVICE_MEMORY_INFO;
     /// True when the API can do asynchronous texture downloads.
     static constexpr bool IMPLEMENTS_ASYNC_DOWNLOADS = P::IMPLEMENTS_ASYNC_DOWNLOADS;
+    static constexpr bool HAS_TIMELINE_SYNC_POINTS = P::HAS_TIMELINE_SYNC_POINTS;
 
     static constexpr size_t UNSET_CHANNEL{(std::numeric_limits<size_t>::max)()};
 
 #ifdef YUZU_LEGACY
-    static constexpr s64 TARGET_THRESHOLD = 3_GiB;
+    static constexpr u64 RECLAIM_HEADROOM = 384_MiB;
 #else
-    static constexpr s64 TARGET_THRESHOLD = 4_GiB;
+    static constexpr u64 RECLAIM_HEADROOM = 512_MiB;
 #endif
 
-    static constexpr s64 DEFAULT_EXPECTED_MEMORY = 1_GiB + 125_MiB;
-    static constexpr s64 DEFAULT_CRITICAL_MEMORY = 1_GiB + 625_MiB;
-    static constexpr size_t GC_EMERGENCY_COUNTS = 2;
+    static constexpr u64 FALLBACK_MEMORY_BUDGET = 2_GiB;
+    static constexpr u32 USAGE_REFRESH_INTERVAL = 16;
+    static constexpr u64 RECLAIM_GUARD_FRAMES = 8;
+    static constexpr u64 RECLAIM_TARGET_PERCENT = 88;
 
     using Runtime = typename P::Runtime;
     using Image = typename P::Image;
@@ -154,6 +156,8 @@ public:
     /// Notify the cache that a new frame has been queued
     void TickFrame();
 
+    u64 ReclaimMemory(u64 target_bytes, bool allow_download);
+
     /// Return a constant reference to the given image view id
     [[nodiscard]] const ImageView& GetImageView(ImageViewId id) const noexcept;
 
@@ -293,8 +297,17 @@ private:
 
     void OnGPUASRegister(size_t map_id) final override;
 
-    /// Runs the Garbage Collector.
-    void RunGarbageCollector();
+    u64 ImageSizeBytes(const ImageBase& image);
+
+    u64 DeviceUsage(bool force_refresh);
+
+    void EnsureHeadroom(bool allow_download);
+
+    void QueueEvictionDownload(Image& image);
+
+    void TickEvictionDownloads(u64 completed_sync_point);
+
+    void FlushEvictionDownloads();
 
     /// Find or create an image view in the guest descriptor table
     ImageViewId VisitImageView(u32 index, bool compute);
@@ -451,9 +464,11 @@ private:
     bool has_deleted_images = false;
     bool is_rescaling = false;
     u64 total_used_memory = 0;
-    u64 minimum_memory;
-    u64 expected_memory;
-    u64 critical_memory;
+    u64 memory_budget = 0;
+    u64 cached_device_usage = 0;
+    u32 usage_refresh_countdown = 0;
+    bool in_reclaim = false;
+    bool reclaim_stalled = false;
     size_t gpu_unswizzle_maxsize = 0;
     size_t swizzle_chunk_size = 0;
     u32 swizzle_slices_per_batch = 0;
@@ -491,14 +506,19 @@ private:
     };
     Common::LeastRecentlyUsedCache<LRUItemParams> lru_cache;
 
- #ifdef YUZU_LEGACY
-    static constexpr size_t TICKS_TO_DESTROY = 6;
- #else
-    static constexpr size_t TICKS_TO_DESTROY = 8;
-#endif
-    DelayedDestructionRing<Image, TICKS_TO_DESTROY> sentenced_images;
-    DelayedDestructionRing<ImageView, TICKS_TO_DESTROY> sentenced_image_view;
-    DelayedDestructionRing<Framebuffer, TICKS_TO_DESTROY> sentenced_framebuffers;
+    DeferredDestructionQueue<Image> sentenced_images;
+    DeferredDestructionQueue<ImageView> sentenced_image_view;
+    DeferredDestructionQueue<Framebuffer> sentenced_framebuffers;
+
+    struct PendingEvictionDownload {
+        AsyncBuffer staging;
+        Tegra::MemoryManager* gpu_memory;
+        boost::container::small_vector<VideoCommon::BufferImageCopy, 16> copies;
+        VideoCommon::ImageInfo info;
+        GPUVAddr gpu_addr;
+        u64 sync_point;
+    };
+    std::deque<PendingEvictionDownload> pending_eviction_downloads;
 
     ankerl::unordered_dense::map<GPUVAddr, ImageAllocId> image_allocs_table;
 
@@ -509,7 +529,8 @@ private:
     u64 frame_tick = 0;
     u64 last_sampler_gc_frame = (std::numeric_limits<u64>::max)();
 
-    Common::ThreadWorker texture_decode_worker{1, "TextureDecoder"};
+    Common::ThreadWorker texture_decode_worker{1, "TextureDecoder", {},
+                                               Common::ThreadPlacement::Background};
     std::vector<std::unique_ptr<AsyncDecodeContext>> async_decodes;
 
     std::deque<PendingUnswizzle> unswizzle_queue;

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
@@ -747,7 +747,7 @@ boost::container::small_vector<ImageCopy, 16> MakeShrinkImageCopies(const ImageI
 
     const bool is_dst_3d = dst.type == ImageType::e3D;
     if (is_dst_3d) {
-        ASSERT(src.type == ImageType::e3D);
+        ASSERT(src.type == ImageType::e3D || src.resources.layers == 1);
         ASSERT(src.resources.levels == 1);
     }
     const bool both_2d{src.type == ImageType::e2D && dst.type == ImageType::e2D};

src/video_core/textures/astc.cpp

@@ -1769,12 +1769,6 @@ static void DecompressBlock(std::span<const u8, 16> inBuf, const u32 blockWidth,
         return;
     }
 
-    if (weightParams.GetNumWeightValues() > 64) {
-        assert(false && "Too many weights in the weight grid");
-        FillError(outBuf, blockWidth, blockHeight);
-        return;
-    }
-
     // Read num partitions
     u32 nPartitions = strm.ReadBits<2>() + 1;
     assert(nPartitions <= 4);
@@ -1811,11 +1805,6 @@ static void DecompressBlock(std::span<const u8, 16> inBuf, const u32 blockWidth,
 
     // Remaining bits are color endpoint data...
     u32 nWeightBits = weightParams.GetPackedBitSize();
-    if (nWeightBits < 24 || nWeightBits > 96) {
-        assert(false && "Invalid weight bit count");
-        FillError(outBuf, blockWidth, blockHeight);
-        return;
-    }
     s32 remainingBits = 128 - nWeightBits - static_cast<int>(strm.GetBitsRead());
 
     // Consider extra bits prior to texel data...

src/video_core/textures/workers.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 2023 yuzu Emulator Project
@@ -10,7 +10,8 @@ namespace Tegra::Texture {
 
 Common::ThreadWorker& GetThreadWorkers() {
     static Common::ThreadWorker workers{(std::max)(std::thread::hardware_concurrency(), 2U) / 2,
-                                        "ImageTranscode"};
+                                        "ImageTranscode", {},
+                                        Common::ThreadPlacement::Background};
 
     return workers;
 }

src/video_core/vulkan_common/vulkan_device.cpp

@@ -5,6 +5,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <algorithm>
+#include <array>
 #include <bitset>
 #include <chrono>
 #include <optional>
@@ -506,8 +507,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
     if (is_qualcomm) {
         must_emulate_scaled_formats = true;
         LOG_WARNING(Render_Vulkan, "Qualcomm drivers require scaled vertex format emulation.");
-        has_broken_descriptor_aliasing = true;
-        LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken descriptor aliasing.");
         LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken custom border color.");
         RemoveExtensionFeature(extensions.custom_border_color, features.custom_border_color,
                                VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
@@ -715,37 +714,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
         RemoveExtensionFeature(extensions.vertex_input_dynamic_state, features.vertex_input_dynamic_state, VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME);
     }
 
-    // Descriptors feature list
-    {
-        auto& descriptor_indexing = features.descriptor_indexing;
-        descriptor_indexing.shaderInputAttachmentArrayDynamicIndexing = false;
-        descriptor_indexing.shaderUniformTexelBufferArrayDynamicIndexing = false;
-        descriptor_indexing.shaderStorageTexelBufferArrayDynamicIndexing = false;
-        descriptor_indexing.shaderUniformBufferArrayNonUniformIndexing = false;
-        descriptor_indexing.shaderStorageBufferArrayNonUniformIndexing = false;
-        descriptor_indexing.shaderInputAttachmentArrayNonUniformIndexing = false;
-        descriptor_indexing.descriptorBindingUniformBufferUpdateAfterBind = false;
-        descriptor_indexing.descriptorBindingSampledImageUpdateAfterBind = false;
-        descriptor_indexing.descriptorBindingStorageImageUpdateAfterBind = false;
-        descriptor_indexing.descriptorBindingStorageBufferUpdateAfterBind = false;
-        descriptor_indexing.descriptorBindingUniformTexelBufferUpdateAfterBind = false;
-        descriptor_indexing.descriptorBindingStorageTexelBufferUpdateAfterBind = false;
-        descriptor_indexing.descriptorBindingUpdateUnusedWhilePending = false;
-        descriptor_indexing.descriptorBindingVariableDescriptorCount = false;
-        descriptor_indexing.runtimeDescriptorArray = false;
-    }
-
-    // VK_EXT_descriptor_buffer requires VK_KHR_buffer_device_address
-    if (extensions.descriptor_buffer && !features.buffer_device_address.bufferDeviceAddress) {
-        LOG_WARNING(Render_Vulkan, "Descriptor buffer needs buffer device address, disabling.");
-        RemoveExtensionFeature(extensions.descriptor_buffer, features.descriptor_buffer,
-                               VK_EXT_DESCRIPTOR_BUFFER_EXTENSION_NAME);
-    }
-    if (!extensions.descriptor_buffer) {
-        RemoveExtensionFeature(extensions.buffer_device_address, features.buffer_device_address,
-                               VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME);
-    }
-
     logical = vk::Device::Create(physical, queue_cis, ExtensionListForVulkan(loaded_extensions), first_next, dld);
 
     graphics_queue = logical.GetQueue(graphics_family);
@@ -759,16 +727,13 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
     if (extensions.memory_budget) {
         flags |= VMA_ALLOCATOR_CREATE_EXT_MEMORY_BUDGET_BIT;
     }
-    if (extensions.buffer_device_address) {
-        flags |= VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT;
-    }
     const VmaAllocatorCreateInfo allocator_info{
             .flags = flags,
             .physicalDevice = physical,
             .device = *logical,
             .preferredLargeHeapBlockSize = is_integrated
                                            ? (64u * 1024u * 1024u)
-                                           : (256u * 1024u * 1024u),
+                                           : (128u * 1024u * 1024u),
             .pAllocationCallbacks = nullptr,
             .pDeviceMemoryCallbacks = nullptr,
             .pHeapSizeLimit = nullptr,
@@ -780,12 +745,32 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
 
     vk::Check(vmaCreateAllocator(&allocator_info, &allocator));
 
+    {
+        const auto& limits = properties.properties.limits;
+        LOG_INFO(Render_Vulkan, "MSAA sample count support:");
+        LOG_INFO(Render_Vulkan, "  framebufferColorSampleCounts:   {:#x}",
+                 limits.framebufferColorSampleCounts);
+        LOG_INFO(Render_Vulkan, "  framebufferDepthSampleCounts:   {:#x}",
+                 limits.framebufferDepthSampleCounts);
+        LOG_INFO(Render_Vulkan, "  framebufferStencilSampleCounts: {:#x}",
+                 limits.framebufferStencilSampleCounts);
+        LOG_INFO(Render_Vulkan, "  sampledImageColorSampleCounts:  {:#x}",
+                 limits.sampledImageColorSampleCounts);
+        LOG_INFO(Render_Vulkan, "  sampledImageDepthSampleCounts:  {:#x}",
+                 limits.sampledImageDepthSampleCounts);
+        LOG_INFO(Render_Vulkan, "  sampledImageIntegerSampleCounts:{:#x}",
+                 limits.sampledImageIntegerSampleCounts);
+        LOG_INFO(Render_Vulkan, "  storageImageSampleCounts:       {:#x}",
+                 limits.storageImageSampleCounts);
+    }
+
     // Initialize GPU logging if enabled
     InitializeGPULogging();
 }
 
 Device::~Device() {
     ShutdownGPULogging();
+    vk::FlushDeletionQueue();
     vmaDestroyAllocator(allocator);
 }
 
@@ -853,16 +838,18 @@ bool Device::ComputeIsOptimalAstcSupported() const {
         VK_FORMAT_ASTC_12x10_UNORM_BLOCK, VK_FORMAT_ASTC_12x10_SRGB_BLOCK,
         VK_FORMAT_ASTC_12x12_UNORM_BLOCK, VK_FORMAT_ASTC_12x12_SRGB_BLOCK,
     };
-    if (!features.features.textureCompressionASTC_LDR) {
+    if (!features.features.textureCompressionASTC_LDR ||
+        !features.texture_compression_astc_hdr.textureCompressionASTC_HDR) {
         return false;
     }
-    const VkFormatFeatureFlags format_feature_usage{
-        VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT | VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_LINEAR_BIT |
-        VK_FORMAT_FEATURE_TRANSFER_SRC_BIT | VK_FORMAT_FEATURE_TRANSFER_DST_BIT};
+    const auto format_feature_usage{VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT |
+                                    VK_FORMAT_FEATURE_BLIT_SRC_BIT |
+                                    VK_FORMAT_FEATURE_BLIT_DST_BIT |
+                                    VK_FORMAT_FEATURE_TRANSFER_SRC_BIT |
+                                    VK_FORMAT_FEATURE_TRANSFER_DST_BIT};
     for (const auto format : astc_formats) {
         const auto physical_format_properties{physical.GetFormatProperties(format)};
-        if ((physical_format_properties.optimalTilingFeatures & format_feature_usage) !=
-            format_feature_usage) {
+        if ((physical_format_properties.optimalTilingFeatures & format_feature_usage) == 0) {
             return false;
         }
     }
@@ -1008,10 +995,6 @@ bool Device::GetSuitability(bool requires_swapchain) {
         CHECK_EXTENSION(VK_KHR_SWAPCHAIN_EXTENSION_NAME);
     }
 
-    if (instance_version < VK_API_VERSION_1_2) {
-        CHECK_EXTENSION(VK_EXT_DESCRIPTOR_INDEXING_EXTENSION_NAME);
-    }
-
 #undef LOG_EXTENSION
 #undef CHECK_EXTENSION
 
@@ -1122,11 +1105,6 @@ bool Device::GetSuitability(bool requires_swapchain) {
             VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR;
         SetNext(next, properties.push_descriptor);
     }
-    if (extensions.descriptor_buffer) {
-        properties.descriptor_buffer.sType =
-            VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_BUFFER_PROPERTIES_EXT;
-        SetNext(next, properties.descriptor_buffer);
-    }
     if (extensions.subgroup_size_control || features.subgroup_size_control.subgroupSizeControl) {
         properties.subgroup_size_control.sType =
             VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_PROPERTIES;
@@ -1253,15 +1231,20 @@ void Device::RemoveUnsuitableExtensions() {
     RemoveExtensionFeatureIfUnsuitable(extensions.depth_bias_control, features.depth_bias_control,
                                        VK_EXT_DEPTH_BIAS_CONTROL_EXTENSION_NAME);
 
+    // VK_EXT_depth_clamp_zero_one
+    extensions.depth_clamp_zero_one = features.depth_clamp_zero_one.depthClampZeroOne;
+    RemoveExtensionFeatureIfUnsuitable(extensions.depth_clamp_zero_one,
+                                       features.depth_clamp_zero_one,
+                                       VK_EXT_DEPTH_CLAMP_ZERO_ONE_EXTENSION_NAME);
+
     // VK_EXT_depth_clip_control
     extensions.depth_clip_control = features.depth_clip_control.depthClipControl;
     RemoveExtensionFeatureIfUnsuitable(extensions.depth_clip_control, features.depth_clip_control,
                                        VK_EXT_DEPTH_CLIP_CONTROL_EXTENSION_NAME);
-
-    // VK_EXT_descriptor_buffer
-    extensions.descriptor_buffer = features.descriptor_buffer.descriptorBuffer;
-    RemoveExtensionFeatureIfUnsuitable(extensions.descriptor_buffer, features.descriptor_buffer,
-                                       VK_EXT_DESCRIPTOR_BUFFER_EXTENSION_NAME);
+    // VK_EXT_depth_clip_enable
+    extensions.depth_clip_enable = features.depth_clip_enable.depthClipEnable;
+    RemoveExtensionFeatureIfUnsuitable(extensions.depth_clip_enable, features.depth_clip_enable,
+                                       VK_EXT_DEPTH_CLIP_ENABLE_EXTENSION_NAME);
 
     // VK_EXT_extended_dynamic_state
     extensions.extended_dynamic_state = features.extended_dynamic_state.extendedDynamicState;
@@ -1493,6 +1476,23 @@ std::optional<size_t> Device::GetSamplerHeapBudget() const {
     return sampler_heap_budget;
 }
 
+Device::MemoryBudgetInfo Device::GetMemoryBudgetInfo() const {
+    std::array<VmaBudget, VK_MAX_MEMORY_HEAPS> budgets{};
+    vmaGetHeapBudgets(allocator, budgets.data());
+    MemoryBudgetInfo info{};
+    for (const size_t heap : valid_heap_memory) {
+        info.usage += budgets[heap].usage;
+        info.budget += budgets[heap].budget;
+        info.block_bytes += budgets[heap].statistics.blockBytes;
+        info.allocation_bytes += budgets[heap].statistics.allocationBytes;
+    }
+    return info;
+}
+
+void Device::TickAllocatorFrame() const {
+    vmaSetCurrentFrameIndex(allocator, ++allocator_frame_index);
+}
+
 u64 Device::GetDeviceMemoryUsage() const {
     VkPhysicalDeviceMemoryBudgetPropertiesEXT budget;
     budget.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_BUDGET_PROPERTIES_EXT;
@@ -1542,9 +1542,12 @@ void Device::CollectPhysicalMemoryInfo() {
         device_access_memory -= reserve_memory;
         if (Settings::values.vram_usage_mode.GetValue() != Settings::VramUsageMode::Aggressive) {
             // Account for resolution scaling in memory limits
-            const size_t normal_memory = 6_GiB;
-            const size_t scaler_memory = 1_GiB * Settings::values.resolution_info.ScaleUp(1);
-            device_access_memory = std::min<u64>(device_access_memory, normal_memory + scaler_memory);
+            const u64 normal_memory = 6_GiB;
+            const u64 scaler_memory = 1_GiB * Settings::values.resolution_info.ScaleUp(1);
+            const u64 baseline = normal_memory + scaler_memory;
+            const u64 proportional = (device_access_memory / 4) * 3;
+            device_access_memory =
+                std::min<u64>(device_access_memory, std::max<u64>(baseline, proportional));
         }
     }
 }

src/video_core/vulkan_common/vulkan_device.h

@@ -36,7 +36,6 @@ VK_DEFINE_HANDLE(VmaAllocator)
     FEATURE(EXT, DescriptorIndexing, DESCRIPTOR_INDEXING, descriptor_indexing)                     \
     FEATURE(EXT, HostQueryReset, HOST_QUERY_RESET, host_query_reset)                               \
     FEATURE(KHR, 8BitStorage, 8BIT_STORAGE, bit8_storage)                                          \
-    FEATURE(KHR, BufferDeviceAddress, BUFFER_DEVICE_ADDRESS, buffer_device_address)                \
     FEATURE(KHR, TimelineSemaphore, TIMELINE_SEMAPHORE, timeline_semaphore)
 
 #define FOR_EACH_VK_FEATURE_1_3(FEATURE)                                                           \
@@ -55,8 +54,9 @@ VK_DEFINE_HANDLE(VmaAllocator)
     FEATURE(EXT, ColorWriteEnable, COLOR_WRITE_ENABLE, color_write_enable)                         \
     FEATURE(EXT, CustomBorderColor, CUSTOM_BORDER_COLOR, custom_border_color)                      \
     FEATURE(EXT, DepthBiasControl, DEPTH_BIAS_CONTROL, depth_bias_control)                         \
+    FEATURE(EXT, DepthClampZeroOne, DEPTH_CLAMP_ZERO_ONE, depth_clamp_zero_one)                   \
     FEATURE(EXT, DepthClipControl, DEPTH_CLIP_CONTROL, depth_clip_control)                         \
-    FEATURE(EXT, DescriptorBuffer, DESCRIPTOR_BUFFER, descriptor_buffer)                           \
+    FEATURE(EXT, DepthClipEnable, DEPTH_CLIP_ENABLE, depth_clip_enable)                            \
     FEATURE(EXT, ExtendedDynamicState, EXTENDED_DYNAMIC_STATE, extended_dynamic_state)             \
     FEATURE(EXT, ExtendedDynamicState2, EXTENDED_DYNAMIC_STATE_2, extended_dynamic_state2)         \
     FEATURE(EXT, ExtendedDynamicState3, EXTENDED_DYNAMIC_STATE_3, extended_dynamic_state3)         \
@@ -74,12 +74,13 @@ VK_DEFINE_HANDLE(VmaAllocator)
     FEATURE(KHR, PipelineExecutableProperties, PIPELINE_EXECUTABLE_PROPERTIES,                     \
             pipeline_executable_properties)                                                        \
     FEATURE(KHR, WorkgroupMemoryExplicitLayout, WORKGROUP_MEMORY_EXPLICIT_LAYOUT,                  \
-            workgroup_memory_explicit_layout)
+            workgroup_memory_explicit_layout)                                                      \
+    FEATURE(EXT, TextureCompressionASTCHDR, TEXTURE_COMPRESSION_ASTC_HDR,                          \
+            texture_compression_astc_hdr)
 
 
 // Define miscellaneous extensions which may be used by the implementation here.
 #define FOR_EACH_VK_EXTENSION(EXTENSION)                                                           \
-    EXTENSION(EXT, ASTC_DECODE_MODE, astc_decode_mode)                                             \
     EXTENSION(EXT, CONDITIONAL_RENDERING, conditional_rendering)                                   \
     EXTENSION(EXT, CONSERVATIVE_RASTERIZATION, conservative_rasterization)                         \
     EXTENSION(EXT, DEPTH_RANGE_UNRESTRICTED, depth_range_unrestricted)                             \
@@ -177,8 +178,6 @@ VK_DEFINE_HANDLE(VmaAllocator)
     FEATURE_NAME(depth_bias_control, depthBiasControl)                                             \
     FEATURE_NAME(depth_bias_control, leastRepresentableValueForceUnormRepresentation)              \
     FEATURE_NAME(depth_bias_control, depthBiasExact)                                               \
-    FEATURE_NAME(descriptor_indexing, descriptorBindingPartiallyBound)                             \
-    FEATURE_NAME(descriptor_indexing, shaderSampledImageArrayNonUniformIndexing)                   \
     FEATURE_NAME(extended_dynamic_state, extendedDynamicState)                                     \
     FEATURE_NAME(format_a4b4g4r4, formatA4B4G4R4)                                                  \
     FEATURE_NAME(robust_image_access, robustImageAccess)                                           \
@@ -258,6 +257,17 @@ public:
         return allocator;
     }
 
+    struct MemoryBudgetInfo {
+        u64 usage;
+        u64 budget;
+        u64 block_bytes;
+        u64 allocation_bytes;
+    };
+
+    MemoryBudgetInfo GetMemoryBudgetInfo() const;
+
+    void TickAllocatorFrame() const;
+
     /// Returns the logical device.
     const vk::Device& GetLogical() const {
         return logical;
@@ -387,7 +397,7 @@ FN_MAX_LIMIT_LIST
 
     /// Returns true if descriptor aliasing is natively supported.
     bool IsDescriptorAliasingSupported() const {
-        return !has_broken_descriptor_aliasing;
+        return GetDriverID() != VK_DRIVER_ID_QUALCOMM_PROPRIETARY;
     }
 
     bool IsSampledImageArrayNonUniformIndexingSupported() const {
@@ -470,26 +480,6 @@ FN_MAX_LIMIT_LIST
         return properties.push_descriptor.maxPushDescriptors;
     }
 
-    /// Returns true if robust buffer access is enabled on the device.
-    bool IsRobustBufferAccessEnabled() const {
-        return features.features.robustBufferAccess == VK_TRUE;
-    }
-
-    /// Returns true if the device supports descriptor buffers.
-    bool IsExtDescriptorBufferSupported() const {
-        return extensions.descriptor_buffer;
-    }
-
-    /// Returns the descriptor buffer properties of the device.
-    const VkPhysicalDeviceDescriptorBufferPropertiesEXT& DescriptorBufferProperties() const {
-        return properties.descriptor_buffer;
-    }
-
-    /// Returns true if the device supports buffer device address.
-    bool IsBufferDeviceAddressSupported() const {
-        return extensions.buffer_device_address;
-    }
-
     /// Returns true if formatless image load is supported.
     bool IsFormatlessImageLoadSupported() const {
         return features.features.shaderStorageImageReadWithoutFormat;
@@ -624,6 +614,16 @@ FN_MAX_LIMIT_LIST
         return extensions.depth_clip_control;
     }
 
+    /// Returns true if the device supports VK_EXT_depth_clamp_zero_one.
+    bool IsExtDepthClampZeroOneSupported() const {
+        return extensions.depth_clamp_zero_one;
+    }
+
+    /// Returns true if the device supports VK_EXT_depth_clip_enable.
+    bool IsExtDepthClipEnableSupported() const {
+        return extensions.depth_clip_enable;
+    }
+
     /// Returns true if the device supports VK_EXT_depth_bias_control.
     bool IsExtDepthBiasControlSupported() const {
         return extensions.depth_bias_control;
@@ -756,6 +756,38 @@ FN_MAX_LIMIT_LIST
         return features.line_rasterization.stippledRectangularLines != VK_FALSE;
     }
 
+    VkLineRasterizationModeEXT GetLineRasterizationMode(bool wants_smooth) const {
+        if (wants_smooth && SupportsSmoothLines()) {
+            return VK_LINE_RASTERIZATION_MODE_RECTANGULAR_SMOOTH_EXT;
+        }
+        if (SupportsRectangularLines()) {
+            return VK_LINE_RASTERIZATION_MODE_RECTANGULAR_EXT;
+        }
+        return VK_LINE_RASTERIZATION_MODE_DEFAULT_EXT;
+    }
+
+    bool SupportsStippleForMode(VkLineRasterizationModeEXT mode) const {
+        switch (mode) {
+        case VK_LINE_RASTERIZATION_MODE_RECTANGULAR_SMOOTH_EXT:
+            return features.line_rasterization.stippledSmoothLines != VK_FALSE;
+        case VK_LINE_RASTERIZATION_MODE_BRESENHAM_EXT:
+            return features.line_rasterization.stippledBresenhamLines != VK_FALSE;
+        default:
+            return features.line_rasterization.stippledRectangularLines != VK_FALSE;
+        }
+    }
+
+    float ClampLineWidth(float width) const {
+        if (!features.features.wideLines) {
+            return 1.0f;
+        }
+        const auto& range = properties.properties.limits.lineWidthRange;
+        if (!(width >= range[0])) {
+            return range[0];
+        }
+        return width > range[1] ? range[1] : width;
+    }
+
     bool SupportsAlphaToOne() const {
         return features.features.alphaToOne != VK_FALSE;
     }
@@ -838,15 +870,6 @@ FN_MAX_LIMIT_LIST
         return extensions.conditional_rendering;
     }
 
-    bool IsExtAstcDecodeModeSupported() const {
-        return extensions.astc_decode_mode;
-    }
-
-    /// Returns true if descriptor bindings is partially bound.
-    bool IsDescriptorBindingPartiallyBoundSupported() const {
-        return features.descriptor_indexing.descriptorBindingPartiallyBound;
-    }
-
     bool HasTimelineSemaphore() const;
 
     /// Returns true if the device supports VK_KHR_synchronization2.
@@ -907,6 +930,10 @@ FN_MAX_LIMIT_LIST
 
     u64 GetDeviceMemoryUsage() const;
 
+    VkSampleCountFlags GetStorageImageSampleCounts() const {
+        return properties.properties.limits.storageImageSampleCounts;
+    }
+
     u32 GetSetsPerPool() const {
         return sets_per_pool;
     }
@@ -1091,6 +1118,7 @@ private:
 private:
     VkInstance instance;         ///< Vulkan instance.
     VmaAllocator allocator;      ///< VMA allocator.
+    mutable u32 allocator_frame_index{};
     vk::DeviceDispatch dld;      ///< Device function pointers.
     vk::PhysicalDevice physical; ///< Physical device.
     vk::Device logical;          ///< Logical device.
@@ -1138,7 +1166,6 @@ private:
         VkPhysicalDeviceSubgroupProperties subgroup_properties{};
         VkPhysicalDeviceFloatControlsProperties float_controls{};
         VkPhysicalDevicePushDescriptorPropertiesKHR push_descriptor{};
-        VkPhysicalDeviceDescriptorBufferPropertiesEXT descriptor_buffer{};
         VkPhysicalDeviceSubgroupSizeControlProperties subgroup_size_control{};
         VkPhysicalDeviceTransformFeedbackPropertiesEXT transform_feedback{};
         VkPhysicalDeviceMaintenance5PropertiesKHR maintenance5{};
@@ -1163,7 +1190,6 @@ private:
     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
     bool has_broken_parallel_compiling{};      ///< Has broken parallel shader compiling.
     bool has_renderdoc{};                      ///< Has RenderDoc attached
     bool has_nsight_graphics{};                ///< Has Nsight Graphics attached

src/video_core/vulkan_common/vulkan_memory_allocator.cpp

@@ -30,26 +30,6 @@ namespace Vulkan {
 
 // 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;
@@ -86,125 +66,11 @@ namespace Vulkan {
         }
 
 
-// This avoids calling vkGetBufferMemoryRequirements* directly.
-        template<typename T>
-        static VkBuffer GetVkHandleFromBuffer(const T &buf) {
-            if constexpr (requires { static_cast<VkBuffer>(buf); }) {
-                return static_cast<VkBuffer>(buf);
-            } else if constexpr (requires {{ buf.GetHandle() } -> std::convertible_to<VkBuffer>; }) {
-                return buf.GetHandle();
-            } else if constexpr (requires {{ buf.Handle() } -> std::convertible_to<VkBuffer>; }) {
-                return buf.Handle();
-            } else if constexpr (requires {{ buf.vk_handle() } -> std::convertible_to<VkBuffer>; }) {
-                return buf.vk_handle();
-            } else {
-                static_assert(sizeof(T) == 0, "Cannot extract VkBuffer handle from vk::Buffer");
-                return VK_NULL_HANDLE;
-            }
-        }
-
     } // 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<uintptr_t>(memory),
-                static_cast<u64>(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<u8> MemoryCommit::Map()
-    {
-        if (!allocation) return {};
-        if (!mapped_ptr) {
-            if (vmaMapMemory(allocator, allocation, &mapped_ptr) != VK_SUCCESS) return {};
-        }
-        const size_t n = static_cast<size_t>(std::min<VkDeviceSize>(size,
-                                                                    (std::numeric_limits<size_t>::max)()));
-        return std::span<u8>{static_cast<u8 *>(mapped_ptr), n};
-    }
-
-    std::span<const u8> MemoryCommit::Map() const
-    {
-        if (!allocation) return {};
-        if (!mapped_ptr) {
-            void *p = nullptr;
-            if (vmaMapMemory(allocator, allocation, &p) != VK_SUCCESS) return {};
-            const_cast<MemoryCommit *>(this)->mapped_ptr = p;
-        }
-        const size_t n = static_cast<size_t>(std::min<VkDeviceSize>(size,
-                                                                    (std::numeric_limits<size_t>::max)()));
-        return std::span<const u8>{static_cast<const u8 *>(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<uintptr_t>(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())
@@ -224,6 +90,21 @@ namespace Vulkan {
 
     MemoryAllocator::~MemoryAllocator() = default;
 
+    void MemoryAllocator::SetReclaimCallback(ReclaimCallback callback) {
+        reclaim_callback = std::move(callback);
+        vk::SetAllocatorOwnerThread();
+    }
+
+    bool MemoryAllocator::ReclaimAtLeast(u64 hint_bytes) const {
+        if (!reclaim_callback || in_reclaim) {
+            return false;
+        }
+        in_reclaim = true;
+        const u64 freed = reclaim_callback(hint_bytes);
+        in_reclaim = false;
+        return freed > 0;
+    }
+
     vk::Image MemoryAllocator::CreateImage(const VkImageCreateInfo &ci) const
     {
         const VmaAllocationCreateInfo alloc_ci = {
@@ -240,7 +121,26 @@ namespace Vulkan {
         VkImage handle{};
         VmaAllocation allocation{};
         VmaAllocationInfo alloc_info{};
-        vk::Check(vmaCreateImage(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info));
+        DEBUG_ASSERT(vk::OnAllocatorOwnerThread());
+
+        VkResult res = vmaCreateImage(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info);
+
+        if (res != VK_SUCCESS && ReclaimAtLeast(IMAGE_RECLAIM_HINT)) {
+            res = vmaCreateImage(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info);
+        }
+
+        if (res != VK_SUCCESS) {
+            auto relaxed_ci = alloc_ci;
+            relaxed_ci.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
+            res = vmaCreateImage(allocator, &ci, &relaxed_ci, &handle, &allocation, &alloc_info);
+
+            if (res != VK_SUCCESS) {
+                relaxed_ci.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
+                res = vmaCreateImage(allocator, &ci, &relaxed_ci, &handle, &allocation, &alloc_info);
+            }
+        }
+
+        vk::Check(res);
 
         // Log GPU memory allocation for images
         if (GPU::Logging::IsActive() &&
@@ -277,7 +177,28 @@ namespace Vulkan {
         VmaAllocation allocation{};
         VkMemoryPropertyFlags property_flags{};
 
-        vk::Check(vmaCreateBuffer(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info));
+        DEBUG_ASSERT(vk::OnAllocatorOwnerThread());
+
+        VkResult res = vmaCreateBuffer(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info);
+
+        if (res != VK_SUCCESS && ReclaimAtLeast(ci.size)) {
+            res = vmaCreateBuffer(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info);
+        }
+
+        if (res != VK_SUCCESS) {
+            auto relaxed_ci = alloc_ci;
+            relaxed_ci.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
+            res = vmaCreateBuffer(allocator, &ci, &relaxed_ci, &handle, &allocation, &alloc_info);
+
+            if (res != VK_SUCCESS &&
+                (relaxed_ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) {
+                relaxed_ci.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
+                res = vmaCreateBuffer(allocator, &ci, &relaxed_ci, &handle, &allocation,
+                                      &alloc_info);
+            }
+        }
+
+        vk::Check(res);
         vmaGetAllocationMemoryProperties(allocator, allocation, &property_flags);
 
         // Log GPU memory allocation for buffers
@@ -299,77 +220,4 @@ namespace Vulkan {
                           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

src/video_core/vulkan_common/vulkan_memory_allocator.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 2019 yuzu Emulator Project
@@ -6,6 +6,7 @@
 
 #pragma once
 
+#include <functional>
 #include <memory>
 #include <span>
 #include <vector>
@@ -39,51 +40,6 @@ 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<u8> Map();
-
-        [[nodiscard]] std::span<const u8> 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 {
@@ -107,36 +63,21 @@ namespace Vulkan {
 
         vk::Buffer CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsage usage) 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);
+        using ReclaimCallback = std::function<u64(u64)>;
 
-        /// Commits memory required by the buffer and binds it (for buffers created outside VMA).
-        MemoryCommit Commit(const vk::Buffer &buffer, MemoryUsage usage);
+        void SetReclaimCallback(ReclaimCallback callback);
 
     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;
-            }
-        }
+        bool ReclaimAtLeast(u64 hint_bytes) const;
+
+        static constexpr u64 IMAGE_RECLAIM_HINT = 64ULL * 1024 * 1024;
 
         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};
+        ReclaimCallback reclaim_callback;
+        mutable bool in_reclaim{false};
     };
 
 } // namespace Vulkan

src/video_core/vulkan_common/vulkan_wrapper.cpp

@@ -5,11 +5,16 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <algorithm>
+#include <atomic>
+#include <limits>
 #include <memory>
+#include <mutex>
 #include <optional>
+#include <thread>
 #include <utility>
 #include <vector>
 
+#include "common/assert.h"
 #include "common/common_types.h"
 #include "common/logging.h"
 #include "video_core/vulkan_common/vk_enum_string_helper.h"
@@ -20,6 +25,60 @@ namespace Vulkan::vk {
 
 namespace {
 
+std::thread::id allocator_owner_thread;
+
+template <typename HandleType>
+struct PendingRelease {
+    VmaAllocator allocator;
+    HandleType handle;
+    VmaAllocation allocation;
+    u64 timeline;
+};
+
+std::mutex deletion_mutex;
+std::atomic<u64> deletion_timeline{1};
+std::vector<PendingRelease<VkImage>> pending_images;
+std::vector<PendingRelease<VkBuffer>> pending_buffers;
+
+template <typename HandleType>
+void PushPendingRelease(std::vector<PendingRelease<HandleType>>& pending, VmaAllocator allocator,
+                        HandleType handle, VmaAllocation allocation) noexcept {
+    std::scoped_lock lock{deletion_mutex};
+    pending.push_back(PendingRelease<HandleType>{
+        .allocator = allocator,
+        .handle = handle,
+        .allocation = allocation,
+        .timeline = deletion_timeline.load(std::memory_order_acquire),
+    });
+}
+
+template <typename HandleType>
+void ExtractReleased(std::vector<PendingRelease<HandleType>>& pending,
+                     std::vector<PendingRelease<HandleType>>& released, u64 completed_value) {
+    const auto split = std::partition(pending.begin(), pending.end(),
+                                      [completed_value](const PendingRelease<HandleType>& entry) {
+                                          return entry.timeline > completed_value;
+                                      });
+    released.assign(split, pending.end());
+    pending.erase(split, pending.end());
+}
+
+void DrainDeletionQueue(u64 completed_value) noexcept {
+    std::vector<PendingRelease<VkImage>> images;
+    std::vector<PendingRelease<VkBuffer>> buffers;
+    {
+        std::scoped_lock lock{deletion_mutex};
+        ExtractReleased(pending_images, images, completed_value);
+        ExtractReleased(pending_buffers, buffers, completed_value);
+    }
+    for (const auto& entry : images) {
+        vmaDestroyImage(entry.allocator, entry.handle, entry.allocation);
+    }
+    for (const auto& entry : buffers) {
+        vmaDestroyBuffer(entry.allocator, entry.handle, entry.allocation);
+    }
+}
+
 template <typename Func>
 void SortPhysicalDevices(std::vector<VkPhysicalDevice>& devices, const InstanceDispatch& dld,
                          Func&& func) {
@@ -237,12 +296,6 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
     X(vkUnmapMemory);
     X(vkUpdateDescriptorSetWithTemplate);
     X(vkUpdateDescriptorSets);
-    X(vkGetBufferDeviceAddress);
-    X(vkGetDescriptorSetLayoutSizeEXT);
-    X(vkGetDescriptorSetLayoutBindingOffsetEXT);
-    X(vkGetDescriptorEXT);
-    X(vkCmdBindDescriptorBuffersEXT);
-    X(vkCmdSetDescriptorBufferOffsetsEXT);
     X(vkWaitForFences);
     X(vkWaitSemaphores);
 
@@ -508,13 +561,35 @@ DebugReportCallback Instance::CreateDebugReportCallback(
     return DebugReportCallback(object, handle, *dld);
 }
 
+void SetAllocatorOwnerThread() {
+    allocator_owner_thread = std::this_thread::get_id();
+}
+
+bool OnAllocatorOwnerThread() noexcept {
+    return allocator_owner_thread == std::thread::id{} ||
+           allocator_owner_thread == std::this_thread::get_id();
+}
+
+void SetDeletionTimeline(u64 value) noexcept {
+    deletion_timeline.store(value, std::memory_order_release);
+}
+
+void TickDeletionQueue(u64 completed_value) noexcept {
+    DEBUG_ASSERT(OnAllocatorOwnerThread());
+    DrainDeletionQueue(completed_value);
+}
+
+void FlushDeletionQueue() noexcept {
+    DrainDeletionQueue((std::numeric_limits<u64>::max)());
+}
+
 void Image::SetObjectNameEXT(const char* name) const {
     SetObjectName(dld, owner, handle, VK_OBJECT_TYPE_IMAGE, name);
 }
 
 void Image::Release() const noexcept {
     if (handle) {
-        vmaDestroyImage(allocator, handle, allocation);
+        PushPendingRelease(pending_images, allocator, handle, allocation);
     }
 }
 
@@ -536,7 +611,7 @@ void Buffer::SetObjectNameEXT(const char* name) const {
 
 void Buffer::Release() const noexcept {
     if (handle) {
-        vmaDestroyBuffer(allocator, handle, allocation);
+        PushPendingRelease(pending_buffers, allocator, handle, allocation);
     }
 }
 

src/video_core/vulkan_common/vulkan_wrapper.h

@@ -131,6 +131,16 @@ private:
     VkResult result;
 };
 
+void SetAllocatorOwnerThread();
+
+[[nodiscard]] bool OnAllocatorOwnerThread() noexcept;
+
+void SetDeletionTimeline(u64 value) noexcept;
+
+void TickDeletionQueue(u64 completed_value) noexcept;
+
+void FlushDeletionQueue() noexcept;
+
 /// Throws a Vulkan exception if result is not success.
 inline void Check(VkResult result) {
     if (result != VK_SUCCESS) {
@@ -352,12 +362,6 @@ struct DeviceDispatch : InstanceDispatch {
     PFN_vkSetDebugUtilsObjectTagEXT vkSetDebugUtilsObjectTagEXT{};
     PFN_vkUnmapMemory vkUnmapMemory{};
     PFN_vkUpdateDescriptorSetWithTemplate vkUpdateDescriptorSetWithTemplate{};
-    PFN_vkGetBufferDeviceAddress vkGetBufferDeviceAddress{};
-    PFN_vkGetDescriptorSetLayoutSizeEXT vkGetDescriptorSetLayoutSizeEXT{};
-    PFN_vkGetDescriptorSetLayoutBindingOffsetEXT vkGetDescriptorSetLayoutBindingOffsetEXT{};
-    PFN_vkGetDescriptorEXT vkGetDescriptorEXT{};
-    PFN_vkCmdBindDescriptorBuffersEXT vkCmdBindDescriptorBuffersEXT{};
-    PFN_vkCmdSetDescriptorBufferOffsetsEXT vkCmdSetDescriptorBufferOffsetsEXT{};
     PFN_vkUpdateDescriptorSets vkUpdateDescriptorSets{};
     PFN_vkWaitForFences vkWaitForFences{};
     PFN_vkWaitSemaphores vkWaitSemaphores{};
@@ -799,11 +803,6 @@ public:
         return !mapped.empty();
     }
 
-    /// Returns true if host writes are visible to the device.
-    bool IsHostCoherent() const noexcept {
-        return is_coherent;
-    }
-
     void Flush() const;
 
     void Invalidate() const;
@@ -1098,34 +1097,6 @@ public:
         dld->vkUpdateDescriptorSetWithTemplate(handle, set, update_template, data);
     }
 
-    [[nodiscard]] VkDeviceAddress GetBufferDeviceAddress(VkBuffer buffer) const noexcept {
-        const VkBufferDeviceAddressInfo info{
-            .sType = VK_STRUCTURE_TYPE_BUFFER_DEVICE_ADDRESS_INFO,
-            .pNext = nullptr,
-            .buffer = buffer,
-        };
-        return dld->vkGetBufferDeviceAddress(handle, &info);
-    }
-
-    [[nodiscard]] VkDeviceSize GetDescriptorSetLayoutSizeEXT(
-        VkDescriptorSetLayout layout) const noexcept {
-        VkDeviceSize size{};
-        dld->vkGetDescriptorSetLayoutSizeEXT(handle, layout, &size);
-        return size;
-    }
-
-    [[nodiscard]] VkDeviceSize GetDescriptorSetLayoutBindingOffsetEXT(
-        VkDescriptorSetLayout layout, u32 binding) const noexcept {
-        VkDeviceSize offset{};
-        dld->vkGetDescriptorSetLayoutBindingOffsetEXT(handle, layout, binding, &offset);
-        return offset;
-    }
-
-    void GetDescriptorEXT(const VkDescriptorGetInfoEXT& info, size_t size,
-                          void* descriptor) const noexcept {
-        dld->vkGetDescriptorEXT(handle, &info, size, descriptor);
-    }
-
     VkResult AcquireNextImageKHR(VkSwapchainKHR swapchain, u64 timeout, VkSemaphore semaphore,
                                  VkFence fence, u32* image_index) const noexcept {
         return dld->vkAcquireNextImageKHR(handle, swapchain, timeout, semaphore, fence,
@@ -1439,18 +1410,6 @@ public:
         PipelineBarrier(src_stage_mask, dst_stage_mask, dependency_flags, {}, {}, image_barrier);
     }
 
-    void BindDescriptorBuffersEXT(Span<VkDescriptorBufferBindingInfoEXT> bindings) const noexcept {
-        dld->vkCmdBindDescriptorBuffersEXT(handle, bindings.size(), bindings.data());
-    }
-
-    void SetDescriptorBufferOffsetsEXT(VkPipelineBindPoint bind_point, VkPipelineLayout layout,
-                                       u32 first_set, Span<u32> buffer_indices,
-                                       Span<VkDeviceSize> offsets) const noexcept {
-        dld->vkCmdSetDescriptorBufferOffsetsEXT(handle, bind_point, layout, first_set,
-                                                buffer_indices.size(), buffer_indices.data(),
-                                                offsets.data());
-    }
-
     void CopyBufferToImage(VkBuffer src_buffer, VkImage dst_image, VkImageLayout dst_image_layout,
                            Span<VkBufferImageCopy> regions) const noexcept {
         dld->vkCmdCopyBufferToImage(handle, src_buffer, dst_image, dst_image_layout, regions.size(),