[vulkan] 3rd Vulkan Global Maintenance (#4189)

Following the philosophy on the previous Vulkan maintenance PR's, this one is reviewing the video_core to resolve VUID's related to the missing handling/ safechecks (UBO's case on missing StorageBufferAccess 8/16 bits on QCOM drivers, resulting on bad compute pipeline compiled), wiring topology representative to EDS2 configuration (following #4117), refactored the sampled image access via new memeber function for type "typeless" which adds a handling on the integer mistmatch (uint to sint, float to uint) + fixing bugs on current texture sampling (int) including the depth/stencil path resolve, maintenance to previous changes on query cache (resolving bugs from #3853), reduced the amount of binding BindVertexBufferEXT2 when a new tick/ frame is presented (with this being a reason for performance reduction on games where vertex polutes the stage, like BOTW/TOTK where grass is draw with vertex); implemented color border swizzle and color write enable for future improvements on EDS3, added initial implementation on Synchronization2 starting a path to ensure an access to VK 1.3 features safely and other smaller fixes along the road.

Special Thanks:

-> Mr. Smolio Gidolard (@gidoly)

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4189
This commit is contained in:
CamilleLaVey
2026-07-09 04:22:43 +02:00
committed by crueter
parent 41762940d6
commit 8225151a44
34 changed files with 739 additions and 133 deletions
+36 -22
View File
@@ -95,6 +95,12 @@ constexpr std::array VK_FORMAT_A4B4G4R4_UNORM_PACK16{
VK_FORMAT_UNDEFINED,
};
constexpr std::array B10G11R11_UFLOAT_PACK32{
VK_FORMAT_R16G16B16A16_SFLOAT,
VK_FORMAT_A8B8G8R8_SRGB_PACK32,
VK_FORMAT_UNDEFINED,
};
} // namespace Alternatives
template <typename T>
@@ -127,6 +133,8 @@ constexpr const VkFormat* GetFormatAlternatives(VkFormat format) {
return Alternatives::VK_FORMAT_R32G32B32_SFLOAT.data();
case VK_FORMAT_A4B4G4R4_UNORM_PACK16_EXT:
return Alternatives::VK_FORMAT_A4B4G4R4_UNORM_PACK16.data();
case VK_FORMAT_B10G11R11_UFLOAT_PACK32:
return Alternatives::B10G11R11_UFLOAT_PACK32.data();
default:
return nullptr;
}
@@ -492,18 +500,14 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
CollectToolingInfo();
if (is_qualcomm) {
LOG_WARNING(Render_Vulkan,
"Qualcomm drivers require scaled vertex format emulation");
LOG_WARNING(Render_Vulkan, "Qualcomm drivers require scaled vertex format emulation");
must_emulate_scaled_formats = true;
LOG_WARNING(Render_Vulkan,
"Qualcomm drivers have broken provoking vertex");
RemoveExtension(extensions.provoking_vertex, VK_EXT_PROVOKING_VERTEX_EXTENSION_NAME);
LOG_WARNING(Render_Vulkan,
"Qualcomm drivers have slow push descriptor implementation");
RemoveExtension(extensions.push_descriptor, VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME);
LOG_WARNING(Render_Vulkan,
"Disabling shader float controls and 64-bit integer features on Qualcomm proprietary drivers");
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken CustomBorderColor.");
RemoveExtensionFeature(extensions.custom_border_color, features.custom_border_color,
VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken shader float controls.");
RemoveExtension(extensions.shader_float_controls, VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME);
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken shader atomic int64.");
RemoveExtensionFeature(extensions.shader_atomic_int64, features.shader_atomic_int64,
VK_KHR_SHADER_ATOMIC_INT64_EXTENSION_NAME);
features.shader_atomic_int64.shaderBufferInt64Atomics = false;
@@ -561,7 +565,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
features.shader_float16_int8.shaderFloat16 = false;
}
// Mali/ NVIDIA proprietary drivers: Shader stencil export not supported
// Use hardware depth/stencil blits instead when available
if (!extensions.shader_stencil_export) {
LOG_INFO(Render_Vulkan,
@@ -657,14 +660,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
}
const auto dyna_state = Settings::values.dyna_state.GetValue();
// Base dynamic states (VIEWPORT, SCISSOR, DEPTH_BIAS, etc.) are ALWAYS active in vk_graphics_pipeline.cpp
// This slider controls EXTENDED dynamic states with accumulative levels per Vulkan specs:
// Level 0 = Core Dynamic States only (Vulkan 1.0)
// Level 1 = Core + VK_EXT_extended_dynamic_state
// Level 2 = Core + VK_EXT_extended_dynamic_state + VK_EXT_extended_dynamic_state2
// Level 3 = Core + VK_EXT_extended_dynamic_state + VK_EXT_extended_dynamic_state2 + VK_EXT_extended_dynamic_state3
switch (dyna_state) {
case Settings::ExtendedDynamicState::Disabled:
// Level 0: Disable all extended dynamic state extensions
@@ -699,8 +694,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
break;
}
// VK_EXT_vertex_input_dynamic_state is independent from EDS
// It can be enabled even without extended_dynamic_state
// VK_EXT_vertex_input_dynamic_state
if (!Settings::values.vertex_input_dynamic_state.GetValue()) {
RemoveExtensionFeature(extensions.vertex_input_dynamic_state, features.vertex_input_dynamic_state, VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME);
}
@@ -793,7 +787,6 @@ void Device::SaveShader(std::span<const u32> spirv) const {
}
bool Device::ComputeIsOptimalAstcSupported() const {
// Verify hardware supports all ASTC formats with optimal tiling to avoid software conversion
static constexpr std::array<VkFormat, 28> astc_formats = {
VK_FORMAT_ASTC_4x4_UNORM_BLOCK, VK_FORMAT_ASTC_4x4_SRGB_BLOCK,
VK_FORMAT_ASTC_5x4_UNORM_BLOCK, VK_FORMAT_ASTC_5x4_SRGB_BLOCK,
@@ -1170,6 +1163,11 @@ bool Device::GetSuitability(bool requires_swapchain) {
}
void Device::RemoveUnsuitableExtensions() {
// VK_EXT_color_write_enable
extensions.color_write_enable = features.color_write_enable.colorWriteEnable;
RemoveExtensionFeatureIfUnsuitable(extensions.color_write_enable, features.color_write_enable,
VK_EXT_COLOR_WRITE_ENABLE_EXTENSION_NAME);
// VK_EXT_custom_border_color
if (extensions.custom_border_color) {
extensions.custom_border_color =
@@ -1179,6 +1177,17 @@ void Device::RemoveUnsuitableExtensions() {
RemoveExtensionFeatureIfUnsuitable(extensions.custom_border_color, features.custom_border_color,
VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
// VK_EXT_border_color_swizzle
if (extensions.border_color_swizzle) {
extensions.border_color_swizzle =
extensions.custom_border_color &&
features.border_color_swizzle.borderColorSwizzle &&
features.border_color_swizzle.borderColorSwizzleFromImage;
}
RemoveExtensionFeatureIfUnsuitable(extensions.border_color_swizzle,
features.border_color_swizzle,
VK_EXT_BORDER_COLOR_SWIZZLE_EXTENSION_NAME);
// VK_EXT_depth_bias_control
extensions.depth_bias_control =
features.depth_bias_control.depthBiasControl &&
@@ -1375,6 +1384,11 @@ void Device::RemoveUnsuitableExtensions() {
// VK_KHR_maintenance8
extensions.maintenance8 = loaded_extensions.contains(VK_KHR_MAINTENANCE_8_EXTENSION_NAME);
RemoveExtensionIfUnsuitable(extensions.maintenance8, VK_KHR_MAINTENANCE_8_EXTENSION_NAME);
// VK_KHR_synchronization2
extensions.synchronization2 = features.synchronization2.synchronization2;
RemoveExtensionFeatureIfUnsuitable(extensions.synchronization2, features.synchronization2,
VK_KHR_SYNCHRONIZATION_2_EXTENSION_NAME);
}
void Device::SetupFamilies(VkSurfaceKHR surface) {
+40 -2
View File
@@ -43,12 +43,15 @@ VK_DEFINE_HANDLE(VmaAllocator)
FEATURE(EXT, ShaderDemoteToHelperInvocation, SHADER_DEMOTE_TO_HELPER_INVOCATION, \
shader_demote_to_helper_invocation) \
FEATURE(EXT, SubgroupSizeControl, SUBGROUP_SIZE_CONTROL, subgroup_size_control) \
FEATURE(KHR, Maintenance4, MAINTENANCE_4, maintenance4)
FEATURE(KHR, Maintenance4, MAINTENANCE_4, maintenance4) \
FEATURE(KHR, Synchronization2, SYNCHRONIZATION_2, synchronization2)
#define FOR_EACH_VK_FEATURE_1_4(FEATURE)
// Define all features which may be used by the implementation and require an extension here.
#define FOR_EACH_VK_FEATURE_EXT(FEATURE) \
FEATURE(EXT, BorderColorSwizzle, BORDER_COLOR_SWIZZLE, border_color_swizzle) \
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, DepthClipControl, DEPTH_CLIP_CONTROL, depth_clip_control) \
@@ -180,6 +183,7 @@ VK_DEFINE_HANDLE(VmaAllocator)
FEATURE_NAME(robustness2, nullDescriptor) \
FEATURE_NAME(shader_float16_int8, shaderFloat16) \
FEATURE_NAME(shader_float16_int8, shaderInt8) \
FEATURE_NAME(synchronization2, synchronization2) \
FEATURE_NAME(timeline_semaphore, timelineSemaphore) \
FEATURE_NAME(transform_feedback, transformFeedback) \
FEATURE_NAME(uniform_buffer_standard_layout, uniformBufferStandardLayout) \
@@ -383,6 +387,16 @@ FN_MAX_LIMIT_LIST
return features.shader_float16_int8.shaderInt8;
}
/// Returns true if the device allows 8-bit integer members in uniform/storage buffers.
bool IsUniformAndStorageBuffer8BitAccessSupported() const {
return features.bit8_storage.uniformAndStorageBuffer8BitAccess;
}
/// Returns true if the device allows 16-bit integer members in uniform/storage buffers.
bool IsUniformAndStorageBuffer16BitAccessSupported() const {
return features.bit16_storage.uniformAndStorageBuffer16BitAccess;
}
/// Returns true if the device supports binding multisample images as storage images.
bool IsStorageImageMultisampleSupported() const {
return features.features.shaderStorageImageMultisample;
@@ -509,7 +523,6 @@ FN_MAX_LIMIT_LIST
}
/// Returns true if the device supports VK_EXT_shader_stencil_export.
/// Note: Most Mali/NVIDIA drivers don't support this. Use hardware blits as fallback.
bool IsExtShaderStencilExportSupported() const {
return extensions.shader_stencil_export;
}
@@ -546,6 +559,11 @@ FN_MAX_LIMIT_LIST
return extensions.subgroup_size_control;
}
/// Returns true if vkResetQueryPool (host-side query reset) is supported.
bool IsHostQueryResetSupported() const {
return features.host_query_reset.hostQueryReset != VK_FALSE;
}
/// Returns true if the device supports VK_EXT_transform_feedback.
bool IsExtTransformFeedbackSupported() const {
return extensions.transform_feedback;
@@ -582,6 +600,21 @@ FN_MAX_LIMIT_LIST
return features.custom_border_color.customBorderColorWithoutFormat;
}
/// Returns true if the device supports VK_EXT_color_write_enable.
bool IsExtColorWriteEnableSupported() const {
return extensions.color_write_enable;
}
/// Returns true if the device supports VK_EXT_border_color_swizzle.
bool IsExtBorderColorSwizzleSupported() const {
return extensions.border_color_swizzle;
}
/// Returns true if borderColorSwizzleFromImage is available.
bool IsBorderColorSwizzleFromImageSupported() const {
return features.border_color_swizzle.borderColorSwizzleFromImage;
}
/// Returns true if the device supports VK_EXT_extended_dynamic_state.
bool IsExtExtendedDynamicStateSupported() const {
return extensions.extended_dynamic_state;
@@ -722,6 +755,11 @@ FN_MAX_LIMIT_LIST
bool HasTimelineSemaphore() const;
/// Returns true if the device supports VK_KHR_synchronization2.
bool HasSynchronization2() const {
return extensions.synchronization2;
}
/// Returns the minimum supported version of SPIR-V.
u32 SupportedSpirvVersion() const {
if (instance_version >= VK_API_VERSION_1_3) {
@@ -123,6 +123,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
X(vkCmdEndDebugUtilsLabelEXT);
X(vkCmdFillBuffer);
X(vkCmdPipelineBarrier);
X(vkCmdPipelineBarrier2);
X(vkCmdPushConstants);
X(vkCmdPushDescriptorSetWithTemplateKHR);
X(vkCmdSetBlendConstants);
@@ -161,8 +162,10 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
X(vkCmdSetStencilTestEnableEXT);
X(vkCmdSetVertexInputEXT);
X(vkCmdSetColorWriteMaskEXT);
X(vkCmdSetColorWriteEnableEXT);
X(vkCmdSetColorBlendEnableEXT);
X(vkCmdSetColorBlendEquationEXT);
X(vkCmdResetQueryPool);
X(vkCmdResolveImage);
X(vkCreateBuffer);
X(vkCreateBufferView);
@@ -226,6 +229,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
X(vkGetSemaphoreCounterValue);
X(vkMapMemory);
X(vkQueueSubmit);
X(vkQueueSubmit2);
X(vkResetFences);
X(vkResetQueryPool);
X(vkSetDebugUtilsObjectNameEXT);
@@ -252,6 +256,14 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
Proc(dld.vkCmdDrawIndirectCount, dld, "vkCmdDrawIndirectCountKHR", device);
Proc(dld.vkCmdDrawIndexedIndirectCount, dld, "vkCmdDrawIndexedIndirectCountKHR", device);
}
// Synchronization2 is core in Vulkan 1.3, otherwise requires VK_KHR_synchronization2
if (!dld.vkCmdPipelineBarrier2) {
Proc(dld.vkCmdPipelineBarrier2, dld, "vkCmdPipelineBarrier2KHR", device);
}
if (!dld.vkQueueSubmit2) {
Proc(dld.vkQueueSubmit2, dld, "vkQueueSubmit2KHR", device);
}
#undef X
}
@@ -6,6 +6,7 @@
#pragma once
#include <array>
#include <exception>
#include <limits>
#include <memory>
@@ -237,8 +238,10 @@ struct DeviceDispatch : InstanceDispatch {
PFN_vkCmdEndTransformFeedbackEXT vkCmdEndTransformFeedbackEXT{};
PFN_vkCmdFillBuffer vkCmdFillBuffer{};
PFN_vkCmdPipelineBarrier vkCmdPipelineBarrier{};
PFN_vkCmdPipelineBarrier2 vkCmdPipelineBarrier2{};
PFN_vkCmdPushConstants vkCmdPushConstants{};
PFN_vkCmdPushDescriptorSetWithTemplateKHR vkCmdPushDescriptorSetWithTemplateKHR{};
PFN_vkCmdResetQueryPool vkCmdResetQueryPool{};
PFN_vkCmdResolveImage vkCmdResolveImage{};
PFN_vkCmdSetBlendConstants vkCmdSetBlendConstants{};
PFN_vkCmdSetCullModeEXT vkCmdSetCullModeEXT{};
@@ -275,6 +278,7 @@ struct DeviceDispatch : InstanceDispatch {
PFN_vkCmdSetVertexInputEXT vkCmdSetVertexInputEXT{};
PFN_vkCmdSetViewport vkCmdSetViewport{};
PFN_vkCmdSetColorWriteMaskEXT vkCmdSetColorWriteMaskEXT{};
PFN_vkCmdSetColorWriteEnableEXT vkCmdSetColorWriteEnableEXT{};
PFN_vkCmdSetColorBlendEnableEXT vkCmdSetColorBlendEnableEXT{};
PFN_vkCmdSetColorBlendEquationEXT vkCmdSetColorBlendEquationEXT{};
PFN_vkCmdWaitEvents vkCmdWaitEvents{};
@@ -340,6 +344,7 @@ struct DeviceDispatch : InstanceDispatch {
PFN_vkGetSemaphoreCounterValue vkGetSemaphoreCounterValue{};
PFN_vkMapMemory vkMapMemory{};
PFN_vkQueueSubmit vkQueueSubmit{};
PFN_vkQueueSubmit2 vkQueueSubmit2{};
PFN_vkResetFences vkResetFences{};
PFN_vkResetQueryPool vkResetQueryPool{};
PFN_vkSetDebugUtilsObjectNameEXT vkSetDebugUtilsObjectNameEXT{};
@@ -819,6 +824,12 @@ public:
return dld->vkQueueSubmit(queue, submit_infos.size(), submit_infos.data(), fence);
}
/// Submits using VK_KHR_synchronization2 / Vulkan 1.3 vkQueueSubmit2.
VkResult Submit2(Span<VkSubmitInfo2> submit_infos,
VkFence fence = VK_NULL_HANDLE) const noexcept {
return dld->vkQueueSubmit2(queue, submit_infos.size(), submit_infos.data(), fence);
}
VkResult Present(const VkPresentInfoKHR& present_info) const noexcept {
return dld->vkQueuePresentKHR(queue, &present_info);
}
@@ -1180,6 +1191,10 @@ public:
dld->vkCmdEndQuery(handle, query_pool, query);
}
void ResetQueryPool(VkQueryPool query_pool, u32 first, u32 count) const noexcept {
dld->vkCmdResetQueryPool(handle, query_pool, first, count);
}
void BindDescriptorSets(VkPipelineBindPoint bind_point, VkPipelineLayout layout, u32 first,
Span<VkDescriptorSet> sets, Span<u32> dynamic_offsets) const noexcept {
dld->vkCmdBindDescriptorSets(handle, bind_point, layout, first, sets.size(), sets.data(),
@@ -1287,6 +1302,72 @@ public:
VkDependencyFlags dependency_flags, Span<VkMemoryBarrier> memory_barriers,
Span<VkBufferMemoryBarrier> buffer_barriers,
Span<VkImageMemoryBarrier> image_barriers) const noexcept {
static constexpr u32 MaxBarriers = 16;
if (dld->vkCmdPipelineBarrier2 && memory_barriers.size() <= MaxBarriers &&
buffer_barriers.size() <= MaxBarriers && image_barriers.size() <= MaxBarriers) {
const auto src_stage_mask2 = static_cast<VkPipelineStageFlags2>(src_stage_mask);
const auto dst_stage_mask2 = static_cast<VkPipelineStageFlags2>(dst_stage_mask);
std::array<VkMemoryBarrier2, MaxBarriers> memory_barriers2;
for (u32 i = 0; i < memory_barriers.size(); ++i) {
memory_barriers2[i] = VkMemoryBarrier2{
.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2,
.pNext = nullptr,
.srcStageMask = src_stage_mask2,
.srcAccessMask = static_cast<VkAccessFlags2>(memory_barriers[i].srcAccessMask),
.dstStageMask = dst_stage_mask2,
.dstAccessMask = static_cast<VkAccessFlags2>(memory_barriers[i].dstAccessMask),
};
}
std::array<VkBufferMemoryBarrier2, MaxBarriers> buffer_barriers2;
for (u32 i = 0; i < buffer_barriers.size(); ++i) {
const auto& barrier = buffer_barriers[i];
buffer_barriers2[i] = VkBufferMemoryBarrier2{
.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER_2,
.pNext = nullptr,
.srcStageMask = src_stage_mask2,
.srcAccessMask = static_cast<VkAccessFlags2>(barrier.srcAccessMask),
.dstStageMask = dst_stage_mask2,
.dstAccessMask = static_cast<VkAccessFlags2>(barrier.dstAccessMask),
.srcQueueFamilyIndex = barrier.srcQueueFamilyIndex,
.dstQueueFamilyIndex = barrier.dstQueueFamilyIndex,
.buffer = barrier.buffer,
.offset = barrier.offset,
.size = barrier.size,
};
}
std::array<VkImageMemoryBarrier2, MaxBarriers> image_barriers2;
for (u32 i = 0; i < image_barriers.size(); ++i) {
const auto& barrier = image_barriers[i];
image_barriers2[i] = VkImageMemoryBarrier2{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2,
.pNext = nullptr,
.srcStageMask = src_stage_mask2,
.srcAccessMask = static_cast<VkAccessFlags2>(barrier.srcAccessMask),
.dstStageMask = dst_stage_mask2,
.dstAccessMask = static_cast<VkAccessFlags2>(barrier.dstAccessMask),
.oldLayout = barrier.oldLayout,
.newLayout = barrier.newLayout,
.srcQueueFamilyIndex = barrier.srcQueueFamilyIndex,
.dstQueueFamilyIndex = barrier.dstQueueFamilyIndex,
.image = barrier.image,
.subresourceRange = barrier.subresourceRange,
};
}
const VkDependencyInfo dependency_info{
.sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO,
.pNext = nullptr,
.dependencyFlags = dependency_flags,
.memoryBarrierCount = memory_barriers.size(),
.pMemoryBarriers = memory_barriers2.data(),
.bufferMemoryBarrierCount = buffer_barriers.size(),
.pBufferMemoryBarriers = buffer_barriers2.data(),
.imageMemoryBarrierCount = image_barriers.size(),
.pImageMemoryBarriers = image_barriers2.data(),
};
dld->vkCmdPipelineBarrier2(handle, &dependency_info);
return;
}
dld->vkCmdPipelineBarrier(handle, src_stage_mask, dst_stage_mask, dependency_flags,
memory_barriers.size(), memory_barriers.data(),
buffer_barriers.size(), buffer_barriers.data(),
@@ -1510,6 +1591,10 @@ public:
dld->vkCmdSetColorWriteMaskEXT(handle, first, masks.size(), masks.data());
}
void SetColorWriteEnableEXT(Span<VkBool32> enables) const noexcept {
dld->vkCmdSetColorWriteEnableEXT(handle, enables.size(), enables.data());
}
void SetColorBlendEnableEXT(u32 first, Span<VkBool32> enables) const noexcept {
dld->vkCmdSetColorBlendEnableEXT(handle, first, enables.size(), enables.data());
}