Browse Source

[vulkan] 4th Vulkan Global Maintenance (#4212)

The approach on this PR moves around refactoring/ updating residual work; main issue was the whole logic behind msaa upload/download images with the constant image copy per pass required; which was gated via shaderStorageImageMultisample, which is not available on all the platforms supported including Android, making them to rely into a more heavier approach to convert msaa image copy into a non msaa image copy, not only losing precision in the conversion, but also creating bugs around the resolve of the image itself, if the supported path had to copy 2 times the same image to be actually presentable, when the fallback was enabled 3 copies were performed, not only decreasing the performance but also increasing race conditions due to the certain capabilities processed in this chain (storage_views). In order to resolve them more naturally, the constant copies were removed from the now "common" path along establishing a proper color resolve for Android and actually making them more direct.

Yet this opened a new pack of outdated handling on our current backend; since we now pass msaa images more directly (whenever a game request a msaa type of image/ which seems to be not common + each fragment resolves images on how it's actually required, adding missing depth/stencil), how we often load and store them becomes a critical issue, since our current configuration renderpass/framebuffer forces to always load and load clear whenever a new attachment it's passing through the renderpass, making this really heavier on memory bandwidth limited devices as SteamDeck and others that doesn't rely on the inmediate rendering mode (Tilers vs iGPU); to prevent over allocating data, I had to re-structure part of the current renderpass and framebuffer to actually reuse part of the data/ attachments called, with flags as DON'T_CARE, simplifying partially on how we actually call attachments including their barriers and conditions (occlusion, etc), the DON'T_CARE it's specifically useful to not only prevent calling new loads (which is a flush and heavier on tiler devices) and not only reusing the actual data store, which works on the future time, this saves memory bandwidth and gives space to actually avoid the constant pressure on SYSMEM/GMEM shader resolver path.

This maintenance also improved the current removal features logic on QCOM drivers, closing gaps between implementations on descriptors (VK_EXT_indexing_descriptor), adjusting the whole WMEL to avoid being so stricter if certain capabilities are missing to actually use this feature, leading to test and play more on how QCOM drivers are actually behaving; meanwhile the main objective is to open path for a robust and bigger implementation in the future, the changes made here also contains small bug fixes on SPIR-V logic, adding missing cases to resolve shaders (FP64 -> FP16), adding more robustness on how narrow features are actually being used and how to wired them whenever they're not available, meanwhile there are still some issues in regards on how Tegra vs Adreno can preserve unorms at FP32 and FP16, leaving better understanding on base for what VK_EXT_shader_float_controls is actually doing most of the times (flushes NaN's and Denorms equally).

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4212
remotes/1784422634639613560/master
CamilleLaVey 7 days ago
committed by crueter
parent
commit
eb9280dedf
No known key found for this signature in database GPG Key ID: 425ACD2D4830EBC6
  1. 18
      src/shader_recompiler/backend/spirv/emit_spirv.cpp
  2. 7
      src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp
  3. 6
      src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp
  4. 97
      src/shader_recompiler/backend/spirv/emit_spirv_image.cpp
  5. 20
      src/shader_recompiler/backend/spirv/emit_spirv_memory.cpp
  6. 24
      src/shader_recompiler/backend/spirv/emit_spirv_shared_memory.cpp
  7. 26
      src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
  8. 6
      src/shader_recompiler/backend/spirv/spirv_emit_context.h
  9. 3
      src/shader_recompiler/ir_opt/lower_fp64_to_fp32.cpp
  10. 125
      src/shader_recompiler/ir_opt/lower_int64_to_int32.cpp
  11. 8
      src/shader_recompiler/profile.h
  12. 5
      src/video_core/host_shaders/CMakeLists.txt
  13. 13
      src/video_core/host_shaders/blit_color_msaa.frag
  14. 12
      src/video_core/host_shaders/blit_depth_msaa.frag
  15. 15
      src/video_core/host_shaders/blit_depth_stencil_msaa.frag
  16. 22
      src/video_core/host_shaders/convert_msaa_to_non_msaa.frag
  17. 21
      src/video_core/host_shaders/convert_non_msaa_to_msaa.frag
  18. 430
      src/video_core/renderer_vulkan/blit_image.cpp
  19. 57
      src/video_core/renderer_vulkan/blit_image.h
  20. 134
      src/video_core/renderer_vulkan/vk_compute_pass.cpp
  21. 19
      src/video_core/renderer_vulkan/vk_compute_pass.h
  22. 4
      src/video_core/renderer_vulkan/vk_compute_pipeline.h
  23. 30
      src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
  24. 13
      src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
  25. 124
      src/video_core/renderer_vulkan/vk_rasterizer.cpp
  26. 52
      src/video_core/renderer_vulkan/vk_render_pass_cache.cpp
  27. 8
      src/video_core/renderer_vulkan/vk_render_pass_cache.h
  28. 106
      src/video_core/renderer_vulkan/vk_scheduler.cpp
  29. 23
      src/video_core/renderer_vulkan/vk_scheduler.h
  30. 468
      src/video_core/renderer_vulkan/vk_texture_cache.cpp
  31. 48
      src/video_core/renderer_vulkan/vk_texture_cache.h
  32. 2
      src/video_core/texture_cache/image_info.cpp
  33. 14
      src/video_core/vulkan_common/vulkan_device.cpp
  34. 47
      src/video_core/vulkan_common/vulkan_device.h

18
src/shader_recompiler/backend/spirv/emit_spirv.cpp

@ -465,12 +465,22 @@ void SetupCapabilities(const Profile& profile, const Info& info, EmitContext& ct
ctx.AddCapability(spv::Capability::ImageGatherExtended);
ctx.AddCapability(spv::Capability::ImageQuery);
ctx.AddCapability(spv::Capability::SampledBuffer);
// TODO: this usage needs to be tracked properly
if (ctx.profile.support_sampled_image_array_nonuniform_indexing) {
if (ctx.profile.supported_spirv < 0x00010400)
if (!ctx.non_uniform_ids.empty()) {
if (ctx.profile.supported_spirv < 0x00010500)
ctx.AddExtension("SPV_EXT_descriptor_indexing");
ctx.AddCapability(spv::Capability::ShaderNonUniform);
ctx.AddCapability(spv::Capability::SampledImageArrayNonUniformIndexing);
if (ctx.uses_nonuniform_sampled_image) {
ctx.AddCapability(spv::Capability::SampledImageArrayNonUniformIndexing);
}
if (ctx.uses_nonuniform_storage_image) {
ctx.AddCapability(spv::Capability::StorageImageArrayNonUniformIndexing);
}
if (ctx.uses_nonuniform_uniform_texel_buffer) {
ctx.AddCapability(spv::Capability::UniformTexelBufferArrayNonUniformIndexing);
}
if (ctx.uses_nonuniform_storage_texel_buffer) {
ctx.AddCapability(spv::Capability::StorageTexelBufferArrayNonUniformIndexing);
}
}
}

7
src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp

@ -1,3 +1,6 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
@ -15,7 +18,7 @@ Id SharedPointer(EmitContext& ctx, Id offset, u32 index_offset = 0) {
if (index_offset > 0) {
index = ctx.OpIAdd(ctx.U32[1], index, ctx.Const(index_offset));
}
return ctx.profile.support_explicit_workgroup_layout
return ctx.uses_explicit_workgroup_layout
? ctx.OpAccessChain(ctx.shared_u32, ctx.shared_memory_u32, ctx.u32_zero_value, index)
: ctx.OpAccessChain(ctx.shared_u32, ctx.shared_memory_u32, index);
}
@ -155,7 +158,7 @@ Id EmitSharedAtomicExchange32(EmitContext& ctx, Id offset, Id value) {
}
Id EmitSharedAtomicExchange64(EmitContext& ctx, Id offset, Id value) {
if (ctx.profile.support_int64_atomics && ctx.profile.support_explicit_workgroup_layout) {
if (ctx.profile.support_shared_int64_atomics && ctx.uses_explicit_workgroup_layout) {
const Id shift_id{ctx.Const(3U)};
const Id index{ctx.OpShiftRightArithmetic(ctx.U32[1], offset, shift_id)};
const Id pointer{

6
src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp

@ -624,12 +624,14 @@ Id EmitRenderArea(EmitContext& ctx) {
}
Id EmitLoadLocal(EmitContext& ctx, Id word_offset) {
const Id pointer{ctx.OpAccessChain(ctx.private_u32, ctx.local_memory, word_offset)};
const Id pointer{
ctx.OpAccessChain(ctx.private_u32, ctx.local_memory, word_offset, ctx.Const(0U))};
return ctx.OpLoad(ctx.U32[1], pointer);
}
void EmitWriteLocal(EmitContext& ctx, Id word_offset, Id value) {
const Id pointer{ctx.OpAccessChain(ctx.private_u32, ctx.local_memory, word_offset)};
const Id pointer{
ctx.OpAccessChain(ctx.private_u32, ctx.local_memory, word_offset, ctx.Const(0U))};
ctx.OpStore(pointer, value);
}

97
src/shader_recompiler/backend/spirv/emit_spirv_image.cpp

@ -15,8 +15,56 @@
namespace Shader::Backend::SPIRV {
namespace {
[[nodiscard]] bool IsNonUniformDescriptor(EmitContext& ctx, const IR::Value& index) noexcept {
return ctx.profile.support_sampled_image_array_nonuniform_indexing && !index.IsImmediate();
enum class NonUniformKind {
SampledImage,
StorageImage,
UniformTexelBuffer,
StorageTexelBuffer,
};
[[nodiscard]] bool IsNonUniformSupported(const Profile& profile, NonUniformKind kind) noexcept {
switch (kind) {
case NonUniformKind::SampledImage:
return profile.support_sampled_image_array_nonuniform_indexing;
case NonUniformKind::StorageImage:
return profile.support_storage_image_array_nonuniform_indexing;
case NonUniformKind::UniformTexelBuffer:
return profile.support_uniform_texel_buffer_array_nonuniform_indexing;
case NonUniformKind::StorageTexelBuffer:
return profile.support_storage_texel_buffer_array_nonuniform_indexing;
}
return false;
}
void DecorateNonUniform(EmitContext& ctx, Id object) {
if (ctx.non_uniform_ids.contains(object.value)) {
return;
}
ctx.Decorate(object, spv::Decoration::NonUniform);
ctx.non_uniform_ids.insert(object.value);
}
[[nodiscard]] bool MarkNonUniform(EmitContext& ctx, Id idx, const IR::Value& index,
NonUniformKind kind) {
if (index.IsImmediate() || !IsNonUniformSupported(ctx.profile, kind)) {
return false;
}
DecorateNonUniform(ctx, idx);
switch (kind) {
case NonUniformKind::SampledImage:
ctx.uses_nonuniform_sampled_image = true;
break;
case NonUniformKind::StorageImage:
ctx.uses_nonuniform_storage_image = true;
break;
case NonUniformKind::UniformTexelBuffer:
ctx.uses_nonuniform_uniform_texel_buffer = true;
break;
case NonUniformKind::StorageTexelBuffer:
ctx.uses_nonuniform_storage_texel_buffer = true;
break;
}
return true;
}
class ImageOperands {
@ -195,12 +243,13 @@ Id Texture(EmitContext& ctx, IR::TextureInstInfo info, [[maybe_unused]] const IR
const TextureDefinition& def{ctx.textures.at(info.descriptor_index)};
if (def.count > 1) {
auto const idx = index.IsImmediate() ? ctx.Const(index.U32()) : ctx.Def(index);
if (!ctx.non_uniform_ids.contains(idx.value) && IsNonUniformDescriptor(ctx, index)) {
ctx.Decorate(idx, spv::Decoration::NonUniform);
ctx.non_uniform_ids.insert(idx.value);
}
const bool non_uniform{MarkNonUniform(ctx, idx, index, NonUniformKind::SampledImage)};
const Id pointer{ctx.OpAccessChain(def.pointer_type, def.id, idx)};
const Id object{ctx.OpLoad(def.sampled_type, pointer)};
if (non_uniform) {
DecorateNonUniform(ctx, pointer);
DecorateNonUniform(ctx, object);
}
return object;
} else {
return ctx.OpLoad(def.sampled_type, def.id);
@ -212,21 +261,30 @@ Id TextureImage(EmitContext& ctx, IR::TextureInstInfo info, const IR::Value& ind
const TextureBufferDefinition& def{ctx.texture_buffers.at(info.descriptor_index)};
if (def.count > 1) {
const Id idx{index.IsImmediate() ? ctx.Const(index.U32()) : ctx.Def(index)};
const bool non_uniform{
MarkNonUniform(ctx, idx, index, NonUniformKind::UniformTexelBuffer)};
const Id ptr{ctx.OpAccessChain(ctx.image_buffer_type, def.id, idx)};
return ctx.OpLoad(ctx.image_buffer_type, ptr);
const Id object{ctx.OpLoad(ctx.image_buffer_type, ptr)};
if (non_uniform) {
DecorateNonUniform(ctx, ptr);
DecorateNonUniform(ctx, object);
}
return object;
}
return ctx.OpLoad(ctx.image_buffer_type, def.id);
} else {
const TextureDefinition& def{ctx.textures.at(info.descriptor_index)};
if (def.count > 1) {
auto const idx = index.IsImmediate() ? ctx.Const(index.U32()) : ctx.Def(index);
if (!ctx.non_uniform_ids.contains(idx.value) && IsNonUniformDescriptor(ctx, index)) {
ctx.Decorate(idx, spv::Decoration::NonUniform);
ctx.non_uniform_ids.insert(idx.value);
}
const bool non_uniform{MarkNonUniform(ctx, idx, index, NonUniformKind::SampledImage)};
const Id ptr = ctx.OpAccessChain(def.pointer_type, def.id, idx);
const Id object = ctx.OpLoad(def.sampled_type, ptr);
const Id image = ctx.OpImage(def.image_type, object);
if (non_uniform) {
DecorateNonUniform(ctx, ptr);
DecorateNonUniform(ctx, object);
DecorateNonUniform(ctx, image);
}
return image;
}
return ctx.OpImage(def.image_type, ctx.OpLoad(def.sampled_type, def.id));
@ -238,16 +296,29 @@ std::pair<Id, bool> Image(EmitContext& ctx, const IR::Value& index, IR::TextureI
const ImageBufferDefinition def{ctx.image_buffers.at(info.descriptor_index)};
if (def.count > 1) {
const Id idx{index.IsImmediate() ? ctx.Const(index.U32()) : ctx.Def(index)};
const bool non_uniform{
MarkNonUniform(ctx, idx, index, NonUniformKind::StorageTexelBuffer)};
const Id ptr{ctx.OpAccessChain(def.pointer_type, def.id, idx)};
return {ctx.OpLoad(def.image_type, ptr), def.is_integer};
const Id image{ctx.OpLoad(def.image_type, ptr)};
if (non_uniform) {
DecorateNonUniform(ctx, ptr);
DecorateNonUniform(ctx, image);
}
return {image, def.is_integer};
}
return {ctx.OpLoad(def.image_type, def.id), def.is_integer};
} else {
const ImageDefinition def{ctx.images.at(info.descriptor_index)};
if (def.count > 1) {
const Id idx{index.IsImmediate() ? ctx.Const(index.U32()) : ctx.Def(index)};
const bool non_uniform{MarkNonUniform(ctx, idx, index, NonUniformKind::StorageImage)};
const Id ptr{ctx.OpAccessChain(def.pointer_type, def.id, idx)};
return {ctx.OpLoad(def.image_type, ptr), def.is_integer};
const Id image{ctx.OpLoad(def.image_type, ptr)};
if (non_uniform) {
DecorateNonUniform(ctx, ptr);
DecorateNonUniform(ctx, image);
}
return {image, def.is_integer};
}
return {ctx.OpLoad(def.image_type, def.id), def.is_integer};
}

20
src/shader_recompiler/backend/spirv/emit_spirv_memory.cpp

@ -159,7 +159,7 @@ void EmitWriteGlobal128(EmitContext& ctx, Id address, Id value) {
}
Id EmitLoadStorageU8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset) {
if (ctx.profile.support_int8 && ctx.profile.support_uniform_and_storage_buffer_8bit &&
if (ctx.profile.support_int8 && ctx.profile.support_storage_buffer_8bit &&
ctx.profile.support_descriptor_aliasing) {
return ctx.OpUConvert(ctx.U32[1],
LoadStorage(ctx, binding, offset, ctx.U8, ctx.storage_types.U8,
@ -171,7 +171,7 @@ Id EmitLoadStorageU8(EmitContext& ctx, const IR::Value& binding, const IR::Value
}
Id EmitLoadStorageS8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset) {
if (ctx.profile.support_int8 && ctx.profile.support_uniform_and_storage_buffer_8bit &&
if (ctx.profile.support_int8 && ctx.profile.support_storage_buffer_8bit &&
ctx.profile.support_descriptor_aliasing) {
return ctx.OpSConvert(ctx.U32[1],
LoadStorage(ctx, binding, offset, ctx.S8, ctx.storage_types.S8,
@ -183,7 +183,7 @@ Id EmitLoadStorageS8(EmitContext& ctx, const IR::Value& binding, const IR::Value
}
Id EmitLoadStorageU16(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset) {
if (ctx.profile.support_int16 && ctx.profile.support_uniform_and_storage_buffer_16bit &&
if (ctx.profile.support_int16 && ctx.profile.support_storage_buffer_16bit &&
ctx.profile.support_descriptor_aliasing) {
return ctx.OpUConvert(ctx.U32[1],
LoadStorage(ctx, binding, offset, ctx.U16, ctx.storage_types.U16,
@ -195,7 +195,7 @@ Id EmitLoadStorageU16(EmitContext& ctx, const IR::Value& binding, const IR::Valu
}
Id EmitLoadStorageS16(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset) {
if (ctx.profile.support_int16 && ctx.profile.support_uniform_and_storage_buffer_16bit &&
if (ctx.profile.support_int16 && ctx.profile.support_storage_buffer_16bit &&
ctx.profile.support_descriptor_aliasing) {
return ctx.OpSConvert(ctx.U32[1],
LoadStorage(ctx, binding, offset, ctx.S16, ctx.storage_types.S16,
@ -234,7 +234,8 @@ Id EmitLoadStorage128(EmitContext& ctx, const IR::Value& binding, const IR::Valu
void EmitWriteStorageU8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
Id value) {
if (ctx.profile.support_int8 && ctx.profile.support_uniform_and_storage_buffer_8bit) {
if (ctx.profile.support_int8 && ctx.profile.support_storage_buffer_8bit &&
ctx.profile.support_descriptor_aliasing) {
WriteStorage(ctx, binding, offset, ctx.OpSConvert(ctx.U8, value), ctx.storage_types.U8,
sizeof(u8), &StorageDefinitions::U8);
} else {
@ -244,7 +245,8 @@ void EmitWriteStorageU8(EmitContext& ctx, const IR::Value& binding, const IR::Va
void EmitWriteStorageS8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
Id value) {
if (ctx.profile.support_int8 && ctx.profile.support_uniform_and_storage_buffer_8bit) {
if (ctx.profile.support_int8 && ctx.profile.support_storage_buffer_8bit &&
ctx.profile.support_descriptor_aliasing) {
WriteStorage(ctx, binding, offset, ctx.OpSConvert(ctx.S8, value), ctx.storage_types.S8,
sizeof(s8), &StorageDefinitions::S8);
} else {
@ -254,7 +256,8 @@ void EmitWriteStorageS8(EmitContext& ctx, const IR::Value& binding, const IR::Va
void EmitWriteStorageU16(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
Id value) {
if (ctx.profile.support_int16 && ctx.profile.support_uniform_and_storage_buffer_16bit) {
if (ctx.profile.support_int16 && ctx.profile.support_storage_buffer_16bit &&
ctx.profile.support_descriptor_aliasing) {
WriteStorage(ctx, binding, offset, ctx.OpSConvert(ctx.U16, value), ctx.storage_types.U16,
sizeof(u16), &StorageDefinitions::U16);
} else {
@ -264,7 +267,8 @@ void EmitWriteStorageU16(EmitContext& ctx, const IR::Value& binding, const IR::V
void EmitWriteStorageS16(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
Id value) {
if (ctx.profile.support_int16 && ctx.profile.support_uniform_and_storage_buffer_16bit) {
if (ctx.profile.support_int16 && ctx.profile.support_storage_buffer_16bit &&
ctx.profile.support_descriptor_aliasing) {
WriteStorage(ctx, binding, offset, ctx.OpSConvert(ctx.S16, value), ctx.storage_types.S16,
sizeof(s16), &StorageDefinitions::S16);
} else {

24
src/shader_recompiler/backend/spirv/emit_spirv_shared_memory.cpp

@ -31,7 +31,7 @@ std::pair<Id, Id> ExtractArgs(EmitContext& ctx, Id offset, u32 mask, u32 count)
} // Anonymous namespace
Id EmitLoadSharedU8(EmitContext& ctx, Id offset) {
if (ctx.profile.support_explicit_workgroup_layout) {
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{
ctx.OpAccessChain(ctx.shared_u8, ctx.shared_memory_u8, ctx.u32_zero_value, offset)};
return ctx.OpUConvert(ctx.U32[1], ctx.OpLoad(ctx.U8, pointer));
@ -42,7 +42,7 @@ Id EmitLoadSharedU8(EmitContext& ctx, Id offset) {
}
Id EmitLoadSharedS8(EmitContext& ctx, Id offset) {
if (ctx.profile.support_explicit_workgroup_layout) {
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{
ctx.OpAccessChain(ctx.shared_u8, ctx.shared_memory_u8, ctx.u32_zero_value, offset)};
return ctx.OpSConvert(ctx.U32[1], ctx.OpLoad(ctx.U8, pointer));
@ -53,7 +53,7 @@ Id EmitLoadSharedS8(EmitContext& ctx, Id offset) {
}
Id EmitLoadSharedU16(EmitContext& ctx, Id offset) {
if (ctx.profile.support_explicit_workgroup_layout) {
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u16, ctx.shared_memory_u16, offset, 1)};
return ctx.OpUConvert(ctx.U32[1], ctx.OpLoad(ctx.U16, pointer));
} else {
@ -63,7 +63,7 @@ Id EmitLoadSharedU16(EmitContext& ctx, Id offset) {
}
Id EmitLoadSharedS16(EmitContext& ctx, Id offset) {
if (ctx.profile.support_explicit_workgroup_layout) {
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u16, ctx.shared_memory_u16, offset, 1)};
return ctx.OpSConvert(ctx.U32[1], ctx.OpLoad(ctx.U16, pointer));
} else {
@ -73,7 +73,7 @@ Id EmitLoadSharedS16(EmitContext& ctx, Id offset) {
}
Id EmitLoadSharedU32(EmitContext& ctx, Id offset) {
if (ctx.profile.support_explicit_workgroup_layout) {
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u32, ctx.shared_memory_u32, offset, 2)};
return ctx.OpLoad(ctx.U32[1], pointer);
} else {
@ -82,7 +82,7 @@ Id EmitLoadSharedU32(EmitContext& ctx, Id offset) {
}
Id EmitLoadSharedU64(EmitContext& ctx, Id offset) {
if (ctx.profile.support_explicit_workgroup_layout) {
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u32x2, ctx.shared_memory_u32x2, offset, 3)};
return ctx.OpLoad(ctx.U32[2], pointer);
} else {
@ -97,7 +97,7 @@ Id EmitLoadSharedU64(EmitContext& ctx, Id offset) {
}
Id EmitLoadSharedU128(EmitContext& ctx, Id offset) {
if (ctx.profile.support_explicit_workgroup_layout) {
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u32x4, ctx.shared_memory_u32x4, offset, 4)};
return ctx.OpLoad(ctx.U32[4], pointer);
}
@ -113,7 +113,7 @@ Id EmitLoadSharedU128(EmitContext& ctx, Id offset) {
}
void EmitWriteSharedU8(EmitContext& ctx, Id offset, Id value) {
if (ctx.profile.support_explicit_workgroup_layout) {
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{
ctx.OpAccessChain(ctx.shared_u8, ctx.shared_memory_u8, ctx.u32_zero_value, offset)};
ctx.OpStore(pointer, ctx.OpUConvert(ctx.U8, value));
@ -123,7 +123,7 @@ void EmitWriteSharedU8(EmitContext& ctx, Id offset, Id value) {
}
void EmitWriteSharedU16(EmitContext& ctx, Id offset, Id value) {
if (ctx.profile.support_explicit_workgroup_layout) {
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u16, ctx.shared_memory_u16, offset, 1)};
ctx.OpStore(pointer, ctx.OpUConvert(ctx.U16, value));
} else {
@ -133,7 +133,7 @@ void EmitWriteSharedU16(EmitContext& ctx, Id offset, Id value) {
void EmitWriteSharedU32(EmitContext& ctx, Id offset, Id value) {
Id pointer{};
if (ctx.profile.support_explicit_workgroup_layout) {
if (ctx.uses_explicit_workgroup_layout) {
pointer = Pointer(ctx, ctx.shared_u32, ctx.shared_memory_u32, offset, 2);
} else {
const Id shift{ctx.Const(2U)};
@ -144,7 +144,7 @@ void EmitWriteSharedU32(EmitContext& ctx, Id offset, Id value) {
}
void EmitWriteSharedU64(EmitContext& ctx, Id offset, Id value) {
if (ctx.profile.support_explicit_workgroup_layout) {
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u32x2, ctx.shared_memory_u32x2, offset, 3)};
ctx.OpStore(pointer, value);
return;
@ -159,7 +159,7 @@ void EmitWriteSharedU64(EmitContext& ctx, Id offset, Id value) {
}
void EmitWriteSharedU128(EmitContext& ctx, Id offset, Id value) {
if (ctx.profile.support_explicit_workgroup_layout) {
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u32x4, ctx.shared_memory_u32x4, offset, 4)};
ctx.OpStore(pointer, value);
return;

26
src/shader_recompiler/backend/spirv/spirv_emit_context.cpp

@ -371,7 +371,7 @@ Id CasFunction(EmitContext& ctx, Operation operation, Id value_type) {
Id CasLoop(EmitContext& ctx, Operation operation, Id array_pointer, Id element_pointer,
Id value_type, Id memory_type, spv::Scope scope) {
const bool is_shared{scope == spv::Scope::Workgroup};
const bool is_struct{!is_shared || ctx.profile.support_explicit_workgroup_layout};
const bool is_struct{!is_shared || ctx.uses_explicit_workgroup_layout};
const Id cas_func{CasFunction(ctx, operation, value_type)};
const Id zero{ctx.u32_zero_value};
const Id scope_id{ctx.Const(static_cast<u32>(scope))};
@ -591,7 +591,8 @@ void EmitContext::DefineLocalMemory(const IR::Program& program) {
return;
}
const u32 num_elements{Common::DivCeil(program.local_memory_size, 4U)};
const Id type{TypeArray(U32[1], Const(num_elements))};
const Id element_type{TypeStruct(U32[1])};
const Id type{TypeArray(element_type, Const(num_elements))};
const Id pointer{TypePointer(spv::StorageClass::Private, type)};
local_memory = AddGlobalVariable(pointer, spv::StorageClass::Private);
if (profile.supported_spirv >= 0x00010400) {
@ -600,6 +601,10 @@ void EmitContext::DefineLocalMemory(const IR::Program& program) {
}
void EmitContext::DefineSharedMemory(const IR::Program& program) {
uses_explicit_workgroup_layout =
profile.support_explicit_workgroup_layout &&
(!program.info.uses_int8 || profile.support_workgroup_layout_8bit_access) &&
(!program.info.uses_int16 || profile.support_workgroup_layout_16bit_access);
if (program.shared_memory_size == 0) {
return;
}
@ -620,18 +625,18 @@ void EmitContext::DefineSharedMemory(const IR::Program& program) {
return std::make_tuple(variable, element_pointer, pointer);
}};
if (profile.support_explicit_workgroup_layout) {
if (uses_explicit_workgroup_layout) {
AddExtension("SPV_KHR_workgroup_memory_explicit_layout");
AddCapability(spv::Capability::WorkgroupMemoryExplicitLayoutKHR);
if (program.info.uses_int8) {
if (program.info.uses_int8 && profile.support_int8) {
AddCapability(spv::Capability::WorkgroupMemoryExplicitLayout8BitAccessKHR);
std::tie(shared_memory_u8, shared_u8, std::ignore) = make(U8, 1);
}
if (program.info.uses_int16) {
if (program.info.uses_int16 && profile.support_int16) {
AddCapability(spv::Capability::WorkgroupMemoryExplicitLayout16BitAccessKHR);
std::tie(shared_memory_u16, shared_u16, std::ignore) = make(U16, 2);
}
if (program.info.uses_int64) {
if (program.info.uses_int64 && profile.support_int64) {
std::tie(shared_memory_u64, shared_u64, std::ignore) = make(U64, 8);
}
std::tie(shared_memory_u32, shared_u32, shared_memory_u32_type) = make(U32[1], 4);
@ -1229,16 +1234,17 @@ void EmitContext::DefineStorageBuffers(const Info& info, u32& binding) {
}
AddExtension("SPV_KHR_storage_buffer_storage_class");
const IR::Type used_types{profile.support_descriptor_aliasing ? info.used_storage_buffer_types
: IR::Type::U32};
if (profile.support_int8 && profile.support_uniform_and_storage_buffer_8bit &&
IR::Type used_types{profile.support_descriptor_aliasing ? info.used_storage_buffer_types
: IR::Type::U32};
used_types |= IR::Type::U32;
if (profile.support_int8 && profile.support_storage_buffer_8bit &&
True(used_types & IR::Type::U8)) {
DefineSsbos(*this, storage_types.U8, &StorageDefinitions::U8, info, binding, U8,
sizeof(u8));
DefineSsbos(*this, storage_types.S8, &StorageDefinitions::S8, info, binding, S8,
sizeof(u8));
}
if (profile.support_int16 && profile.support_uniform_and_storage_buffer_16bit &&
if (profile.support_int16 && profile.support_storage_buffer_16bit &&
True(used_types & IR::Type::U16)) {
DefineSsbos(*this, storage_types.U16, &StorageDefinitions::U16, info, binding, U16,
sizeof(u16));

6
src/shader_recompiler/backend/spirv/spirv_emit_context.h

@ -311,6 +311,7 @@ public:
Id local_memory{};
bool uses_explicit_workgroup_layout{};
Id shared_memory_u8{};
Id shared_memory_u16{};
Id shared_memory_u32{};
@ -371,6 +372,11 @@ public:
// Sirit::Id doesn't play nice with *::set<>
ankerl::unordered_dense::set<u32> non_uniform_ids;
bool uses_nonuniform_sampled_image{};
bool uses_nonuniform_storage_image{};
bool uses_nonuniform_uniform_texel_buffer{};
bool uses_nonuniform_storage_texel_buffer{};
private:
void DefineCommonTypes(const Info& info);
void DefineCommonConstants();

3
src/shader_recompiler/ir_opt/lower_fp64_to_fp32.cpp

@ -1,3 +1,6 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later

125
src/shader_recompiler/ir_opt/lower_int64_to_int32.cpp

@ -1,3 +1,6 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
@ -181,6 +184,72 @@ void ShiftRightArithmetic64To32(IR::Block& block, IR::Inst& inst) {
inst.ReplaceUsesWith(ir.CompositeConstruct(ret_lo, ret_hi));
}
void IAbs64To32(IR::Block& block, IR::Inst& inst) {
IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
const auto [lo, hi]{Unpack(ir, inst.Arg(0))};
const IR::U32 neg_lo{ir.IAdd(ir.BitwiseNot(lo), ir.Imm32(1))};
const IR::U32 carry{IR::U32{ir.Select(ir.GetCarryFromOp(neg_lo), ir.Imm32(1u), ir.Imm32(0u))}};
const IR::U32 neg_hi{ir.IAdd(ir.BitwiseNot(hi), carry)};
const IR::U1 is_negative{ir.INotEqual(ir.BitwiseAnd(hi, ir.Imm32(0x80000000u)), ir.Imm32(0u))};
const IR::U32 ret_lo{IR::U32{ir.Select(is_negative, neg_lo, lo)}};
const IR::U32 ret_hi{IR::U32{ir.Select(is_negative, neg_hi, hi)}};
inst.ReplaceUsesWith(ir.CompositeConstruct(ret_lo, ret_hi));
}
void SelectU64To32(IR::Block& block, IR::Inst& inst) {
IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
const IR::U1 condition{inst.Arg(0)};
const auto [true_lo, true_hi]{Unpack(ir, inst.Arg(1))};
const auto [false_lo, false_hi]{Unpack(ir, inst.Arg(2))};
const IR::U32 ret_lo{IR::U32{ir.Select(condition, true_lo, false_lo)}};
const IR::U32 ret_hi{IR::U32{ir.Select(condition, true_hi, false_hi)}};
inst.ReplaceUsesWith(ir.CompositeConstruct(ret_lo, ret_hi));
}
void UndefU64To32(IR::Block& block, IR::Inst& inst) {
IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
inst.ReplaceUsesWith(ir.CompositeConstruct(ir.Imm32(0u), ir.Imm32(0u)));
}
void ConvertU64U32To32(IR::Block& block, IR::Inst& inst) {
IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
inst.ReplaceUsesWith(ir.CompositeConstruct(IR::U32{inst.Arg(0)}, ir.Imm32(0u)));
}
void ConvertU32U64To32(IR::Block& block, IR::Inst& inst) {
IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
inst.ReplaceUsesWith(Unpack(ir, inst.Arg(0)).first);
}
void IntToFloat64To32(IR::Block& block, IR::Inst& inst, bool is_signed, size_t dest_bitsize) {
IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
const auto [lo, hi]{Unpack(ir, inst.Arg(0))};
const IR::F32 low{ir.ConvertUToF(32, 32, lo)};
const IR::F32 high{is_signed ? IR::F32{ir.ConvertSToF(32, 32, hi)}
: IR::F32{ir.ConvertUToF(32, 32, hi)}};
const IR::F32 combined{ir.FPFma(high, ir.Imm32(4294967296.0f), low)};
if (dest_bitsize == 32) {
inst.ReplaceUsesWith(combined);
} else {
inst.ReplaceUsesWith(ir.FPConvert(dest_bitsize, combined));
}
}
void FloatToInt64To32(IR::Block& block, IR::Inst& inst, bool is_signed, size_t src_bitsize) {
IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
const IR::F32 value{src_bitsize == 32 ? IR::F32{inst.Arg(0)}
: IR::F32{ir.FPConvert(32, IR::F16F32F64{inst.Arg(0)})}};
const IR::F32 high_f{ir.FPFloor(ir.FPMul(value, ir.Imm32(1.0f / 4294967296.0f)))};
const IR::U32 hi{is_signed ? IR::U32{ir.ConvertFToS(32, high_f)}
: IR::U32{ir.ConvertFToU(32, high_f)}};
const IR::F32 low_f{ir.FPFma(high_f, ir.FPNeg(ir.Imm32(4294967296.0f)), value)};
const IR::U32 lo{IR::U32{ir.ConvertFToU(32, low_f)}};
inst.ReplaceUsesWith(ir.CompositeConstruct(lo, hi));
}
void Lower(IR::Block& block, IR::Inst& inst) {
switch (inst.GetOpcode()) {
case IR::Opcode::PackUint2x32:
@ -218,6 +287,62 @@ void Lower(IR::Block& block, IR::Inst& inst) {
return inst.ReplaceOpcode(IR::Opcode::GlobalAtomicXor32x2);
case IR::Opcode::GlobalAtomicExchange64:
return inst.ReplaceOpcode(IR::Opcode::GlobalAtomicExchange32x2);
case IR::Opcode::StorageAtomicIAdd64:
return inst.ReplaceOpcode(IR::Opcode::StorageAtomicIAdd32x2);
case IR::Opcode::StorageAtomicSMin64:
return inst.ReplaceOpcode(IR::Opcode::StorageAtomicSMin32x2);
case IR::Opcode::StorageAtomicUMin64:
return inst.ReplaceOpcode(IR::Opcode::StorageAtomicUMin32x2);
case IR::Opcode::StorageAtomicSMax64:
return inst.ReplaceOpcode(IR::Opcode::StorageAtomicSMax32x2);
case IR::Opcode::StorageAtomicUMax64:
return inst.ReplaceOpcode(IR::Opcode::StorageAtomicUMax32x2);
case IR::Opcode::StorageAtomicAnd64:
return inst.ReplaceOpcode(IR::Opcode::StorageAtomicAnd32x2);
case IR::Opcode::StorageAtomicOr64:
return inst.ReplaceOpcode(IR::Opcode::StorageAtomicOr32x2);
case IR::Opcode::StorageAtomicXor64:
return inst.ReplaceOpcode(IR::Opcode::StorageAtomicXor32x2);
case IR::Opcode::StorageAtomicExchange64:
return inst.ReplaceOpcode(IR::Opcode::StorageAtomicExchange32x2);
case IR::Opcode::BitCastU64F64:
return inst.ReplaceOpcode(IR::Opcode::UnpackDouble2x32);
case IR::Opcode::BitCastF64U64:
return inst.ReplaceOpcode(IR::Opcode::PackDouble2x32);
case IR::Opcode::UndefU64:
return UndefU64To32(block, inst);
case IR::Opcode::SelectU64:
return SelectU64To32(block, inst);
case IR::Opcode::IAbs64:
return IAbs64To32(block, inst);
case IR::Opcode::ConvertU64U32:
return ConvertU64U32To32(block, inst);
case IR::Opcode::ConvertU32U64:
return ConvertU32U64To32(block, inst);
case IR::Opcode::ConvertS64F16:
return FloatToInt64To32(block, inst, true, 16);
case IR::Opcode::ConvertS64F32:
return FloatToInt64To32(block, inst, true, 32);
case IR::Opcode::ConvertS64F64:
return FloatToInt64To32(block, inst, true, 64);
case IR::Opcode::ConvertU64F16:
return FloatToInt64To32(block, inst, false, 16);
case IR::Opcode::ConvertU64F32:
return FloatToInt64To32(block, inst, false, 32);
case IR::Opcode::ConvertU64F64:
return FloatToInt64To32(block, inst, false, 64);
case IR::Opcode::ConvertF16S64:
return IntToFloat64To32(block, inst, true, 16);
case IR::Opcode::ConvertF32S64:
return IntToFloat64To32(block, inst, true, 32);
case IR::Opcode::ConvertF64S64:
return IntToFloat64To32(block, inst, true, 64);
case IR::Opcode::ConvertF16U64:
return IntToFloat64To32(block, inst, false, 16);
case IR::Opcode::ConvertF32U64:
return IntToFloat64To32(block, inst, false, 32);
case IR::Opcode::ConvertF64U64:
return IntToFloat64To32(block, inst, false, 64);
default:
break;
}

8
src/shader_recompiler/profile.h

@ -18,8 +18,10 @@ struct Profile {
bool support_descriptor_aliasing{};
bool support_int8{};
bool support_uniform_and_storage_buffer_8bit{};
bool support_storage_buffer_8bit{};
bool support_int16{};
bool support_uniform_and_storage_buffer_16bit{};
bool support_storage_buffer_16bit{};
bool support_int64{};
bool support_vertex_instance_id{};
bool support_float_controls{};
@ -33,6 +35,8 @@ struct Profile {
bool support_fp32_signed_zero_nan_preserve{};
bool support_fp64_signed_zero_nan_preserve{};
bool support_explicit_workgroup_layout{};
bool support_workgroup_layout_8bit_access{};
bool support_workgroup_layout_16bit_access{};
bool support_vote{};
u32 supported_subgroup_stages{0x7F};
bool support_viewport_index_layer_non_geometry{};
@ -40,6 +44,7 @@ struct Profile {
bool support_typeless_image_loads{};
bool support_demote_to_helper_invocation{};
bool support_int64_atomics{};
bool support_shared_int64_atomics{};
bool support_derivative_control{};
bool support_geometry_shader_passthrough{};
bool support_native_ndc{};
@ -54,6 +59,9 @@ struct Profile {
bool support_multi_viewport{};
bool support_geometry_streams{};
bool support_sampled_image_array_nonuniform_indexing{};
bool support_storage_image_array_nonuniform_indexing{};
bool support_uniform_texel_buffer_array_nonuniform_indexing{};
bool support_storage_texel_buffer_array_nonuniform_indexing{};
bool warp_size_potentially_larger_than_guest{};

5
src/video_core/host_shaders/CMakeLists.txt

@ -17,6 +17,9 @@ set(SHADER_FILES
${CMAKE_CURRENT_SOURCE_DIR}/astc_decoder.comp
${CMAKE_CURRENT_SOURCE_DIR}/blit_color_float.frag
${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_2d.comp
${CMAKE_CURRENT_SOURCE_DIR}/blit_color_msaa.frag
${CMAKE_CURRENT_SOURCE_DIR}/blit_depth_msaa.frag
${CMAKE_CURRENT_SOURCE_DIR}/blit_depth_stencil_msaa.frag
${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_3d.comp
${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_3d_bcn.comp
${CMAKE_CURRENT_SOURCE_DIR}/convert_abgr8_to_d24s8.frag
@ -26,7 +29,9 @@ set(SHADER_FILES
${CMAKE_CURRENT_SOURCE_DIR}/convert_depth_to_float.frag
${CMAKE_CURRENT_SOURCE_DIR}/convert_float_to_depth.frag
${CMAKE_CURRENT_SOURCE_DIR}/convert_msaa_to_non_msaa.comp
${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_s8d24_to_abgr8.frag
${CMAKE_CURRENT_SOURCE_DIR}/full_screen_triangle.vert
${CMAKE_CURRENT_SOURCE_DIR}/fxaa.frag

13
src/video_core/host_shaders/blit_color_msaa.frag

@ -0,0 +1,13 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#version 450 core
layout(binding = 0) uniform sampler2DMS tex;
layout(location = 0) in vec2 texcoord;
layout(location = 0) out vec4 color;
void main() {
color = texelFetch(tex, ivec2(texcoord), gl_SampleID);
}

12
src/video_core/host_shaders/blit_depth_msaa.frag

@ -0,0 +1,12 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#version 450 core
layout(binding = 0) uniform sampler2DMS depth_tex;
layout(location = 0) in vec2 texcoord;
void main() {
gl_FragDepth = texelFetch(depth_tex, ivec2(texcoord), 0).r;
}

15
src/video_core/host_shaders/blit_depth_stencil_msaa.frag

@ -0,0 +1,15 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#version 450 core
#extension GL_ARB_shader_stencil_export : require
layout(binding = 0) uniform sampler2DMS depth_tex;
layout(binding = 1) uniform usampler2DMS stencil_tex;
layout(location = 0) in vec2 texcoord;
void main() {
gl_FragDepth = texelFetch(depth_tex, ivec2(texcoord), 0).r;
gl_FragStencilRefARB = int(texelFetch(stencil_tex, ivec2(texcoord), 0).r);
}

22
src/video_core/host_shaders/convert_msaa_to_non_msaa.frag

@ -0,0 +1,22 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#version 450 core
layout(binding = 0) uniform sampler2DMS msaa_in;
layout(push_constant) uniform PushConstants {
ivec2 dst_offset;
ivec2 src_offset;
ivec2 scale;
};
layout(location = 0) out vec4 frag_color;
void main() {
const ivec2 coord = ivec2(gl_FragCoord.xy) - dst_offset + src_offset;
const ivec2 msaa_coord = coord / scale;
const ivec2 sample_offset = coord % scale;
const int sample_id = sample_offset.x + scale.x * sample_offset.y;
frag_color = texelFetch(msaa_in, msaa_coord, sample_id);
}

21
src/video_core/host_shaders/convert_non_msaa_to_msaa.frag

@ -0,0 +1,21 @@
// 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;
};
layout(location = 0) out vec4 frag_color;
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;
frag_color = texelFetch(img_in, coord, 0);
}

430
src/video_core/renderer_vulkan/blit_image.cpp

@ -8,14 +8,20 @@
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "common/div_ceil.h"
#include "common/settings.h"
#include "video_core/host_shaders/blit_color_float_frag_spv.h"
#include "video_core/host_shaders/blit_color_msaa_frag_spv.h"
#include "video_core/host_shaders/blit_depth_msaa_frag_spv.h"
#include "video_core/host_shaders/blit_depth_stencil_msaa_frag_spv.h"
#include "video_core/host_shaders/convert_abgr8_to_d24s8_frag_spv.h"
#include "video_core/host_shaders/convert_abgr8_to_d32f_frag_spv.h"
#include "video_core/host_shaders/convert_d24s8_to_abgr8_frag_spv.h"
#include "video_core/host_shaders/convert_d32f_to_abgr8_frag_spv.h"
#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_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"
#include "video_core/host_shaders/vulkan_blit_depth_stencil_frag_spv.h"
@ -24,11 +30,13 @@
#include "video_core/host_shaders/vulkan_depthstencil_clear_frag_spv.h"
#include "video_core/renderer_vulkan/blit_image.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_render_pass_cache.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_shader_util.h"
#include "video_core/renderer_vulkan/vk_state_tracker.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/surface.h"
#include "video_core/texture_cache/samples_helper.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@ -74,6 +82,12 @@ struct PushConstants {
std::array<float, 2> tex_offset;
};
struct MSAACopyPushConstants {
std::array<s32, 2> dst_offset;
std::array<s32, 2> src_offset;
std::array<s32, 2> scale;
};
template <u32 binding>
inline constexpr VkDescriptorSetLayoutBinding TEXTURE_DESCRIPTOR_SET_LAYOUT_BINDING{
.binding = binding,
@ -241,6 +255,21 @@ constexpr VkPipelineDepthStencilStateCreateInfo PIPELINE_DEPTH_STENCIL_STATE_CRE
.maxDepthBounds = 0.0f,
};
constexpr VkPipelineDepthStencilStateCreateInfo PIPELINE_DEPTH_ONLY_STATE_CREATE_INFO{
.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 = VkStencilOpState{},
.back = VkStencilOpState{},
.minDepthBounds = 0.0f,
.maxDepthBounds = 0.0f,
};
template <VkFilter filter>
inline constexpr VkSamplerCreateInfo SAMPLER_CREATE_INFO{
.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
@ -474,6 +503,46 @@ void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view)
});
}
[[nodiscard]] VkSampleCountFlagBits SampleCountFlag(u32 num_samples) {
switch (num_samples) {
case 2:
return VK_SAMPLE_COUNT_2_BIT;
case 4:
return VK_SAMPLE_COUNT_4_BIT;
case 8:
return VK_SAMPLE_COUNT_8_BIT;
case 16:
return VK_SAMPLE_COUNT_16_BIT;
default:
return VK_SAMPLE_COUNT_1_BIT;
}
}
[[nodiscard]] vk::ImageView MakeMSAACopyView(const vk::Device& device, VkImage image,
VkFormat format, u32 base_level) {
return device.CreateImageView(VkImageViewCreateInfo{
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.image = image,
.viewType = VK_IMAGE_VIEW_TYPE_2D,
.format = format,
.components{
.r = VK_COMPONENT_SWIZZLE_IDENTITY,
.g = VK_COMPONENT_SWIZZLE_IDENTITY,
.b = VK_COMPONENT_SWIZZLE_IDENTITY,
.a = VK_COMPONENT_SWIZZLE_IDENTITY,
},
.subresourceRange{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.baseMipLevel = base_level,
.levelCount = 1,
.baseArrayLayer = 0,
.layerCount = 1,
},
});
}
void BeginRenderPass(vk::CommandBuffer& cmdbuf, const Framebuffer* framebuffer) {
const VkRenderPass render_pass = framebuffer->RenderPass();
const VkFramebuffer framebuffer_handle = framebuffer->Handle();
@ -514,11 +583,19 @@ BlitImageHelper::BlitImageHelper(const Device& device_, Scheduler& scheduler_,
PUSH_CONSTANT_RANGE<VK_SHADER_STAGE_VERTEX_BIT, sizeof(PushConstants)>))),
clear_color_pipeline_layout(device.GetLogical().CreatePipelineLayout(PipelineLayoutCreateInfo(
nullptr, PUSH_CONSTANT_RANGE<VK_SHADER_STAGE_FRAGMENT_BIT, sizeof(float) * 4>))),
msaa_copy_pipeline_layout(device.GetLogical().CreatePipelineLayout(PipelineLayoutCreateInfo(
one_texture_set_layout.address(),
PUSH_CONSTANT_RANGE<VK_SHADER_STAGE_FRAGMENT_BIT, sizeof(MSAACopyPushConstants)>))),
full_screen_vert(BuildShader(device, FULL_SCREEN_TRIANGLE_VERT_SPV)),
blit_color_to_color_frag(BuildShader(device, BLIT_COLOR_FLOAT_FRAG_SPV)),
blit_color_msaa_frag(BuildShader(device, BLIT_COLOR_MSAA_FRAG_SPV)),
blit_depth_stencil_frag(device.IsExtShaderStencilExportSupported()
? BuildShader(device, VULKAN_BLIT_DEPTH_STENCIL_FRAG_SPV)
: vk::ShaderModule{}),
blit_depth_msaa_frag(BuildShader(device, BLIT_DEPTH_MSAA_FRAG_SPV)),
blit_depth_stencil_msaa_frag(device.IsExtShaderStencilExportSupported()
? BuildShader(device, BLIT_DEPTH_STENCIL_MSAA_FRAG_SPV)
: vk::ShaderModule{}),
clear_color_vert(BuildShader(device, VULKAN_COLOR_CLEAR_VERT_SPV)),
clear_color_frag(BuildShader(device, VULKAN_COLOR_CLEAR_FRAG_SPV)),
clear_stencil_frag(BuildShader(device, VULKAN_DEPTHSTENCIL_CLEAR_FRAG_SPV)),
@ -531,6 +608,8 @@ BlitImageHelper::BlitImageHelper(const Device& device_, Scheduler& scheduler_,
convert_d32f_to_abgr8_frag(BuildShader(device, CONVERT_D32F_TO_ABGR8_FRAG_SPV)),
convert_d24s8_to_abgr8_frag(BuildShader(device, CONVERT_D24S8_TO_ABGR8_FRAG_SPV)),
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)),
linear_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_LINEAR>)),
nearest_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_NEAREST>)) {}
@ -591,6 +670,70 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, VkImageView
});
}
void BlitImageHelper::BlitColorMSAA(const Framebuffer* dst_framebuffer,
const ImageView& src_image_view, const Region2D& dst_region,
const Region2D& src_region) {
const BlitMSAAPipelineKey key{
.renderpass = dst_framebuffer->RenderPass(),
.samples = dst_framebuffer->Samples(),
};
const VkPipelineLayout layout = *one_texture_pipeline_layout;
const VkSampler sampler = *nearest_sampler;
const VkPipeline pipeline = FindOrEmplaceBlitColorMSAAPipeline(key);
const VkImageView src_view = src_image_view.Handle(Shader::TextureType::Color2D);
RecordShaderReadBarrier(scheduler, src_image_view);
scheduler.RequestRenderpass(dst_framebuffer);
scheduler.Record([this, dst_region, src_region, pipeline, layout, sampler,
src_view](vk::CommandBuffer cmdbuf) {
const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit();
UpdateOneTextureDescriptorSet(device, descriptor_set, sampler, src_view);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set,
nullptr);
BindBlitState(cmdbuf, layout, dst_region, src_region);
cmdbuf.Draw(3, 1, 0, 0);
});
scheduler.InvalidateState();
}
void BlitImageHelper::ResolveDepthStencil(const Framebuffer* dst_framebuffer,
ImageView& src_image_view, const Region2D& dst_region,
const Region2D& src_region) {
const bool resolve_stencil =
dst_framebuffer->HasAspectStencilBit() && device.IsExtShaderStencilExportSupported();
const VkPipeline pipeline =
FindOrEmplaceResolveDepthStencilPipeline(dst_framebuffer->RenderPass(), resolve_stencil);
const VkPipelineLayout layout =
resolve_stencil ? *two_textures_pipeline_layout : *one_texture_pipeline_layout;
const VkSampler sampler = *nearest_sampler;
const VkImageView src_depth_view = src_image_view.DepthView();
const VkImageView src_stencil_view =
resolve_stencil ? src_image_view.StencilView() : VK_NULL_HANDLE;
RecordShaderReadBarrier(scheduler, src_image_view);
scheduler.RequestRenderpass(dst_framebuffer);
scheduler.Record([this, dst_region, src_region, pipeline, layout, sampler, src_depth_view,
src_stencil_view, resolve_stencil](vk::CommandBuffer cmdbuf) {
if (resolve_stencil) {
const VkDescriptorSet descriptor_set = two_textures_descriptor_allocator.Commit();
UpdateTwoTexturesDescriptorSet(device, descriptor_set, sampler, src_depth_view,
src_stencil_view);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set,
nullptr);
} else {
const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit();
UpdateOneTextureDescriptorSet(device, descriptor_set, sampler, src_depth_view);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set,
nullptr);
}
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
BindBlitState(cmdbuf, layout, dst_region, src_region);
cmdbuf.Draw(3, 1, 0, 0);
});
scheduler.InvalidateState();
}
void BlitImageHelper::BlitDepthStencil(const Framebuffer* dst_framebuffer,
ImageView& src_image_view,
const Region2D& dst_region, const Region2D& src_region,
@ -739,6 +882,174 @@ void BlitImageHelper::ClearDepthStencil(const Framebuffer* dst_framebuffer, bool
scheduler.InvalidateState();
}
void BlitImageHelper::CopyMSAA(RenderPassCache& render_pass_cache, VkImage dst_image,
VideoCore::Surface::PixelFormat dst_format, VkImage src_image,
VideoCore::Surface::PixelFormat src_format, u32 num_samples,
std::span<const VideoCommon::ImageCopy> copies,
bool msaa_to_non_msaa) {
while (!msaa_copy_resources.empty() && scheduler.IsFree(msaa_copy_resources.front().tick)) {
msaa_copy_resources.pop_front();
}
const auto [samples_x, samples_y] = VideoCommon::SamplesLog2(static_cast<int>(num_samples));
const s32 scale_x = 1 << samples_x;
const s32 scale_y = 1 << samples_y;
const VkSampleCountFlagBits samples =
msaa_to_non_msaa ? VK_SAMPLE_COUNT_1_BIT : SampleCountFlag(num_samples);
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;
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,
};
const VkPipeline pipeline = FindOrEmplaceMSAACopyPipeline(key);
const VkPipelineLayout layout = *msaa_copy_pipeline_layout;
const VkSampler sampler = *nearest_sampler;
const VkFormat src_vk_format =
MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, true, src_format).format;
const VkFormat dst_vk_format =
MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, true, dst_format).format;
for (const VideoCommon::ImageCopy& copy : copies) {
ASSERT(copy.src_subresource.base_layer == 0);
ASSERT(copy.src_subresource.num_layers == 1);
ASSERT(copy.dst_subresource.base_layer == 0);
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));
vk::ImageView dst_view =
MakeMSAACopyView(device.GetLogical(), dst_image, dst_vk_format,
static_cast<u32>(copy.dst_subresource.base_level));
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{
.offset = dst_offset,
.extent = dst_extent,
};
vk::Framebuffer framebuffer = device.GetLogical().CreateFramebuffer(VkFramebufferCreateInfo{
.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.renderPass = renderpass,
.attachmentCount = 1,
.pAttachments = dst_view.address(),
.width = static_cast<u32>(dst_offset.x) + dst_extent.width,
.height = static_cast<u32>(dst_offset.y) + dst_extent.height,
.layers = 1,
});
const MSAACopyPushConstants push_constants{
.dst_offset = {dst_offset.x, dst_offset.y},
.src_offset = {copy.src_offset.x, copy.src_offset.y},
.scale = {scale_x, scale_y},
};
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,
push_constants](vk::CommandBuffer cmdbuf) {
constexpr VkImageSubresourceRange color_range{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.baseMipLevel = 0,
.levelCount = VK_REMAINING_MIP_LEVELS,
.baseArrayLayer = 0,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
};
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,
.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,
},
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,
.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,
},
};
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
0, nullptr, nullptr, pre_barriers);
const VkRenderPassBeginInfo renderpass_bi{
.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
.pNext = nullptr,
.renderPass = renderpass,
.framebuffer = framebuffer_handle,
.renderArea = render_area,
.clearValueCount = 0,
.pClearValues = nullptr,
};
cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE);
const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit();
UpdateOneTextureDescriptorSet(device, descriptor_set, sampler, src_view_handle);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set,
nullptr);
const VkViewport viewport{
.x = static_cast<float>(render_area.offset.x),
.y = static_cast<float>(render_area.offset.y),
.width = static_cast<float>(render_area.extent.width),
.height = static_cast<float>(render_area.extent.height),
.minDepth = 0.0f,
.maxDepth = 1.0f,
};
cmdbuf.SetViewport(0, viewport);
cmdbuf.SetScissor(0, render_area);
cmdbuf.PushConstants(layout, VK_SHADER_STAGE_FRAGMENT_BIT, push_constants);
cmdbuf.Draw(3, 1, 0, 0);
cmdbuf.EndRenderPass();
const VkImageMemoryBarrier post_barrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
.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,
};
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
VK_PIPELINE_STAGE_TRANSFER_BIT,
0, post_barrier);
});
msaa_copy_resources.push_back(MSAACopyResources{
.tick = scheduler.CurrentTick(),
.src_view = std::move(src_view),
.dst_view = std::move(dst_view),
.framebuffer = std::move(framebuffer),
});
}
scheduler.InvalidateState();
}
void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
const ImageView& src_image_view) {
const VkPipelineLayout layout = *one_texture_pipeline_layout;
@ -905,7 +1216,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceDepthStencilPipeline(const BlitImagePip
.pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
.pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
.pDepthStencilState = &PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
.pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO,
.pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_EMPTY_CREATE_INFO,
.pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
.layout = *two_textures_pipeline_layout,
.renderPass = key.renderpass,
@ -1025,14 +1336,63 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearStencilPipeline(
return *clear_stencil_pipelines.back();
}
void BlitImageHelper::ConvertDepthToColorPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass) {
if (pipeline) {
return;
VkPipeline BlitImageHelper::FindOrEmplaceBlitColorMSAAPipeline(const BlitMSAAPipelineKey& key) {
const auto it = std::ranges::find(blit_msaa_color_keys, key);
if (it != blit_msaa_color_keys.end()) {
return *blit_msaa_color_pipelines[std::distance(blit_msaa_color_keys.begin(), it)];
}
VkShaderModule frag_shader = *convert_float_to_depth_frag;
const std::array stages = MakeStages(*full_screen_vert, frag_shader);
blit_msaa_color_keys.push_back(key);
const std::array stages = MakeStages(*full_screen_vert, *blit_color_msaa_frag);
const VkPipelineMultisampleStateCreateInfo multisample_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.rasterizationSamples = key.samples,
.sampleShadingEnable = VK_TRUE,
.minSampleShading = 1.0f,
.pSampleMask = nullptr,
.alphaToCoverageEnable = VK_FALSE,
.alphaToOneEnable = VK_FALSE,
};
const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci = GetPipelineInputAssemblyStateCreateInfo(device);
pipeline = device.GetLogical().CreateGraphicsPipeline(VkGraphicsPipelineCreateInfo{
blit_msaa_color_pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({
.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stageCount = static_cast<u32>(stages.size()),
.pStages = stages.data(),
.pVertexInputState = &PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
.pInputAssemblyState = &input_assembly_ci,
.pTessellationState = nullptr,
.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,
.pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
.layout = *one_texture_pipeline_layout,
.renderPass = key.renderpass,
.subpass = 0,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0,
}));
return *blit_msaa_color_pipelines.back();
}
VkPipeline BlitImageHelper::FindOrEmplaceResolveDepthStencilPipeline(VkRenderPass renderpass,
bool resolve_stencil) {
auto& keys = resolve_stencil ? resolve_depth_stencil_keys : resolve_depth_keys;
auto& pipelines = resolve_stencil ? resolve_depth_stencil_pipelines : resolve_depth_pipelines;
const auto it = std::ranges::find(keys, renderpass);
if (it != keys.end()) {
return *pipelines[std::distance(keys.begin(), it)];
}
keys.push_back(renderpass);
const std::array stages =
MakeStages(*full_screen_vert,
resolve_stencil ? *blit_depth_stencil_msaa_frag : *blit_depth_msaa_frag);
const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci = GetPipelineInputAssemblyStateCreateInfo(device);
pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({
.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
@ -1044,25 +1404,41 @@ void BlitImageHelper::ConvertDepthToColorPipeline(vk::Pipeline& pipeline, VkRend
.pViewportState = &PIPELINE_VIEWPORT_STATE_CREATE_INFO,
.pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
.pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
.pDepthStencilState = &PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
.pDepthStencilState = resolve_stencil ? &PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO
: &PIPELINE_DEPTH_ONLY_STATE_CREATE_INFO,
.pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_EMPTY_CREATE_INFO,
.pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
.layout = *one_texture_pipeline_layout,
.layout = resolve_stencil ? *two_textures_pipeline_layout : *one_texture_pipeline_layout,
.renderPass = renderpass,
.subpass = 0,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0,
});
}));
return *pipelines.back();
}
void BlitImageHelper::ConvertColorToDepthPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass) {
if (pipeline) {
return;
VkPipeline BlitImageHelper::FindOrEmplaceMSAACopyPipeline(const MSAACopyPipelineKey& key) {
const auto it = std::ranges::find(msaa_copy_keys, key);
if (it != msaa_copy_keys.end()) {
return *msaa_copy_pipelines[std::distance(msaa_copy_keys.begin(), it)];
}
VkShaderModule frag_shader = *convert_depth_to_float_frag;
const std::array stages = MakeStages(*full_screen_vert, frag_shader);
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 VkPipelineMultisampleStateCreateInfo multisample_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.rasterizationSamples = key.samples,
.sampleShadingEnable = key.msaa_to_non_msaa ? VK_FALSE : VK_TRUE,
.minSampleShading = key.msaa_to_non_msaa ? 0.0f : 1.0f,
.pSampleMask = nullptr,
.alphaToCoverageEnable = VK_FALSE,
.alphaToOneEnable = VK_FALSE,
};
const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci = GetPipelineInputAssemblyStateCreateInfo(device);
pipeline = device.GetLogical().CreateGraphicsPipeline(VkGraphicsPipelineCreateInfo{
msaa_copy_pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({
.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
@ -1073,16 +1449,25 @@ void BlitImageHelper::ConvertColorToDepthPipeline(vk::Pipeline& pipeline, VkRend
.pTessellationState = nullptr,
.pViewportState = &PIPELINE_VIEWPORT_STATE_CREATE_INFO,
.pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
.pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
.pDepthStencilState = &PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
.pMultisampleState = &multisample_ci,
.pDepthStencilState = nullptr,
.pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO,
.pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
.layout = *one_texture_pipeline_layout,
.renderPass = renderpass,
.layout = *msaa_copy_pipeline_layout,
.renderPass = key.renderpass,
.subpass = 0,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0,
});
}));
return *msaa_copy_pipelines.back();
}
void BlitImageHelper::ConvertDepthToColorPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass) {
ConvertPipeline(pipeline, renderpass, false);
}
void BlitImageHelper::ConvertColorToDepthPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass) {
ConvertPipeline(pipeline, renderpass, true);
}
void BlitImageHelper::ConvertPipelineEx(vk::Pipeline& pipeline, VkRenderPass renderpass,
@ -1106,7 +1491,8 @@ void BlitImageHelper::ConvertPipelineEx(vk::Pipeline& pipeline, VkRenderPass ren
.pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
.pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
.pDepthStencilState = is_target_depth ? &PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO : nullptr,
.pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO,
.pColorBlendState = is_target_depth ? &PIPELINE_COLOR_BLEND_STATE_EMPTY_CREATE_INFO
: &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO,
.pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
.layout = single_texture ? *one_texture_pipeline_layout : *two_textures_pipeline_layout,
.renderPass = renderpass,

57
src/video_core/renderer_vulkan/blit_image.h

@ -6,8 +6,12 @@
#pragma once
#include <deque>
#include <span>
#include "video_core/engines/fermi_2d.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/surface.h"
#include "video_core/texture_cache/types.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@ -20,6 +24,7 @@ using VideoCommon::Region2D;
class Device;
class Framebuffer;
class ImageView;
class RenderPassCache;
class StateTracker;
class Scheduler;
@ -40,6 +45,21 @@ struct BlitDepthStencilPipelineKey {
u32 stencil_ref;
};
struct MSAACopyPipelineKey {
constexpr auto operator<=>(const MSAACopyPipelineKey&) const noexcept = default;
VkRenderPass renderpass;
VkSampleCountFlagBits samples;
bool msaa_to_non_msaa;
};
struct BlitMSAAPipelineKey {
constexpr auto operator<=>(const BlitMSAAPipelineKey&) const noexcept = default;
VkRenderPass renderpass;
VkSampleCountFlagBits samples;
};
class BlitImageHelper {
public:
explicit BlitImageHelper(const Device& device, Scheduler& scheduler,
@ -55,6 +75,12 @@ public:
VkImage src_image, VkSampler src_sampler, const Region2D& dst_region,
const Region2D& src_region, const Extent3D& src_size);
void BlitColorMSAA(const Framebuffer* dst_framebuffer, const ImageView& src_image_view,
const Region2D& dst_region, const Region2D& src_region);
void ResolveDepthStencil(const Framebuffer* dst_framebuffer, ImageView& src_image_view,
const Region2D& dst_region, const Region2D& src_region);
void BlitDepthStencil(const Framebuffer* dst_framebuffer, ImageView& src_image_view,
const Region2D& dst_region, const Region2D& src_region,
Tegra::Engines::Fermi2D::Filter filter,
@ -84,6 +110,12 @@ public:
void ClearDepthStencil(const Framebuffer* dst_framebuffer, bool depth_clear, f32 clear_depth,
u8 stencil_mask, u32 stencil_ref, u32 stencil_compare_mask,
const Region2D& dst_region);
void CopyMSAA(RenderPassCache& render_pass_cache, VkImage dst_image,
VideoCore::Surface::PixelFormat dst_format, VkImage src_image,
VideoCore::Surface::PixelFormat src_format, u32 num_samples,
std::span<const VideoCommon::ImageCopy> copies, bool msaa_to_non_msaa);
private:
void Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
const ImageView& src_image_view);
@ -98,6 +130,10 @@ private:
[[nodiscard]] VkPipeline FindOrEmplaceClearColorPipeline(const BlitImagePipelineKey& key);
[[nodiscard]] VkPipeline FindOrEmplaceClearStencilPipeline(
const BlitDepthStencilPipelineKey& key);
[[nodiscard]] VkPipeline FindOrEmplaceMSAACopyPipeline(const MSAACopyPipelineKey& key);
[[nodiscard]] VkPipeline FindOrEmplaceBlitColorMSAAPipeline(const BlitMSAAPipelineKey& key);
[[nodiscard]] VkPipeline FindOrEmplaceResolveDepthStencilPipeline(VkRenderPass renderpass,
bool resolve_stencil);
void ConvertPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass, bool is_target_depth);
@ -125,9 +161,13 @@ private:
vk::PipelineLayout one_texture_pipeline_layout;
vk::PipelineLayout two_textures_pipeline_layout;
vk::PipelineLayout clear_color_pipeline_layout;
vk::PipelineLayout msaa_copy_pipeline_layout;
vk::ShaderModule full_screen_vert;
vk::ShaderModule blit_color_to_color_frag;
vk::ShaderModule blit_color_msaa_frag;
vk::ShaderModule blit_depth_stencil_frag;
vk::ShaderModule blit_depth_msaa_frag;
vk::ShaderModule blit_depth_stencil_msaa_frag;
vk::ShaderModule clear_color_vert;
vk::ShaderModule clear_color_frag;
vk::ShaderModule clear_stencil_frag;
@ -138,6 +178,8 @@ private:
vk::ShaderModule convert_d32f_to_abgr8_frag;
vk::ShaderModule convert_d24s8_to_abgr8_frag;
vk::ShaderModule convert_s8d24_to_abgr8_frag;
vk::ShaderModule convert_msaa_to_non_msaa_frag;
vk::ShaderModule convert_non_msaa_to_msaa_frag;
vk::Sampler linear_sampler;
vk::Sampler nearest_sampler;
@ -149,6 +191,21 @@ private:
std::vector<vk::Pipeline> clear_color_pipelines;
std::vector<BlitDepthStencilPipelineKey> clear_stencil_keys;
std::vector<vk::Pipeline> clear_stencil_pipelines;
std::vector<MSAACopyPipelineKey> msaa_copy_keys;
std::vector<vk::Pipeline> msaa_copy_pipelines;
std::vector<BlitMSAAPipelineKey> blit_msaa_color_keys;
std::vector<vk::Pipeline> blit_msaa_color_pipelines;
std::vector<VkRenderPass> resolve_depth_keys;
std::vector<vk::Pipeline> resolve_depth_pipelines;
std::vector<VkRenderPass> resolve_depth_stencil_keys;
std::vector<vk::Pipeline> resolve_depth_stencil_pipelines;
struct MSAACopyResources {
u64 tick;
vk::ImageView src_view;
vk::ImageView dst_view;
vk::Framebuffer framebuffer;
};
std::deque<MSAACopyResources> msaa_copy_resources;
vk::Pipeline convert_d32_to_r32_pipeline;
vk::Pipeline convert_r32_to_d32_pipeline;
vk::Pipeline convert_d16_to_r16_pipeline;

134
src/video_core/renderer_vulkan/vk_compute_pass.cpp

@ -17,8 +17,6 @@
#include "common/div_ceil.h"
#include "common/vector_math.h"
#include "video_core/host_shaders/astc_decoder_comp_spv.h"
#include "video_core/host_shaders/convert_msaa_to_non_msaa_comp_spv.h"
#include "video_core/host_shaders/convert_non_msaa_to_msaa_comp_spv.h"
#include "video_core/host_shaders/queries_prefix_scan_sum_comp_spv.h"
#include "video_core/host_shaders/queries_prefix_scan_sum_nosubgroups_comp_spv.h"
#include "video_core/host_shaders/resolve_conditional_render_comp_spv.h"
@ -140,33 +138,6 @@ constexpr DescriptorBankInfo ASTC_BANK_INFO{
.score = 2,
};
constexpr std::array<VkDescriptorSetLayoutBinding, ASTC_NUM_BINDINGS> MSAA_DESCRIPTOR_SET_BINDINGS{{
{
.binding = 0,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
}};
constexpr DescriptorBankInfo MSAA_BANK_INFO{
.uniform_buffers = 0,
.storage_buffers = 0,
.texture_buffers = 0,
.image_buffers = 0,
.textures = 0,
.images = 2,
.score = 2,
};
constexpr VkDescriptorUpdateTemplateEntry INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE{
.dstBinding = 0,
.dstArrayElement = 0,
@ -185,15 +156,6 @@ constexpr VkDescriptorUpdateTemplateEntry QUERIES_SCAN_DESCRIPTOR_UPDATE_TEMPLAT
.stride = sizeof(DescriptorUpdateEntry),
};
constexpr VkDescriptorUpdateTemplateEntry MSAA_DESCRIPTOR_UPDATE_TEMPLATE{
.dstBinding = 0,
.dstArrayElement = 0,
.descriptorCount = 2,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.offset = 0,
.stride = sizeof(DescriptorUpdateEntry),
};
constexpr std::array<VkDescriptorUpdateTemplateEntry, ASTC_NUM_BINDINGS>
ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY{{
{
@ -910,100 +872,4 @@ void BlockLinearUnswizzle3DPass::UnswizzleChunk(
});
}
MSAACopyPass::MSAACopyPass(const Device& device_, Scheduler& scheduler_,
DescriptorPool& descriptor_pool_,
StagingBufferPool& staging_buffer_pool_,
ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
: ComputePass(device_, scheduler_, descriptor_pool_, MSAA_DESCRIPTOR_SET_BINDINGS,
MSAA_DESCRIPTOR_UPDATE_TEMPLATE, MSAA_BANK_INFO, {},
CONVERT_NON_MSAA_TO_MSAA_COMP_SPV),
scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {
const auto make_msaa_pipeline = [this](size_t i, std::span<const u32> code) {
modules[i] = device.GetLogical().CreateShaderModule({
.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.codeSize = static_cast<u32>(code.size_bytes()),
.pCode = code.data(),
});
pipelines[i] = device.GetLogical().CreateComputePipeline(VkComputePipelineCreateInfo{
.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage{
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage = VK_SHADER_STAGE_COMPUTE_BIT,
.module = *modules[i],
.pName = "main",
.pSpecializationInfo = nullptr,
},
.layout = *layout,
.basePipelineHandle = {},
.basePipelineIndex = 0,
});
};
make_msaa_pipeline(0, CONVERT_NON_MSAA_TO_MSAA_COMP_SPV);
make_msaa_pipeline(1, CONVERT_MSAA_TO_NON_MSAA_COMP_SPV);
}
MSAACopyPass::~MSAACopyPass() = default;
void MSAACopyPass::CopyImage(Image& dst_image, Image& src_image,
std::span<const VideoCommon::ImageCopy> copies,
bool msaa_to_non_msaa) {
const VkPipeline msaa_pipeline = *pipelines[msaa_to_non_msaa ? 1 : 0];
scheduler.RequestOutsideRenderPassOperationContext();
for (const VideoCommon::ImageCopy& copy : copies) {
ASSERT(copy.src_subresource.base_layer == 0);
ASSERT(copy.src_subresource.num_layers == 1);
ASSERT(copy.dst_subresource.base_layer == 0);
ASSERT(copy.dst_subresource.num_layers == 1);
compute_pass_descriptor_queue.Acquire(scheduler, 2);
compute_pass_descriptor_queue.AddImage(
src_image.StorageImageView(copy.src_subresource.base_level));
compute_pass_descriptor_queue.AddImage(
dst_image.StorageImageView(copy.dst_subresource.base_level));
const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
const Common::Vec3<u32> num_dispatches = {
Common::DivCeil(copy.extent.width, 8U),
Common::DivCeil(copy.extent.height, 8U),
copy.extent.depth,
};
scheduler.Record([this, dst = dst_image.Handle(), msaa_pipeline, num_dispatches,
descriptor_data](vk::CommandBuffer cmdbuf) {
const VkDescriptorSet set = descriptor_allocator.Commit();
device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, msaa_pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
cmdbuf.Dispatch(num_dispatches.x, num_dispatches.y, num_dispatches.z);
const VkImageMemoryBarrier write_barrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_SHADER_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 = dst,
.subresourceRange{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.baseMipLevel = 0,
.levelCount = VK_REMAINING_MIP_LEVELS,
.baseArrayLayer = 0,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
};
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier);
});
}
}
} // namespace Vulkan

19
src/video_core/renderer_vulkan/vk_compute_pass.h

@ -164,23 +164,4 @@ private:
ComputePassDescriptorQueue& compute_pass_descriptor_queue;
};
class MSAACopyPass final : public ComputePass {
public:
explicit MSAACopyPass(const Device& device_, Scheduler& scheduler_,
DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_,
ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
~MSAACopyPass();
void CopyImage(Image& dst_image, Image& src_image,
std::span<const VideoCommon::ImageCopy> copies, bool msaa_to_non_msaa);
private:
Scheduler& scheduler;
StagingBufferPool& staging_buffer_pool;
ComputePassDescriptorQueue& compute_pass_descriptor_queue;
std::array<vk::ShaderModule, 2> modules;
std::array<vk::Pipeline, 2> pipelines;
};
} // namespace Vulkan

4
src/video_core/renderer_vulkan/vk_compute_pipeline.h

@ -48,6 +48,10 @@ public:
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);
}
private:
const Device& device;
vk::PipelineCache& pipeline_cache;

30
src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@ -124,8 +124,8 @@ PixelFormat DecodeFormat(u8 encoded_format) {
return PixelFormatFromRenderTargetFormat(format);
}
RenderPassKey MakeRenderPassKey(const FixedPipelineState& state) {
RenderPassKey key;
RenderPassKey MakeRenderPassKey(const FixedPipelineState& state, const Device& device) {
RenderPassKey key{};
std::ranges::transform(state.color_formats, key.color_formats.begin(), DecodeFormat);
if (state.depth_enabled != 0) {
const auto depth_format{static_cast<Tegra::DepthFormat>(state.depth_format.Value())};
@ -134,6 +134,11 @@ RenderPassKey MakeRenderPassKey(const FixedPipelineState& state) {
key.depth_format = PixelFormat::Invalid;
}
key.samples = MaxwellToVK::MsaaMode(state.msaa_mode);
const bool has_color = std::ranges::any_of(key.color_formats, [](PixelFormat format) {
return format != PixelFormat::Invalid;
});
key.resolve_color =
key.samples != VK_SAMPLE_COUNT_1_BIT && has_color && device.IsTiler();
return key;
}
@ -285,9 +290,20 @@ GraphicsPipeline::GraphicsPipeline(
descriptor_update_template =
builder.CreateTemplate(set_layout, *pipeline_layout, uses_push_descriptor);
const VkRenderPass render_pass{render_pass_cache.Get(MakeRenderPassKey(key.state))};
const VkRenderPass render_pass{render_pass_cache.Get(MakeRenderPassKey(key.state, device))};
Validate();
MakePipeline(render_pass);
try {
MakePipeline(render_pass);
} catch (const vk::Exception& exception) {
LOG_CRITICAL(Render_Vulkan, "Graphics pipeline build failed: {}", exception.what());
std::scoped_lock lock{build_mutex};
is_built = true;
build_condvar.notify_one();
if (shader_notify) {
shader_notify->MarkShaderComplete();
}
return;
}
if (pipeline_statistics) {
pipeline_statistics->Collect(device, *pipeline);
}
@ -519,6 +535,9 @@ bool GraphicsPipeline::ConfigureImpl(bool is_indexed) {
texture_cache.UpdateRenderTargets(false);
texture_cache.CheckFeedbackLoop(std::span<const VideoCommon::ImageViewInOut>{views.data(),
views.size()});
if (IsBuilt() && !pipeline) {
return false;
}
ConfigureDraw(rescaling, render_area);
return true;
@ -551,6 +570,9 @@ void GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling,
uses_render_area = render_area.uses_render_area,
render_area_data = render_area.words](vk::CommandBuffer cmdbuf) {
if (bind_pipeline) {
if (!pipeline) {
return;
}
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline);
}
cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_ALL_GRAPHICS,

13
src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@ -374,9 +374,11 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
.support_int8 = device.IsInt8Supported(),
.support_uniform_and_storage_buffer_8bit =
device.IsUniformAndStorageBuffer8BitAccessSupported(),
.support_storage_buffer_8bit = device.IsStorageBuffer8BitAccessSupported(),
.support_int16 = device.IsShaderInt16Supported(),
.support_uniform_and_storage_buffer_16bit =
device.IsUniformAndStorageBuffer16BitAccessSupported(),
.support_storage_buffer_16bit = device.IsStorageBuffer16BitAccessSupported(),
.support_int64 = device.IsShaderInt64Supported(),
.support_vertex_instance_id = false,
.support_float_controls = device.IsKhrShaderFloatControlsSupported(),
@ -395,6 +397,10 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
.support_fp64_signed_zero_nan_preserve =
float_control.shaderSignedZeroInfNanPreserveFloat64 != VK_FALSE,
.support_explicit_workgroup_layout = device.IsKhrWorkgroupMemoryExplicitLayoutSupported(),
.support_workgroup_layout_8bit_access =
device.IsWorkgroupMemoryExplicitLayout8BitAccessSupported(),
.support_workgroup_layout_16bit_access =
device.IsWorkgroupMemoryExplicitLayout16BitAccessSupported(),
.support_vote = device.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_VOTE_BIT),
.supported_subgroup_stages = supported_subgroup_stages,
.support_viewport_index_layer_non_geometry =
@ -404,6 +410,7 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
.support_demote_to_helper_invocation =
device.IsExtShaderDemoteToHelperInvocationSupported(),
.support_int64_atomics = device.IsExtShaderAtomicInt64Supported(),
.support_shared_int64_atomics = device.IsSharedInt64AtomicsSupported(),
.support_derivative_control = true,
.support_geometry_shader_passthrough = device.IsNvGeometryShaderPassthroughSupported(),
.support_native_ndc = device.IsExtDepthClipControlSupported(),
@ -412,6 +419,12 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
.support_geometry_streams = device.AreTransformFeedbackGeometryStreamsSupported(),
.support_sampled_image_array_nonuniform_indexing =
device.IsSampledImageArrayNonUniformIndexingSupported(),
.support_storage_image_array_nonuniform_indexing =
device.IsStorageImageArrayNonUniformIndexingSupported(),
.support_uniform_texel_buffer_array_nonuniform_indexing =
device.IsUniformTexelBufferArrayNonUniformIndexingSupported(),
.support_storage_texel_buffer_array_nonuniform_indexing =
device.IsStorageTexelBufferArrayNonUniformIndexingSupported(),
.warp_size_potentially_larger_than_guest = device.IsWarpSizePotentiallyBiggerThanGuest(),

124
src/video_core/renderer_vulkan/vk_rasterizer.cpp

@ -410,7 +410,23 @@ void RasterizerVulkan::Clear(u32 layer_count) {
texture_cache.UpdateRenderTargets(true);
const Framebuffer* const framebuffer = texture_cache.GetFramebuffer();
const VkExtent2D render_area = framebuffer->RenderArea();
scheduler.RequestRenderpass(framebuffer);
constexpr bool ENABLE_DEFERRED_CLEAR = true;
const bool color_full_channels = regs.clear_surface.R && regs.clear_surface.G &&
regs.clear_surface.B && regs.clear_surface.A;
const bool stencil_partial = use_stencil && framebuffer->HasAspectStencilBit() &&
regs.stencil_front_mask != 0xFF && regs.stencil_front_mask != 0;
const bool ds_used = use_depth || use_stencil;
const bool ds_deferrable =
!ds_used || ((!framebuffer->HasAspectDepthBit() || use_depth) &&
(!framebuffer->HasAspectStencilBit() || use_stencil) && !stencil_partial);
const bool can_defer_clear = ENABLE_DEFERRED_CLEAR && !regs.clear_control.use_scissor &&
regs.clear_surface.layer == 0 &&
!scheduler.IsRenderPassActive() &&
(!use_color || color_full_channels) && ds_deferrable;
if (!can_defer_clear) {
scheduler.RequestRenderpass(framebuffer);
}
query_cache.NotifySegment(true);
query_cache.CounterEnable(VideoCommon::QueryType::ZPassPixelCount64, maxwell3d->regs.zpass_pixel_count_enable);
@ -494,15 +510,19 @@ void RasterizerVulkan::Clear(u32 layer_count) {
clear_value.color.int32[i] = s32(f32(s64(int_size - 1) << 1) * (regs.clear_color[i] - 0.5f));
}
if (regs.clear_surface.R && regs.clear_surface.G && regs.clear_surface.B && regs.clear_surface.A) {
scheduler.Record([color_attachment, clear_value, clear_rect](vk::CommandBuffer cmdbuf) {
const VkClearAttachment attachment{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.colorAttachment = color_attachment,
.clearValue = clear_value,
};
cmdbuf.ClearAttachments(attachment, clear_rect);
});
if (color_full_channels) {
if (can_defer_clear) {
scheduler.DeferColorClear(framebuffer, color_attachment, clear_value);
} else {
scheduler.Record([color_attachment, clear_value, clear_rect](vk::CommandBuffer cmdbuf) {
const VkClearAttachment attachment{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.colorAttachment = color_attachment,
.clearValue = clear_value,
};
cmdbuf.ClearAttachments(attachment, clear_rect);
});
}
} else {
u8 color_mask = u8(regs.clear_surface.R | regs.clear_surface.G << 1 | regs.clear_surface.B << 2 | regs.clear_surface.A << 3);
Region2D dst_region = {
@ -536,6 +556,11 @@ void RasterizerVulkan::Clear(u32 layer_count) {
blit_image.ClearDepthStencil(framebuffer, use_depth, regs.clear_depth,
u8(regs.stencil_front_mask), regs.clear_stencil,
regs.stencil_front_func_mask, dst_region);
} else if (can_defer_clear) {
VkClearValue ds_value{};
ds_value.depthStencil.depth = regs.clear_depth;
ds_value.depthStencil.stencil = regs.clear_stencil;
scheduler.DeferDepthStencilClear(framebuffer, ds_value);
} else {
scheduler.Record([clear_depth = regs.clear_depth, clear_stencil = regs.clear_stencil,
clear_rect, aspect_flags](vk::CommandBuffer cmdbuf) {
@ -569,13 +594,20 @@ void RasterizerVulkan::DispatchCompute() {
const auto [buffer, offset] =
buffer_cache.ObtainBuffer(*indirect_address, 12, sync_info, post_op);
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([indirect_buffer = buffer->Handle(),
scheduler.Record([pipeline, indirect_buffer = buffer->Handle(),
indirect_offset = offset](vk::CommandBuffer cmdbuf) {
if (!pipeline->IsBound()) {
return;
}
cmdbuf.DispatchIndirect(indirect_buffer, indirect_offset);
});
return;
}
const std::array<u32, 3> dim{qmd.grid_dim_x, qmd.grid_dim_y, qmd.grid_dim_z};
const std::array<u32, 3> max_dim{device.GetMaxComputeWorkGroupCount()};
if (dim[0] > max_dim[0] || dim[1] > max_dim[1] || dim[2] > max_dim[2]) {
return;
}
scheduler.RequestOutsideRenderPassOperationContext();
static constexpr VkMemoryBarrier READ_BARRIER{
.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
@ -585,7 +617,12 @@ void RasterizerVulkan::DispatchCompute() {
};
scheduler.Record([](vk::CommandBuffer cmdbuf) { cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
0, READ_BARRIER); });
scheduler.Record([dim](vk::CommandBuffer cmdbuf) { cmdbuf.Dispatch(dim[0], dim[1], dim[2]); });
scheduler.Record([pipeline, dim](vk::CommandBuffer cmdbuf) {
if (!pipeline->IsBound()) {
return;
}
cmdbuf.Dispatch(dim[0], dim[1], dim[2]);
});
// Log compute dispatch
if (GPU::Logging::IsActive() &&
@ -1309,15 +1346,22 @@ void RasterizerVulkan::UpdateStencilFaces(Tegra::Engines::Maxwell3D::Regs& regs)
}
if (update_references) {
[&]() {
bool changed;
if (regs.stencil_two_side_enable) {
if (!state_tracker.CheckStencilReferenceFront(regs.stencil_front_ref) &&
!state_tracker.CheckStencilReferenceBack(regs.stencil_back_ref)) {
return;
}
const bool front_changed =
state_tracker.CheckStencilReferenceFront(regs.stencil_front_ref);
const bool back_changed =
state_tracker.CheckStencilReferenceBack(regs.stencil_back_ref);
changed = front_changed || back_changed;
} else {
if (!state_tracker.CheckStencilReferenceFront(regs.stencil_front_ref)) {
return;
}
const bool front_changed =
state_tracker.CheckStencilReferenceFront(regs.stencil_front_ref);
const bool back_changed =
state_tracker.CheckStencilReferenceBack(regs.stencil_front_ref);
changed = front_changed || back_changed;
}
if (!changed) {
return;
}
scheduler.Record([front_ref = regs.stencil_front_ref, back_ref = regs.stencil_back_ref,
two_sided = regs.stencil_two_side_enable](vk::CommandBuffer cmdbuf) {
@ -1334,15 +1378,22 @@ void RasterizerVulkan::UpdateStencilFaces(Tegra::Engines::Maxwell3D::Regs& regs)
}
if (update_write_mask) {
[&]() {
bool changed;
if (regs.stencil_two_side_enable) {
if (!state_tracker.CheckStencilWriteMaskFront(regs.stencil_front_mask) &&
!state_tracker.CheckStencilWriteMaskBack(regs.stencil_back_mask)) {
return;
}
const bool front_changed =
state_tracker.CheckStencilWriteMaskFront(regs.stencil_front_mask);
const bool back_changed =
state_tracker.CheckStencilWriteMaskBack(regs.stencil_back_mask);
changed = front_changed || back_changed;
} else {
if (!state_tracker.CheckStencilWriteMaskFront(regs.stencil_front_mask)) {
return;
}
const bool front_changed =
state_tracker.CheckStencilWriteMaskFront(regs.stencil_front_mask);
const bool back_changed =
state_tracker.CheckStencilWriteMaskBack(regs.stencil_front_mask);
changed = front_changed || back_changed;
}
if (!changed) {
return;
}
scheduler.Record([front_write_mask = regs.stencil_front_mask,
back_write_mask = regs.stencil_back_mask,
@ -1360,15 +1411,22 @@ void RasterizerVulkan::UpdateStencilFaces(Tegra::Engines::Maxwell3D::Regs& regs)
}
if (update_compare_masks) {
[&]() {
bool changed;
if (regs.stencil_two_side_enable) {
if (!state_tracker.CheckStencilCompareMaskFront(regs.stencil_front_func_mask) &&
!state_tracker.CheckStencilCompareMaskBack(regs.stencil_back_func_mask)) {
return;
}
const bool front_changed =
state_tracker.CheckStencilCompareMaskFront(regs.stencil_front_func_mask);
const bool back_changed =
state_tracker.CheckStencilCompareMaskBack(regs.stencil_back_func_mask);
changed = front_changed || back_changed;
} else {
if (!state_tracker.CheckStencilCompareMaskFront(regs.stencil_front_func_mask)) {
return;
}
const bool front_changed =
state_tracker.CheckStencilCompareMaskFront(regs.stencil_front_func_mask);
const bool back_changed =
state_tracker.CheckStencilCompareMaskBack(regs.stencil_front_func_mask);
changed = front_changed || back_changed;
}
if (!changed) {
return;
}
scheduler.Record([front_test_mask = regs.stencil_front_func_mask,
back_test_mask = regs.stencil_back_func_mask,

52
src/video_core/renderer_vulkan/vk_render_pass_cache.cpp

@ -44,7 +44,9 @@ using VideoCore::Surface::SurfaceType;
}
VkAttachmentDescription AttachmentDescription(const Device& device, PixelFormat format,
VkSampleCountFlagBits samples) {
VkSampleCountFlagBits samples,
VkAttachmentLoadOp load_op,
VkAttachmentStoreOp store_op) {
using MaxwellToVK::SurfaceFormat;
const SurfaceType surface_type = GetSurfaceType(format);
@ -55,12 +57,10 @@ using VideoCore::Surface::SurfaceType;
.flags = {},
.format = SurfaceFormat(device, FormatType::Optimal, true, format).format,
.samples = samples,
.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
.storeOp = VK_ATTACHMENT_STORE_OP_STORE,
.stencilLoadOp = has_stencil ? VK_ATTACHMENT_LOAD_OP_LOAD
: VK_ATTACHMENT_LOAD_OP_DONT_CARE,
.stencilStoreOp = has_stencil ? VK_ATTACHMENT_STORE_OP_STORE
: VK_ATTACHMENT_STORE_OP_DONT_CARE,
.loadOp = load_op,
.storeOp = store_op,
.stencilLoadOp = has_stencil ? load_op : VK_ATTACHMENT_LOAD_OP_DONT_CARE,
.stencilStoreOp = has_stencil ? store_op : VK_ATTACHMENT_STORE_OP_DONT_CARE,
.initialLayout = VK_IMAGE_LAYOUT_GENERAL,
.finalLayout = VK_IMAGE_LAYOUT_GENERAL,
};
@ -87,7 +87,14 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) {
.layout = VK_IMAGE_LAYOUT_GENERAL,
};
if (is_valid) {
descriptions.push_back(AttachmentDescription(*device, format, key.samples));
const VkAttachmentLoadOp load_op = (key.color_clear_mask & (1u << index)) != 0
? VK_ATTACHMENT_LOAD_OP_CLEAR
: VK_ATTACHMENT_LOAD_OP_LOAD;
const VkAttachmentStoreOp store_op = (key.color_discard_mask & (1u << index)) != 0
? VK_ATTACHMENT_STORE_OP_DONT_CARE
: VK_ATTACHMENT_STORE_OP_STORE;
descriptions.push_back(
AttachmentDescription(*device, format, key.samples, load_op, store_op));
num_attachments = static_cast<u32>(index + 1);
++num_colors;
}
@ -99,7 +106,32 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) {
.attachment = num_colors,
.layout = VK_IMAGE_LAYOUT_GENERAL,
};
descriptions.push_back(AttachmentDescription(*device, key.depth_format, key.samples));
const VkAttachmentLoadOp depth_load_op = key.depth_stencil_clear
? VK_ATTACHMENT_LOAD_OP_CLEAR
: VK_ATTACHMENT_LOAD_OP_LOAD;
descriptions.push_back(AttachmentDescription(*device, key.depth_format, key.samples,
depth_load_op, VK_ATTACHMENT_STORE_OP_STORE));
}
std::array<VkAttachmentReference, 8> resolve_references{};
const bool do_resolve_color =
key.resolve_color && key.samples != VK_SAMPLE_COUNT_1_BIT && num_colors > 0;
if (do_resolve_color) {
for (size_t index = 0; index < key.color_formats.size(); ++index) {
const PixelFormat format{key.color_formats[index]};
const bool is_valid{format != PixelFormat::Invalid};
resolve_references[index] = VkAttachmentReference{
.attachment = is_valid ? static_cast<u32>(descriptions.size()) : VK_ATTACHMENT_UNUSED,
.layout = VK_IMAGE_LAYOUT_GENERAL,
};
if (is_valid) {
VkAttachmentDescription resolve_desc =
AttachmentDescription(*device, format, VK_SAMPLE_COUNT_1_BIT,
VK_ATTACHMENT_LOAD_OP_DONT_CARE,
VK_ATTACHMENT_STORE_OP_STORE);
resolve_desc.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
descriptions.push_back(resolve_desc);
}
}
}
const VkSubpassDescription subpass{
.flags = 0,
@ -108,7 +140,7 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) {
.pInputAttachments = nullptr,
.colorAttachmentCount = num_attachments,
.pColorAttachments = references.data(),
.pResolveAttachments = nullptr,
.pResolveAttachments = do_resolve_color ? resolve_references.data() : nullptr,
.pDepthStencilAttachment = has_depth ? &depth_reference : nullptr,
.preserveAttachmentCount = 0,
.pPreserveAttachments = nullptr,

8
src/video_core/renderer_vulkan/vk_render_pass_cache.h

@ -20,6 +20,10 @@ struct RenderPassKey {
std::array<VideoCore::Surface::PixelFormat, 8> color_formats;
VideoCore::Surface::PixelFormat depth_format;
VkSampleCountFlagBits samples;
bool resolve_color;
u32 color_clear_mask;
bool depth_stencil_clear;
u32 color_discard_mask;
};
} // namespace Vulkan
@ -30,6 +34,10 @@ struct hash<Vulkan::RenderPassKey> {
[[nodiscard]] size_t operator()(const Vulkan::RenderPassKey& key) const noexcept {
size_t value = static_cast<size_t>(key.depth_format) << 48;
value ^= static_cast<size_t>(key.samples) << 52;
value ^= static_cast<size_t>(key.resolve_color) << 63;
value ^= static_cast<size_t>(key.color_clear_mask) << 54;
value ^= static_cast<size_t>(key.depth_stencil_clear) << 62;
value ^= static_cast<size_t>(key.color_discard_mask) << 24;
for (size_t i = 0; i < key.color_formats.size(); ++i) {
value ^= static_cast<size_t>(key.color_formats[i]) << (i * 6);
}

106
src/video_core/renderer_vulkan/vk_scheduler.cpp

@ -93,30 +93,27 @@ void Scheduler::DispatchWork() {
}
}
void Scheduler::RequestRenderpass(const Framebuffer* framebuffer) {
const VkRenderPass renderpass = framebuffer->RenderPass();
void Scheduler::BeginRenderPassImpl(const Framebuffer* framebuffer, VkRenderPass renderpass,
const VkClearValue* clear_values, u32 clear_value_count) {
const VkFramebuffer framebuffer_handle = framebuffer->Handle();
const VkExtent2D render_area = framebuffer->RenderArea();
if (renderpass == state.renderpass && framebuffer_handle == state.framebuffer &&
render_area.width == state.render_area.width &&
render_area.height == state.render_area.height) {
return;
}
EndRenderPass();
state.renderpass = renderpass;
state.framebuffer = framebuffer_handle;
state.render_area = render_area;
// Log render pass begin
if (GPU::Logging::IsActive() &&
Settings::values.gpu_log_vulkan_calls.GetValue()) {
const std::string render_pass_info = fmt::format(
"renderArea={}x{}, numImages={}",
render_area.width, render_area.height, framebuffer->NumImages());
if (GPU::Logging::IsActive() && Settings::values.gpu_log_vulkan_calls.GetValue()) {
const std::string render_pass_info =
fmt::format("renderArea={}x{}, numImages={}", render_area.width, render_area.height,
framebuffer->NumImages());
GPU::Logging::GPULogger::GetInstance().LogRenderPassBegin(render_pass_info);
}
Record([renderpass, framebuffer_handle, render_area](vk::CommandBuffer cmdbuf) {
std::array<VkClearValue, 9> values{};
for (u32 i = 0; i < clear_value_count && i < values.size(); ++i) {
values[i] = clear_values[i];
}
Record([renderpass, framebuffer_handle, render_area, values, clear_value_count](
vk::CommandBuffer cmdbuf) {
const VkRenderPassBeginInfo renderpass_bi{
.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
.pNext = nullptr,
@ -127,8 +124,8 @@ void Scheduler::RequestRenderpass(const Framebuffer* framebuffer) {
.offset = {.x = 0, .y = 0},
.extent = render_area,
},
.clearValueCount = 0,
.pClearValues = nullptr,
.clearValueCount = clear_value_count,
.pClearValues = clear_value_count != 0 ? values.data() : nullptr,
};
cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE);
});
@ -137,6 +134,80 @@ void Scheduler::RequestRenderpass(const Framebuffer* framebuffer) {
renderpass_image_ranges = framebuffer->ImageRanges();
}
void Scheduler::RealizeDeferredClear() {
if (deferred_clear.framebuffer == nullptr) {
return;
}
const DeferredClear dc = deferred_clear;
deferred_clear = {};
std::array<VkClearValue, 9> clear_values{};
u32 count = 0;
const RenderPassKey& base = dc.framebuffer->RenderPassKeyBase();
for (u32 slot = 0; slot < 8; ++slot) {
if (base.color_formats[slot] == VideoCore::Surface::PixelFormat::Invalid) {
continue;
}
clear_values[count++] = dc.color_values[slot];
}
if (base.depth_format != VideoCore::Surface::PixelFormat::Invalid) {
clear_values[count++] = dc.depth_stencil_value;
}
const u32 color_discard_mask =
dc.framebuffer->DiscardsMsaaColor() ? dc.color_clear_mask : 0u;
const VkRenderPass renderpass = dc.framebuffer->RenderPassVariant(
dc.color_clear_mask, dc.depth_stencil, color_discard_mask);
EndRenderPass();
BeginRenderPassImpl(dc.framebuffer, renderpass, clear_values.data(), count);
}
bool Scheduler::DeferColorClear(const Framebuffer* framebuffer, u32 rt_slot,
const VkClearValue& value) {
if (IsRenderPassActive()) {
return false;
}
if (deferred_clear.framebuffer != nullptr && deferred_clear.framebuffer != framebuffer) {
RealizeDeferredClear();
EndRenderPass();
}
deferred_clear.framebuffer = framebuffer;
deferred_clear.color_clear_mask |= 1u << rt_slot;
deferred_clear.color_values[rt_slot] = value;
return true;
}
bool Scheduler::DeferDepthStencilClear(const Framebuffer* framebuffer, const VkClearValue& value) {
if (IsRenderPassActive()) {
return false;
}
if (deferred_clear.framebuffer != nullptr && deferred_clear.framebuffer != framebuffer) {
RealizeDeferredClear();
EndRenderPass();
}
deferred_clear.framebuffer = framebuffer;
deferred_clear.depth_stencil = true;
deferred_clear.depth_stencil_value = value;
return true;
}
void Scheduler::RequestRenderpass(const Framebuffer* framebuffer) {
if (deferred_clear.framebuffer == framebuffer) {
RealizeDeferredClear();
return;
}
const VkRenderPass renderpass = framebuffer->RenderPass();
const VkFramebuffer framebuffer_handle = framebuffer->Handle();
const VkExtent2D render_area = framebuffer->RenderArea();
if (renderpass == state.renderpass && framebuffer_handle == state.framebuffer &&
render_area.width == state.render_area.width &&
render_area.height == state.render_area.height) {
return;
}
// Ends any active pass and realizes a deferred clear
EndRenderPass();
BeginRenderPassImpl(framebuffer, renderpass, nullptr, 0);
}
void Scheduler::RequestOutsideRenderPassOperationContext() {
EndRenderPass();
}
@ -308,6 +379,7 @@ void Scheduler::EndPendingOperations() {
void Scheduler::EndRenderPass()
{
RealizeDeferredClear();
if (!state.renderpass) {
return;
}

23
src/video_core/renderer_vulkan/vk_scheduler.h

@ -59,6 +59,12 @@ public:
/// Requests to begin a renderpass.
void RequestRenderpass(const Framebuffer* framebuffer);
/// Defers a full-attachment color clear so it becomes the next render pass.
bool DeferColorClear(const Framebuffer* framebuffer, u32 rt_slot, const VkClearValue& value);
/// Defers a full depth/stencil clear so it becomes the next render pass.
bool DeferDepthStencilClear(const Framebuffer* framebuffer, const VkClearValue& value);
/// Requests the current execution context to be able to execute operations only allowed outside
/// of a renderpass.
void RequestOutsideRenderPassOperationContext();
@ -251,6 +257,21 @@ private:
bool needs_state_enable_refresh = false;
};
struct DeferredClear {
const Framebuffer* framebuffer = nullptr;
u32 color_clear_mask = 0;
std::array<VkClearValue, 8> color_values{};
bool depth_stencil = false;
VkClearValue depth_stencil_value{};
};
/// Begins a render pass for the given framebuffer, optionally with clear values.
void BeginRenderPassImpl(const Framebuffer* framebuffer, VkRenderPass renderpass,
const VkClearValue* clear_values, u32 clear_value_count);
/// If a deferred clear is pending.
void RealizeDeferredClear();
void WorkerThread(std::stop_token stop_token);
void AllocateWorkerCommandBuffer();
@ -276,6 +297,8 @@ private:
vk::CommandBuffer current_cmdbuf;
vk::CommandBuffer current_upload_cmdbuf;
DeferredClear deferred_clear;
std::unique_ptr<CommandChunk> chunk;
std::function<void()> on_submit;

468
src/video_core/renderer_vulkan/vk_texture_cache.cpp

@ -52,6 +52,9 @@ using VideoCore::Surface::IsPixelFormatInteger;
using VideoCore::Surface::SurfaceType;
namespace {
constexpr bool ENABLE_MSAA_RESOLVE_CONSUME = true;
constexpr bool ENABLE_MSAA_COLOR_DISCARD = true;
constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
if (color == std::array<float, 4>{0, 0, 0, 0}) {
return VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
@ -909,9 +912,6 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched
astc_decoder_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool,
compute_pass_descriptor_queue, memory_allocator);
}
if (device.IsStorageImageMultisampleSupported()) {
msaa_copy_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, compute_pass_descriptor_queue);
}
if (!device.IsKhrImageFormatListSupported()) {
return;
}
@ -995,12 +995,84 @@ VkBuffer TextureCacheRuntime::GetTemporaryBuffer(size_t needed_size) {
return *buffers[level];
}
VkImageView TextureCacheRuntime::GetOrCreateResolveShadow(VkImage msaa_image, VkFormat format,
VkExtent2D extent, u32 layers) {
ResolveShadow& shadow = resolve_shadows[msaa_image];
if (shadow.image && shadow.format == format && shadow.extent.width == extent.width &&
shadow.extent.height == extent.height && shadow.layers == layers) {
shadow.up_to_date = true;
return *shadow.view;
}
shadow.image = memory_allocator.CreateImage(VkImageCreateInfo{
.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.imageType = VK_IMAGE_TYPE_2D,
.format = format,
.extent = {extent.width, extent.height, 1},
.mipLevels = 1,
.arrayLayers = layers,
.samples = VK_SAMPLE_COUNT_1_BIT,
.tiling = VK_IMAGE_TILING_OPTIMAL,
.usage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_SAMPLED_BIT |
VK_IMAGE_USAGE_TRANSFER_SRC_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
});
shadow.view = device.GetLogical().CreateImageView(VkImageViewCreateInfo{
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.image = *shadow.image,
.viewType = layers > 1 ? VK_IMAGE_VIEW_TYPE_2D_ARRAY : VK_IMAGE_VIEW_TYPE_2D,
.format = format,
.components{},
.subresourceRange{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.baseMipLevel = 0,
.levelCount = 1,
.baseArrayLayer = 0,
.layerCount = layers,
},
});
shadow.format = format;
shadow.extent = extent;
shadow.layers = layers;
shadow.up_to_date = true;
return *shadow.view;
}
const TextureCacheRuntime::ResolveShadow* TextureCacheRuntime::GetValidResolveShadow(
VkImage msaa_image) const {
const auto it = resolve_shadows.find(msaa_image);
if (it == resolve_shadows.end() || !it->second.up_to_date || !it->second.image) {
return nullptr;
}
return &it->second;
}
void TextureCacheRuntime::InvalidateResolveShadow(VkImage msaa_image) {
const auto it = resolve_shadows.find(msaa_image);
if (it != resolve_shadows.end()) {
it->second.up_to_date = false;
}
}
void TextureCacheRuntime::EraseResolveShadow(VkImage msaa_image) {
resolve_shadows.erase(msaa_image);
}
void TextureCacheRuntime::BarrierFeedbackLoop() {
scheduler.RequestOutsideRenderPassOperationContext();
}
void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src,
std::span<const VideoCommon::ImageCopy> copies) {
if (ENABLE_MSAA_RESOLVE_CONSUME) {
InvalidateResolveShadow(dst.Handle());
}
boost::container::small_vector<VkBufferImageCopy, 16> vk_in_copies(copies.size());
boost::container::small_vector<VkBufferImageCopy, 16> vk_out_copies(copies.size());
const VkImageAspectFlags src_aspect_mask = src.AspectMask();
@ -1147,6 +1219,15 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst
return;
}
ASSERT(src.format == dst.format);
if (is_src_msaa && !is_dst_msaa &&
(aspect_mask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) != 0) {
if ((aspect_mask & VK_IMAGE_ASPECT_DEPTH_BIT) == 0) {
UNIMPLEMENTED_MSG("Stencil-only MSAA resolve is not supported");
return;
}
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] {
@ -1172,19 +1253,20 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst
ASSERT(!(is_dst_msaa && !is_src_msaa));
ASSERT(operation == Fermi2D::Operation::SrcCopy);
const VkImage dst_image = dst.ImageHandle();
const VkImage src_image = src.ImageHandle();
const VkImageSubresourceLayers dst_layers = MakeSubresourceLayers(&dst);
const VkImageSubresourceLayers src_layers = MakeSubresourceLayers(&src);
const bool is_msaa_to_msaa = is_src_msaa && is_dst_msaa;
// NVIDIA 510+ and Intel crash on MSAA->MSAA blits (scaling operations)
// Fall back to 3D helpers for MSAA scaling
if (is_msaa_to_msaa && aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT) {
blit_image_helper.BlitColorMSAA(dst_framebuffer, src, dst_region, src_region);
return;
}
if (is_msaa_to_msaa && device.CantBlitMSAA()) {
// This should be handled by NeedsScaleHelper() and use 3D helpers instead
UNIMPLEMENTED_MSG("MSAA to MSAA blit not supported on this driver");
UNIMPLEMENTED_MSG("MSAA to MSAA depth-stencil blit is not supported on this driver");
return;
}
const VkImage dst_image = dst.ImageHandle();
const VkImage src_image = src.ImageHandle();
const VkImageSubresourceLayers dst_layers = MakeSubresourceLayers(&dst);
const VkImageSubresourceLayers src_layers = MakeSubresourceLayers(&src);
const bool is_resolve = is_src_msaa && !is_dst_msaa;
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([filter, dst_region, src_region, dst_image, src_image, dst_layers, src_layers,
@ -1442,6 +1524,9 @@ bool TextureCacheRuntime::IsFormatScalable(PixelFormat format) {
void TextureCacheRuntime::CopyImage(Image& dst, Image& src,
std::span<const VideoCommon::ImageCopy> copies) {
if (ENABLE_MSAA_RESOLVE_CONSUME) {
InvalidateResolveShadow(dst.Handle());
}
// As per the size-compatible formats section of vulkan, copy manually via ReinterpretImage
// these images that aren't size-compatible
if (BytesPerBlock(src.info.format) != BytesPerBlock(dst.info.format)) {
@ -1555,10 +1640,117 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src,
void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src,
std::span<const VideoCommon::ImageCopy> copies) {
const bool msaa_to_non_msaa = src.info.num_samples > 1 && dst.info.num_samples == 1;
if (msaa_copy_pass) {
return msaa_copy_pass->CopyImage(dst, src, copies, msaa_to_non_msaa);
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.");
return;
}
if (ENABLE_MSAA_RESOLVE_CONSUME && msaa_to_non_msaa && copies.size() == 1 &&
src.info.format == dst.info.format) {
const VideoCommon::ImageCopy& copy = copies.front();
const ResolveShadow* const shadow = GetValidResolveShadow(src.Handle());
if (shadow != nullptr && copy.src_offset.x == 0 && copy.src_offset.y == 0 &&
copy.src_subresource.base_level == 0 &&
static_cast<u32>(copy.extent.width) <= shadow->extent.width &&
static_cast<u32>(copy.extent.height) <= shadow->extent.height) {
const VkImage shadow_image = *shadow->image;
const VkImage dst_image = dst.Handle();
const VkImageCopy region{
.srcSubresource{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.mipLevel = 0,
.baseArrayLayer = static_cast<u32>(copy.src_subresource.base_layer),
.layerCount = static_cast<u32>(copy.src_subresource.num_layers),
},
.srcOffset = {0, 0, 0},
.dstSubresource{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.mipLevel = static_cast<u32>(copy.dst_subresource.base_level),
.baseArrayLayer = static_cast<u32>(copy.dst_subresource.base_layer),
.layerCount = static_cast<u32>(copy.dst_subresource.num_layers),
},
.dstOffset = {copy.dst_offset.x, copy.dst_offset.y, copy.dst_offset.z},
.extent = {copy.extent.width, copy.extent.height, 1},
};
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([shadow_image, dst_image, region](vk::CommandBuffer cmdbuf) {
const std::array pre_barriers{
VkImageMemoryBarrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
.oldLayout = VK_IMAGE_LAYOUT_GENERAL,
.newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = shadow_image,
.subresourceRange{VK_IMAGE_ASPECT_COLOR_BIT, 0, VK_REMAINING_MIP_LEVELS, 0,
VK_REMAINING_ARRAY_LAYERS},
},
VkImageMemoryBarrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT |
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
VK_ACCESS_TRANSFER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
.oldLayout = VK_IMAGE_LAYOUT_GENERAL,
.newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = dst_image,
.subresourceRange{VK_IMAGE_ASPECT_COLOR_BIT, 0, VK_REMAINING_MIP_LEVELS, 0,
VK_REMAINING_ARRAY_LAYERS},
},
};
const std::array post_barriers{
VkImageMemoryBarrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = 0,
.dstAccessMask = 0,
.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
.newLayout = VK_IMAGE_LAYOUT_GENERAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = shadow_image,
.subresourceRange{VK_IMAGE_ASPECT_COLOR_BIT, 0, VK_REMAINING_MIP_LEVELS, 0,
VK_REMAINING_ARRAY_LAYERS},
},
VkImageMemoryBarrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_SHADER_READ_BIT |
VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
.newLayout = VK_IMAGE_LAYOUT_GENERAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = dst_image,
.subresourceRange{VK_IMAGE_ASPECT_COLOR_BIT, 0, VK_REMAINING_MIP_LEVELS, 0,
VK_REMAINING_ARRAY_LAYERS},
},
};
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0, nullptr, nullptr,
pre_barriers);
cmdbuf.CopyImage(shadow_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, region);
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, nullptr, nullptr,
post_barriers);
});
return;
}
}
UNIMPLEMENTED_MSG("Copying images with different samples is not supported.");
blit_image_helper.CopyMSAA(render_pass_cache, dst.Handle(), dst.info.format, src.Handle(),
src.info.format, num_samples, copies, msaa_to_non_msaa);
}
u64 TextureCacheRuntime::GetDeviceLocalMemory() const {
@ -1577,7 +1769,11 @@ std::optional<size_t> TextureCacheRuntime::GetSamplerHeapBudget() const {
return device.GetSamplerHeapBudget();
}
void TextureCacheRuntime::TickFrame() {}
void TextureCacheRuntime::TickFrame() {
std::erase_if(pending_msaa_images, [this](const auto& pending) {
return scheduler.IsFree(pending.first);
});
}
Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu_addr_,
VAddr cpu_addr_)
@ -1632,7 +1828,16 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
Image::Image(const VideoCommon::NullImageParams& params) : VideoCommon::ImageBase{params} {}
Image::~Image() = default;
Image::~Image() {
if (ENABLE_MSAA_RESOLVE_CONSUME && runtime != nullptr) {
if (original_image) {
runtime->EraseResolveShadow(*original_image);
}
if (scaled_image) {
runtime->EraseResolveShadow(*scaled_image);
}
}
}
void Image::AllocateComputeUnswizzleBuffer(u32 max_slices) {
using VideoCore::Surface::BytesPerBlock;
@ -1679,82 +1884,75 @@ void Image::AllocateComputeUnswizzleBuffer(u32 max_slices) {
void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
std::span<const VideoCommon::BufferImageCopy> copies) {
// TODO: Move this to another API
if (ENABLE_MSAA_RESOLVE_CONSUME && runtime != nullptr) {
runtime->InvalidateResolveShadow(Handle());
}
const bool is_rescaled = True(flags & ImageFlagBits::Rescaled);
if (is_rescaled) {
ScaleDown(true);
}
// Handle MSAA upload if necessary
/* WARNING, TODO: This code uses some hacks, besides being fundamentally ugly
since tropic didn't want to touch it for a long time, so it needs a rewrite from someone
better than me at vulkan. */
// CHANGE: Gate the MSAA path more strictly and only use it for color, when the pass and device
// support are available. Avoid running the MSAA path when prerequisites aren't met,
// preventing validation and runtime issues.
const bool wants_msaa_upload = info.num_samples > 1
&& (aspect_mask & VK_IMAGE_ASPECT_COLOR_BIT) != 0
&& runtime->CanUploadMSAA() && runtime->msaa_copy_pass.has_value()
&& runtime->device.IsStorageImageMultisampleSupported();
&& !VideoCore::Surface::IsPixelFormatInteger(info.format);
if (wants_msaa_upload) {
// Create a temporary non-MSAA image to upload the data first
ImageInfo temp_info = info;
temp_info.num_samples = 1;
// CHANGE: Build a fresh VkImageCreateInfo with robust usage flags for the temp image.
// Using the target image's usage as-is could miss STORAGE/TRANSFER bits and trigger
// validation errors.
VkImageCreateInfo image_ci = MakeImageCreateInfo(runtime->device, temp_info);
image_ci.usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT |
VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_STORAGE_BIT;
// CHANGE: The previous stack-allocated wrapper was destroyed at function exit,
// which could destroy VkImage before the GPU used it.
auto temp_wrapper = std::make_shared<Image>(*runtime, temp_info, 0, 0);
temp_wrapper->original_image = runtime->memory_allocator.CreateImage(image_ci);
temp_wrapper->current_image = &Image::original_image;
temp_wrapper->aspect_mask = aspect_mask;
temp_wrapper->initialized = true;
// Upload to the temporary non-MSAA image
image_ci.format =
MaxwellToVK::SurfaceFormat(runtime->device, FormatType::Optimal, true, info.format)
.format;
image_ci.usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT;
vk::Image temp_image = runtime->memory_allocator.CreateImage(image_ci);
scheduler->RequestOutsideRenderPassOperationContext();
auto vk_copies = TransformBufferImageCopies(copies, offset, temp_wrapper->aspect_mask);
auto vk_copies = TransformBufferImageCopies(copies, offset, aspect_mask);
const VkBuffer src_buffer = buffer;
const VkImage temp_vk_image = *temp_wrapper->original_image;
const VkImageAspectFlags vk_aspect_mask = temp_wrapper->aspect_mask;
const VkImage temp_vk_image = *temp_image;
const VkImageAspectFlags vk_aspect_mask = aspect_mask;
scheduler->Record([src_buffer, temp_vk_image, vk_aspect_mask, vk_copies,
keep = temp_wrapper](vk::CommandBuffer cmdbuf) {
scheduler->Record([src_buffer, temp_vk_image, vk_aspect_mask,
vk_copies](vk::CommandBuffer cmdbuf) {
CopyBufferToImage(cmdbuf, src_buffer, temp_vk_image, vk_aspect_mask, false, VideoCommon::FixSmallVectorADL(vk_copies));
});
// Use MSAACopyPass to convert from non-MSAA to MSAA
const auto [samples_x, samples_y] = VideoCommon::SamplesLog2(info.num_samples);
std::vector<VideoCommon::ImageCopy> image_copies;
image_copies.reserve(copies.size());
for (const auto& copy : copies) {
VideoCommon::ImageCopy image_copy{};
image_copy.src_offset = {0, 0, 0}; // Use zero offset for source
image_copy.dst_offset = copy.image_offset;
image_copy.src_offset = {0, 0, 0};
image_copy.dst_offset = {copy.image_offset.x >> samples_x,
copy.image_offset.y >> samples_y, copy.image_offset.z};
image_copy.src_subresource = copy.image_subresource;
image_copy.dst_subresource = copy.image_subresource;
image_copy.extent = copy.image_extent;
image_copy.extent = {copy.image_extent.width >> samples_x,
copy.image_extent.height >> samples_y, copy.image_extent.depth};
image_copies.push_back(image_copy);
}
runtime->msaa_copy_pass->CopyImage(*this, *temp_wrapper, image_copies,
/*msaa_to_non_msaa=*/false);
std::exchange(initialized, true);
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));
const u64 tick = scheduler->Flush();
scheduler->Wait(tick);
if (is_rescaled) {
ScaleUp();
}
return;
}
if (info.num_samples > 1) {
LOG_WARNING(Render_Vulkan, "MSAA upload not implemented for format {}", info.format);
if (is_rescaled) {
ScaleUp();
}
return;
}
// Regular non-MSAA upload (original behavior preserved)
scheduler->RequestOutsideRenderPassOperationContext();
auto vk_copies = TransformBufferImageCopies(copies, offset, aspect_mask);
const VkBuffer src_buffer = buffer;
@ -1794,22 +1992,45 @@ void Image::DownloadMemory(std::span<VkBuffer> buffers_span, std::span<size_t> o
ScaleDown();
}
// RE-USE MSAA UPLOAD CODE BUT NOW FOR DOWNLOAD
if (info.num_samples > 1 && runtime->msaa_copy_pass) {
// TODO: Depth/stencil formats need special handling
if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT) {
if (info.num_samples > 1) {
if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT &&
!VideoCore::Surface::IsPixelFormatInteger(info.format)) {
ImageInfo temp_info = info;
temp_info.num_samples = 1;
VkImageCreateInfo image_ci = MakeImageCreateInfo(runtime->device, temp_info);
image_ci.usage = original_image.UsageFlags();
image_ci.format =
MaxwellToVK::SurfaceFormat(runtime->device, FormatType::Optimal, true, info.format)
.format;
image_ci.usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
vk::Image temp_image = runtime->memory_allocator.CreateImage(image_ci);
const VkImage temp_vk_image = *temp_image;
Image temp_wrapper(*runtime, temp_info, 0, 0);
temp_wrapper.original_image = std::move(temp_image);
temp_wrapper.current_image = &Image::original_image;
temp_wrapper.aspect_mask = aspect_mask;
temp_wrapper.initialized = true;
scheduler->RequestOutsideRenderPassOperationContext();
scheduler->Record([temp_vk_image](vk::CommandBuffer cmdbuf) {
const VkImageMemoryBarrier init_barrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = 0,
.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED,
.newLayout = VK_IMAGE_LAYOUT_GENERAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = temp_vk_image,
.subresourceRange{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.baseMipLevel = 0,
.levelCount = VK_REMAINING_MIP_LEVELS,
.baseArrayLayer = 0,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
};
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, 0,
init_barrier);
});
std::vector<VideoCommon::ImageCopy> image_copies;
for (const auto& copy : copies) {
@ -1822,7 +2043,9 @@ void Image::DownloadMemory(std::span<VkBuffer> buffers_span, std::span<size_t> o
image_copies.push_back(image_copy);
}
runtime->msaa_copy_pass->CopyImage(temp_wrapper, *this, image_copies, true);
runtime->blit_image_helper.CopyMSAA(runtime->render_pass_cache, temp_vk_image,
info.format, Handle(), info.format,
info.num_samples, image_copies, true);
boost::container::small_vector<VkBuffer, 8> buffers_vector{};
boost::container::small_vector<boost::container::small_vector<VkBufferImageCopy, 16>, 8>
@ -1834,7 +2057,7 @@ void Image::DownloadMemory(std::span<VkBuffer> buffers_span, std::span<size_t> o
}
scheduler->RequestOutsideRenderPassOperationContext();
scheduler->Record([buffers = std::move(buffers_vector), image = *temp_wrapper.original_image,
scheduler->Record([buffers = std::move(buffers_vector), image = temp_vk_image,
aspect_mask_ = aspect_mask, vk_copies](vk::CommandBuffer cmdbuf) {
const VkImageMemoryBarrier read_barrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
@ -1889,6 +2112,8 @@ void Image::DownloadMemory(std::span<VkBuffer> buffers_span, std::span<size_t> o
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE,
0, memory_write_barrier, nullptr, image_write_barrier);
});
runtime->pending_msaa_images.emplace_back(scheduler->CurrentTick(),
std::move(temp_image));
return;
}
} else {
@ -2074,10 +2299,11 @@ bool Image::BlitScaleHelper(bool scale_up) {
blit_view = std::make_unique<ImageView>(*runtime, view_info, NULL_IMAGE_ID, *this);
}
const u32 src_width = scale_up ? info.size.width : scaled_width;
const u32 src_height = scale_up ? info.size.height : scaled_height;
const u32 dst_width = scale_up ? scaled_width : info.size.width;
const u32 dst_height = scale_up ? scaled_height : info.size.height;
const auto [samples_x, samples_y] = VideoCommon::SamplesLog2(info.num_samples);
const u32 src_width = (scale_up ? info.size.width : scaled_width) >> samples_x;
const u32 src_height = (scale_up ? info.size.height : scaled_height) >> samples_y;
const u32 dst_width = (scale_up ? scaled_width : info.size.width) >> samples_x;
const u32 dst_height = (scale_up ? scaled_height : info.size.height) >> samples_y;
const Region2D src_region{
.start = {0, 0},
.end = {s32(src_width), s32(src_height)},
@ -2087,17 +2313,23 @@ bool Image::BlitScaleHelper(bool scale_up) {
.end = {s32(dst_width), s32(dst_height)},
};
const VkExtent2D extent{
.width = (std::max)(scaled_width, info.size.width),
.height = (std::max)(scaled_height, info.size.height),
.width = (std::max)(scaled_width, info.size.width) >> samples_x,
.height = (std::max)(scaled_height, info.size.height) >> samples_y,
};
auto* view_ptr = blit_view.get();
if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT) {
if (!blit_framebuffer)
blit_framebuffer.emplace(*runtime, view_ptr, nullptr, extent, scale_up);
runtime->blit_image_helper.BlitColor(&*blit_framebuffer, *blit_view,
dst_region, src_region, operation, BLIT_OPERATION);
} else if (aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
if (info.num_samples > 1) {
runtime->blit_image_helper.BlitColorMSAA(&*blit_framebuffer, *blit_view,
dst_region, src_region);
} else {
runtime->blit_image_helper.BlitColor(&*blit_framebuffer, *blit_view,
dst_region, src_region, operation, BLIT_OPERATION);
}
} else if (aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT) &&
info.num_samples == 1) {
if (!blit_framebuffer)
blit_framebuffer.emplace(*runtime, nullptr, view_ptr, extent, scale_up);
runtime->blit_image_helper.BlitDepthStencil(&*blit_framebuffer, *blit_view,
@ -2113,7 +2345,8 @@ bool Image::BlitScaleHelper(bool scale_up) {
bool Image::NeedsScaleHelper() const {
const auto& device = runtime->device;
const bool needs_msaa_helper = info.num_samples > 1 && device.CantBlitMSAA();
const bool needs_msaa_helper = info.num_samples > 1 &&
(device.CantBlitMSAA() || aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT);
if (needs_msaa_helper) {
return true;
}
@ -2515,11 +2748,76 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
renderpass_key.depth_format = PixelFormat::Invalid;
}
renderpass_key.samples = samples;
const bool do_resolve_color =
samples != VK_SAMPLE_COUNT_1_BIT && num_colors > 0 && runtime.device.IsTiler();
renderpass_key.resolve_color = do_resolve_color;
discard_msaa_color =
ENABLE_MSAA_RESOLVE_CONSUME && ENABLE_MSAA_COLOR_DISCARD && do_resolve_color;
renderpass = runtime.render_pass_cache.Get(renderpass_key);
render_pass_key = renderpass_key;
render_pass_cache = &runtime.render_pass_cache;
render_area.width = (std::min)(render_area.width, width);
render_area.height = (std::min)(render_area.height, height);
if (do_resolve_color) {
const u32 layers = static_cast<u32>((std::max)(num_layers, 1));
for (size_t index = 0; index < NUM_RT; ++index) {
const PixelFormat format = renderpass_key.color_formats[index];
if (format == PixelFormat::Invalid) {
continue;
}
const VkFormat vk_format =
MaxwellToVK::SurfaceFormat(runtime.device, FormatType::Optimal, true, format).format;
if (ENABLE_MSAA_RESOLVE_CONSUME) {
const VkImage msaa_image = images[rt_map[index]];
attachments.push_back(runtime.GetOrCreateResolveShadow(msaa_image, vk_format,
render_area, layers));
continue;
}
VkImageCreateInfo resolve_ci{
.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.imageType = VK_IMAGE_TYPE_2D,
.format = vk_format,
.extent = {render_area.width, render_area.height, 1},
.mipLevels = 1,
.arrayLayers = layers,
.samples = VK_SAMPLE_COUNT_1_BIT,
.tiling = VK_IMAGE_TILING_OPTIMAL,
.usage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_SAMPLED_BIT |
VK_IMAGE_USAGE_TRANSFER_SRC_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
};
vk::Image resolve_image = runtime.memory_allocator.CreateImage(resolve_ci);
vk::ImageView resolve_view =
runtime.device.GetLogical().CreateImageView(VkImageViewCreateInfo{
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.image = *resolve_image,
.viewType = layers > 1 ? VK_IMAGE_VIEW_TYPE_2D_ARRAY : VK_IMAGE_VIEW_TYPE_2D,
.format = vk_format,
.components{},
.subresourceRange{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.baseMipLevel = 0,
.levelCount = 1,
.baseArrayLayer = 0,
.layerCount = layers,
},
});
attachments.push_back(*resolve_view);
resolve_images.push_back(std::move(resolve_image));
resolve_image_views.push_back(std::move(resolve_view));
}
}
num_color_buffers = static_cast<u32>(num_colors);
framebuffer = runtime.device.GetLogical().CreateFramebuffer({
.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
@ -2534,6 +2832,18 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
});
}
VkRenderPass Framebuffer::RenderPassVariant(u32 color_clear_mask, bool depth_stencil_clear,
u32 color_discard_mask) const {
if (color_clear_mask == 0 && !depth_stencil_clear && color_discard_mask == 0) {
return renderpass;
}
RenderPassKey key = render_pass_key;
key.color_clear_mask = color_clear_mask;
key.depth_stencil_clear = depth_stencil_clear;
key.color_discard_mask = color_discard_mask;
return render_pass_cache->Get(key);
}
void TextureCacheRuntime::AccelerateImageUpload(
Image& image, const StagingBufferRef& map,
std::span<const VideoCommon::SwizzleParameters> swizzles,

48
src/video_core/renderer_vulkan/vk_texture_cache.h

@ -12,6 +12,7 @@
#include "shader_recompiler/shader_info.h"
#include "video_core/renderer_vulkan/vk_compute_pass.h"
#include "video_core/renderer_vulkan/vk_render_pass_cache.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/texture_cache/image_view_base.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
@ -87,7 +88,7 @@ public:
}
bool CanUploadMSAA() const noexcept {
return msaa_copy_pass.operator bool();
return true;
}
void AccelerateImageUpload(Image&, const StagingBufferRef&,
@ -110,6 +111,24 @@ public:
[[nodiscard]] VkBuffer GetTemporaryBuffer(size_t needed_size);
struct ResolveShadow {
vk::Image image;
vk::ImageView view;
VkFormat format = VK_FORMAT_UNDEFINED;
VkExtent2D extent{};
u32 layers = 0;
bool up_to_date = false;
};
[[nodiscard]] VkImageView GetOrCreateResolveShadow(VkImage msaa_image, VkFormat format,
VkExtent2D extent, u32 layers);
[[nodiscard]] const ResolveShadow* GetValidResolveShadow(VkImage msaa_image) const;
void InvalidateResolveShadow(VkImage msaa_image);
void EraseResolveShadow(VkImage msaa_image);
std::span<const VkFormat> ViewFormats(PixelFormat format) {
return view_formats[static_cast<std::size_t>(format)];
}
@ -130,12 +149,13 @@ public:
std::optional<ASTCDecoderPass> astc_decoder_pass;
std::optional<BlockLinearUnswizzle3DPass> bl3d_unswizzle_pass;
std::optional<MSAACopyPass> msaa_copy_pass;
const Settings::ResolutionScalingInfo& resolution;
std::array<std::vector<VkFormat>, VideoCore::Surface::MaxPixelFormat> view_formats;
static constexpr size_t indexing_slots = 8 * sizeof(size_t);
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;
};
class Framebuffer {
@ -166,6 +186,13 @@ public:
return renderpass;
}
[[nodiscard]] const RenderPassKey& RenderPassKeyBase() const noexcept {
return render_pass_key;
}
[[nodiscard]] VkRenderPass RenderPassVariant(u32 color_clear_mask, bool depth_stencil_clear,
u32 color_discard_mask) const;
[[nodiscard]] VkExtent2D RenderArea() const noexcept {
return render_area;
}
@ -206,6 +233,18 @@ public:
return is_rescaled;
}
[[nodiscard]] bool HasResolveColor() const noexcept {
return !resolve_images.empty();
}
[[nodiscard]] VkImage ResolveColorImage(size_t index) const noexcept {
return index < resolve_images.size() ? *resolve_images[index] : VK_NULL_HANDLE;
}
[[nodiscard]] bool DiscardsMsaaColor() const noexcept {
return discard_msaa_color;
}
private:
vk::Framebuffer framebuffer;
VkRenderPass renderpass{};
@ -219,6 +258,11 @@ private:
bool has_depth{};
bool has_stencil{};
bool is_rescaled{};
std::vector<vk::Image> resolve_images;
std::vector<vk::ImageView> resolve_image_views;
RenderPassKey render_pass_key{};
RenderPassCache* render_pass_cache{nullptr};
bool discard_msaa_color{};
};
class Image : public VideoCommon::ImageBase {

2
src/video_core/texture_cache/image_info.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

14
src/video_core/vulkan_common/vulkan_device.cpp

@ -504,8 +504,8 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
CollectToolingInfo();
if (is_qualcomm) {
LOG_WARNING(Render_Vulkan, "Qualcomm drivers require scaled vertex format emulation");
must_emulate_scaled_formats = true;
LOG_WARNING(Render_Vulkan, "Qualcomm drivers require scaled vertex format emulation.");
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);
@ -523,11 +523,10 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
features.shader_atomic_int64.shaderBufferInt64Atomics = false;
features.shader_atomic_int64.shaderSharedInt64Atomics = false;
features.features.shaderInt64 = false;
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken storage buffer access.");
features.bit8_storage.storageBuffer8BitAccess = false;
features.bit8_storage.uniformAndStorageBuffer8BitAccess = false;
features.bit16_storage.storageBuffer16BitAccess = false;
features.bit16_storage.uniformAndStorageBuffer16BitAccess = false;
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken workgroup memory explicit layout.");
RemoveExtensionFeature(extensions.workgroup_memory_explicit_layout,
features.workgroup_memory_explicit_layout,
VK_KHR_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_EXTENSION_NAME);
#if defined(__ANDROID__) && defined(ARCHITECTURE_arm64)
// BCn patching only safe on Android 9+ (API 28+). Older versions crash on driver load.
@ -1356,10 +1355,7 @@ void Device::RemoveUnsuitableExtensions() {
// VK_KHR_workgroup_memory_explicit_layout
extensions.workgroup_memory_explicit_layout =
features.features.shaderInt16 &&
features.workgroup_memory_explicit_layout.workgroupMemoryExplicitLayout &&
features.workgroup_memory_explicit_layout.workgroupMemoryExplicitLayout8BitAccess &&
features.workgroup_memory_explicit_layout.workgroupMemoryExplicitLayout16BitAccess &&
features.workgroup_memory_explicit_layout.workgroupMemoryExplicitLayoutScalarBlockLayout;
RemoveExtensionFeatureIfUnsuitable(extensions.workgroup_memory_explicit_layout,
features.workgroup_memory_explicit_layout,

47
src/video_core/vulkan_common/vulkan_device.h

@ -305,6 +305,19 @@ public:
return properties.driver.driverID;
}
/// Returns true for tile-based deferred renderers.
bool IsTiler() const {
switch (GetDriverID()) {
case VK_DRIVER_ID_QUALCOMM_PROPRIETARY:
case VK_DRIVER_ID_ARM_PROPRIETARY:
case VK_DRIVER_ID_SAMSUNG_PROPRIETARY:
case VK_DRIVER_ID_MESA_TURNIP:
return true;
default:
return false;
}
}
bool ShouldBoostClocks() const;
/// Returns uniform buffer alignment requirement.
@ -322,6 +335,11 @@ public:
return properties.properties.limits.maxStorageBufferRange;
}
std::array<u32, 3> GetMaxComputeWorkGroupCount() const {
const auto& count = properties.properties.limits.maxComputeWorkGroupCount;
return {count[0], count[1], count[2]};
}
/// Returns the maximum size for push constants.
VkDeviceSize GetMaxPushConstantsSize() const {
return properties.properties.limits.maxPushConstantsSize;
@ -374,6 +392,18 @@ FN_MAX_LIMIT_LIST
return features.descriptor_indexing.shaderSampledImageArrayNonUniformIndexing;
}
bool IsStorageImageArrayNonUniformIndexingSupported() const {
return features.descriptor_indexing.shaderStorageImageArrayNonUniformIndexing;
}
bool IsUniformTexelBufferArrayNonUniformIndexingSupported() const {
return features.descriptor_indexing.shaderUniformTexelBufferArrayNonUniformIndexing;
}
bool IsStorageTexelBufferArrayNonUniformIndexingSupported() const {
return features.descriptor_indexing.shaderStorageTexelBufferArrayNonUniformIndexing;
}
/// Returns true if the device supports float64 natively.
bool IsFloat64Supported() const {
return features.features.shaderFloat64;
@ -513,6 +543,18 @@ FN_MAX_LIMIT_LIST
return extensions.workgroup_memory_explicit_layout;
}
bool IsWorkgroupMemoryExplicitLayout8BitAccessSupported() const {
return extensions.workgroup_memory_explicit_layout &&
features.workgroup_memory_explicit_layout.workgroupMemoryExplicitLayout8BitAccess &&
features.shader_float16_int8.shaderInt8;
}
bool IsWorkgroupMemoryExplicitLayout16BitAccessSupported() const {
return extensions.workgroup_memory_explicit_layout &&
features.workgroup_memory_explicit_layout.workgroupMemoryExplicitLayout16BitAccess &&
features.features.shaderInt16;
}
/// Returns true if the device supports VK_KHR_image_format_list.
bool IsKhrImageFormatListSupported() const {
return extensions.image_format_list || instance_version >= VK_API_VERSION_1_2;
@ -765,6 +807,11 @@ FN_MAX_LIMIT_LIST
return extensions.shader_atomic_int64;
}
bool IsSharedInt64AtomicsSupported() const {
return extensions.shader_atomic_int64 &&
features.shader_atomic_int64.shaderSharedInt64Atomics;
}
bool IsExtConditionalRendering() const {
return extensions.conditional_rendering;
}

Loading…
Cancel
Save