Mirrors/yuzu
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

shader: Interact texture buffers with buffer cache

Browse source
This commit is contained in:
ReinUsesLisp 2021-04-06 20:14:55 -03:00 • committed by ameerj
parent 56b92bd89c
commit e9a91bc5cc
17 changed files with 334 additions and 149 deletions
Download patch file Download diff file Expand all files Collapse all files

26
src/video_core/renderer_vulkan/pipeline_helper.h
View file

@ -24,7 +24,8 @@ struct TextureHandle {
[[likely]] if (via_header_index) {
image = data;
sampler = data;
} else {
}
else {
const Tegra::Texture::TextureHandle handle{data};
image = handle.tic_id;
sampler = via_header_index ? image : handle.tsc_id.Value();
@ -90,12 +91,12 @@ public:
for ([[maybe_unused]] const auto& desc : info.storage_buffers_descriptors) {
Add(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, stage);
}
for ([[maybe_unused]] const auto& desc : info.texture_buffer_descriptors) {
Add(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, stage);
}
for ([[maybe_unused]] const auto& desc : info.texture_descriptors) {
Add(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, stage);
}
for (const auto& desc : info.texture_buffer_descriptors) {
Add(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, stage);
}
}
private:
@ -156,20 +157,15 @@ inline VideoCommon::ImageViewType CastType(Shader::TextureType type) {
return {};
}
inline void PushImageDescriptors(const Shader::Info& info, const VkSampler* samplers,
const ImageId* image_view_ids, TextureCache& texture_cache,
VKUpdateDescriptorQueue& update_descriptor_queue, size_t& index) {
inline void PushImageDescriptors(const Shader::Info& info, const VkSampler*& samplers,
const ImageId*& image_view_ids, TextureCache& texture_cache,
VKUpdateDescriptorQueue& update_descriptor_queue) {
image_view_ids += info.texture_buffer_descriptors.size();
for (const auto& desc : info.texture_descriptors) {
const VkSampler sampler{samplers[index]};
ImageView& image_view{texture_cache.GetImageView(image_view_ids[index])};
const VkSampler sampler{*(samplers++)};
ImageView& image_view{texture_cache.GetImageView(*(image_view_ids++))};
const VkImageView vk_image_view{image_view.Handle(CastType(desc.type))};
update_descriptor_queue.AddSampledImage(vk_image_view, sampler);
++index;
}
for (const auto& desc : info.texture_buffer_descriptors) {
ImageView& image_view{texture_cache.GetImageView(image_view_ids[index])};
update_descriptor_queue.AddTexelBuffer(image_view.BufferView());
++index;
}
}

57
src/video_core/renderer_vulkan/vk_buffer_cache.cpp
View file

@ -67,25 +67,50 @@ Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params)
Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_,
VAddr cpu_addr_, u64 size_bytes_)
: VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_) {
buffer = runtime.device.GetLogical().CreateBuffer(VkBufferCreateInfo{
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.size = SizeBytes(),
.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT |
VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT |
VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
});
: VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_),
device{&runtime.device},
buffer{device->GetLogical().CreateBuffer({
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.size = SizeBytes(),
.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT |
VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT |
VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
})},
commit{runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal)} {
if (runtime.device.HasDebuggingToolAttached()) {
buffer.SetObjectNameEXT(fmt::format("Buffer 0x{:x}", CpuAddr()).c_str());
}
commit = runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal);
}
VkBufferView Buffer::View(u32 offset, u32 size, VideoCore::Surface::PixelFormat format) {
const auto it{std::ranges::find_if(views, [offset, size, format](const BufferView& view) {
return offset == view.offset && size == view.size && format == view.format;
})};
if (it != views.end()) {
return *it->handle;
}
views.push_back({
.offset = offset,
.size = size,
.format = format,
.handle = device->GetLogical().CreateBufferView({
.sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.buffer = *buffer,
.format = MaxwellToVK::SurfaceFormat(*device, FormatType::Buffer, false, format).format,
.offset = offset,
.range = size,
}),
});
return *views.back().handle;
}
BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_allocator_,

18
src/video_core/renderer_vulkan/vk_buffer_cache.h
View file

@ -9,6 +9,7 @@
#include "video_core/renderer_vulkan/vk_compute_pass.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/surface.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@ -26,6 +27,8 @@ public:
explicit Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_,
VAddr cpu_addr_, u64 size_bytes_);
[[nodiscard]] VkBufferView View(u32 offset, u32 size, VideoCore::Surface::PixelFormat format);
[[nodiscard]] VkBuffer Handle() const noexcept {
return *buffer;
}
@ -35,8 +38,17 @@ public:
}
private:
struct BufferView {
u32 offset;
u32 size;
VideoCore::Surface::PixelFormat format;
vk::BufferView handle;
};
const Device* device{};
vk::Buffer buffer;
MemoryCommit commit;
std::vector<BufferView> views;
};
class BufferCacheRuntime {
@ -87,6 +99,11 @@ public:
BindBuffer(buffer, offset, size);
}
void BindTextureBuffer(Buffer& buffer, u32 offset, u32 size,
VideoCore::Surface::PixelFormat format) {
update_descriptor_queue.AddTexelBuffer(buffer.View(offset, size, format));
}
private:
void BindBuffer(VkBuffer buffer, u32 offset, u32 size) {
update_descriptor_queue.AddBuffer(buffer, offset, size);
@ -123,6 +140,7 @@ struct BufferCacheParams {
static constexpr bool HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT = false;
static constexpr bool NEEDS_BIND_UNIFORM_INDEX = false;
static constexpr bool NEEDS_BIND_STORAGE_INDEX = false;
static constexpr bool NEEDS_BIND_TEXTURE_BUFFER_INDEX = false;
static constexpr bool USE_MEMORY_MAPS = true;
};

30
src/video_core/renderer_vulkan/vk_compute_pipeline.cpp
View file

@ -80,8 +80,6 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
desc.is_written);
++ssbo_index;
}
buffer_cache.UpdateComputeBuffers();
buffer_cache.BindHostComputeBuffers();
texture_cache.SynchronizeComputeDescriptors();
@ -99,6 +97,10 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
const u32 raw_handle{gpu_memory.Read<u32>(addr)};
return TextureHandle(raw_handle, via_header_index);
}};
for (const auto& desc : info.texture_buffer_descriptors) {
const TextureHandle handle{read_handle(desc.cbuf_index, desc.cbuf_offset)};
image_view_indices.push_back(handle.image);
}
for (const auto& desc : info.texture_descriptors) {
const TextureHandle handle{read_handle(desc.cbuf_index, desc.cbuf_offset)};
image_view_indices.push_back(handle.image);
@ -106,16 +108,26 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
Sampler* const sampler = texture_cache.GetComputeSampler(handle.sampler);
samplers.push_back(sampler->Handle());
}
for (const auto& desc : info.texture_buffer_descriptors) {
const TextureHandle handle{read_handle(desc.cbuf_index, desc.cbuf_offset)};
image_view_indices.push_back(handle.image);
}
const std::span indices_span(image_view_indices.data(), image_view_indices.size());
texture_cache.FillComputeImageViews(indices_span, image_view_ids);
size_t image_index{};
PushImageDescriptors(info, samplers.data(), image_view_ids.data(), texture_cache,
update_descriptor_queue, image_index);
buffer_cache.UnbindComputeTextureBuffers();
ImageId* texture_buffer_ids{image_view_ids.data()};
size_t index{};
for (const auto& desc : info.texture_buffer_descriptors) {
ASSERT(desc.count == 1);
ImageView& image_view = texture_cache.GetImageView(*texture_buffer_ids);
buffer_cache.BindComputeTextureBuffer(index, image_view.GpuAddr(), image_view.BufferSize(),
image_view.format);
++texture_buffer_ids;
++index;
}
buffer_cache.UpdateComputeBuffers();
buffer_cache.BindHostComputeBuffers();
const VkSampler* samplers_it{samplers.data()};
const ImageId* views_it{image_view_ids.data()};
PushImageDescriptors(info, samplers_it, views_it, texture_cache, update_descriptor_queue);
if (!is_built.load(std::memory_order::relaxed)) {
// Wait for the pipeline to be built

33
src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
View file

@ -175,6 +175,10 @@ void GraphicsPipeline::Configure(bool is_indexed) {
const u32 raw_handle{gpu_memory.Read<u32>(addr)};
return TextureHandle(raw_handle, via_header_index);
}};
for (const auto& desc : info.texture_buffer_descriptors) {
const TextureHandle handle{read_handle(desc.cbuf_index, desc.cbuf_offset)};
image_view_indices.push_back(handle.image);
}
for (const auto& desc : info.texture_descriptors) {
const TextureHandle handle{read_handle(desc.cbuf_index, desc.cbuf_offset)};
image_view_indices.push_back(handle.image);
@ -182,24 +186,37 @@ void GraphicsPipeline::Configure(bool is_indexed) {
Sampler* const sampler{texture_cache.GetGraphicsSampler(handle.sampler)};
samplers.push_back(sampler->Handle());
}
for (const auto& desc : info.texture_buffer_descriptors) {
const TextureHandle handle{read_handle(desc.cbuf_index, desc.cbuf_offset)};
image_view_indices.push_back(handle.image);
}
}
const std::span indices_span(image_view_indices.data(), image_view_indices.size());
buffer_cache.UpdateGraphicsBuffers(is_indexed);
texture_cache.FillGraphicsImageViews(indices_span, image_view_ids);
ImageId* texture_buffer_index{image_view_ids.data()};
for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
const Shader::Info& info{stage_infos[stage]};
buffer_cache.UnbindGraphicsTextureBuffers(stage);
size_t index{};
for (const auto& desc : info.texture_buffer_descriptors) {
ASSERT(desc.count == 1);
ImageView& image_view = texture_cache.GetImageView(*texture_buffer_index);
buffer_cache.BindGraphicsTextureBuffer(stage, index, image_view.GpuAddr(),
image_view.BufferSize(), image_view.format);
++index;
++texture_buffer_index;
}
texture_buffer_index += info.texture_descriptors.size();
}
buffer_cache.UpdateGraphicsBuffers(is_indexed);
buffer_cache.BindHostGeometryBuffers(is_indexed);
update_descriptor_queue.Acquire();
size_t index{};
const VkSampler* samplers_it{samplers.data()};
const ImageId* views_it{image_view_ids.data()};
for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
buffer_cache.BindHostStageBuffers(stage);
PushImageDescriptors(stage_infos[stage], samplers.data(), image_view_ids.data(),
texture_cache, update_descriptor_queue, index);
PushImageDescriptors(stage_infos[stage], samplers_it, views_it, texture_cache,
update_descriptor_queue);
}
texture_cache.UpdateRenderTargets(false);
scheduler.RequestRenderpass(texture_cache.GetFramebuffer());

63
src/video_core/renderer_vulkan/vk_texture_cache.cpp
View file

@ -15,10 +15,10 @@
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_compute_pass.h"
#include "video_core/renderer_vulkan/vk_rasterizer.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_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/vk_render_pass_cache.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@ -162,25 +162,6 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
return device.GetLogical().CreateImage(MakeImageCreateInfo(device, info));
}
[[nodiscard]] vk::Buffer MakeBuffer(const Device& device, const ImageInfo& info) {
if (info.type != ImageType::Buffer) {
return vk::Buffer{};
}
const size_t bytes_per_block = VideoCore::Surface::BytesPerBlock(info.format);
return device.GetLogical().CreateBuffer(VkBufferCreateInfo{
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.size = info.size.width * bytes_per_block,
.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT |
VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
});
}
[[nodiscard]] VkImageAspectFlags ImageAspectMask(PixelFormat format) {
switch (VideoCore::Surface::GetFormatType(format)) {
case VideoCore::Surface::SurfaceType::ColorTexture:
@ -813,13 +794,9 @@ u64 TextureCacheRuntime::GetDeviceLocalMemory() const {
Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_addr_,
VAddr cpu_addr_)
: VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_), scheduler{&runtime.scheduler},
image(MakeImage(runtime.device, info)), buffer(MakeBuffer(runtime.device, info)),
image(MakeImage(runtime.device, info)),
commit(runtime.memory_allocator.Commit(image, MemoryUsage::DeviceLocal)),
aspect_mask(ImageAspectMask(info.format)) {
if (image) {
commit = runtime.memory_allocator.Commit(image, MemoryUsage::DeviceLocal);
} else {
commit = runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal);
}
if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) {
if (Settings::values.accelerate_astc.GetValue()) {
flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
@ -828,11 +805,7 @@ Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_
}
}
if (runtime.device.HasDebuggingToolAttached()) {
if (image) {
image.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
} else {
buffer.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
}
image.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
}
static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO,
@ -884,19 +857,6 @@ void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImag
});
}
void Image::UploadMemory(const StagingBufferRef& map,
std::span<const VideoCommon::BufferCopy> copies) {
// TODO: Move this to another API
scheduler->RequestOutsideRenderPassOperationContext();
std::vector vk_copies = TransformBufferCopies(copies, map.offset);
const VkBuffer src_buffer = map.buffer;
const VkBuffer dst_buffer = *buffer;
scheduler->Record([src_buffer, dst_buffer, vk_copies](vk::CommandBuffer cmdbuf) {
// TODO: Barriers
cmdbuf.CopyBuffer(src_buffer, dst_buffer, vk_copies);
});
}
void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) {
std::vector vk_copies = TransformBufferImageCopies(copies, map.offset, aspect_mask);
scheduler->RequestOutsideRenderPassOperationContext();
@ -1032,19 +992,16 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
UNIMPLEMENTED();
break;
case VideoCommon::ImageViewType::Buffer:
buffer_view = device->GetLogical().CreateBufferView(VkBufferViewCreateInfo{
.sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.buffer = image.Buffer(),
.format = format_info.format,
.offset = 0, // TODO: Redesign buffer cache to support this
.range = image.guest_size_bytes,
});
UNREACHABLE();
break;
}
}
ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info,
const VideoCommon::ImageViewInfo& view_info, GPUVAddr gpu_addr_)
: VideoCommon::ImageViewBase{info, view_info}, gpu_addr{gpu_addr_},
buffer_size{VideoCommon::CalculateGuestSizeInBytes(info)} {}
ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams& params)
: VideoCommon::ImageViewBase{params} {}

30
src/video_core/renderer_vulkan/vk_texture_cache.h
View file

@ -41,9 +41,9 @@ struct TextureCacheRuntime {
void Finish();
[[nodiscard]] StagingBufferRef UploadStagingBuffer(size_t size);
StagingBufferRef UploadStagingBuffer(size_t size);
[[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size);
StagingBufferRef DownloadStagingBuffer(size_t size);
void BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src,
const Region2D& dst_region, const Region2D& src_region,
@ -54,7 +54,7 @@ struct TextureCacheRuntime {
void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view);
[[nodiscard]] bool CanAccelerateImageUpload(Image&) const noexcept {
bool CanAccelerateImageUpload(Image&) const noexcept {
return false;
}
@ -92,8 +92,6 @@ public:
void UploadMemory(const StagingBufferRef& map,
std::span<const VideoCommon::BufferImageCopy> copies);
void UploadMemory(const StagingBufferRef& map, std::span<const VideoCommon::BufferCopy> copies);
void DownloadMemory(const StagingBufferRef& map,
std::span<const VideoCommon::BufferImageCopy> copies);
@ -101,10 +99,6 @@ public:
return *image;
}
[[nodiscard]] VkBuffer Buffer() const noexcept {
return *buffer;
}
[[nodiscard]] VkImageAspectFlags AspectMask() const noexcept {
return aspect_mask;
}
@ -121,7 +115,6 @@ public:
private:
VKScheduler* scheduler;
vk::Image image;
vk::Buffer buffer;
MemoryCommit commit;
vk::ImageView image_view;
std::vector<vk::ImageView> storage_image_views;
@ -132,6 +125,8 @@ private:
class ImageView : public VideoCommon::ImageViewBase {
public:
explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageViewInfo&, ImageId, Image&);
explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo&,
const VideoCommon::ImageViewInfo&, GPUVAddr);
explicit ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams&);
[[nodiscard]] VkImageView DepthView();
@ -142,10 +137,6 @@ public:
return *image_views[static_cast<size_t>(query_type)];
}
[[nodiscard]] VkBufferView BufferView() const noexcept {
return *buffer_view;
}
[[nodiscard]] VkImage ImageHandle() const noexcept {
return image_handle;
}
@ -162,6 +153,14 @@ public:
return samples;
}
[[nodiscard]] GPUVAddr GpuAddr() const noexcept {
return gpu_addr;
}
[[nodiscard]] u32 BufferSize() const noexcept {
return buffer_size;
}
private:
[[nodiscard]] vk::ImageView MakeDepthStencilView(VkImageAspectFlags aspect_mask);
@ -169,11 +168,12 @@ private:
std::array<vk::ImageView, VideoCommon::NUM_IMAGE_VIEW_TYPES> image_views;
vk::ImageView depth_view;
vk::ImageView stencil_view;
vk::BufferView buffer_view;
VkImage image_handle = VK_NULL_HANDLE;
VkImageView render_target = VK_NULL_HANDLE;
PixelFormat image_format = PixelFormat::Invalid;
VkSampleCountFlagBits samples = VK_SAMPLE_COUNT_1_BIT;
GPUVAddr gpu_addr = 0;
u32 buffer_size = 0;
};
class ImageAlloc : public VideoCommon::ImageAllocBase {};