vk_rasterizer: Implement storage texels

This is the equivalent of an image buffer on OpenGL.

- Used by Octopath Traveler
This commit is contained in:
ReinUsesLisp 2020-06-01 21:41:07 -03:00
parent 3a6714ab7f
commit 4a6b9a1a71
8 changed files with 121 additions and 53 deletions

View File

@ -53,8 +53,9 @@ vk::DescriptorSetLayout VKComputePipeline::CreateDescriptorSetLayout() const {
};
add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, entries.const_buffers.size());
add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, entries.global_buffers.size());
add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, entries.texel_buffers.size());
add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, entries.uniform_texels.size());
add_bindings(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, entries.samplers.size());
add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, entries.storage_texels.size());
add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, entries.images.size());
VkDescriptorSetLayoutCreateInfo ci;

View File

@ -42,6 +42,7 @@ vk::DescriptorPool* VKDescriptorPool::AllocateNewPool() {
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, num_sets * 60},
{VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, num_sets * 64},
{VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, num_sets * 64},
{VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, num_sets * 64},
{VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, num_sets * 40}};
VkDescriptorPoolCreateInfo ci;

View File

@ -45,6 +45,7 @@ constexpr VkDescriptorType UNIFORM_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
constexpr VkDescriptorType STORAGE_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
constexpr VkDescriptorType UNIFORM_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER;
constexpr VkDescriptorType COMBINED_IMAGE_SAMPLER = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
constexpr VkDescriptorType STORAGE_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER;
constexpr VkDescriptorType STORAGE_IMAGE = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
constexpr VideoCommon::Shader::CompilerSettings compiler_settings{
@ -104,8 +105,9 @@ u32 FillDescriptorLayout(const ShaderEntries& entries,
u32 binding = base_binding;
AddBindings<UNIFORM_BUFFER>(bindings, binding, flags, entries.const_buffers);
AddBindings<STORAGE_BUFFER>(bindings, binding, flags, entries.global_buffers);
AddBindings<UNIFORM_TEXEL_BUFFER>(bindings, binding, flags, entries.texel_buffers);
AddBindings<UNIFORM_TEXEL_BUFFER>(bindings, binding, flags, entries.uniform_texels);
AddBindings<COMBINED_IMAGE_SAMPLER>(bindings, binding, flags, entries.samplers);
AddBindings<STORAGE_TEXEL_BUFFER>(bindings, binding, flags, entries.storage_texels);
AddBindings<STORAGE_IMAGE>(bindings, binding, flags, entries.images);
return binding;
}
@ -377,16 +379,17 @@ void AddEntry(std::vector<VkDescriptorUpdateTemplateEntry>& template_entries, u3
return;
}
if constexpr (descriptor_type == UNIFORM_TEXEL_BUFFER) {
// Nvidia has a bug where updating multiple uniform texels at once causes the driver to
// crash.
if constexpr (descriptor_type == UNIFORM_TEXEL_BUFFER ||
descriptor_type == STORAGE_TEXEL_BUFFER) {
// Nvidia has a bug where updating multiple texels at once causes the driver to crash.
// Note: Fixed in driver Windows 443.24, Linux 440.66.15
for (u32 i = 0; i < count; ++i) {
VkDescriptorUpdateTemplateEntry& entry = template_entries.emplace_back();
entry.dstBinding = binding + i;
entry.dstArrayElement = 0;
entry.descriptorCount = 1;
entry.descriptorType = descriptor_type;
entry.offset = offset + i * entry_size;
entry.offset = static_cast<std::size_t>(offset + i * entry_size);
entry.stride = entry_size;
}
} else if (count > 0) {
@ -407,8 +410,9 @@ void FillDescriptorUpdateTemplateEntries(
std::vector<VkDescriptorUpdateTemplateEntryKHR>& template_entries) {
AddEntry<UNIFORM_BUFFER>(template_entries, offset, binding, entries.const_buffers);
AddEntry<STORAGE_BUFFER>(template_entries, offset, binding, entries.global_buffers);
AddEntry<UNIFORM_TEXEL_BUFFER>(template_entries, offset, binding, entries.texel_buffers);
AddEntry<UNIFORM_TEXEL_BUFFER>(template_entries, offset, binding, entries.uniform_texels);
AddEntry<COMBINED_IMAGE_SAMPLER>(template_entries, offset, binding, entries.samplers);
AddEntry<STORAGE_TEXEL_BUFFER>(template_entries, offset, binding, entries.storage_texels);
AddEntry<STORAGE_IMAGE>(template_entries, offset, binding, entries.images);
}

View File

@ -468,8 +468,9 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
const auto& entries = pipeline.GetEntries();
SetupComputeConstBuffers(entries);
SetupComputeGlobalBuffers(entries);
SetupComputeTexelBuffers(entries);
SetupComputeUniformTexels(entries);
SetupComputeTextures(entries);
SetupComputeStorageTexels(entries);
SetupComputeImages(entries);
buffer_cache.Unmap();
@ -787,8 +788,9 @@ void RasterizerVulkan::SetupShaderDescriptors(
const auto& entries = shader->GetEntries();
SetupGraphicsConstBuffers(entries, stage);
SetupGraphicsGlobalBuffers(entries, stage);
SetupGraphicsTexelBuffers(entries, stage);
SetupGraphicsUniformTexels(entries, stage);
SetupGraphicsTextures(entries, stage);
SetupGraphicsStorageTexels(entries, stage);
SetupGraphicsImages(entries, stage);
}
texture_cache.GuardSamplers(false);
@ -976,12 +978,12 @@ void RasterizerVulkan::SetupGraphicsGlobalBuffers(const ShaderEntries& entries,
}
}
void RasterizerVulkan::SetupGraphicsTexelBuffers(const ShaderEntries& entries, std::size_t stage) {
void RasterizerVulkan::SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage) {
MICROPROFILE_SCOPE(Vulkan_Textures);
const auto& gpu = system.GPU().Maxwell3D();
for (const auto& entry : entries.texel_buffers) {
for (const auto& entry : entries.uniform_texels) {
const auto image = GetTextureInfo(gpu, entry, stage).tic;
SetupTexelBuffer(image, entry);
SetupUniformTexels(image, entry);
}
}
@ -996,6 +998,15 @@ void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, std::
}
}
void RasterizerVulkan::SetupGraphicsStorageTexels(const ShaderEntries& entries, std::size_t stage) {
MICROPROFILE_SCOPE(Vulkan_Textures);
const auto& gpu = system.GPU().Maxwell3D();
for (const auto& entry : entries.storage_texels) {
const auto image = GetTextureInfo(gpu, entry, stage).tic;
SetupStorageTexel(image, entry);
}
}
void RasterizerVulkan::SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage) {
MICROPROFILE_SCOPE(Vulkan_Images);
const auto& gpu = system.GPU().Maxwell3D();
@ -1028,12 +1039,12 @@ void RasterizerVulkan::SetupComputeGlobalBuffers(const ShaderEntries& entries) {
}
}
void RasterizerVulkan::SetupComputeTexelBuffers(const ShaderEntries& entries) {
void RasterizerVulkan::SetupComputeUniformTexels(const ShaderEntries& entries) {
MICROPROFILE_SCOPE(Vulkan_Textures);
const auto& gpu = system.GPU().KeplerCompute();
for (const auto& entry : entries.texel_buffers) {
for (const auto& entry : entries.uniform_texels) {
const auto image = GetTextureInfo(gpu, entry, ComputeShaderIndex).tic;
SetupTexelBuffer(image, entry);
SetupUniformTexels(image, entry);
}
}
@ -1048,6 +1059,15 @@ void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) {
}
}
void RasterizerVulkan::SetupComputeStorageTexels(const ShaderEntries& entries) {
MICROPROFILE_SCOPE(Vulkan_Textures);
const auto& gpu = system.GPU().KeplerCompute();
for (const auto& entry : entries.storage_texels) {
const auto image = GetTextureInfo(gpu, entry, ComputeShaderIndex).tic;
SetupStorageTexel(image, entry);
}
}
void RasterizerVulkan::SetupComputeImages(const ShaderEntries& entries) {
MICROPROFILE_SCOPE(Vulkan_Images);
const auto& gpu = system.GPU().KeplerCompute();
@ -1097,8 +1117,8 @@ void RasterizerVulkan::SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAdd
update_descriptor_queue.AddBuffer(buffer, offset, size);
}
void RasterizerVulkan::SetupTexelBuffer(const Tegra::Texture::TICEntry& tic,
const TexelBufferEntry& entry) {
void RasterizerVulkan::SetupUniformTexels(const Tegra::Texture::TICEntry& tic,
const UniformTexelEntry& entry) {
const auto view = texture_cache.GetTextureSurface(tic, entry);
ASSERT(view->IsBufferView());
@ -1120,6 +1140,14 @@ void RasterizerVulkan::SetupTexture(const Tegra::Texture::FullTextureInfo& textu
sampled_views.push_back(ImageView{std::move(view), image_layout});
}
void RasterizerVulkan::SetupStorageTexel(const Tegra::Texture::TICEntry& tic,
const StorageTexelEntry& entry) {
const auto view = texture_cache.GetImageSurface(tic, entry);
ASSERT(view->IsBufferView());
update_descriptor_queue.AddTexelBuffer(view->GetBufferView());
}
void RasterizerVulkan::SetupImage(const Tegra::Texture::TICEntry& tic, const ImageEntry& entry) {
auto view = texture_cache.GetImageSurface(tic, entry);

View File

@ -193,12 +193,15 @@ private:
/// Setup global buffers in the graphics pipeline.
void SetupGraphicsGlobalBuffers(const ShaderEntries& entries, std::size_t stage);
/// Setup texel buffers in the graphics pipeline.
void SetupGraphicsTexelBuffers(const ShaderEntries& entries, std::size_t stage);
/// Setup uniform texels in the graphics pipeline.
void SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage);
/// Setup textures in the graphics pipeline.
void SetupGraphicsTextures(const ShaderEntries& entries, std::size_t stage);
/// Setup storage texels in the graphics pipeline.
void SetupGraphicsStorageTexels(const ShaderEntries& entries, std::size_t stage);
/// Setup images in the graphics pipeline.
void SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage);
@ -209,11 +212,14 @@ private:
void SetupComputeGlobalBuffers(const ShaderEntries& entries);
/// Setup texel buffers in the compute pipeline.
void SetupComputeTexelBuffers(const ShaderEntries& entries);
void SetupComputeUniformTexels(const ShaderEntries& entries);
/// Setup textures in the compute pipeline.
void SetupComputeTextures(const ShaderEntries& entries);
/// Setup storage texels in the compute pipeline.
void SetupComputeStorageTexels(const ShaderEntries& entries);
/// Setup images in the compute pipeline.
void SetupComputeImages(const ShaderEntries& entries);
@ -222,10 +228,12 @@ private:
void SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAddr address);
void SetupTexelBuffer(const Tegra::Texture::TICEntry& image, const TexelBufferEntry& entry);
void SetupUniformTexels(const Tegra::Texture::TICEntry& image, const UniformTexelEntry& entry);
void SetupTexture(const Tegra::Texture::FullTextureInfo& texture, const SamplerEntry& entry);
void SetupStorageTexel(const Tegra::Texture::TICEntry& tic, const StorageTexelEntry& entry);
void SetupImage(const Tegra::Texture::TICEntry& tic, const ImageEntry& entry);
void UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs);

View File

@ -400,8 +400,9 @@ private:
u32 binding = specialization.base_binding;
binding = DeclareConstantBuffers(binding);
binding = DeclareGlobalBuffers(binding);
binding = DeclareTexelBuffers(binding);
binding = DeclareUniformTexels(binding);
binding = DeclareSamplers(binding);
binding = DeclareStorageTexels(binding);
binding = DeclareImages(binding);
const Id main = OpFunction(t_void, {}, TypeFunction(t_void));
@ -889,7 +890,7 @@ private:
return binding;
}
u32 DeclareTexelBuffers(u32 binding) {
u32 DeclareUniformTexels(u32 binding) {
for (const auto& sampler : ir.GetSamplers()) {
if (!sampler.is_buffer) {
continue;
@ -910,7 +911,7 @@ private:
Decorate(id, spv::Decoration::Binding, binding++);
Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET);
texel_buffers.emplace(sampler.index, TexelBuffer{image_type, id});
uniform_texels.emplace(sampler.index, TexelBuffer{image_type, id});
}
return binding;
}
@ -945,8 +946,27 @@ private:
return binding;
}
u32 DeclareStorageTexels(u32 binding) {
for (const auto& image : ir.GetImages()) {
if (image.type != Tegra::Shader::ImageType::TextureBuffer) {
continue;
}
DeclareImage(image, binding);
}
return binding;
}
u32 DeclareImages(u32 binding) {
for (const auto& image : ir.GetImages()) {
if (image.type == Tegra::Shader::ImageType::TextureBuffer) {
continue;
}
DeclareImage(image, binding);
}
return binding;
}
void DeclareImage(const Image& image, u32& binding) {
const auto [dim, arrayed] = GetImageDim(image);
constexpr int depth = 0;
constexpr bool ms = false;
@ -967,8 +987,6 @@ private:
images.emplace(image.index, StorageImage{image_type, id});
}
return binding;
}
bool IsRenderTargetEnabled(u32 rt) const {
for (u32 component = 0; component < 4; ++component) {
@ -1674,7 +1692,7 @@ private:
const auto& meta = std::get<MetaTexture>(operation.GetMeta());
const u32 index = meta.sampler.index;
if (meta.sampler.is_buffer) {
const auto& entry = texel_buffers.at(index);
const auto& entry = uniform_texels.at(index);
return OpLoad(entry.image_type, entry.image);
} else {
const auto& entry = sampled_images.at(index);
@ -2794,15 +2812,16 @@ private:
std::unordered_map<u8, GenericVaryingDescription> output_attributes;
std::map<u32, Id> constant_buffers;
std::map<GlobalMemoryBase, Id> global_buffers;
std::map<u32, TexelBuffer> texel_buffers;
std::map<u32, TexelBuffer> uniform_texels;
std::map<u32, SampledImage> sampled_images;
std::map<u32, TexelBuffer> storage_texels;
std::map<u32, StorageImage> images;
std::array<Id, Maxwell::NumRenderTargets> frag_colors{};
Id instance_index{};
Id vertex_index{};
Id base_instance{};
Id base_vertex{};
std::array<Id, Maxwell::NumRenderTargets> frag_colors{};
Id frag_depth{};
Id frag_coord{};
Id front_facing{};
@ -3058,14 +3077,18 @@ ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir) {
}
for (const auto& sampler : ir.GetSamplers()) {
if (sampler.is_buffer) {
entries.texel_buffers.emplace_back(sampler);
entries.uniform_texels.emplace_back(sampler);
} else {
entries.samplers.emplace_back(sampler);
}
}
for (const auto& image : ir.GetImages()) {
if (image.type == Tegra::Shader::ImageType::TextureBuffer) {
entries.storage_texels.emplace_back(image);
} else {
entries.images.emplace_back(image);
}
}
for (const auto& attribute : ir.GetInputAttributes()) {
if (IsGenericAttribute(attribute)) {
entries.attributes.insert(GetGenericAttributeLocation(attribute));

View File

@ -21,8 +21,9 @@ class VKDevice;
namespace Vulkan {
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
using TexelBufferEntry = VideoCommon::Shader::Sampler;
using UniformTexelEntry = VideoCommon::Shader::Sampler;
using SamplerEntry = VideoCommon::Shader::Sampler;
using StorageTexelEntry = VideoCommon::Shader::Image;
using ImageEntry = VideoCommon::Shader::Image;
constexpr u32 DESCRIPTOR_SET = 0;
@ -66,13 +67,15 @@ private:
struct ShaderEntries {
u32 NumBindings() const {
return static_cast<u32>(const_buffers.size() + global_buffers.size() +
texel_buffers.size() + samplers.size() + images.size());
uniform_texels.size() + samplers.size() + storage_texels.size() +
images.size());
}
std::vector<ConstBufferEntry> const_buffers;
std::vector<GlobalBufferEntry> global_buffers;
std::vector<TexelBufferEntry> texel_buffers;
std::vector<UniformTexelEntry> uniform_texels;
std::vector<SamplerEntry> samplers;
std::vector<StorageTexelEntry> storage_texels;
std::vector<ImageEntry> images;
std::set<u32> attributes;
std::array<bool, Maxwell::NumClipDistances> clip_distances{};

View File

@ -100,8 +100,8 @@ vk::Buffer CreateBuffer(const VKDevice& device, const SurfaceParams& params,
ci.pNext = nullptr;
ci.flags = 0;
ci.size = static_cast<VkDeviceSize>(host_memory_size);
ci.usage = VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_SRC_BIT |
VK_BUFFER_USAGE_TRANSFER_DST_BIT;
ci.usage = VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT |
VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT;
ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
ci.queueFamilyIndexCount = 0;
ci.pQueueFamilyIndices = nullptr;