OpenPunk/yuzu
1
0
Fork 0
mirror of https://github.com/yuzu-emu/yuzu.git synced 2024-11-15 11:40:05 +00:00
Code Issues Projects Releases Wiki Activity

astc_decoder: Refactor for style and more efficient memory use

Browse Source
This commit is contained in:
ameerj 2021-03-25 16:53:51 -04:00
parent 2f30c10584
commit 2f83d9a61b
9 changed files with 504 additions and 2258 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
src/video_core/CMakeLists.txt
View File

@ -236,7 +236,6 @@ add_library(video_core STATIC
texture_cache/types.h
texture_cache/util.cpp
texture_cache/util.h
textures/astc.cpp
textures/astc.h
textures/decoders.cpp
textures/decoders.h

569
src/video_core/host_shaders/astc_decoder.comp
View File

File diff suppressed because it is too large Load Diff

11
src/video_core/renderer_opengl/gl_texture_cache.cpp
View File

@ -307,7 +307,7 @@ void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4
[[nodiscard]] bool CanBeAccelerated(const TextureCacheRuntime& runtime,
const VideoCommon::ImageInfo& info) {
return (!runtime.HasNativeASTC() && IsPixelFormatASTC(info.format));
return !runtime.HasNativeASTC() && IsPixelFormatASTC(info.format);
// Disable other accelerated uploads for now as they don't implement swizzled uploads
return false;
switch (info.type) {
@ -568,12 +568,13 @@ void TextureCacheRuntime::BlitFramebuffer(Framebuffer* dst, Framebuffer* src,
void TextureCacheRuntime::AccelerateImageUpload(Image& image, const ImageBufferMap& map,
std::span<const SwizzleParameters> swizzles) {
if (IsPixelFormatASTC(image.info.format)) {
return util_shaders.ASTCDecode(image, map, swizzles);
}
switch (image.info.type) {
case ImageType::e2D:
return util_shaders.BlockLinearUpload2D(image, map, swizzles);
if (IsPixelFormatASTC(image.info.format)) {
return util_shaders.ASTCDecode(image, map, swizzles);
} else {
return util_shaders.BlockLinearUpload2D(image, map, swizzles);
}
case ImageType::e3D:
return util_shaders.BlockLinearUpload3D(image, map, swizzles);
case ImageType::Linear:

102
src/video_core/renderer_opengl/util_shaders.cpp
View File

@ -2,11 +2,7 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <bit>
#include <fstream>
#include <span>
#include <streambuf>
#include <string>
#include <string_view>
#include <glad/glad.h>
@ -24,7 +20,6 @@
#include "video_core/renderer_opengl/gl_shader_manager.h"
#include "video_core/renderer_opengl/gl_texture_cache.h"
#include "video_core/renderer_opengl/util_shaders.h"
#include "video_core/surface.h"
#include "video_core/texture_cache/accelerated_swizzle.h"
#include "video_core/texture_cache/types.h"
#include "video_core/texture_cache/util.h"
@ -36,6 +31,7 @@ namespace OpenGL {
using namespace HostShaders;
using namespace Tegra::Texture::ASTC;
using VideoCommon::Extent2D;
using VideoCommon::Extent3D;
using VideoCommon::ImageCopy;
using VideoCommon::ImageType;
@ -69,33 +65,15 @@ UtilShaders::UtilShaders(ProgramManager& program_manager_)
pitch_unswizzle_program(MakeProgram(PITCH_UNSWIZZLE_COMP)),
copy_bgra_program(MakeProgram(OPENGL_COPY_BGRA_COMP)),
copy_bc4_program(MakeProgram(OPENGL_COPY_BC4_COMP)) {
MakeBuffers();
const auto swizzle_table = Tegra::Texture::MakeSwizzleTable();
swizzle_table_buffer.Create();
astc_buffer.Create();
glNamedBufferStorage(swizzle_table_buffer.handle, sizeof(swizzle_table), &swizzle_table, 0);
glNamedBufferStorage(astc_buffer.handle, sizeof(ASTC_BUFFER_DATA), &ASTC_BUFFER_DATA, 0);
}
UtilShaders::~UtilShaders() = default;
void UtilShaders::MakeBuffers() {
const auto swizzle_table = Tegra::Texture::MakeSwizzleTable();
swizzle_table_buffer.Create();
glNamedBufferStorage(swizzle_table_buffer.handle, sizeof(swizzle_table), &swizzle_table, 0);
astc_encodings_buffer.Create();
glNamedBufferStorage(astc_encodings_buffer.handle, sizeof(EncodingsValues), &EncodingsValues,
0);
replicate_6_to_8_buffer.Create();
glNamedBufferStorage(replicate_6_to_8_buffer.handle, sizeof(REPLICATE_6_BIT_TO_8_TABLE),
&REPLICATE_6_BIT_TO_8_TABLE, 0);
replicate_7_to_8_buffer.Create();
glNamedBufferStorage(replicate_7_to_8_buffer.handle, sizeof(REPLICATE_7_BIT_TO_8_TABLE),
&REPLICATE_7_BIT_TO_8_TABLE, 0);
replicate_8_to_8_buffer.Create();
glNamedBufferStorage(replicate_8_to_8_buffer.handle, sizeof(REPLICATE_8_BIT_TO_8_TABLE),
&REPLICATE_8_BIT_TO_8_TABLE, 0);
replicate_byte_to_16_buffer.Create();
glNamedBufferStorage(replicate_byte_to_16_buffer.handle, sizeof(REPLICATE_BYTE_TO_16_TABLE),
&REPLICATE_BYTE_TO_16_TABLE, 0);
}
void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map,
std::span<const VideoCommon::SwizzleParameters> swizzles) {
static constexpr GLuint BINDING_SWIZZLE_BUFFER = 0;
@ -108,47 +86,51 @@ void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map,
static constexpr GLuint BINDING_BYTE_TO_16_BUFFER = 6;
static constexpr GLuint BINDING_OUTPUT_IMAGE = 0;
static constexpr GLuint LOC_NUM_IMAGE_BLOCKS = 0;
static constexpr GLuint LOC_BLOCK_DIMS = 1;
const Extent3D tile_size = {
VideoCore::Surface::DefaultBlockWidth(image.info.format),
VideoCore::Surface::DefaultBlockHeight(image.info.format),
const Extent2D tile_size{
.width = VideoCore::Surface::DefaultBlockWidth(image.info.format),
.height = VideoCore::Surface::DefaultBlockHeight(image.info.format),
};
program_manager.BindHostCompute(astc_decoder_program.handle);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_ENC_BUFFER, astc_encodings_buffer.handle);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_6_TO_8_BUFFER,
replicate_6_to_8_buffer.handle);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_7_TO_8_BUFFER,
replicate_7_to_8_buffer.handle);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_8_TO_8_BUFFER,
replicate_8_to_8_buffer.handle);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_BYTE_TO_16_BUFFER,
replicate_byte_to_16_buffer.handle);
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_ENC_BUFFER, astc_buffer.handle,
offsetof(AstcBufferData, encoding_values),
sizeof(AstcBufferData::encoding_values));
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_6_TO_8_BUFFER, astc_buffer.handle,
offsetof(AstcBufferData, replicate_6_to_8),
sizeof(AstcBufferData::replicate_6_to_8));
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_7_TO_8_BUFFER, astc_buffer.handle,
offsetof(AstcBufferData, replicate_7_to_8),
sizeof(AstcBufferData::replicate_7_to_8));
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_8_TO_8_BUFFER, astc_buffer.handle,
offsetof(AstcBufferData, replicate_8_to_8),
sizeof(AstcBufferData::replicate_8_to_8));
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_BYTE_TO_16_BUFFER, astc_buffer.handle,
offsetof(AstcBufferData, replicate_byte_to_16),
sizeof(AstcBufferData::replicate_byte_to_16));
glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes);
glUniform2ui(LOC_BLOCK_DIMS, tile_size.width, tile_size.height);
glUniform2ui(1, tile_size.width, tile_size.height);
// Ensure buffer data is valid before dispatching
glFlush();
for (const SwizzleParameters& swizzle : swizzles) {
const size_t input_offset = swizzle.buffer_offset + map.offset;
const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U);
const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 32U);
const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
ASSERT(params.origin == (std::array<u32, 3>{0, 0, 0}));
ASSERT(params.destination == (std::array<s32, 3>{0, 0, 0}));
glUniform1ui(2, params.bytes_per_block_log2);
glUniform1ui(3, params.layer_stride);
glUniform1ui(4, params.block_size);
glUniform1ui(5, params.x_shift);
glUniform1ui(6, params.block_height);
glUniform1ui(7, params.block_height_mask);
glBindImageTexture(BINDING_OUTPUT_IMAGE, image.StorageHandle(), swizzle.level, GL_TRUE, 0,
GL_WRITE_ONLY, GL_RGBA8);
const size_t input_offset = swizzle.buffer_offset + map.offset;
const auto num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U);
const auto num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 32U);
glUniform2ui(LOC_NUM_IMAGE_BLOCKS, swizzle.num_tiles.width, swizzle.num_tiles.height);
// To unswizzle the ASTC data
const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
glUniform3uiv(2, 1, params.origin.data());
glUniform3iv(3, 1, params.destination.data());
glUniform1ui(4, params.bytes_per_block_log2);
glUniform1ui(5, params.layer_stride);
glUniform1ui(6, params.block_size);
glUniform1ui(7, params.x_shift);
glUniform1ui(8, params.block_height);
glUniform1ui(9, params.block_height_mask);
// ASTC texture data
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.buffer, input_offset,
image.guest_size_bytes - swizzle.buffer_offset);

8
src/video_core/renderer_opengl/util_shaders.h
View File

@ -40,8 +40,6 @@ public:
explicit UtilShaders(ProgramManager& program_manager);
~UtilShaders();
void MakeBuffers();
void ASTCDecode(Image& image, const ImageBufferMap& map,
std::span<const VideoCommon::SwizzleParameters> swizzles);
@ -64,11 +62,7 @@ private:
ProgramManager& program_manager;
OGLBuffer swizzle_table_buffer;
OGLBuffer astc_encodings_buffer;
OGLBuffer replicate_6_to_8_buffer;
OGLBuffer replicate_7_to_8_buffer;
OGLBuffer replicate_8_to_8_buffer;
OGLBuffer replicate_byte_to_16_buffer;
OGLBuffer astc_buffer;
OGLProgram astc_decoder_program;
OGLProgram block_linear_unswizzle_2d_program;

175
src/video_core/renderer_vulkan/vk_compute_pass.cpp
View File

@ -35,13 +35,13 @@ using namespace Tegra::Texture::ASTC;
namespace {
constexpr u32 ASTC_BINDING_SWIZZLE_BUFFER = 0;
constexpr u32 ASTC_BINDING_INPUT_BUFFER = 1;
constexpr u32 ASTC_BINDING_ENC_BUFFER = 2;
constexpr u32 ASTC_BINDING_6_TO_8_BUFFER = 3;
constexpr u32 ASTC_BINDING_7_TO_8_BUFFER = 4;
constexpr u32 ASTC_BINDING_8_TO_8_BUFFER = 5;
constexpr u32 ASTC_BINDING_BYTE_TO_16_BUFFER = 6;
constexpr u32 ASTC_BINDING_INPUT_BUFFER = 0;
constexpr u32 ASTC_BINDING_ENC_BUFFER = 1;
constexpr u32 ASTC_BINDING_6_TO_8_BUFFER = 2;
constexpr u32 ASTC_BINDING_7_TO_8_BUFFER = 3;
constexpr u32 ASTC_BINDING_8_TO_8_BUFFER = 4;
constexpr u32 ASTC_BINDING_BYTE_TO_16_BUFFER = 5;
constexpr u32 ASTC_BINDING_SWIZZLE_BUFFER = 6;
constexpr u32 ASTC_BINDING_OUTPUT_IMAGE = 7;
VkPushConstantRange BuildComputePushConstantRange(std::size_t size) {
@ -74,56 +74,56 @@ std::array<VkDescriptorSetLayoutBinding, 2> BuildInputOutputDescriptorSetBinding
std::array<VkDescriptorSetLayoutBinding, 8> BuildASTCDescriptorSetBindings() {
return {{
{
.binding = ASTC_BINDING_SWIZZLE_BUFFER, // Swizzle buffer
.binding = ASTC_BINDING_INPUT_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_INPUT_BUFFER, // ASTC Img data buffer
.binding = ASTC_BINDING_ENC_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_ENC_BUFFER, // Encodings buffer
.binding = ASTC_BINDING_6_TO_8_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_6_TO_8_BUFFER, // BINDING_6_TO_8_BUFFER
.binding = ASTC_BINDING_7_TO_8_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_7_TO_8_BUFFER, // BINDING_7_TO_8_BUFFER
.binding = ASTC_BINDING_8_TO_8_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_8_TO_8_BUFFER, // BINDING_8_TO_8_BUFFER
.binding = ASTC_BINDING_BYTE_TO_16_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_BYTE_TO_16_BUFFER, // BINDING_BYTE_TO_16_BUFFER
.binding = ASTC_BINDING_SWIZZLE_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_OUTPUT_IMAGE, // Output image
.binding = ASTC_BINDING_OUTPUT_IMAGE,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
@ -145,20 +145,12 @@ VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() {
std::array<VkDescriptorUpdateTemplateEntryKHR, 8> BuildASTCPassDescriptorUpdateTemplateEntry() {
return {{
{
.dstBinding = ASTC_BINDING_SWIZZLE_BUFFER,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = 0 * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
.dstBinding = ASTC_BINDING_INPUT_BUFFER,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = 1 * sizeof(DescriptorUpdateEntry),
.offset = ASTC_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
@ -166,7 +158,7 @@ std::array<VkDescriptorUpdateTemplateEntryKHR, 8> BuildASTCPassDescriptorUpdateT
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = 2 * sizeof(DescriptorUpdateEntry),
.offset = ASTC_BINDING_ENC_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
@ -174,7 +166,7 @@ std::array<VkDescriptorUpdateTemplateEntryKHR, 8> BuildASTCPassDescriptorUpdateT
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = 3 * sizeof(DescriptorUpdateEntry),
.offset = ASTC_BINDING_6_TO_8_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
@ -182,7 +174,7 @@ std::array<VkDescriptorUpdateTemplateEntryKHR, 8> BuildASTCPassDescriptorUpdateT
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = 4 * sizeof(DescriptorUpdateEntry),
.offset = ASTC_BINDING_7_TO_8_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
@ -190,7 +182,7 @@ std::array<VkDescriptorUpdateTemplateEntryKHR, 8> BuildASTCPassDescriptorUpdateT
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = 5 * sizeof(DescriptorUpdateEntry),
.offset = ASTC_BINDING_8_TO_8_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
@ -198,7 +190,15 @@ std::array<VkDescriptorUpdateTemplateEntryKHR, 8> BuildASTCPassDescriptorUpdateT
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = 6 * sizeof(DescriptorUpdateEntry),
.offset = ASTC_BINDING_BYTE_TO_16_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
.dstBinding = ASTC_BINDING_SWIZZLE_BUFFER,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = ASTC_BINDING_SWIZZLE_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
@ -206,16 +206,20 @@ std::array<VkDescriptorUpdateTemplateEntryKHR, 8> BuildASTCPassDescriptorUpdateT
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.offset = 7 * sizeof(DescriptorUpdateEntry),
.offset = ASTC_BINDING_OUTPUT_IMAGE * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
}};
}
struct AstcPushConstants {
std::array<u32, 2> num_image_blocks;
std::array<u32, 2> blocks_dims;
VideoCommon::Accelerated::BlockLinearSwizzle2DParams params;
u32 bytes_per_block_log2;
u32 layer_stride;
u32 block_size;
u32 x_shift;
u32 block_height;
u32 block_height_mask;
};
struct AstcBufferData {
@ -419,11 +423,12 @@ ASTCDecoderPass::ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_,
ASTCDecoderPass::~ASTCDecoderPass() = default;
void ASTCDecoderPass::MakeDataBuffer() {
constexpr size_t TOTAL_BUFFER_SIZE = sizeof(ASTC_BUFFER_DATA) + sizeof(SWIZZLE_TABLE);
data_buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.size = sizeof(ASTC_BUFFER_DATA),
.size = TOTAL_BUFFER_SIZE,
.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
@ -431,15 +436,19 @@ void ASTCDecoderPass::MakeDataBuffer() {
});
data_buffer_commit = memory_allocator.Commit(data_buffer, MemoryUsage::Upload);
const auto staging_ref =
staging_buffer_pool.Request(sizeof(ASTC_BUFFER_DATA), MemoryUsage::Upload);
const auto staging_ref = staging_buffer_pool.Request(TOTAL_BUFFER_SIZE, MemoryUsage::Upload);
std::memcpy(staging_ref.mapped_span.data(), &ASTC_BUFFER_DATA, sizeof(ASTC_BUFFER_DATA));
scheduler.Record([src = staging_ref.buffer, dst = *data_buffer](vk::CommandBuffer cmdbuf) {
// Tack on the swizzle table at the end of the buffer
std::memcpy(staging_ref.mapped_span.data() + sizeof(ASTC_BUFFER_DATA), &SWIZZLE_TABLE,
sizeof(SWIZZLE_TABLE));
scheduler.Record([src = staging_ref.buffer, offset = staging_ref.offset, dst = *data_buffer,
TOTAL_BUFFER_SIZE](vk::CommandBuffer cmdbuf) {
cmdbuf.CopyBuffer(src, dst,
VkBufferCopy{
.srcOffset = 0,
.srcOffset = offset,
.dstOffset = 0,
.size = sizeof(ASTC_BUFFER_DATA),
.size = TOTAL_BUFFER_SIZE,
});
cmdbuf.PipelineBarrier(
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0,
@ -448,61 +457,58 @@ void ASTCDecoderPass::MakeDataBuffer() {
.pNext = nullptr,
.srcAccessMask = 0,
.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
},
{}, {});
});
});
}
void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
std::span<const VideoCommon::SwizzleParameters> swizzles) {
using namespace VideoCommon::Accelerated;
const VideoCommon::Extent2D tile_size{
.width = VideoCore::Surface::DefaultBlockWidth(image.info.format),
.height = VideoCore::Surface::DefaultBlockHeight(image.info.format),
const std::array<u32, 2> block_dims{
VideoCore::Surface::DefaultBlockWidth(image.info.format),
VideoCore::Surface::DefaultBlockHeight(image.info.format),
};
scheduler.RequestOutsideRenderPassOperationContext();
if (!data_buffer) {
MakeDataBuffer();
}
const VkPipeline vk_pipeline = *pipeline;
const VkImageAspectFlags aspect_mask = image.AspectMask();
const VkImage vk_image = image.Handle();
const bool is_initialized = image.ExchangeInitialization();
scheduler.Record([vk_image, aspect_mask, is_initialized](vk::CommandBuffer cmdbuf) {
const VkImageMemoryBarrier image_barrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT,
.oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
.newLayout = VK_IMAGE_LAYOUT_GENERAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = vk_image,
.subresourceRange{
.aspectMask = aspect_mask,
.baseMipLevel = 0,
.levelCount = VK_REMAINING_MIP_LEVELS,
.baseArrayLayer = 0,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
};
cmdbuf.PipelineBarrier(0, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier);
});
const std::array<u32, 2> block_dims{tile_size.width, tile_size.height};
scheduler.Record(
[vk_pipeline, vk_image, aspect_mask, is_initialized](vk::CommandBuffer cmdbuf) {
const VkImageMemoryBarrier image_barrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT,
.oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
.newLayout = VK_IMAGE_LAYOUT_GENERAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = vk_image,
.subresourceRange{
.aspectMask = aspect_mask,
.baseMipLevel = 0,
.levelCount = VK_REMAINING_MIP_LEVELS,
.baseArrayLayer = 0,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
};
cmdbuf.PipelineBarrier(is_initialized ? VK_PIPELINE_STAGE_ALL_COMMANDS_BIT : 0,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline);
});
for (const VideoCommon::SwizzleParameters& swizzle : swizzles) {
const size_t input_offset = swizzle.buffer_offset + map.offset;
const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U);
const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 32U);
const u32 num_dispatches_z = image.info.resources.layers;
const std::array num_image_blocks{swizzle.num_tiles.width, swizzle.num_tiles.height};
const u32 layer_image_size =
image.guest_size_bytes - static_cast<u32>(swizzle.buffer_offset);
update_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(*data_buffer,
offsetof(AstcBufferData, swizzle_table_buffer),
sizeof(AstcBufferData::swizzle_table_buffer));
update_descriptor_queue.AddBuffer(map.buffer, input_offset, layer_image_size);
update_descriptor_queue.AddBuffer(map.buffer, input_offset,
image.guest_size_bytes - swizzle.buffer_offset);
update_descriptor_queue.AddBuffer(*data_buffer, offsetof(AstcBufferData, encoding_values),
sizeof(AstcBufferData::encoding_values));
update_descriptor_queue.AddBuffer(*data_buffer, offsetof(AstcBufferData, replicate_6_to_8),
@ -514,18 +520,28 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
update_descriptor_queue.AddBuffer(*data_buffer,
offsetof(AstcBufferData, replicate_byte_to_16),
sizeof(AstcBufferData::replicate_byte_to_16));
update_descriptor_queue.AddBuffer(*data_buffer, sizeof(AstcBufferData),
sizeof(SWIZZLE_TABLE));
update_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
const VkPipelineLayout vk_layout = *layout;
const VkPipeline vk_pipeline = *pipeline;
// To unswizzle the ASTC data
const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
scheduler.Record([vk_layout, vk_pipeline, buffer = map.buffer, num_dispatches_x,
num_dispatches_y, num_dispatches_z, num_image_blocks, block_dims, params,
set, input_offset](vk::CommandBuffer cmdbuf) {
const AstcPushConstants uniforms{num_image_blocks, block_dims, params};
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline);
ASSERT(params.origin == (std::array<u32, 3>{0, 0, 0}));
ASSERT(params.destination == (std::array<s32, 3>{0, 0, 0}));
scheduler.Record([vk_layout, num_dispatches_x, num_dispatches_y, num_dispatches_z,
block_dims, params, set](vk::CommandBuffer cmdbuf) {
const AstcPushConstants uniforms{
.blocks_dims = block_dims,
.bytes_per_block_log2 = params.bytes_per_block_log2,
.layer_stride = params.layer_stride,
.block_size = params.block_size,
.x_shift = params.x_shift,
.block_height = params.block_height,
.block_height_mask = params.block_height_mask,
};
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, vk_layout, 0, set, {});
cmdbuf.PushConstants(vk_layout, VK_SHADER_STAGE_COMPUTE_BIT, uniforms);
cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_dispatches_z);
@ -550,7 +566,8 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
};
cmdbuf.PipelineBarrier(0, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier);
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, image_barrier);
});
}

14
src/video_core/texture_cache/util.cpp
View File

@ -47,7 +47,6 @@
#include "video_core/texture_cache/formatter.h"
#include "video_core/texture_cache/samples_helper.h"
#include "video_core/texture_cache/util.h"
#include "video_core/textures/astc.h"
#include "video_core/textures/decoders.h"
namespace VideoCommon {
@ -879,17 +878,8 @@ void ConvertImage(std::span<const u8> input, const ImageInfo& info, std::span<u8
ASSERT(copy.image_extent == mip_size);
ASSERT(copy.buffer_row_length == Common::AlignUp(mip_size.width, tile_size.width));
ASSERT(copy.buffer_image_height == Common::AlignUp(mip_size.height, tile_size.height));
if (IsPixelFormatASTC(info.format)) {
ASSERT(copy.image_extent.depth == 1);
Tegra::Texture::ASTC::Decompress(input.subspan(copy.buffer_offset),
copy.image_extent.width, copy.image_extent.height,
copy.image_subresource.num_layers, tile_size.width,
tile_size.height, output.subspan(output_offset));
} else {
DecompressBC4(input.subspan(copy.buffer_offset), copy.image_extent,
output.subspan(output_offset));
}
DecompressBC4(input.subspan(copy.buffer_offset), copy.image_extent,
output.subspan(output_offset));
copy.buffer_offset = output_offset;
copy.buffer_row_length = mip_size.width;
copy.buffer_image_height = mip_size.height;

1710
src/video_core/textures/astc.cpp
View File

File diff suppressed because it is too large Load Diff

172
src/video_core/textures/astc.h
View File

@ -4,20 +4,12 @@
#pragma once
#include <cstdint>
#include <bit>
#include "common/common_types.h"
namespace Tegra::Texture::ASTC {
/// Count the number of bits set in a number.
constexpr u32 Popcnt(u32 n) {
u32 c = 0;
for (; n; c++) {
n &= n - 1;
}
return c;
}
enum class IntegerEncoding { JustBits, Qus32, Trit };
enum class IntegerEncoding { JustBits, Quint, Trit };
struct IntegerEncodedValue {
constexpr IntegerEncodedValue() = default;
@ -29,55 +21,55 @@ struct IntegerEncodedValue {
return encoding == other.encoding && num_bits == other.num_bits;
}
// Returns the number of bits required to encode nVals values.
u32 GetBitLength(u32 nVals) const {
u32 totalBits = num_bits * nVals;
// Returns the number of bits required to encode num_vals values.
u32 GetBitLength(u32 num_vals) const {
u32 total_bits = num_bits * num_vals;
if (encoding == IntegerEncoding::Trit) {
totalBits += (nVals * 8 + 4) / 5;
} else if (encoding == IntegerEncoding::Qus32) {
totalBits += (nVals * 7 + 2) / 3;
total_bits += (num_vals * 8 + 4) / 5;
} else if (encoding == IntegerEncoding::Quint) {
total_bits += (num_vals * 7 + 2) / 3;
}
return totalBits;
return total_bits;
}
IntegerEncoding encoding{};
u32 num_bits = 0;
u32 bit_value = 0;
union {
u32 qus32_value = 0;
u32 quint_value = 0;
u32 trit_value;
};
};
// Returns a new instance of this struct that corresponds to the
// can take no more than maxval values
static constexpr IntegerEncodedValue CreateEncoding(u32 maxVal) {
while (maxVal > 0) {
u32 check = maxVal + 1;
// can take no more than mav_value values
constexpr IntegerEncodedValue CreateEncoding(u32 mav_value) {
while (mav_value > 0) {
u32 check = mav_value + 1;
// Is maxVal a power of two?
// Is mav_value a power of two?
if (!(check & (check - 1))) {
return IntegerEncodedValue(IntegerEncoding::JustBits, Popcnt(maxVal));
return IntegerEncodedValue(IntegerEncoding::JustBits, std::popcount(mav_value));
}
// Is maxVal of the type 3*2^n - 1?
// Is mav_value of the type 3*2^n - 1?
if ((check % 3 == 0) && !((check / 3) & ((check / 3) - 1))) {
return IntegerEncodedValue(IntegerEncoding::Trit, Popcnt(check / 3 - 1));
return IntegerEncodedValue(IntegerEncoding::Trit, std::popcount(check / 3 - 1));
}
// Is maxVal of the type 5*2^n - 1?
// Is mav_value of the type 5*2^n - 1?
if ((check % 5 == 0) && !((check / 5) & ((check / 5) - 1))) {
return IntegerEncodedValue(IntegerEncoding::Qus32, Popcnt(check / 5 - 1));
return IntegerEncodedValue(IntegerEncoding::Quint, std::popcount(check / 5 - 1));
}
// Apparently it can't be represented with a bounded integer sequence...
// just iterate.
maxVal--;
mav_value--;
}
return IntegerEncodedValue(IntegerEncoding::JustBits, 0);
}
static constexpr std::array<IntegerEncodedValue, 256> MakeEncodedValues() {
constexpr std::array<IntegerEncodedValue, 256> MakeEncodedValues() {
std::array<IntegerEncodedValue, 256> encodings{};
for (std::size_t i = 0; i < encodings.size(); ++i) {
encodings[i] = CreateEncoding(static_cast<u32>(i));
@ -85,41 +77,38 @@ static constexpr std::array<IntegerEncodedValue, 256> MakeEncodedValues() {
return encodings;
}
static constexpr std::array<IntegerEncodedValue, 256> EncodingsValues = MakeEncodedValues();
constexpr std::array<IntegerEncodedValue, 256> EncodingsValues = MakeEncodedValues();
// Replicates low numBits such that [(toBit - 1):(toBit - 1 - fromBit)]
// is the same as [(numBits - 1):0] and repeats all the way down.
// Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)]
// is the same as [(num_bits - 1):0] and repeats all the way down.
template <typename IntType>
static constexpr IntType Replicate(IntType val, u32 numBits, u32 toBit) {
if (numBits == 0) {
constexpr IntType Replicate(IntType val, u32 num_bits, u32 to_bit) {
if (num_bits == 0 || to_bit == 0) {
return 0;
}
if (toBit == 0) {
return 0;
}
const IntType v = val & static_cast<IntType>((1 << numBits) - 1);
const IntType v = val & static_cast<IntType>((1 << num_bits) - 1);
IntType res = v;
u32 reslen = numBits;
while (reslen < toBit) {
u32 reslen = num_bits;
while (reslen < to_bit) {
u32 comp = 0;
if (numBits > toBit - reslen) {
u32 newshift = toBit - reslen;
comp = numBits - newshift;
numBits = newshift;
if (num_bits > to_bit - reslen) {
u32 newshift = to_bit - reslen;
comp = num_bits - newshift;
num_bits = newshift;
}
res = static_cast<IntType>(res << numBits);
res = static_cast<IntType>(res << num_bits);
res = static_cast<IntType>(res | (v >> comp));
reslen += numBits;
reslen += num_bits;
}
return res;
}
static constexpr std::size_t NumReplicateEntries(u32 num_bits) {
constexpr std::size_t NumReplicateEntries(u32 num_bits) {
return std::size_t(1) << num_bits;
}
template <typename IntType, u32 num_bits, u32 to_bit>
static constexpr auto MakeReplicateTable() {
constexpr auto MakeReplicateTable() {
std::array<IntType, NumReplicateEntries(num_bits)> table{};
for (IntType value = 0; value < static_cast<IntType>(std::size(table)); ++value) {
table[value] = Replicate(value, num_bits, to_bit);
@ -127,78 +116,17 @@ static constexpr auto MakeReplicateTable() {
return table;
}
static constexpr auto REPLICATE_BYTE_TO_16_TABLE = MakeReplicateTable<u32, 8, 16>();
static constexpr u32 ReplicateByteTo16(std::size_t value) {
return REPLICATE_BYTE_TO_16_TABLE[value];
}
constexpr auto REPLICATE_BYTE_TO_16_TABLE = MakeReplicateTable<u32, 8, 16>();
constexpr auto REPLICATE_6_BIT_TO_8_TABLE = MakeReplicateTable<u32, 6, 8>();
constexpr auto REPLICATE_7_BIT_TO_8_TABLE = MakeReplicateTable<u32, 7, 8>();
constexpr auto REPLICATE_8_BIT_TO_8_TABLE = MakeReplicateTable<u32, 8, 8>();
static constexpr auto REPLICATE_BIT_TO_7_TABLE = MakeReplicateTable<u32, 1, 7>();
static constexpr u32 ReplicateBitTo7(std::size_t value) {
return REPLICATE_BIT_TO_7_TABLE[value];
}
static constexpr auto REPLICATE_BIT_TO_9_TABLE = MakeReplicateTable<u32, 1, 9>();
static constexpr u32 ReplicateBitTo9(std::size_t value) {
return REPLICATE_BIT_TO_9_TABLE[value];
}
static constexpr auto REPLICATE_1_BIT_TO_8_TABLE = MakeReplicateTable<u32, 1, 8>();
static constexpr auto REPLICATE_2_BIT_TO_8_TABLE = MakeReplicateTable<u32, 2, 8>();
static constexpr auto REPLICATE_3_BIT_TO_8_TABLE = MakeReplicateTable<u32, 3, 8>();
static constexpr auto REPLICATE_4_BIT_TO_8_TABLE = MakeReplicateTable<u32, 4, 8>();
static constexpr auto REPLICATE_5_BIT_TO_8_TABLE = MakeReplicateTable<u32, 5, 8>();
static constexpr auto REPLICATE_6_BIT_TO_8_TABLE = MakeReplicateTable<u32, 6, 8>();
static constexpr auto REPLICATE_7_BIT_TO_8_TABLE = MakeReplicateTable<u32, 7, 8>();
static constexpr auto REPLICATE_8_BIT_TO_8_TABLE = MakeReplicateTable<u32, 8, 8>();
/// Use a precompiled table with the most common usages, if it's not in the expected range, fallback
/// to the runtime implementation
static constexpr u32 FastReplicateTo8(u32 value, u32 num_bits) {
switch (num_bits) {
case 1:
return REPLICATE_1_BIT_TO_8_TABLE[value];
case 2:
return REPLICATE_2_BIT_TO_8_TABLE[value];
case 3:
return REPLICATE_3_BIT_TO_8_TABLE[value];
case 4:
return REPLICATE_4_BIT_TO_8_TABLE[value];
case 5:
return REPLICATE_5_BIT_TO_8_TABLE[value];
case 6:
return REPLICATE_6_BIT_TO_8_TABLE[value];
case 7:
return REPLICATE_7_BIT_TO_8_TABLE[value];
case 8:
return REPLICATE_8_BIT_TO_8_TABLE[value];
default:
return Replicate(value, num_bits, 8);
}
}
static constexpr auto REPLICATE_1_BIT_TO_6_TABLE = MakeReplicateTable<u32, 1, 6>();
static constexpr auto REPLICATE_2_BIT_TO_6_TABLE = MakeReplicateTable<u32, 2, 6>();
static constexpr auto REPLICATE_3_BIT_TO_6_TABLE = MakeReplicateTable<u32, 3, 6>();
static constexpr auto REPLICATE_4_BIT_TO_6_TABLE = MakeReplicateTable<u32, 4, 6>();
static constexpr auto REPLICATE_5_BIT_TO_6_TABLE = MakeReplicateTable<u32, 5, 6>();
static constexpr u32 FastReplicateTo6(u32 value, u32 num_bits) {
switch (num_bits) {
case 1:
return REPLICATE_1_BIT_TO_6_TABLE[value];
case 2:
return REPLICATE_2_BIT_TO_6_TABLE[value];
case 3:
return REPLICATE_3_BIT_TO_6_TABLE[value];
case 4:
return REPLICATE_4_BIT_TO_6_TABLE[value];
case 5:
return REPLICATE_5_BIT_TO_6_TABLE[value];
default:
return Replicate(value, num_bits, 6);
}
}
void Decompress(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
uint32_t block_width, uint32_t block_height, std::span<uint8_t> output);
struct AstcBufferData {
decltype(EncodingsValues) encoding_values = EncodingsValues;
decltype(REPLICATE_6_BIT_TO_8_TABLE) replicate_6_to_8 = REPLICATE_6_BIT_TO_8_TABLE;
decltype(REPLICATE_7_BIT_TO_8_TABLE) replicate_7_to_8 = REPLICATE_7_BIT_TO_8_TABLE;
decltype(REPLICATE_8_BIT_TO_8_TABLE) replicate_8_to_8 = REPLICATE_8_BIT_TO_8_TABLE;
decltype(REPLICATE_BYTE_TO_16_TABLE) replicate_byte_to_16 = REPLICATE_BYTE_TO_16_TABLE;
} constexpr ASTC_BUFFER_DATA;
} // namespace Tegra::Texture::ASTC