early-access version 1457

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 1456.
+This is the source code for early-access 1457.
 
 ## Legal Notice
 

src/video_core/CMakeLists.txt

@@ -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

src/video_core/host_shaders/CMakeLists.txt

@@ -1,4 +1,5 @@
 set(SHADER_FILES
+    astc_decoder.comp
     block_linear_unswizzle_2d.comp
     block_linear_unswizzle_3d.comp
     convert_depth_to_float.frag

src/video_core/host_shaders/StringShaderHeader.cmake

@@ -6,7 +6,27 @@ get_filename_component(CONTENTS_NAME ${SOURCE_FILE} NAME)
 string(REPLACE "." "_" CONTENTS_NAME ${CONTENTS_NAME})
 string(TOUPPER ${CONTENTS_NAME} CONTENTS_NAME)
 
-file(READ ${SOURCE_FILE} CONTENTS)
+FILE(READ ${SOURCE_FILE} line_contents)
+
+# Replace double quotes with single quotes,
+# as double quotes will be used to wrap the lines
+STRING(REGEX REPLACE "\"" "'" line_contents "${line_contents}")
+
+# CMake separates list elements with semicolons, but semicolons
+# are used extensively in the shader code.
+# Replace with a temporary marker, to be reverted later.
+STRING(REGEX REPLACE ";" "{{SEMICOLON}}" line_contents "${line_contents}")
+
+# Make every line an individual element in the CMake list.
+STRING(REGEX REPLACE "\n" ";" line_contents "${line_contents}")
+
+# Build the shader string, wrapping each line in double quotes.
+foreach(line IN LISTS line_contents)
+    string(CONCAT CONTENTS "${CONTENTS}" \"${line}\\n\"\n)
+endforeach()
+
+# Revert the original semicolons in the source.
+STRING(REGEX REPLACE "{{SEMICOLON}}" ";" CONTENTS "${CONTENTS}")
 
 get_filename_component(OUTPUT_DIR ${HEADER_FILE} DIRECTORY)
 make_directory(${OUTPUT_DIR})

src/video_core/host_shaders/source_shader.h.in

@@ -4,6 +4,8 @@
 
 namespace HostShaders {
 
-constexpr std::string_view @CONTENTS_NAME@ = R"(@CONTENTS@)";
+constexpr std::string_view @CONTENTS_NAME@ = {
+@CONTENTS@
+};
 
 } // namespace HostShaders

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -307,7 +307,8 @@ void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4
 
 [[nodiscard]] bool CanBeAccelerated(const TextureCacheRuntime& runtime,
                                     const VideoCommon::ImageInfo& info) {
-    // Disable accelerated uploads for now as they don't implement swizzled uploads
+    return (!runtime.HasNativeASTC() && IsPixelFormatASTC(info.format));
+    // Disable other accelerated uploads for now as they don't implement swizzled uploads
     return false;
     switch (info.type) {
     case ImageType::e2D:
@@ -579,6 +580,9 @@ 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);
@@ -611,6 +615,10 @@ FormatProperties TextureCacheRuntime::FormatInfo(ImageType type, GLenum internal
     }
 }
 
+bool TextureCacheRuntime::HasNativeASTC() const noexcept {
+    return device.HasASTC();
+}
+
 TextureCacheRuntime::StagingBuffers::StagingBuffers(GLenum storage_flags_, GLenum map_flags_)
     : storage_flags{storage_flags_}, map_flags{map_flags_} {}
 

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -90,6 +90,8 @@ public:
         return has_broken_texture_view_formats;
     }
 
+    bool HasNativeASTC() const noexcept;
+
 private:
     struct StagingBuffers {
         explicit StagingBuffers(GLenum storage_flags_, GLenum map_flags_);

src/video_core/renderer_opengl/util_shaders.cpp

@@ -2,7 +2,6 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <bit>
 #include <span>
 #include <string_view>
 
@@ -11,6 +10,7 @@
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "common/div_ceil.h"
+#include "video_core/host_shaders/astc_decoder_comp.h"
 #include "video_core/host_shaders/block_linear_unswizzle_2d_comp.h"
 #include "video_core/host_shaders/block_linear_unswizzle_3d_comp.h"
 #include "video_core/host_shaders/opengl_copy_bc4_comp.h"
@@ -21,16 +21,18 @@
 #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"
+#include "video_core/textures/astc.h"
 #include "video_core/textures/decoders.h"
 
 namespace OpenGL {
 
 using namespace HostShaders;
+using namespace Tegra::Texture::ASTC;
 
+using VideoCommon::Extent2D;
 using VideoCommon::Extent3D;
 using VideoCommon::ImageCopy;
 using VideoCommon::ImageType;
@@ -54,20 +56,90 @@ OGLProgram MakeProgram(std::string_view source) {
 } // Anonymous namespace
 
 UtilShaders::UtilShaders(ProgramManager& program_manager_)
-    : program_manager{program_manager_},
+    : program_manager{program_manager_}, astc_decoder_program(MakeProgram(ASTC_DECODER_COMP)),
       block_linear_unswizzle_2d_program(MakeProgram(BLOCK_LINEAR_UNSWIZZLE_2D_COMP)),
       block_linear_unswizzle_3d_program(MakeProgram(BLOCK_LINEAR_UNSWIZZLE_3D_COMP)),
       pitch_unswizzle_program(MakeProgram(PITCH_UNSWIZZLE_COMP)),
-      copy_bc4_program(MakeProgram(OPENGL_COPY_BC4_COMP)),
       copy_bgr16_program(MakeProgram(OPENGL_COPY_BGR16_COMP)),
-      copy_bgra_program(MakeProgram(OPENGL_COPY_BGRA_COMP)) {
+      copy_bgra_program(MakeProgram(OPENGL_COPY_BGRA_COMP)),
+      copy_bc4_program(MakeProgram(OPENGL_COPY_BC4_COMP)) {
     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::ASTCDecode(Image& image, const ImageBufferMap& map,
+                             std::span<const VideoCommon::SwizzleParameters> swizzles) {
+    static constexpr GLuint BINDING_SWIZZLE_BUFFER = 0;
+    static constexpr GLuint BINDING_INPUT_BUFFER = 1;
+    static constexpr GLuint BINDING_ENC_BUFFER = 2;
+
+    static constexpr GLuint BINDING_6_TO_8_BUFFER = 3;
+    static constexpr GLuint BINDING_7_TO_8_BUFFER = 4;
+    static constexpr GLuint BINDING_8_TO_8_BUFFER = 5;
+    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 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);
+    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);
+    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}));
+
+        glUniform2ui(LOC_NUM_IMAGE_BLOCKS, swizzle.num_tiles.width, swizzle.num_tiles.height);
+        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);
+
+        // ASTC texture data
+        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.buffer, input_offset,
+                          image.guest_size_bytes - swizzle.buffer_offset);
+
+        glDispatchCompute(num_dispatches_x, num_dispatches_y, image.info.resources.layers);
+    }
+    program_manager.RestoreGuestCompute();
+}
+
 void UtilShaders::BlockLinearUpload2D(Image& image, const ImageBufferMap& map,
                                       std::span<const SwizzleParameters> swizzles) {
     static constexpr Extent3D WORKGROUP_SIZE{32, 32, 1};

src/video_core/renderer_opengl/util_shaders.h

@@ -24,6 +24,9 @@ public:
     explicit UtilShaders(ProgramManager& program_manager);
     ~UtilShaders();
 
+    void ASTCDecode(Image& image, const ImageBufferMap& map,
+                    std::span<const VideoCommon::SwizzleParameters> swizzles);
+
     void BlockLinearUpload2D(Image& image, const ImageBufferMap& map,
                              std::span<const VideoCommon::SwizzleParameters> swizzles);
 
@@ -43,13 +46,15 @@ private:
     ProgramManager& program_manager;
 
     OGLBuffer swizzle_table_buffer;
+    OGLBuffer astc_buffer;
 
+    OGLProgram astc_decoder_program;
     OGLProgram block_linear_unswizzle_2d_program;
     OGLProgram block_linear_unswizzle_3d_program;
     OGLProgram pitch_unswizzle_program;
-    OGLProgram copy_bc4_program;
     OGLProgram copy_bgr16_program;
     OGLProgram copy_bgra_program;
+    OGLProgram copy_bc4_program;
 };
 
 GLenum StoreFormat(u32 bytes_per_block);

src/video_core/renderer_vulkan/maxwell_to_vk.cpp

@@ -166,7 +166,7 @@ struct FormatTuple {
     {VK_FORMAT_R16G16_SINT, Attachable | Storage},             // R16G16_SINT
     {VK_FORMAT_R16G16_SNORM, Attachable | Storage},            // R16G16_SNORM
     {VK_FORMAT_UNDEFINED},                                     // R32G32B32_FLOAT
-    {VK_FORMAT_R8G8B8A8_SRGB, Attachable},                     // A8B8G8R8_SRGB
+    {VK_FORMAT_A8B8G8R8_SRGB_PACK32, Attachable},              // A8B8G8R8_SRGB
     {VK_FORMAT_R8G8_UNORM, Attachable | Storage},              // R8G8_UNORM
     {VK_FORMAT_R8G8_SNORM, Attachable | Storage},              // R8G8_SNORM
     {VK_FORMAT_R8G8_SINT, Attachable | Storage},               // R8G8_SINT

src/video_core/renderer_vulkan/vk_compute_pass.cpp

@@ -11,18 +11,39 @@
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "common/div_ceil.h"
+#include "video_core/host_shaders/astc_decoder_comp_spv.h"
 #include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h"
 #include "video_core/host_shaders/vulkan_uint8_comp_spv.h"
 #include "video_core/renderer_vulkan/vk_compute_pass.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.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_update_descriptor.h"
+#include "video_core/texture_cache/accelerated_swizzle.h"
+#include "video_core/texture_cache/types.h"
+#include "video_core/textures/astc.h"
+#include "video_core/textures/decoders.h"
 #include "video_core/vulkan_common/vulkan_device.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
 namespace Vulkan {
+
+using Tegra::Texture::SWIZZLE_TABLE;
+using Tegra::Texture::ASTC::EncodingsValues;
+using namespace Tegra::Texture::ASTC;
+
 namespace {
+
+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) {
     return {
         .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
@@ -50,6 +71,67 @@ std::array<VkDescriptorSetLayoutBinding, 2> BuildInputOutputDescriptorSetBinding
     }};
 }
 
+std::array<VkDescriptorSetLayoutBinding, 8> BuildASTCDescriptorSetBindings() {
+    return {{
+        {
+            .binding = ASTC_BINDING_INPUT_BUFFER,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .descriptorCount = 1,
+            .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+            .pImmutableSamplers = nullptr,
+        },
+        {
+            .binding = ASTC_BINDING_ENC_BUFFER,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .descriptorCount = 1,
+            .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+            .pImmutableSamplers = nullptr,
+        },
+        {
+            .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_7_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,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .descriptorCount = 1,
+            .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+            .pImmutableSamplers = nullptr,
+        },
+        {
+            .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_SWIZZLE_BUFFER,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .descriptorCount = 1,
+            .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+            .pImmutableSamplers = nullptr,
+        },
+        {
+            .binding = ASTC_BINDING_OUTPUT_IMAGE,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
+            .descriptorCount = 1,
+            .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+            .pImmutableSamplers = nullptr,
+        },
+    }};
+}
+
 VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() {
     return {
         .dstBinding = 0,
@@ -61,6 +143,95 @@ VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() {
     };
 }
 
+std::array<VkDescriptorUpdateTemplateEntryKHR, 8> BuildASTCPassDescriptorUpdateTemplateEntry() {
+    return {{
+        {
+            .dstBinding = ASTC_BINDING_INPUT_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .offset = ASTC_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+        {
+            .dstBinding = ASTC_BINDING_ENC_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .offset = ASTC_BINDING_ENC_BUFFER * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+        {
+            .dstBinding = ASTC_BINDING_6_TO_8_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .offset = ASTC_BINDING_6_TO_8_BUFFER * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+        {
+            .dstBinding = ASTC_BINDING_7_TO_8_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .offset = ASTC_BINDING_7_TO_8_BUFFER * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+        {
+            .dstBinding = ASTC_BINDING_8_TO_8_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .offset = ASTC_BINDING_8_TO_8_BUFFER * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+        {
+            .dstBinding = ASTC_BINDING_BYTE_TO_16_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .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),
+        },
+        {
+            .dstBinding = ASTC_BINDING_OUTPUT_IMAGE,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
+            .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;
+    u32 bytes_per_block_log2;
+    u32 layer_stride;
+    u32 block_size;
+    u32 x_shift;
+    u32 block_height;
+    u32 block_height_mask;
+};
+
+struct AstcBufferData {
+    decltype(SWIZZLE_TABLE) swizzle_table_buffer = SWIZZLE_TABLE;
+    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;
+
 } // Anonymous namespace
 
 VKComputePass::VKComputePass(const Device& device, VKDescriptorPool& descriptor_pool,
@@ -238,4 +409,170 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
     return {staging.buffer, staging.offset};
 }
 
+ASTCDecoderPass::ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_,
+                                 VKDescriptorPool& descriptor_pool_,
+                                 StagingBufferPool& staging_buffer_pool_,
+                                 VKUpdateDescriptorQueue& update_descriptor_queue_,
+                                 MemoryAllocator& memory_allocator_)
+    : VKComputePass(device_, descriptor_pool_, BuildASTCDescriptorSetBindings(),
+                    BuildASTCPassDescriptorUpdateTemplateEntry(),
+                    BuildComputePushConstantRange(sizeof(AstcPushConstants)),
+                    ASTC_DECODER_COMP_SPV),
+      device{device_}, scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
+      update_descriptor_queue{update_descriptor_queue_}, memory_allocator{memory_allocator_} {}
+
+ASTCDecoderPass::~ASTCDecoderPass() = default;
+
+void ASTCDecoderPass::MakeDataBuffer() {
+    constexpr auto 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 = TOTAL_BUFFER_SIZE,
+        .usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
+        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+        .queueFamilyIndexCount = 0,
+        .pQueueFamilyIndices = nullptr,
+    });
+    data_buffer_commit = memory_allocator.Commit(data_buffer, 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));
+    // 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 = offset,
+                              .dstOffset = 0,
+                              .size = TOTAL_BUFFER_SIZE,
+                          });
+        cmdbuf.PipelineBarrier(
+            VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0,
+            VkMemoryBarrier{
+                .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+                .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),
+    };
+    scheduler.RequestOutsideRenderPassOperationContext();
+    if (!data_buffer) {
+        MakeDataBuffer();
+    }
+    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};
+    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};
+
+        update_descriptor_queue.Acquire();
+
+        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),
+                                          sizeof(AstcBufferData::replicate_6_to_8));
+        update_descriptor_queue.AddBuffer(*data_buffer, offsetof(AstcBufferData, replicate_7_to_8),
+                                          sizeof(AstcBufferData::replicate_7_to_8));
+        update_descriptor_queue.AddBuffer(*data_buffer, offsetof(AstcBufferData, replicate_8_to_8),
+                                          sizeof(AstcBufferData::replicate_8_to_8));
+        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);
+        ASSERT(params.origin == (std::array<u32, 3>{0, 0, 0}));
+        ASSERT(params.destination == (std::array<s32, 3>{0, 0, 0}));
+
+        scheduler.Record([vk_layout, vk_pipeline, num_dispatches_x, num_dispatches_y,
+                          num_dispatches_z, num_image_blocks, block_dims, params,
+                          set](vk::CommandBuffer cmdbuf) {
+            const AstcPushConstants uniforms{
+                .num_image_blocks = num_image_blocks,
+                .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.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline);
+            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);
+        });
+    }
+    scheduler.Record([vk_image, aspect_mask](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 = VK_IMAGE_LAYOUT_GENERAL,
+            .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);
+    });
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_compute_pass.h

@@ -11,14 +11,21 @@
 #include "common/common_types.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
+#include "video_core/vulkan_common/vulkan_memory_allocator.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
+namespace VideoCommon {
+struct SwizzleParameters;
+}
+
 namespace Vulkan {
 
 class Device;
 class StagingBufferPool;
 class VKScheduler;
 class VKUpdateDescriptorQueue;
+class Image;
+struct StagingBufferRef;
 
 class VKComputePass {
 public:
@@ -77,4 +84,29 @@ private:
     VKUpdateDescriptorQueue& update_descriptor_queue;
 };
 
+class ASTCDecoderPass final : public VKComputePass {
+public:
+    explicit ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_,
+                             VKDescriptorPool& descriptor_pool_,
+                             StagingBufferPool& staging_buffer_pool_,
+                             VKUpdateDescriptorQueue& update_descriptor_queue_,
+                             MemoryAllocator& memory_allocator_);
+    ~ASTCDecoderPass();
+
+    void Assemble(Image& image, const StagingBufferRef& map,
+                  std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+private:
+    void MakeDataBuffer();
+
+    const Device& device;
+    VKScheduler& scheduler;
+    StagingBufferPool& staging_buffer_pool;
+    VKUpdateDescriptorQueue& update_descriptor_queue;
+    MemoryAllocator& memory_allocator;
+
+    vk::Buffer data_buffer;
+    MemoryCommit data_buffer_commit;
+};
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -241,7 +241,10 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
       staging_pool(device, memory_allocator, scheduler), descriptor_pool(device, scheduler),
       update_descriptor_queue(device, scheduler),
       blit_image(device, scheduler, state_tracker, descriptor_pool),
-      texture_cache_runtime{device, scheduler, memory_allocator, staging_pool, blit_image},
+      astc_decoder_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue,
+                        memory_allocator),
+      texture_cache_runtime{device,       scheduler,  memory_allocator,
+                            staging_pool, blit_image, astc_decoder_pass},
       texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory),
       buffer_cache_runtime(device, memory_allocator, scheduler, staging_pool,
                            update_descriptor_queue, descriptor_pool),

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -174,6 +174,7 @@ private:
     VKDescriptorPool descriptor_pool;
     VKUpdateDescriptorQueue update_descriptor_queue;
     BlitImageHelper blit_image;
+    ASTCDecoderPass astc_decoder_pass;
 
     TextureCacheRuntime texture_cache_runtime;
     TextureCache texture_cache;

src/video_core/renderer_vulkan/vk_resource_pool.cpp

@@ -17,21 +17,21 @@ ResourcePool::~ResourcePool() = default;
 size_t ResourcePool::CommitResource() {
     // Refresh semaphore to query updated results
     master_semaphore.Refresh();
-
+    const u64 gpu_tick = master_semaphore.KnownGpuTick();
-    const auto search = [this](size_t begin, size_t end) -> std::optional<size_t> {
+    const auto search = [this, gpu_tick](size_t begin, size_t end) -> std::optional<size_t> {
         for (size_t iterator = begin; iterator < end; ++iterator) {
-            if (master_semaphore.IsFree(ticks[iterator])) {
+            if (gpu_tick >= ticks[iterator]) {
                 ticks[iterator] = master_semaphore.CurrentTick();
                 return iterator;
             }
         }
-        return {};
+        return std::nullopt;
     };
     // Try to find a free resource from the hinted position to the end.
-    auto found = search(free_iterator, ticks.size());
+    std::optional<size_t> found = search(hint_iterator, ticks.size());
     if (!found) {
         // Search from beginning to the hinted position.
-        found = search(0, free_iterator);
+        found = search(0, hint_iterator);
         if (!found) {
             // Both searches failed, the pool is full; handle it.
             const size_t free_resource = ManageOverflow();
@@ -41,7 +41,7 @@ size_t ResourcePool::CommitResource() {
         }
     }
     // Free iterator is hinted to the resource after the one that's been commited.
-    free_iterator = (*found + 1) % ticks.size();
+    hint_iterator = (*found + 1) % ticks.size();
     return *found;
 }
 

src/video_core/renderer_vulkan/vk_resource_pool.h

@@ -36,7 +36,7 @@ private:
 
     MasterSemaphore& master_semaphore;
     size_t grow_step = 0;     ///< Number of new resources created after an overflow
-    size_t free_iterator = 0; ///< Hint to where the next free resources is likely to be found
+    size_t hint_iterator = 0; ///< Hint to where the next free resources is likely to be found
     std::vector<u64> ticks;   ///< Ticks for each resource
 };
 

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -10,6 +10,7 @@
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/renderer_vulkan/blit_image.h"
 #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_scheduler.h"
 #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
@@ -807,7 +808,7 @@ Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_
         commit = runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal);
     }
     if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) {
-        flags |= VideoCommon::ImageFlagBits::Converted;
+        flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
     }
     if (runtime.device.HasDebuggingToolAttached()) {
         if (image) {
@@ -816,6 +817,38 @@ Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_
             buffer.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
         }
     }
+    static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{
+        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO,
+        .pNext = nullptr,
+        .usage = VK_IMAGE_USAGE_STORAGE_BIT,
+    };
+    if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) {
+        const auto& device = runtime.device.GetLogical();
+        storage_image_views.reserve(info.resources.levels);
+        for (s32 level = 0; level < info.resources.levels; ++level) {
+            storage_image_views.push_back(device.CreateImageView(VkImageViewCreateInfo{
+                .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
+                .pNext = &storage_image_view_usage_create_info,
+                .flags = 0,
+                .image = *image,
+                .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY,
+                .format = VK_FORMAT_A8B8G8R8_UNORM_PACK32,
+                .components{
+                    .r = VK_COMPONENT_SWIZZLE_IDENTITY,
+                    .g = VK_COMPONENT_SWIZZLE_IDENTITY,
+                    .b = VK_COMPONENT_SWIZZLE_IDENTITY,
+                    .a = VK_COMPONENT_SWIZZLE_IDENTITY,
+                },
+                .subresourceRange{
+                    .aspectMask = aspect_mask,
+                    .baseMipLevel = static_cast<u32>(level),
+                    .levelCount = 1,
+                    .baseArrayLayer = 0,
+                    .layerCount = VK_REMAINING_ARRAY_LAYERS,
+                },
+            }));
+        }
+    }
 }
 
 void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) {
@@ -918,7 +951,6 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
         }
     }
     const auto format_info = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, true, format);
-    const VkFormat vk_format = format_info.format;
     const VkImageViewUsageCreateInfo image_view_usage{
         .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO,
         .pNext = nullptr,
@@ -930,7 +962,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
         .flags = 0,
         .image = image.Handle(),
         .viewType = VkImageViewType{},
-        .format = vk_format,
+        .format = format_info.format,
         .components{
             .r = ComponentSwizzle(swizzle[0]),
             .g = ComponentSwizzle(swizzle[1]),
@@ -982,7 +1014,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
             .pNext = nullptr,
             .flags = 0,
             .buffer = image.Buffer(),
-            .format = vk_format,
+            .format = format_info.format,
             .offset = 0, // TODO: Redesign buffer cache to support this
             .range = image.guest_size_bytes,
         });
@@ -1167,4 +1199,13 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM
     }
 }
 
+void TextureCacheRuntime::AccelerateImageUpload(
+    Image& image, const StagingBufferRef& map,
+    std::span<const VideoCommon::SwizzleParameters> swizzles) {
+    if (IsPixelFormatASTC(image.info.format)) {
+        return astc_decoder_pass.Assemble(image, map, swizzles);
+    }
+    UNREACHABLE();
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -20,6 +20,7 @@ using VideoCommon::Offset2D;
 using VideoCommon::RenderTargets;
 using VideoCore::Surface::PixelFormat;
 
+class ASTCDecoderPass;
 class BlitImageHelper;
 class Device;
 class Image;
@@ -60,6 +61,7 @@ struct TextureCacheRuntime {
     MemoryAllocator& memory_allocator;
     StagingBufferPool& staging_buffer_pool;
     BlitImageHelper& blit_image_helper;
+    ASTCDecoderPass& astc_decoder_pass;
     std::unordered_map<RenderPassKey, vk::RenderPass> renderpass_cache{};
 
     void Finish();
@@ -83,9 +85,7 @@ struct TextureCacheRuntime {
     }
 
     void AccelerateImageUpload(Image&, const StagingBufferRef&,
-                               std::span<const VideoCommon::SwizzleParameters>) {
+                               std::span<const VideoCommon::SwizzleParameters>);
-        UNREACHABLE();
-    }
 
     void InsertUploadMemoryBarrier() {}
 
@@ -116,15 +116,26 @@ public:
         return *buffer;
     }
 
-    [[nodiscard]] VkImageCreateFlags AspectMask() const noexcept {
+    [[nodiscard]] VkImageAspectFlags AspectMask() const noexcept {
         return aspect_mask;
     }
 
+    [[nodiscard]] VkImageView StorageImageView(s32 level) const noexcept {
+        return *storage_image_views[level];
+    }
+
+    /// Returns true when the image is already initialized and mark it as initialized
+    [[nodiscard]] bool ExchangeInitialization() noexcept {
+        return std::exchange(initialized, true);
+    }
+
 private:
     VKScheduler* scheduler;
     vk::Image image;
     vk::Buffer buffer;
     MemoryCommit commit;
+    vk::ImageView image_view;
+    std::vector<vk::ImageView> storage_image_views;
     VkImageAspectFlags aspect_mask = 0;
     bool initialized = false;
 };

src/video_core/texture_cache/accelerated_swizzle.h

@@ -13,8 +13,8 @@
 namespace VideoCommon::Accelerated {
 
 struct BlockLinearSwizzle2DParams {
-    std::array<u32, 3> origin;
+    alignas(16) std::array<u32, 3> origin;
-    std::array<s32, 3> destination;
+    alignas(16) std::array<s32, 3> destination;
     u32 bytes_per_block_log2;
     u32 layer_stride;
     u32 block_size;

src/video_core/texture_cache/util.cpp

@@ -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));
-
+        DecompressBC4(input.subspan(copy.buffer_offset), copy.image_extent,
-        if (IsPixelFormatASTC(info.format)) {
+                      output.subspan(output_offset));
-            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));
-        }
         copy.buffer_offset = output_offset;
         copy.buffer_row_length = mip_size.width;
         copy.buffer_image_height = mip_size.height;

src/video_core/textures/astc.h

@@ -8,6 +8,204 @@
 
 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 };
+
+struct IntegerEncodedValue {
+    constexpr IntegerEncodedValue() = default;
+
+    constexpr IntegerEncodedValue(IntegerEncoding encoding_, u32 num_bits_)
+        : encoding{encoding_}, num_bits{num_bits_} {}
+
+    constexpr bool MatchesEncoding(const IntegerEncodedValue& other) const {
+        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;
+        if (encoding == IntegerEncoding::Trit) {
+            totalBits += (nVals * 8 + 4) / 5;
+        } else if (encoding == IntegerEncoding::Qus32) {
+            totalBits += (nVals * 7 + 2) / 3;
+        }
+        return totalBits;
+    }
+
+    IntegerEncoding encoding{};
+    u32 num_bits = 0;
+    u32 bit_value = 0;
+    union {
+        u32 qus32_value = 0;
+        u32 trit_value;
+    };
+};
+
+// Returns a new instance of this struct that corresponds to the
+// can take no more than maxval values
+constexpr IntegerEncodedValue CreateEncoding(u32 maxVal) {
+    while (maxVal > 0) {
+        u32 check = maxVal + 1;
+
+        // Is maxVal a power of two?
+        if (!(check & (check - 1))) {
+            return IntegerEncodedValue(IntegerEncoding::JustBits, Popcnt(maxVal));
+        }
+
+        // Is maxVal of the type 3*2^n - 1?
+        if ((check % 3 == 0) && !((check / 3) & ((check / 3) - 1))) {
+            return IntegerEncodedValue(IntegerEncoding::Trit, Popcnt(check / 3 - 1));
+        }
+
+        // Is maxVal of the type 5*2^n - 1?
+        if ((check % 5 == 0) && !((check / 5) & ((check / 5) - 1))) {
+            return IntegerEncodedValue(IntegerEncoding::Qus32, Popcnt(check / 5 - 1));
+        }
+
+        // Apparently it can't be represented with a bounded integer sequence...
+        // just iterate.
+        maxVal--;
+    }
+    return IntegerEncodedValue(IntegerEncoding::JustBits, 0);
+}
+
+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));
+    }
+    return encodings;
+}
+
+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.
+template <typename IntType>
+constexpr IntType Replicate(IntType val, u32 numBits, u32 toBit) {
+    if (numBits == 0) {
+        return 0;
+    }
+    if (toBit == 0) {
+        return 0;
+    }
+    const IntType v = val & static_cast<IntType>((1 << numBits) - 1);
+    IntType res = v;
+    u32 reslen = numBits;
+    while (reslen < toBit) {
+        u32 comp = 0;
+        if (numBits > toBit - reslen) {
+            u32 newshift = toBit - reslen;
+            comp = numBits - newshift;
+            numBits = newshift;
+        }
+        res = static_cast<IntType>(res << numBits);
+        res = static_cast<IntType>(res | (v >> comp));
+        reslen += numBits;
+    }
+    return res;
+}
+
+constexpr std::size_t NumReplicateEntries(u32 num_bits) {
+    return std::size_t(1) << num_bits;
+}
+
+template <typename IntType, u32 num_bits, u32 to_bit>
+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);
+    }
+    return table;
+}
+
+constexpr auto REPLICATE_BYTE_TO_16_TABLE = MakeReplicateTable<u32, 8, 16>();
+constexpr u32 ReplicateByteTo16(std::size_t value) {
+    return REPLICATE_BYTE_TO_16_TABLE[value];
+}
+
+constexpr auto REPLICATE_BIT_TO_7_TABLE = MakeReplicateTable<u32, 1, 7>();
+constexpr u32 ReplicateBitTo7(std::size_t value) {
+    return REPLICATE_BIT_TO_7_TABLE[value];
+}
+
+constexpr auto REPLICATE_BIT_TO_9_TABLE = MakeReplicateTable<u32, 1, 9>();
+constexpr u32 ReplicateBitTo9(std::size_t value) {
+    return REPLICATE_BIT_TO_9_TABLE[value];
+}
+
+constexpr auto REPLICATE_1_BIT_TO_8_TABLE = MakeReplicateTable<u32, 1, 8>();
+constexpr auto REPLICATE_2_BIT_TO_8_TABLE = MakeReplicateTable<u32, 2, 8>();
+constexpr auto REPLICATE_3_BIT_TO_8_TABLE = MakeReplicateTable<u32, 3, 8>();
+constexpr auto REPLICATE_4_BIT_TO_8_TABLE = MakeReplicateTable<u32, 4, 8>();
+constexpr auto REPLICATE_5_BIT_TO_8_TABLE = MakeReplicateTable<u32, 5, 8>();
+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>();
+/// Use a precompiled table with the most common usages, if it's not in the expected range, fallback
+/// to the runtime implementation
+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);
+    }
+}
+
+constexpr auto REPLICATE_1_BIT_TO_6_TABLE = MakeReplicateTable<u32, 1, 6>();
+constexpr auto REPLICATE_2_BIT_TO_6_TABLE = MakeReplicateTable<u32, 2, 6>();
+constexpr auto REPLICATE_3_BIT_TO_6_TABLE = MakeReplicateTable<u32, 3, 6>();
+constexpr auto REPLICATE_4_BIT_TO_6_TABLE = MakeReplicateTable<u32, 4, 6>();
+constexpr auto REPLICATE_5_BIT_TO_6_TABLE = MakeReplicateTable<u32, 5, 6>();
+
+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);
+    }
+}
+
+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;
+
 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);
 

src/video_core/textures/decoders.cpp

@@ -17,26 +17,7 @@
 #include "video_core/textures/texture.h"
 
 namespace Tegra::Texture {
-
 namespace {
-/**
- * This table represents the internal swizzle of a gob, in format 16 bytes x 2 sector packing.
- * Calculates the offset of an (x, y) position within a swizzled texture.
- * Taken from the Tegra X1 Technical Reference Manual. pages 1187-1188
- */
-constexpr SwizzleTable MakeSwizzleTableConst() {
-    SwizzleTable table{};
-    for (u32 y = 0; y < table.size(); ++y) {
-        for (u32 x = 0; x < table[0].size(); ++x) {
-            table[y][x] = ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 +
-                          (y % 2) * 16 + (x % 16);
-        }
-    }
-    return table;
-}
-
-constexpr SwizzleTable SWIZZLE_TABLE = MakeSwizzleTableConst();
-
 template <bool TO_LINEAR>
 void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width,
              u32 height, u32 depth, u32 block_height, u32 block_depth, u32 stride_alignment) {
@@ -91,10 +72,6 @@ void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixe
 }
 } // Anonymous namespace
 
-SwizzleTable MakeSwizzleTable() {
-    return SWIZZLE_TABLE;
-}
-
 void UnswizzleTexture(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel,
                       u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth,
                       u32 stride_alignment) {

src/video_core/textures/decoders.h

@@ -23,8 +23,22 @@ constexpr u32 GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_
 
 using SwizzleTable = std::array<std::array<u32, GOB_SIZE_X>, GOB_SIZE_Y>;
 
-/// Returns a z-order swizzle table
+/**
-SwizzleTable MakeSwizzleTable();
+ * This table represents the internal swizzle of a gob, in format 16 bytes x 2 sector packing.
+ * Calculates the offset of an (x, y) position within a swizzled texture.
+ * Taken from the Tegra X1 Technical Reference Manual. pages 1187-1188
+ */
+constexpr SwizzleTable MakeSwizzleTable() {
+    SwizzleTable table{};
+    for (u32 y = 0; y < table.size(); ++y) {
+        for (u32 x = 0; x < table[0].size(); ++x) {
+            table[y][x] = ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 +
+                          (y % 2) * 16 + (x % 16);
+        }
+    }
+    return table;
+}
+constexpr SwizzleTable SWIZZLE_TABLE = MakeSwizzleTable();
 
 /// Unswizzles a block linear texture into linear memory.
 void UnswizzleTexture(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel,