renderer_vulkan: Accelerate ASTC decoding

Co-Authored-By: Rodrigo Locatti <reinuseslisp@airmail.cc>

src/video_core/host_shaders/astc_decoder.comp

@@ -16,7 +16,7 @@
 #define BINDING_7_TO_8_BUFFER 4
 #define BINDING_8_TO_8_BUFFER 5
 #define BINDING_BYTE_TO_16_BUFFER 6
-#define BINDING_OUTPUT_IMAGE 3
+#define BINDING_OUTPUT_IMAGE 7
 
 #else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
 
@@ -85,7 +85,26 @@ layout(binding = BINDING_SWIZZLE_BUFFER, std430) readonly buffer SwizzleTable {
 layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU32 {
     uint astc_data[];
 };
-layout(binding = BINDING_OUTPUT_IMAGE) uniform writeonly image2D dest_image;
+
+// ASTC Encodings data
+layout(binding = BINDING_ENC_BUFFER, std430) readonly buffer EncodingsValues {
+    EncodingData encoding_values[];
+};
+// ASTC Precompiled tables
+layout(binding = BINDING_6_TO_8_BUFFER, std430) readonly buffer REPLICATE_6_BIT_TO_8 {
+    uint REPLICATE_6_BIT_TO_8_TABLE[];
+};
+layout(binding = BINDING_7_TO_8_BUFFER, std430) readonly buffer REPLICATE_7_BIT_TO_8 {
+    uint REPLICATE_7_BIT_TO_8_TABLE[];
+};
+layout(binding = BINDING_8_TO_8_BUFFER, std430) readonly buffer REPLICATE_8_BIT_TO_8 {
+    uint REPLICATE_8_BIT_TO_8_TABLE[];
+};
+layout(binding = BINDING_BYTE_TO_16_BUFFER, std430) readonly buffer REPLICATE_BYTE_TO_16 {
+    uint REPLICATE_BYTE_TO_16_TABLE[];
+};
+
+layout(binding = BINDING_OUTPUT_IMAGE, rgba8) uniform writeonly image2D dest_image;
 
 const uint GOB_SIZE_X = 64;
 const uint GOB_SIZE_Y = 8;
@@ -109,23 +128,6 @@ uint ReadTexel(uint offset) {
     return bitfieldExtract(astc_data[offset / 4], int((offset * 8) & 24), 8);
 }
 
-// ASTC Encodings data
-layout(binding = BINDING_ENC_BUFFER, std430) readonly buffer EncodingsValues {
-    EncodingData encoding_values[256];
-};
-// ASTC Precompiled tables
-layout(binding = BINDING_6_TO_8_BUFFER, std430) readonly buffer REPLICATE_6_BIT_TO_8 {
-    uint REPLICATE_6_BIT_TO_8_TABLE[];
-};
-layout(binding = BINDING_7_TO_8_BUFFER, std430) readonly buffer REPLICATE_7_BIT_TO_8 {
-    uint REPLICATE_7_BIT_TO_8_TABLE[];
-};
-layout(binding = BINDING_8_TO_8_BUFFER, std430) readonly buffer REPLICATE_8_BIT_TO_8 {
-    uint REPLICATE_8_BIT_TO_8_TABLE[];
-};
-layout(binding = BINDING_BYTE_TO_16_BUFFER, std430) readonly buffer REPLICATE_BYTE_TO_16 {
-    uint REPLICATE_BYTE_TO_16_TABLE[];
-};
 
 const int BLOCK_SIZE_IN_BYTES = 16;
 
@@ -1275,8 +1277,7 @@ void main() {
     offset += (pos.x >> GOB_SIZE_X_SHIFT) << x_shift;
     offset += swizzle;
 
-    const ivec3 invocation_destination = ivec3(gl_GlobalInvocationID + destination);
+    const ivec3 coord = ivec3(gl_GlobalInvocationID * uvec3(block_dims, 1.0));
-    const ivec3 coord = ivec3(invocation_destination * uvec3(block_dims, 1.0));
     uint block_index =
         layer * num_image_blocks.x * num_image_blocks.y + pos.y * num_image_blocks.x + pos.x;
     current_index = 0;

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 +11,38 @@
 #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;
+
 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_OUTPUT_IMAGE = 7;
+
 VkPushConstantRange BuildComputePushConstantRange(std::size_t size) {
     return {
         .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
@@ -50,6 +70,67 @@ std::array<VkDescriptorSetLayoutBinding, 2> BuildInputOutputDescriptorSetBinding
     }};
 }
 
+std::array<VkDescriptorSetLayoutBinding, 8> BuildASTCDescriptorSetBindings() {
+    return {{
+        {
+            .binding = ASTC_BINDING_SWIZZLE_BUFFER, // Swizzle 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
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .descriptorCount = 1,
+            .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+            .pImmutableSamplers = nullptr,
+        },
+        {
+            .binding = ASTC_BINDING_ENC_BUFFER, // Encodings 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
+            .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
+            .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
+            .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
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .descriptorCount = 1,
+            .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+            .pImmutableSamplers = nullptr,
+        },
+        {
+            .binding = ASTC_BINDING_OUTPUT_IMAGE, // 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 +142,90 @@ 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),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+        {
+            .dstBinding = ASTC_BINDING_ENC_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .offset = 2 * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+        {
+            .dstBinding = ASTC_BINDING_6_TO_8_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .offset = 3 * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+        {
+            .dstBinding = ASTC_BINDING_7_TO_8_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .offset = 4 * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+        {
+            .dstBinding = ASTC_BINDING_8_TO_8_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .offset = 5 * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+        {
+            .dstBinding = ASTC_BINDING_BYTE_TO_16_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .offset = 6 * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+        {
+            .dstBinding = ASTC_BINDING_OUTPUT_IMAGE,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
+            .offset = 7 * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+    }};
+}
+
+struct AstcPushConstants {
+    std::array<u32, 2> num_image_blocks;
+    std::array<u32, 2> blocks_dims;
+    u32 layer;
+    VideoCommon::Accelerated::BlockLinearSwizzle2DParams params;
+};
+
+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 +403,137 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
     return {staging.buffer, staging.offset};
 }
 
+using namespace Tegra::Texture::ASTC;
+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() {
+    data_buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
+        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .size = sizeof(ASTC_BUFFER_DATA),
+        .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(sizeof(ASTC_BUFFER_DATA), 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) {
+        cmdbuf.CopyBuffer(src, dst,
+                          VkBufferCopy{
+                              .srcOffset = 0,
+                              .dstOffset = 0,
+                              .size = sizeof(ASTC_BUFFER_DATA),
+                          });
+        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 std::array<u32, 2> block_dims{tile_size.width, tile_size.height};
+    for (s32 layer = 0; layer < image.info.resources.layers; layer++) {
+        for (const VideoCommon::SwizzleParameters& swizzle : swizzles) {
+            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);
+            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, image.guest_size_bytes);
+            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.AddImage(image.StorageImageView());
+
+            const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
+            // To unswizzle the ASTC data
+            const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
+            scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = map.buffer,
+                              num_dispatches_x, num_dispatches_y, layer_image_size,
+                              num_image_blocks, block_dims, layer, params, set,
+                              image = image.Handle(), input_offset,
+                              aspect_mask = image.AspectMask()](vk::CommandBuffer cmdbuf) {
+                const AstcPushConstants uniforms{num_image_blocks, block_dims, layer, params};
+
+                cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
+                cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {});
+                cmdbuf.PushConstants(layout, VK_SHADER_STAGE_COMPUTE_BIT, uniforms);
+                cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, 1);
+
+                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,
+                    .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+                    .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+                    .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                    .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                    .image = image,
+                    .subresourceRange{
+                        .aspectMask = aspect_mask,
+                        .baseMipLevel = 0,
+                        .levelCount = VK_REMAINING_MIP_LEVELS,
+                        .baseArrayLayer = 0,
+                        .layerCount = VK_REMAINING_ARRAY_LAYERS,
+                    },
+                };
+                cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+                                       VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, 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

@@ -173,6 +173,7 @@ private:
     VKDescriptorPool descriptor_pool;
     VKUpdateDescriptorQueue update_descriptor_queue;
     BlitImageHelper blit_image;
+    ASTCDecoderPass astc_decoder_pass;
 
     GraphicsPipelineCacheKey graphics_key;
 

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 +817,34 @@ 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()) {
+        storage_image_view = runtime.device.GetLogical().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,
+            .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 = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        });
+    }
 }
 
 void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) {
@@ -918,7 +947,6 @@ 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 +958,7 @@ 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 +1010,7 @@ 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 +1195,13 @@ 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() {}
 
@@ -125,11 +125,17 @@ public:
         return aspect_mask;
     }
 
+    [[nodiscard]] VkImageView StorageImageView() const noexcept {
+        return *storage_image_view;
+    }
+
 private:
     VKScheduler* scheduler;
     vk::Image image;
     vk::Buffer buffer;
     MemoryCommit commit;
+    vk::ImageView image_view;
+    vk::ImageView storage_image_view;
     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/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,