[vulkan] Correction on unconditional/ harcoded makeview 2D array vs compute shader 3D views

This reverts commit a9c4c8aefd.

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4157
Reviewed-by: CamilleLaVey <camillelavey99@gmail.com>
Reviewed-by: MaranBr <maranbr@eden-emu.dev>

src/video_core/host_shaders/block_linear_unswizzle_2d.comp

@@ -11,9 +11,8 @@
 #define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
 #define END_PUSH_CONSTANTS };
 #define UNIFORM(n)
-#define BINDING_SWIZZLE_BUFFER 0
-#define BINDING_INPUT_BUFFER 1
-#define BINDING_OUTPUT_IMAGE 2
+#define BINDING_INPUT_BUFFER 0
+#define BINDING_OUTPUT_IMAGE 1
 
 #else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
 
@@ -26,8 +25,7 @@
 #define BEGIN_PUSH_CONSTANTS
 #define END_PUSH_CONSTANTS
 #define UNIFORM(n) layout (location = n) uniform
-#define BINDING_SWIZZLE_BUFFER 0
-#define BINDING_INPUT_BUFFER 1
+#define BINDING_INPUT_BUFFER 0
 #define BINDING_OUTPUT_IMAGE 0
 
 #endif
@@ -43,10 +41,6 @@ UNIFORM(6) uint block_height;
 UNIFORM(7) uint block_height_mask;
 END_PUSH_CONSTANTS
 
-layout(binding = BINDING_SWIZZLE_BUFFER, std430) readonly buffer SwizzleTable {
-    uint swizzle_table[];
-};
-
 #if HAS_EXTENDED_TYPES
 layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU8 { uint8_t u8data[]; };
 layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU16 { uint16_t u16data[]; };
@@ -71,9 +65,19 @@ const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_SHI
 
 const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1, GOB_SIZE_Y - 1);
 
+uint SwizzleTable(uint pos) {
+    const uint t[8] = uint[](
+        0x12100200, 0x13110301, 0x16140604, 0x17150705,
+        0x1a180a08, 0x1b190b09, 0x1e1c0e0c, 0x1f1d0f0d
+    );
+    const uint i = pos >> 4;
+    const uint h = (t[i / 4] >> ((i % 4) * 8)) & 0xff;
+    return (h << 4) | (pos & 0xf);
+}
+
 uint SwizzleOffset(uvec2 pos) {
     pos = pos & SWIZZLE_MASK;
-    return swizzle_table[pos.y * 64 + pos.x];
+    return SwizzleTable(pos.y * 64 + pos.x);
 }
 
 uvec4 ReadTexel(uint offset) {

src/video_core/host_shaders/block_linear_unswizzle_3d.comp

@@ -11,9 +11,8 @@
 #define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
 #define END_PUSH_CONSTANTS };
 #define UNIFORM(n)
-#define BINDING_SWIZZLE_BUFFER 0
-#define BINDING_INPUT_BUFFER 1
-#define BINDING_OUTPUT_IMAGE 2
+#define BINDING_INPUT_BUFFER 0
+#define BINDING_OUTPUT_IMAGE 1
 
 #else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
 
@@ -26,8 +25,7 @@
 #define BEGIN_PUSH_CONSTANTS
 #define END_PUSH_CONSTANTS
 #define UNIFORM(n) layout (location = n) uniform
-#define BINDING_SWIZZLE_BUFFER 0
-#define BINDING_INPUT_BUFFER 1
+#define BINDING_INPUT_BUFFER 0
 #define BINDING_OUTPUT_IMAGE 0
 
 #endif
@@ -45,10 +43,6 @@ UNIFORM(8) uint block_depth;
 UNIFORM(9) uint block_depth_mask;
 END_PUSH_CONSTANTS
 
-layout(binding = BINDING_SWIZZLE_BUFFER, std430) readonly buffer SwizzleTable {
-    uint swizzle_table[];
-};
-
 #if HAS_EXTENDED_TYPES
 layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU8 { uint8_t u8data[]; };
 layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU16 { uint16_t u16data[]; };
@@ -73,9 +67,19 @@ const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_SHI
 
 const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1, GOB_SIZE_Y - 1);
 
+uint SwizzleTable(uint pos) {
+    const uint t[8] = uint[](
+        0x12100200, 0x13110301, 0x16140604, 0x17150705,
+        0x1a180a08, 0x1b190b09, 0x1e1c0e0c, 0x1f1d0f0d
+    );
+    const uint i = pos >> 4;
+    const uint h = (t[i / 4] >> ((i % 4) * 8)) & 0xff;
+    return (h << 4) | (pos & 0xf);
+}
+
 uint SwizzleOffset(uvec2 pos) {
     pos = pos & SWIZZLE_MASK;
-    return swizzle_table[pos.y * 64 + pos.x];
+    return SwizzleTable(pos.y * 64 + pos.x);
 }
 
 uvec4 ReadTexel(uint offset) {

src/video_core/host_shaders/block_linear_unswizzle_3d_bcn.comp

@@ -10,9 +10,8 @@
     #define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
     #define END_PUSH_CONSTANTS };
     #define UNIFORM(n)
-    #define BINDING_SWIZZLE_BUFFER 0
-    #define BINDING_INPUT_BUFFER   1
-    #define BINDING_OUTPUT_BUFFER  2
+    #define BINDING_INPUT_BUFFER   0
+    #define BINDING_OUTPUT_BUFFER  1
 #else
     #extension GL_NV_gpu_shader5 : enable
     #ifdef GL_NV_gpu_shader5
@@ -23,7 +22,6 @@
     #define BEGIN_PUSH_CONSTANTS
     #define END_PUSH_CONSTANTS
     #define UNIFORM(n) layout(location = n) uniform
-    #define BINDING_SWIZZLE_BUFFER 0
     #define BINDING_INPUT_BUFFER   1
     #define BINDING_OUTPUT_BUFFER  0
 #endif
@@ -66,13 +64,9 @@ END_PUSH_CONSTANTS
 #endif
 
 // --- Buffers ---
-layout(binding = BINDING_SWIZZLE_BUFFER, std430) readonly buffer SwizzleTable {
-    uint swizzle_table[];
-};
-
 #if HAS_EXTENDED_TYPES
-    layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU8   { uint8_t  u8data[];  };
-    layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU16  { uint16_t u16data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU8   { uint8_t  u8data[];  };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU16  { uint16_t u16data[]; };
 #endif
 layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU32  { uint   u32data[];  };
 layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU64  { uvec2  u64data[];  };
@@ -96,10 +90,20 @@ const uint GOB_SIZE_Z_SHIFT = 0;
 const uint GOB_SIZE_SHIFT   = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_SHIFT;
 const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1u, GOB_SIZE_Y - 1u);
 
+uint SwizzleTable(uint pos) {
+    const uint t[8] = uint[](
+        0x12100200, 0x13110301, 0x16140604, 0x17150705,
+        0x1a180a08, 0x1b190b09, 0x1e1c0e0c, 0x1f1d0f0d
+    );
+    const uint i = pos >> 4;
+    const uint h = (t[i / 4] >> ((i % 4) * 8)) & 0xff;
+    return (h << 4) | (pos & 0xf);
+}
+
 // --- Helpers ---
 uint SwizzleOffset(uvec2 pos) {
     pos &= SWIZZLE_MASK;
-    return swizzle_table[pos.y * 64u + pos.x];
+    return SwizzleTable(pos.y * 64u + pos.x);
 }
 
 uvec4 ReadTexel(uint offset) {

src/video_core/renderer_opengl/util_shaders.cpp

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -56,10 +59,8 @@ UtilShaders::UtilShaders(ProgramManager& program_manager_)
       copy_bc4_program(MakeProgram(OPENGL_COPY_BC4_COMP)),
       convert_s8d24_program(MakeProgram(OPENGL_CONVERT_S8D24_COMP)),
       convert_ms_to_nonms_program(MakeProgram(CONVERT_MSAA_TO_NON_MSAA_COMP)),
-      convert_nonms_to_ms_program(MakeProgram(CONVERT_NON_MSAA_TO_MSAA_COMP)) {
-    const auto swizzle_table = Tegra::Texture::MakeSwizzleTable();
-    swizzle_table_buffer.Create();
-    glNamedBufferStorage(swizzle_table_buffer.handle, sizeof(swizzle_table), &swizzle_table, 0);
+      convert_nonms_to_ms_program(MakeProgram(CONVERT_NON_MSAA_TO_MSAA_COMP))
+{
 }
 
 UtilShaders::~UtilShaders() = default;
@@ -116,13 +117,11 @@ void UtilShaders::ASTCDecode(Image& image, const StagingBufferMap& map,
 void UtilShaders::BlockLinearUpload2D(Image& image, const StagingBufferMap& map,
                                       std::span<const SwizzleParameters> swizzles) {
     static constexpr Extent3D WORKGROUP_SIZE{32, 32, 1};
-    static constexpr GLuint BINDING_SWIZZLE_BUFFER = 0;
-    static constexpr GLuint BINDING_INPUT_BUFFER = 1;
+    static constexpr GLuint BINDING_INPUT_BUFFER = 0;
     static constexpr GLuint BINDING_OUTPUT_IMAGE = 0;
 
     program_manager.BindComputeProgram(block_linear_unswizzle_2d_program.handle);
     glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes);
-    glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle);
 
     const GLenum store_format = StoreFormat(BytesPerBlock(image.info.format));
     for (const SwizzleParameters& swizzle : swizzles) {
@@ -153,14 +152,11 @@ void UtilShaders::BlockLinearUpload2D(Image& image, const StagingBufferMap& map,
 void UtilShaders::BlockLinearUpload3D(Image& image, const StagingBufferMap& map,
                                       std::span<const SwizzleParameters> swizzles) {
     static constexpr Extent3D WORKGROUP_SIZE{16, 8, 8};
-
-    static constexpr GLuint BINDING_SWIZZLE_BUFFER = 0;
-    static constexpr GLuint BINDING_INPUT_BUFFER = 1;
+    static constexpr GLuint BINDING_INPUT_BUFFER = 0;
     static constexpr GLuint BINDING_OUTPUT_IMAGE = 0;
 
     glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes);
     program_manager.BindComputeProgram(block_linear_unswizzle_3d_program.handle);
-    glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle);
 
     const GLenum store_format = StoreFormat(BytesPerBlock(image.info.format));
     for (const SwizzleParameters& swizzle : swizzles) {

src/video_core/renderer_opengl/util_shaders.h

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -45,9 +48,6 @@ public:
 
 private:
     ProgramManager& program_manager;
-
-    OGLBuffer swizzle_table_buffer;
-
     OGLProgram astc_decoder_program;
     OGLProgram block_linear_unswizzle_2d_program;
     OGLProgram block_linear_unswizzle_3d_program;

src/video_core/renderer_vulkan/pipeline_helper.h

@@ -203,7 +203,11 @@ inline void PushImageDescriptors(TextureCache& texture_cache,
             const VideoCommon::ImageViewId image_view_id{(views++)->id};
             const VideoCommon::SamplerId sampler_id{*(samplers++)};
             ImageView& image_view{texture_cache.GetImageView(image_view_id)};
-            const VkImageView vk_image_view{image_view.Handle(desc.type)};
+            VkImageView vk_image_view{image_view.Handle(desc.type)};
+            if (vk_image_view == VK_NULL_HANDLE) {
+                const VkImageView null_image_view{texture_cache.GetImageView(VideoCommon::NULL_IMAGE_VIEW_ID).Handle(desc.type)};
+                if (null_image_view != VK_NULL_HANDLE) vk_image_view = null_image_view;
+            }
             const Sampler& sampler{texture_cache.GetSampler(sampler_id)};
             const bool use_fallback_sampler{sampler.HasAddedAnisotropy() &&
                                             !image_view.SupportsAnisotropy()};

src/video_core/renderer_vulkan/renderer_vulkan.cpp

@@ -169,6 +169,7 @@ try
 }
 
 RendererVulkan::~RendererVulkan() {
+    scheduler.WaitWorker();
     scheduler.RegisterOnSubmit([] {});
     void(device.GetLogical().WaitIdle());
 }

src/video_core/renderer_vulkan/vk_compute_pass.cpp

@@ -25,6 +25,9 @@
 #include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h"
 #include "video_core/host_shaders/vulkan_uint8_comp_spv.h"
 #include "video_core/host_shaders/block_linear_unswizzle_3d_bcn_comp_spv.h"
+#include "video_core/host_shaders/block_linear_unswizzle_2d_comp_spv.h"
+#include "video_core/host_shaders/block_linear_unswizzle_3d_comp_spv.h"
+#include "video_core/host_shaders/pitch_unswizzle_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"
@@ -232,6 +235,26 @@ struct QueriesPrefixScanPushConstants {
 struct ConditionalRenderingResolvePushConstants {
     u32 compare_to_zero;
 };
+
+struct BlockLinear3DImagePushConstants {
+    alignas(16) std::array<u32, 3> origin;
+    alignas(16) std::array<s32, 3> destination;
+    u32 bytes_per_block_log2;
+    u32 slice_size;
+    u32 block_size;
+    u32 x_shift;
+    u32 block_height;
+    u32 block_height_mask;
+    u32 block_depth;
+    u32 block_depth_mask;
+};
+
+struct PitchUnswizzlePushConstants {
+    std::array<u32, 2> origin;
+    std::array<s32, 2> destination;
+    u32 bytes_per_block;
+    u32 pitch;
+};
 } // Anonymous namespace
 
 ComputePass::ComputePass(const Device& device_, Scheduler& scheduler, DescriptorPool& descriptor_pool,
@@ -653,71 +676,311 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
     scheduler.Finish();
 }
 
-constexpr u32 BL3D_BINDING_SWIZZLE_TABLE = 0;
-constexpr u32 BL3D_BINDING_INPUT_BUFFER  = 1;
-constexpr u32 BL3D_BINDING_OUTPUT_BUFFER = 2;
+BlockLinearUnswizzleImage2DPass::BlockLinearUnswizzleImage2DPass(
+    const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+    StagingBufferPool& staging_buffer_pool_,
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
+    : ComputePass(device_, scheduler_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS,
+                  ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO,
+                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(BlockLinearSwizzle2DParams)>,
+                  BLOCK_LINEAR_UNSWIZZLE_2D_COMP_SPV),
+      scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
+      compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
 
-constexpr std::array<VkDescriptorSetLayoutBinding, 3> BL3D_DESCRIPTOR_SET_BINDINGS{{
-    {
-        .binding = BL3D_BINDING_SWIZZLE_TABLE,
-        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, // swizzle_table[]
-        .descriptorCount = 1,
-        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
-        .pImmutableSamplers = nullptr,
-    },
-    {
-        .binding = BL3D_BINDING_INPUT_BUFFER,
-        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, // block-linear input
-        .descriptorCount = 1,
-        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
-        .pImmutableSamplers = nullptr,
-    },
-    {
-        .binding = BL3D_BINDING_OUTPUT_BUFFER,
-        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-        .descriptorCount = 1,
-        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
-        .pImmutableSamplers = nullptr,
-    },
-}};
+BlockLinearUnswizzleImage2DPass::~BlockLinearUnswizzleImage2DPass() = default;
 
-constexpr DescriptorBankInfo BL3D_BANK_INFO{
-    .uniform_buffers = 0,
-    .storage_buffers = 3,
-    .texture_buffers = 0,
-    .image_buffers = 0,
-    .textures = 0,
-    .images = 0,
-    .score = 3,
-};
+void BlockLinearUnswizzleImage2DPass::Unswizzle(
+    Image& image, const StagingBufferRef& map,
+    std::span<const VideoCommon::SwizzleParameters> swizzles) {
+    using namespace VideoCommon::Accelerated;
+    scheduler.RequestOutsideRenderPassOperationContext();
+    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_pipeline, vk_image, aspect_mask,
+                      is_initialized](vk::CommandBuffer cmdbuf) {
+        const VkImageMemoryBarrier image_barrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = static_cast<VkAccessFlags>(is_initialized ? VK_ACCESS_SHADER_WRITE_BIT
+                                                                       : VK_ACCESS_NONE),
+            .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_GRAPHICS_COMPUTE_TRANSFER
+                              : VkPipelineStageFlags(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT),
+                       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;
 
-constexpr std::array<VkDescriptorUpdateTemplateEntry, 3>
-    BL3D_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY{{
-        {
-            .dstBinding = BL3D_BINDING_SWIZZLE_TABLE,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-            .offset = BL3D_BINDING_SWIZZLE_TABLE * sizeof(DescriptorUpdateEntry),
-            .stride = sizeof(DescriptorUpdateEntry),
-        },
-        {
-            .dstBinding = BL3D_BINDING_INPUT_BUFFER,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-            .offset = BL3D_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry),
-            .stride = sizeof(DescriptorUpdateEntry),
-        },
-        {
-            .dstBinding = BL3D_BINDING_OUTPUT_BUFFER,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-            .offset = BL3D_BINDING_OUTPUT_BUFFER * sizeof(DescriptorUpdateEntry),
-            .stride = sizeof(DescriptorUpdateEntry),
-        }
-    }};
+        compute_pass_descriptor_queue.Acquire(scheduler, 2);
+        compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset,
+                                                image.guest_size_bytes - swizzle.buffer_offset);
+        compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
+        const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
+
+        const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
+        scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, params,
+                          descriptor_data](vk::CommandBuffer cmdbuf) {
+            const VkDescriptorSet set = descriptor_allocator.Commit();
+            device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
+            cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
+            cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params);
+            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 |
+                             VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_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(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+                       vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, 0, image_barrier);
+    });
+    scheduler.Finish();
+}
+
+BlockLinearUnswizzleImage3DPass::BlockLinearUnswizzleImage3DPass(
+    const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+    StagingBufferPool& staging_buffer_pool_,
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
+    : ComputePass(device_, scheduler_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS,
+                  ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO,
+                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(BlockLinear3DImagePushConstants)>,
+                  BLOCK_LINEAR_UNSWIZZLE_3D_COMP_SPV),
+      scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
+      compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
+
+BlockLinearUnswizzleImage3DPass::~BlockLinearUnswizzleImage3DPass() = default;
+
+void BlockLinearUnswizzleImage3DPass::Unswizzle(
+    Image& image, const StagingBufferRef& map,
+    std::span<const VideoCommon::SwizzleParameters> swizzles) {
+    using namespace VideoCommon::Accelerated;
+    scheduler.RequestOutsideRenderPassOperationContext();
+    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_pipeline, vk_image, aspect_mask,
+                      is_initialized](vk::CommandBuffer cmdbuf) {
+        const VkImageMemoryBarrier image_barrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = static_cast<VkAccessFlags>(is_initialized ? VK_ACCESS_SHADER_WRITE_BIT
+                                                                       : VK_ACCESS_NONE),
+            .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_GRAPHICS_COMPUTE_TRANSFER
+                              : VkPipelineStageFlags(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT),
+                       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, 16U);
+        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U);
+        const u32 num_dispatches_z = Common::DivCeil(swizzle.num_tiles.depth, 8U);
+
+        compute_pass_descriptor_queue.Acquire(scheduler, 2);
+        compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset,
+                                                image.guest_size_bytes - swizzle.buffer_offset);
+        compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
+        const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
+
+        const auto p = MakeBlockLinearSwizzle3DParams(swizzle, image.info);
+        const BlockLinear3DImagePushConstants params{
+            .origin = p.origin,
+            .destination = p.destination,
+            .bytes_per_block_log2 = p.bytes_per_block_log2,
+            .slice_size = p.slice_size,
+            .block_size = p.block_size,
+            .x_shift = p.x_shift,
+            .block_height = p.block_height,
+            .block_height_mask = p.block_height_mask,
+            .block_depth = p.block_depth,
+            .block_depth_mask = p.block_depth_mask,
+        };
+        scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, params,
+                          descriptor_data](vk::CommandBuffer cmdbuf) {
+            const VkDescriptorSet set = descriptor_allocator.Commit();
+            device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
+            cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
+            cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params);
+            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 |
+                             VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_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(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+                       vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, 0, image_barrier);
+    });
+    scheduler.Finish();
+}
+
+PitchUnswizzlePass::PitchUnswizzlePass(const Device& device_, Scheduler& scheduler_,
+                                       DescriptorPool& descriptor_pool_,
+                                       StagingBufferPool& staging_buffer_pool_,
+                                       ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
+    : ComputePass(device_, scheduler_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS,
+                  ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO,
+                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(PitchUnswizzlePushConstants)>,
+                  PITCH_UNSWIZZLE_COMP_SPV),
+      scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
+      compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
+
+PitchUnswizzlePass::~PitchUnswizzlePass() = default;
+
+void PitchUnswizzlePass::Unswizzle(Image& image, const StagingBufferRef& map,
+                                   std::span<const VideoCommon::SwizzleParameters> swizzles) {
+    scheduler.RequestOutsideRenderPassOperationContext();
+    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_pipeline, vk_image, aspect_mask,
+                      is_initialized](vk::CommandBuffer cmdbuf) {
+        const VkImageMemoryBarrier image_barrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = static_cast<VkAccessFlags>(is_initialized ? VK_ACCESS_SHADER_WRITE_BIT
+                                                                       : VK_ACCESS_NONE),
+            .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_GRAPHICS_COMPUTE_TRANSFER
+                              : VkPipelineStageFlags(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT),
+                       VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier);
+        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline);
+    });
+    const u32 bytes_per_block = VideoCore::Surface::BytesPerBlock(image.info.format);
+    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);
+
+        compute_pass_descriptor_queue.Acquire(scheduler, 2);
+        compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset,
+                                                image.guest_size_bytes - swizzle.buffer_offset);
+        compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
+        const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
+
+        const PitchUnswizzlePushConstants params{
+            .origin = {0, 0},
+            .destination = {0, 0},
+            .bytes_per_block = bytes_per_block,
+            .pitch = image.info.pitch,
+        };
+        scheduler.Record([this, num_dispatches_x, num_dispatches_y, params,
+                          descriptor_data](vk::CommandBuffer cmdbuf) {
+            const VkDescriptorSet set = descriptor_allocator.Commit();
+            device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
+            cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
+            cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params);
+            cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, 1);
+        });
+    }
+    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 |
+                             VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_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(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+                       vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, 0, image_barrier);
+    });
+    scheduler.Finish();
+}
+
+constexpr u32 BL3D_BINDING_INPUT_BUFFER  = 0;
+constexpr u32 BL3D_BINDING_OUTPUT_BUFFER = 1;
 
 struct alignas(16) BlockLinearUnswizzle3DPushConstants {
     u32 blocks_dim[3];           // Offset 0
@@ -745,11 +1008,50 @@ BlockLinearUnswizzle3DPass::BlockLinearUnswizzle3DPass(
     DescriptorPool& descriptor_pool_,
     StagingBufferPool& staging_buffer_pool_,
     ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
-    : ComputePass(
-          device_, scheduler_, descriptor_pool_,
-          BL3D_DESCRIPTOR_SET_BINDINGS,
-          BL3D_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY,
-          BL3D_BANK_INFO,
+    : ComputePass(device_, scheduler_, descriptor_pool_,
+        std::array<VkDescriptorSetLayoutBinding, 2>{{
+    {
+        .binding = BL3D_BINDING_INPUT_BUFFER,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, // block-linear input
+        .descriptorCount = 1,
+        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+        .pImmutableSamplers = nullptr,
+    },
+    {
+        .binding = BL3D_BINDING_OUTPUT_BUFFER,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .descriptorCount = 1,
+        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+        .pImmutableSamplers = nullptr,
+    },
+        }},
+        std::array<VkDescriptorUpdateTemplateEntry, 2>{{
+        {
+            .dstBinding = BL3D_BINDING_INPUT_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .offset = BL3D_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+        {
+            .dstBinding = BL3D_BINDING_OUTPUT_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .offset = BL3D_BINDING_OUTPUT_BUFFER * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        }
+        }},
+        DescriptorBankInfo{
+            .uniform_buffers = 0,
+            .storage_buffers = 2,
+            .texture_buffers = 0,
+            .image_buffers = 0,
+            .textures = 0,
+            .images = 0,
+            .score = 2,
+        },
           COMPUTE_PUSH_CONSTANT_RANGE<sizeof(BlockLinearUnswizzle3DPushConstants)>,
           BLOCK_LINEAR_UNSWIZZLE_3D_BCN_COMP_SPV),
       scheduler{scheduler_},
@@ -822,8 +1124,6 @@ void BlockLinearUnswizzle3DPass::UnswizzleChunk(
     pc.blocks_dim[2] = z_count; // Only process the count
 
     compute_pass_descriptor_queue.Acquire(scheduler, 3);
-    compute_pass_descriptor_queue.AddBuffer(*image.runtime->swizzle_table_buffer, 0,
-                                           image.runtime->swizzle_table_size);
     compute_pass_descriptor_queue.AddBuffer(swizzled.buffer,
                                            sw.buffer_offset + swizzled.offset,
                                            image.guest_size_bytes - sw.buffer_offset);
@@ -859,6 +1159,23 @@ void BlockLinearUnswizzle3DPass::UnswizzleChunk(
             return;
         }
 
+        if (!is_first_chunk) {
+            const VkBufferMemoryBarrier reuse_barrier{
+                .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
+                .pNext = nullptr,
+                .srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
+                .dstAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
+                .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                .buffer = out_buffer,
+                .offset = 0,
+                .size = VK_WHOLE_SIZE,
+            };
+            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
+                                   VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, nullptr,
+                                   reuse_barrier, nullptr);
+        }
+
         device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
         cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
         cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});

src/video_core/renderer_vulkan/vk_compute_pass.h

@@ -25,6 +25,7 @@ struct SwizzleParameters;
 
 namespace Vulkan {
 
+using VideoCommon::Accelerated::BlockLinearSwizzle2DParams;
 using VideoCommon::Accelerated::BlockLinearSwizzle3DParams;
 
 class Device;
@@ -164,6 +165,56 @@ private:
     ComputePassDescriptorQueue& compute_pass_descriptor_queue;
 };
 
+class BlockLinearUnswizzleImage2DPass final : public ComputePass {
+public:
+    explicit BlockLinearUnswizzleImage2DPass(const Device& device_, Scheduler& scheduler_,
+                                             DescriptorPool& descriptor_pool_,
+                                             StagingBufferPool& staging_buffer_pool_,
+                                             ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
+    ~BlockLinearUnswizzleImage2DPass();
+
+    void Unswizzle(Image& image, const StagingBufferRef& map,
+                   std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+private:
+    Scheduler& scheduler;
+    StagingBufferPool& staging_buffer_pool;
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
+};
+
+class BlockLinearUnswizzleImage3DPass final : public ComputePass {
+public:
+    explicit BlockLinearUnswizzleImage3DPass(const Device& device_, Scheduler& scheduler_,
+                                             DescriptorPool& descriptor_pool_,
+                                             StagingBufferPool& staging_buffer_pool_,
+                                             ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
+    ~BlockLinearUnswizzleImage3DPass();
+
+    void Unswizzle(Image& image, const StagingBufferRef& map,
+                   std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+private:
+    Scheduler& scheduler;
+    StagingBufferPool& staging_buffer_pool;
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
+};
+
+class PitchUnswizzlePass final : public ComputePass {
+public:
+    explicit PitchUnswizzlePass(const Device& device_, Scheduler& scheduler_,
+                                DescriptorPool& descriptor_pool_,
+                                StagingBufferPool& staging_buffer_pool_,
+                                ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
+    ~PitchUnswizzlePass();
+
+    void Unswizzle(Image& image, const StagingBufferRef& map,
+                   std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+private:
+    Scheduler& scheduler;
+    StagingBufferPool& staging_buffer_pool;
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
+};
 
 class MSAACopyPass final : public ComputePass {
 public:

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -195,8 +195,42 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
     return allocator.CreateImage(image_ci);
 }
 
+[[nodiscard]] VkFormat UnswizzleStorageFormat(u32 bytes_per_block) {
+    switch (bytes_per_block) {
+    case 1:
+        return VK_FORMAT_R8_UINT;
+    case 2:
+        return VK_FORMAT_R16_UINT;
+    case 4:
+        return VK_FORMAT_R32_UINT;
+    case 8:
+        return VK_FORMAT_R32G32_UINT;
+    case 16:
+        return VK_FORMAT_R32G32B32A32_UINT;
+    default:
+        ASSERT_MSG(false, "Invalid bytes_per_block={} for accelerated unswizzle", bytes_per_block);
+        return VK_FORMAT_R32_UINT;
+    }
+}
+
+[[nodiscard]] bool IsUnswizzleStorageFormatSupported(const Device& device, u32 bytes_per_block) {
+    switch (bytes_per_block) {
+    case 1:
+        return device.IsStorageBuffer8BitAccessSupported();
+    case 2:
+        return device.IsStorageBuffer16BitAccessSupported();
+    case 4:
+    case 8:
+    case 16:
+        return true;
+    default:
+        return false;
+    }
+}
+
 [[nodiscard]] vk::ImageView MakeStorageView(const vk::Device& device, u32 level, VkImage image,
-                                            VkFormat format) {
+                                            VkFormat format,
+                                            VkImageViewType view_type = VK_IMAGE_VIEW_TYPE_2D_ARRAY) {
     static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{
         .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO,
         .pNext = nullptr,
@@ -207,7 +241,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
         .pNext = &storage_image_view_usage_create_info,
         .flags = 0,
         .image = image,
-        .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY,
+        .viewType = view_type,
         .format = format,
         .components{
             .r = VK_COMPONENT_SWIZZLE_IDENTITY,
@@ -887,6 +921,12 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched
     if (device.IsStorageImageMultisampleSupported()) {
         msaa_copy_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, compute_pass_descriptor_queue);
     }
+    bl_unswizzle_2d_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool,
+                                 compute_pass_descriptor_queue);
+    bl_unswizzle_3d_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool,
+                                 compute_pass_descriptor_queue);
+    pitch_unswizzle_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool,
+                                 compute_pass_descriptor_queue);
     if (!device.IsKhrImageFormatListSupported()) {
         return;
     }
@@ -894,6 +934,11 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched
         const auto image_format = static_cast<PixelFormat>(index_a);
         if (IsPixelFormatASTC(image_format) && !device.IsOptimalAstcSupported()) {
             view_formats[index_a].push_back(VK_FORMAT_A8B8G8R8_UNORM_PACK32);
+        } else if (!IsPixelFormatASTC(image_format) && !IsPixelFormatBCn(image_format)) {
+            const u32 bpp = VideoCore::Surface::BytesPerBlock(image_format);
+            if (IsUnswizzleStorageFormatSupported(device, bpp)) {
+                view_formats[index_a].push_back(UnswizzleStorageFormat(bpp));
+            }
         }
         for (size_t index_b = 0; index_b < VideoCore::Surface::MaxPixelFormat; index_b++) {
             const auto view_format = static_cast<PixelFormat>(index_b);
@@ -909,40 +954,6 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched
         bl3d_unswizzle_pass.emplace(device, scheduler, descriptor_pool,
                                    staging_buffer_pool, compute_pass_descriptor_queue);
     }
-
-    // --- Create swizzle table buffer ---
-    {
-        auto table = Tegra::Texture::MakeSwizzleTable();
-
-        swizzle_table_size = static_cast<VkDeviceSize>(table.size() * sizeof(table[0]));
-
-        auto staging = staging_buffer_pool.Request(swizzle_table_size, MemoryUsage::Upload);
-        std::memcpy(staging.mapped_span.data(), table.data(), static_cast<size_t>(swizzle_table_size));
-
-        VkBufferCreateInfo ci{
-            .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
-            .size = swizzle_table_size,
-            .usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT |
-                     VK_BUFFER_USAGE_TRANSFER_DST_BIT |
-                     VK_BUFFER_USAGE_TRANSFER_SRC_BIT,
-            .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
-        };
-        swizzle_table_buffer = memory_allocator.CreateBuffer(ci, MemoryUsage::DeviceLocal);
-
-        scheduler.RequestOutsideRenderPassOperationContext();
-        scheduler.Record([staging_buf = staging.buffer,
-                          dst_buf = *swizzle_table_buffer,
-                          size = swizzle_table_size,
-                          src_off = staging.offset](vk::CommandBuffer cmdbuf) {
-
-            const VkBufferCopy region{
-                .srcOffset = src_off,
-                .dstOffset = 0,
-                .size = size,
-            };
-            cmdbuf.CopyBuffer(staging_buf, dst_buf, region);
-        });
-    }
 }
 
 void TextureCacheRuntime::Finish() {
@@ -1614,6 +1625,16 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
         flags |= VideoCommon::ImageFlagBits::Converted;
         flags |= VideoCommon::ImageFlagBits::CostlyLoad;
     }
+    if (!IsPixelFormatASTC(info.format) && !IsPixelFormatBCn(info.format) &&
+        VideoCore::Surface::GetFormatType(info.format) ==
+            VideoCore::Surface::SurfaceType::ColorTexture &&
+        (info.type == ImageType::e2D || info.type == ImageType::e3D ||
+         info.type == ImageType::Linear)) {
+        if (IsUnswizzleStorageFormatSupported(runtime->device,
+                                              VideoCore::Surface::BytesPerBlock(info.format))) {
+            flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
+        }
+    }
     if (runtime->device.HasDebuggingToolAttached()) {
         original_image.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
     }
@@ -1627,6 +1648,19 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
             storage_image_views[level] =
                 MakeStorageView(device, level, *original_image, VK_FORMAT_A8B8G8R8_UNORM_PACK32);
         }
+    } else if (True(flags & VideoCommon::ImageFlagBits::AcceleratedUpload)) {
+        const auto& device = runtime->device.GetLogical();
+        const VkFormat storage_format =
+            UnswizzleStorageFormat(VideoCore::Surface::BytesPerBlock(info.format));
+        const VkImageViewType storage_view_type = info.type == ImageType::e3D
+                                                       ? VK_IMAGE_VIEW_TYPE_3D
+                                                   : info.type == ImageType::Linear
+                                                       ? VK_IMAGE_VIEW_TYPE_2D
+                                                       : VK_IMAGE_VIEW_TYPE_2D_ARRAY;
+        for (s32 level = 0; level < info.resources.levels; ++level) {
+            storage_image_views[level] =
+                MakeStorageView(device, level, *original_image, storage_format, storage_view_type);
+        }
     }
 }
 
@@ -2482,21 +2516,24 @@ void TextureCacheRuntime::AccelerateImageUpload(
         return astc_decoder_pass->Assemble(image, map, swizzles);
     }
 
-    if (!Settings::values.gpu_unswizzle_enabled.GetValue() || !bl3d_unswizzle_pass) {
-        if (IsPixelFormatBCn(image.info.format) && image.info.type == ImageType::e3D) {
-            ASSERT_MSG(false, "GPU unswizzle is disabled for BCn 3D texture");
+    if (IsPixelFormatBCn(image.info.format)) {
+        if (Settings::values.gpu_unswizzle_enabled.GetValue() && bl3d_unswizzle_pass &&
+            image.info.type == ImageType::e3D && image.info.resources.levels == 1 &&
+            image.info.resources.layers == 1) {
+            return bl3d_unswizzle_pass->Unswizzle(image, map, swizzles, z_start, z_count);
         }
-        ASSERT(false);
+        ASSERT(false && "GPU unswizzle is disabled for BCn 3D texture");
         return;
     }
 
-    if (bl3d_unswizzle_pass &&
-        IsPixelFormatBCn(image.info.format) &&
-        image.info.type == ImageType::e3D &&
-        image.info.resources.levels == 1 &&
-        image.info.resources.layers == 1) {
-
-        return bl3d_unswizzle_pass->Unswizzle(image, map, swizzles, z_start, z_count);
+    if (image.info.type == ImageType::e2D) {
+        return bl_unswizzle_2d_pass->Unswizzle(image, map, swizzles);
+    }
+    if (image.info.type == ImageType::e3D) {
+        return bl_unswizzle_3d_pass->Unswizzle(image, map, swizzles);
+    }
+    if (image.info.type == ImageType::Linear) {
+        return pitch_unswizzle_pass->Unswizzle(image, map, swizzles);
     }
 
     ASSERT(false);

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
@@ -130,9 +130,9 @@ public:
     std::optional<ASTCDecoderPass> astc_decoder_pass;
 
     std::optional<BlockLinearUnswizzle3DPass> bl3d_unswizzle_pass;
-    vk::Buffer swizzle_table_buffer;
-    VkDeviceSize swizzle_table_size = 0;
-
+    std::optional<BlockLinearUnswizzleImage2DPass> bl_unswizzle_2d_pass;
+    std::optional<BlockLinearUnswizzleImage3DPass> bl_unswizzle_3d_pass;
+    std::optional<PitchUnswizzlePass> pitch_unswizzle_pass;
     std::optional<MSAACopyPass> msaa_copy_pass;
     const Settings::ResolutionScalingInfo& resolution;
     std::array<std::vector<VkFormat>, VideoCore::Surface::MaxPixelFormat> view_formats;

src/video_core/shader_environment.cpp

@@ -255,8 +255,7 @@ std::optional<u64> GenericEnvironment::TryFindSize() {
     static constexpr u64 SELF_BRANCH_A = 0xE2400FFFFF87000FULL;
     static constexpr u64 SELF_BRANCH_B = 0xE2400FFFFF07000FULL;
 
-    static constexpr u64 MESA_EXIT_MASK = 0xFFF00000000F001FULL;
-    static constexpr u64 MESA_EXIT_VALUE = (0xE30ULL << 52) | (0x7ULL << 16) | 0xFULL;
+    static constexpr u64 EXIT_VALUE = 0xE30000000007000FULL;
 
     code.resize(MAXIMUM_SIZE / INST_SIZE);
 
@@ -271,7 +270,7 @@ std::optional<u64> GenericEnvironment::TryFindSize() {
             if (inst == SELF_BRANCH_A || inst == SELF_BRANCH_B) {
                 return offset + index;
             }
-            if ((inst & MESA_EXIT_MASK) == MESA_EXIT_VALUE) {
+            if (!is_proprietary_driver && inst == EXIT_VALUE) {
                 return offset + index + INST_SIZE;
             }
         }

src/video_core/textures/decoders.h

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -23,24 +26,6 @@ constexpr u32 GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_
 constexpr u32 SWIZZLE_X_BITS = 0b100101111;
 constexpr u32 SWIZZLE_Y_BITS = 0b011010000;
 
-using SwizzleTable = std::array<std::array<u32, GOB_SIZE_X>, GOB_SIZE_Y>;
-
-/**
- * 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;
-}
-
 /// Unswizzles a block linear texture into linear memory.
 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,

src/video_core/vulkan_common/vulkan_device.h

@@ -888,6 +888,16 @@ FN_MAX_LIMIT_LIST
                features.bit16_storage.storageBuffer16BitAccess;
     }
 
+    /// Returns true if the device supports reading 8-bit values from a storage buffer.
+    bool IsStorageBuffer8BitAccessSupported() const {
+        return features.bit8_storage.storageBuffer8BitAccess;
+    }
+
+    /// Returns true if the device supports reading 16-bit values from a storage buffer.
+    bool IsStorageBuffer16BitAccessSupported() const {
+        return features.bit16_storage.storageBuffer16BitAccess;
+    }
+
     [[nodiscard]] static constexpr bool CheckBrokenCompute(VkDriverId driver_id,
                                                            u32 driver_version) {
         if (driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS) {