[TEST] Adjustments on pipeline hash collisions

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt

@@ -586,6 +586,7 @@ abstract class SettingsItem(
                     IntSetting.FSR_SHARPENING_SLIDER,
                     titleId = R.string.fsr_sharpness,
                     descriptionId = R.string.fsr_sharpness_description,
+                    max = 200,
                     units = "%"
                 )
             )

src/android/app/src/main/java/org/yuzu/yuzu_emu/fragments/EmulationFragment.kt

@@ -1182,7 +1182,7 @@ class EmulationFragment : Fragment(), SurfaceHolder.Callback {
                     container,
                     IntSetting.FSR_SHARPENING_SLIDER,
                     minValue = 0,
-                    maxValue = 100,
+                    maxValue = 200,
                     units = "%"
                 )
             }

src/video_core/buffer_cache/buffer_cache.h

@@ -7,6 +7,7 @@
 #pragma once
 
 #include <algorithm>
+#include <bit>
 #include <memory>
 #include <numeric>
 
@@ -809,46 +810,46 @@ void BufferCache<P>::BindHostVertexBuffers() {
 
     if (use_optimized_vertex_buffers) {
         auto& flags = maxwell3d->dirty.flags;
-        u32 enabled_mask = enabled_vertex_buffers_mask;
+        const u32 enabled_mask = enabled_vertex_buffers_mask;
-        HostBindings<Buffer> bindings{};
+        bool any_dirty = false;
-        u32 last_index = (std::numeric_limits<u32>::max)();
+        u32 pending_mask = enabled_mask;
-        const auto flush_bindings = [&]() {
+        while (pending_mask != 0) {
-            if (bindings.buffers.empty()) {
+            const u32 index = std::countr_zero(pending_mask);
-                return;
+            pending_mask &= (pending_mask - 1);
-            }
-            bindings.max_index = bindings.min_index + static_cast<u32>(bindings.buffers.size());
-            runtime.BindVertexBuffers(bindings);
-            bindings = HostBindings<Buffer>{};
-            last_index = (std::numeric_limits<u32>::max)();
-        };
-        while (enabled_mask != 0) {
-            const u32 index = std::countr_zero(enabled_mask);
-            enabled_mask &= (enabled_mask - 1);
             const Binding& binding = VertexBufferSlot(index);
             Buffer& buffer = slot_buffers[binding.buffer_id];
             TouchBuffer(buffer, binding.buffer_id);
             SynchronizeBuffer(buffer, binding.device_addr, binding.size);
-            if (!flags[Dirty::VertexBuffer0 + index]) {
+            any_dirty |= flags[Dirty::VertexBuffer0 + index];
-                flush_bindings();
+        }
-                continue;
+        if (enabled_mask == 0 || !any_dirty) {
-            }
+            return;
+        }
+        const u32 min_index = static_cast<u32>(std::countr_zero(enabled_mask));
+        const u32 max_index = 32u - static_cast<u32>(std::countl_zero(enabled_mask));
+        HostBindings<Buffer> bindings{};
+        bindings.min_index = min_index;
+        bindings.max_index = max_index;
+        for (u32 index = min_index; index < max_index; ++index) {
             flags[Dirty::VertexBuffer0 + index] = false;
             const u32 stride = maxwell3d->regs.vertex_streams[index].stride;
+            if ((enabled_mask & (1u << index)) == 0) {
+                bindings.buffers.push_back(&slot_buffers[NULL_BUFFER_ID]);
+                bindings.offsets.push_back(0);
+                bindings.sizes.push_back(0);
+                bindings.strides.push_back(stride);
+                continue;
+            }
+            const Binding& binding = VertexBufferSlot(index);
+            Buffer& buffer = slot_buffers[binding.buffer_id];
             const u32 offset = buffer.Offset(binding.device_addr);
             buffer.MarkUsage(offset, binding.size);
-            if (!bindings.buffers.empty() && index != last_index + 1) {
-                flush_bindings();
-            }
-            if (bindings.buffers.empty()) {
-                bindings.min_index = index;
-            }
             bindings.buffers.push_back(&buffer);
             bindings.offsets.push_back(offset);
             bindings.sizes.push_back(binding.size);
             bindings.strides.push_back(stride);
-            last_index = index;
         }
-        flush_bindings();
+        runtime.BindVertexBuffers(bindings);
     } else {
         HostBindings<typename P::Buffer> host_bindings;
         bool any_valid{false};

src/video_core/host_shaders/CMakeLists.txt

@@ -17,11 +17,13 @@ set(SHADER_FILES
     ${CMAKE_CURRENT_SOURCE_DIR}/astc_decoder.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/blit_color_float.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_2d.comp
+    ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_2d_buffer.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/blit_color_msaa.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/blit_depth_msaa.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/blit_depth_stencil_msaa.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_3d.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_3d_bcn.comp
+    ${CMAKE_CURRENT_SOURCE_DIR}/block_linear_unswizzle_3d_buffer.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_abgr8_to_d24s8.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_abgr8_to_d32f.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_d32f_to_abgr8.frag

src/video_core/host_shaders/block_linear_unswizzle_2d_buffer.comp

@@ -0,0 +1,104 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#version 430
+
+#extension GL_EXT_shader_16bit_storage : require
+#extension GL_EXT_shader_8bit_storage : require
+
+#define BINDING_INPUT_BUFFER 0
+#define BINDING_OUTPUT_BUFFER 1
+
+layout(push_constant) uniform PushConstants {
+    uvec3 dim;
+    uint bytes_per_block_log2;
+
+    uvec3 origin;
+    uint layer_stride;
+
+    uint block_size;
+    uint x_shift;
+    uint block_height;
+    uint block_height_mask;
+} pc;
+
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU32 { uint u32data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU64 { uvec2 u64data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU128 { uvec4 u128data[]; };
+
+layout(binding = BINDING_OUTPUT_BUFFER, std430) writeonly buffer OutputBuffer {
+    uint out_u32[];
+};
+
+layout(local_size_x = 16, local_size_y = 8, local_size_z = 1) in;
+
+const uint GOB_SIZE_X = 64;
+const uint GOB_SIZE_Y = 8;
+
+const uint GOB_SIZE_X_SHIFT = 6;
+const uint GOB_SIZE_Y_SHIFT = 3;
+const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_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);
+}
+
+uint SwizzleOffset(uvec2 pos) {
+    pos = pos & SWIZZLE_MASK;
+    return SwizzleTable(pos.y * 64u + pos.x);
+}
+
+uvec4 ReadTexel(uint offset) {
+    switch (pc.bytes_per_block_log2) {
+    case 2u:
+        return uvec4(u32data[offset / 4u], 0u, 0u, 0u);
+    case 3u:
+        return uvec4(u64data[offset / 8u], 0u, 0u);
+    case 4u:
+        return u128data[offset / 16u];
+    }
+    return uvec4(0u);
+}
+
+void main() {
+    uvec3 coord = gl_GlobalInvocationID;
+    if (coord.x >= pc.dim.x || coord.y >= pc.dim.y || coord.z >= pc.dim.z) {
+        return;
+    }
+
+    uvec3 pos = coord + pc.origin;
+    pos.x <<= pc.bytes_per_block_log2;
+
+    uint swizzle = SwizzleOffset(pos.xy);
+    uint block_y = pos.y >> GOB_SIZE_Y_SHIFT;
+
+    uint offset = 0u;
+    offset += pos.z * pc.layer_stride;
+    offset += (block_y >> pc.block_height) * pc.block_size;
+    offset += (block_y & pc.block_height_mask) << GOB_SIZE_SHIFT;
+    offset += (pos.x >> GOB_SIZE_X_SHIFT) << pc.x_shift;
+    offset += swizzle;
+
+    uvec4 texel = ReadTexel(offset);
+
+    uint words = 1u << (pc.bytes_per_block_log2 - 2u);
+    uint linear_index = coord.x + coord.y * pc.dim.x + coord.z * pc.dim.x * pc.dim.y;
+    uint out_idx = linear_index * words;
+
+    out_u32[out_idx] = texel.x;
+    if (words > 1u) {
+        out_u32[out_idx + 1u] = texel.y;
+    }
+    if (words > 2u) {
+        out_u32[out_idx + 2u] = texel.z;
+        out_u32[out_idx + 3u] = texel.w;
+    }
+}

src/video_core/host_shaders/block_linear_unswizzle_3d_buffer.comp

@@ -0,0 +1,105 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#version 430
+
+#define BINDING_INPUT_BUFFER 0
+#define BINDING_OUTPUT_BUFFER 1
+
+layout(push_constant) uniform PushConstants {
+    uvec3 dim;
+    uint bytes_per_block_log2;
+
+    uvec3 origin;
+    uint slice_size;
+
+    uint block_size;
+    uint x_shift;
+    uint block_height;
+    uint block_height_mask;
+
+    uint block_depth;
+    uint block_depth_mask;
+} pc;
+
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU32 { uint u32data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU64 { uvec2 u64data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) buffer InputBufferU128 { uvec4 u128data[]; };
+
+layout(binding = BINDING_OUTPUT_BUFFER, std430) writeonly buffer OutputBuffer {
+    uint out_u32[];
+};
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 4) in;
+
+const uint GOB_SIZE_X = 64;
+const uint GOB_SIZE_Y = 8;
+
+const uint GOB_SIZE_X_SHIFT = 6;
+const uint GOB_SIZE_Y_SHIFT = 3;
+const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_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);
+}
+
+uint SwizzleOffset(uvec2 pos) {
+    pos = pos & SWIZZLE_MASK;
+    return SwizzleTable(pos.y * 64u + pos.x);
+}
+
+uvec4 ReadTexel(uint offset) {
+    switch (pc.bytes_per_block_log2) {
+    case 2u:
+        return uvec4(u32data[offset / 4u], 0u, 0u, 0u);
+    case 3u:
+        return uvec4(u64data[offset / 8u], 0u, 0u);
+    case 4u:
+        return u128data[offset / 16u];
+    }
+    return uvec4(0u);
+}
+
+void main() {
+    uvec3 coord = gl_GlobalInvocationID;
+    if (coord.x >= pc.dim.x || coord.y >= pc.dim.y || coord.z >= pc.dim.z) {
+        return;
+    }
+
+    uvec3 pos = coord + pc.origin;
+    pos.x <<= pc.bytes_per_block_log2;
+
+    uint swizzle = SwizzleOffset(pos.xy);
+    uint block_y = pos.y >> GOB_SIZE_Y_SHIFT;
+
+    uint offset = 0u;
+    offset += (pos.z >> pc.block_depth) * pc.slice_size;
+    offset += (pos.z & pc.block_depth_mask) << (GOB_SIZE_SHIFT + pc.block_height);
+    offset += (block_y >> pc.block_height) * pc.block_size;
+    offset += (block_y & pc.block_height_mask) << GOB_SIZE_SHIFT;
+    offset += (pos.x >> GOB_SIZE_X_SHIFT) << pc.x_shift;
+    offset += swizzle;
+
+    uvec4 texel = ReadTexel(offset);
+
+    uint words = 1u << (pc.bytes_per_block_log2 - 2u);
+    uint linear_index = coord.x + coord.y * pc.dim.x + coord.z * pc.dim.x * pc.dim.y;
+    uint out_idx = linear_index * words;
+
+    out_u32[out_idx] = texel.x;
+    if (words > 1u) {
+        out_u32[out_idx + 1u] = texel.y;
+    }
+    if (words > 2u) {
+        out_u32[out_idx + 2u] = texel.z;
+        out_u32[out_idx + 3u] = texel.w;
+    }
+}

src/video_core/host_shaders/sgsr1_shader.vert

@@ -7,6 +7,7 @@ layout(push_constant) uniform constants {
     vec2 scale;
     vec2 size;
     vec2 resize_factor;
+    vec2 crop_offset;
     float edge_sharpness;
 };
 layout(location = 0) out highp vec2 texcoord;
@@ -15,5 +16,5 @@ void main() {
     float x = float((gl_VertexIndex & 1) << 2);
     float y = float((gl_VertexIndex & 2) << 1);
     gl_Position = vec4(x - 1.0f, y - 1.0f, 0.0, 1.0f) * vec4(sign(resize_factor), 1.f, 1.f);
-    texcoord = vec2(x, y) * abs(resize_factor) * 0.5;
+    texcoord = crop_offset + vec2(x, y) * abs(resize_factor) * 0.5;
 }

src/video_core/host_shaders/sgsr1_shader_mobile.frag

@@ -14,6 +14,7 @@ layout(push_constant) uniform constants {
     vec2 scale;
     vec2 size;
     vec2 resize_factor;
+    vec2 crop_offset;
     float edge_sharpness;
 };
 layout(set = 0, binding = 0) uniform sampler2D sampler0;

src/video_core/host_shaders/sgsr1_shader_mobile_edge_direction.frag

@@ -13,6 +13,7 @@
 layout( push_constant ) uniform constants {
     vec4 ViewportInfo[1];
     vec2 ResizeFactor;
+    vec2 CropOffset;
     float EdgeSharpness;
 };
 layout(set = 0, binding = 0) uniform sampler2D ps0;

src/video_core/renderer_vulkan/blit_image.cpp

@@ -1191,7 +1191,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceColorPipeline(const BlitImagePipelineKe
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *blit_color_pipelines.back();
 }
 
@@ -1223,7 +1223,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceDepthStencilPipeline(const BlitImagePip
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *blit_depth_stencil_pipelines.back();
 }
 
@@ -1276,7 +1276,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearColorPipeline(const BlitImagePipel
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *clear_color_pipelines.back();
 }
 
@@ -1332,7 +1332,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearStencilPipeline(
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *clear_stencil_pipelines.back();
 }
 
@@ -1375,7 +1375,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceBlitColorMSAAPipeline(const BlitMSAAPip
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *blit_msaa_color_pipelines.back();
 }
 
@@ -1413,7 +1413,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceResolveDepthStencilPipeline(VkRenderPas
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *pipelines.back();
 }
 
@@ -1458,7 +1458,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceMSAACopyPipeline(const MSAACopyPipeline
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    }));
+    }, device.StaticPipelineCache()));
     return *msaa_copy_pipelines.back();
 }
 
@@ -1499,7 +1499,7 @@ void BlitImageHelper::ConvertPipelineEx(vk::Pipeline& pipeline, VkRenderPass ren
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    });
+    }, device.StaticPipelineCache());
 }
 
 void BlitImageHelper::ConvertPipelineColorTargetEx(vk::Pipeline& pipeline, VkRenderPass renderpass,
@@ -1542,7 +1542,7 @@ void BlitImageHelper::ConvertPipeline(vk::Pipeline& pipeline, VkRenderPass rende
         .subpass = 0,
         .basePipelineHandle = VK_NULL_HANDLE,
         .basePipelineIndex = 0,
-    });
+    }, device.StaticPipelineCache());
 }
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/present/sgsr.cpp

@@ -1,6 +1,8 @@
 // SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
+#include <algorithm>
+
 #include "common/common_types.h"
 #include "common/div_ceil.h"
 #include "common/settings.h"
@@ -17,7 +19,7 @@
 
 namespace Vulkan {
 
-using PushConstants = std::array<u32, 4 + 2 + 1>;
+using PushConstants = std::array<u32, 4 + 2 + 2 + 1>;
 
 SGSR::SGSR(const Device& device, MemoryAllocator& memory_allocator, size_t image_count, VkExtent2D extent, bool edge_dir)
     : m_memory_allocator{memory_allocator}
@@ -100,26 +102,28 @@ VkImageView SGSR::Draw(const Device& device, Scheduler& scheduler, size_t image_
 
     const f32 input_image_width = f32(input_image_extent.width);
     const f32 input_image_height = f32(input_image_extent.height);
-    const f32 viewport_width = (crop_rect.right - crop_rect.left) * input_image_width;
+    const f32 crop_width = (crop_rect.right - crop_rect.left) * input_image_width;
-    const f32 viewport_height = (crop_rect.bottom - crop_rect.top) * input_image_height;
+    const f32 crop_height = (crop_rect.bottom - crop_rect.top) * input_image_height;
-    // expected [0, 2]
+    static constexpr f32 EDGE_SHARPNESS_MAX = 2.0f;
-    const f32 sharpening = f32(Settings::values.fsr_sharpening_slider.GetValue()) / 100.0f;
+    const f32 edge_sharpness =
+        EDGE_SHARPNESS_MAX - f32(Settings::values.fsr_sharpening_slider.GetValue()) / 200.0f;
 
-    // p = (tex * viewport) / input = [0,n] (normalized texcoords)
-    // p * input = [0,1024], [0,768]
     // layout( push_constant ) uniform constants {
     //     highp vec4 ViewportInfo[1];
     //     highp vec2 ResizeFactor;
+    //     highp vec2 CropOffset;
     //     highp float EdgeSharpness;
     // };
     PushConstants viewport_con{};
-    viewport_con[0] = std::bit_cast<u32>(std::abs(1.f / viewport_width));
+    viewport_con[0] = std::bit_cast<u32>(1.f / input_image_width);
-    viewport_con[1] = std::bit_cast<u32>(std::abs(1.f / viewport_height));
+    viewport_con[1] = std::bit_cast<u32>(1.f / input_image_height);
-    viewport_con[2] = std::bit_cast<u32>(std::abs(viewport_width));
+    viewport_con[2] = std::bit_cast<u32>(input_image_width);
-    viewport_con[3] = std::bit_cast<u32>(std::abs(viewport_height));
+    viewport_con[3] = std::bit_cast<u32>(input_image_height);
-    viewport_con[4] = std::bit_cast<u32>(viewport_width / input_image_width);
+    viewport_con[4] = std::bit_cast<u32>(crop_width / input_image_width);
-    viewport_con[5] = std::bit_cast<u32>(viewport_height / input_image_height);
+    viewport_con[5] = std::bit_cast<u32>(crop_height / input_image_height);
-    viewport_con[6] = std::bit_cast<u32>(sharpening);
+    viewport_con[6] = std::bit_cast<u32>((std::min)(crop_rect.left, crop_rect.right));
+    viewport_con[7] = std::bit_cast<u32>((std::min)(crop_rect.top, crop_rect.bottom));
+    viewport_con[8] = std::bit_cast<u32>(edge_sharpness);
 
     UploadImages(device, scheduler);
     UpdateDescriptorSets(device, source_image_view, image_index);

src/video_core/renderer_vulkan/present/util.cpp

@@ -491,7 +491,7 @@ static vk::Pipeline CreateWrappedPipelineImpl(
         .subpass = 0,
         .basePipelineHandle = 0,
         .basePipelineIndex = 0,
-    });
+    }, device.StaticPipelineCache());
 }
 
 vk::Pipeline CreateWrappedPipeline(const Device& device, vk::RenderPass& renderpass,

src/video_core/renderer_vulkan/vk_compute_pass.cpp

@@ -22,7 +22,9 @@
 #include "video_core/host_shaders/resolve_conditional_render_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/host_shaders/block_linear_unswizzle_2d_buffer_comp_spv.h"
 #include "video_core/host_shaders/block_linear_unswizzle_3d_bcn_comp_spv.h"
+#include "video_core/host_shaders/block_linear_unswizzle_3d_buffer_comp_spv.h"
 #include "video_core/renderer_vulkan/vk_compute_pass.h"
 #include "video_core/surface.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
@@ -268,7 +270,7 @@ ComputePass::ComputePass(const Device& device_, Scheduler& scheduler, Descriptor
         .layout = *layout,
         .basePipelineHandle = {},
         .basePipelineIndex = 0,
-    });
+    }, device.StaticPipelineCache());
 }
 
 ComputePass::~ComputePass() = default;
@@ -872,4 +874,479 @@ void BlockLinearUnswizzle3DPass::UnswizzleChunk(
     });
 }
 
+namespace {
+constexpr u32 BL2D_BINDING_INPUT_BUFFER = 0;
+constexpr u32 BL2D_BINDING_OUTPUT_BUFFER = 1;
+
+struct alignas(16) BlockLinearUnswizzle2DPushConstants {
+    std::array<u32, 3> dim;
+    u32 bytes_per_block_log2;
+    std::array<u32, 3> origin;
+    u32 layer_stride;
+    u32 block_size;
+    u32 x_shift;
+    u32 block_height;
+    u32 block_height_mask;
+};
+static_assert(sizeof(BlockLinearUnswizzle2DPushConstants) <= 128);
+
+constexpr std::array<VkDescriptorSetLayoutBinding, 2> BL2D_BINDINGS{{
+    {
+        .binding = BL2D_BINDING_INPUT_BUFFER,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .descriptorCount = 1,
+        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+        .pImmutableSamplers = nullptr,
+    },
+    {
+        .binding = BL2D_BINDING_OUTPUT_BUFFER,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .descriptorCount = 1,
+        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+        .pImmutableSamplers = nullptr,
+    },
+}};
+
+constexpr std::array<VkDescriptorUpdateTemplateEntry, 2> BL2D_TEMPLATE{{
+    {
+        .dstBinding = BL2D_BINDING_INPUT_BUFFER,
+        .dstArrayElement = 0,
+        .descriptorCount = 1,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .offset = BL2D_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry),
+        .stride = sizeof(DescriptorUpdateEntry),
+    },
+    {
+        .dstBinding = BL2D_BINDING_OUTPUT_BUFFER,
+        .dstArrayElement = 0,
+        .descriptorCount = 1,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .offset = BL2D_BINDING_OUTPUT_BUFFER * sizeof(DescriptorUpdateEntry),
+        .stride = sizeof(DescriptorUpdateEntry),
+    },
+}};
+
+constexpr DescriptorBankInfo BL2D_BANK_INFO{
+    .uniform_buffers = 0,
+    .storage_buffers = 2,
+    .texture_buffers = 0,
+    .image_buffers = 0,
+    .textures = 0,
+    .images = 0,
+    .score = 2,
+};
+} // Anonymous namespace
+
+BlockLinearUnswizzle2DPass::BlockLinearUnswizzle2DPass(
+    const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+    StagingBufferPool& staging_buffer_pool_,
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
+    : ComputePass(device_, scheduler_, descriptor_pool_, BL2D_BINDINGS, BL2D_TEMPLATE,
+                  BL2D_BANK_INFO,
+                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(BlockLinearUnswizzle2DPushConstants)>,
+                  BLOCK_LINEAR_UNSWIZZLE_2D_BUFFER_COMP_SPV),
+      scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
+      compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
+
+BlockLinearUnswizzle2DPass::~BlockLinearUnswizzle2DPass() = default;
+
+bool BlockLinearUnswizzle2DPass::IsSupported(const VideoCommon::ImageInfo& info) {
+    if (info.type != VideoCommon::ImageType::e2D) {
+        return false;
+    }
+    if (info.resources.levels != 1 || info.resources.layers != 1) {
+        return false;
+    }
+    if (info.num_samples > 1) {
+        return false;
+    }
+    if (VideoCore::Surface::IsPixelFormatASTC(info.format) ||
+        VideoCore::Surface::IsPixelFormatBCn(info.format)) {
+        return false;
+    }
+    const u32 bytes_per_block = VideoCore::Surface::BytesPerBlock(info.format);
+    if (bytes_per_block != 4 && bytes_per_block != 8 && bytes_per_block != 16) {
+        return false;
+    }
+    return VideoCore::Surface::DefaultBlockWidth(info.format) == 1 &&
+           VideoCore::Surface::DefaultBlockHeight(info.format) == 1;
+}
+
+void BlockLinearUnswizzle2DPass::Unswizzle(
+    Image& image, const StagingBufferRef& swizzled,
+    std::span<const VideoCommon::SwizzleParameters> swizzles) {
+    if (swizzles.empty()) {
+        return;
+    }
+    const VideoCommon::SwizzleParameters& sw = swizzles.front();
+    const auto params = VideoCommon::Accelerated::MakeBlockLinearSwizzle2DParams(sw, image.info);
+
+    const u32 width = sw.num_tiles.width;
+    const u32 height = sw.num_tiles.height;
+    const u32 depth = image.info.resources.layers;
+    const u32 bytes_per_block = 1u << params.bytes_per_block_log2;
+    const VkDeviceSize output_size =
+        static_cast<VkDeviceSize>(width) * height * depth * bytes_per_block;
+
+    const StagingBufferRef output =
+        staging_buffer_pool.Request(static_cast<size_t>(output_size), MemoryUsage::DeviceLocal);
+
+    BlockLinearUnswizzle2DPushConstants pc{};
+    pc.dim = {width, height, depth};
+    pc.bytes_per_block_log2 = params.bytes_per_block_log2;
+    pc.origin = params.origin;
+    pc.layer_stride = params.layer_stride;
+    pc.block_size = params.block_size;
+    pc.x_shift = params.x_shift;
+    pc.block_height = params.block_height;
+    pc.block_height_mask = params.block_height_mask;
+
+    scheduler.RequestOutsideRenderPassOperationContext();
+
+    compute_pass_descriptor_queue.Acquire(scheduler, 2);
+    compute_pass_descriptor_queue.AddBuffer(swizzled.buffer, sw.buffer_offset + swizzled.offset,
+                                            image.guest_size_bytes - sw.buffer_offset);
+    compute_pass_descriptor_queue.AddBuffer(output.buffer, output.offset, output_size);
+
+    const void* descriptor_data = compute_pass_descriptor_queue.UpdateData();
+    const VkDescriptorSet set = descriptor_allocator.Commit();
+
+    const u32 gx = Common::DivCeil(width, 16u);
+    const u32 gy = Common::DivCeil(height, 8u);
+    const bool is_initialized = image.ExchangeInitialization();
+
+    const VkBuffer out_buffer = output.buffer;
+    const VkDeviceSize out_offset = output.offset;
+    const VkImage dst_image = image.Handle();
+    const VkImageAspectFlags aspect = image.AspectMask();
+
+    scheduler.Record([this, set, descriptor_data, pc, gx, gy, depth, output_size, out_buffer,
+                      out_offset, dst_image, aspect, width, height,
+                      is_initialized](vk::CommandBuffer cmdbuf) {
+        if (dst_image == VK_NULL_HANDLE || out_buffer == VK_NULL_HANDLE) {
+            return;
+        }
+        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, {});
+        cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(pc), &pc);
+        cmdbuf.Dispatch(gx, gy, depth);
+
+        const VkBufferMemoryBarrier buffer_barrier{
+            .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .buffer = out_buffer,
+            .offset = out_offset,
+            .size = output_size,
+        };
+        const VkImageMemoryBarrier pre_copy{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = static_cast<VkAccessFlags>(is_initialized ? VK_ACCESS_SHADER_READ_BIT
+                                                                      : VK_ACCESS_NONE),
+            .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
+            .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = dst_image,
+            .subresourceRange{
+                .aspectMask = aspect,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
+                                   (is_initialized ? vk::PIPELINE_STAGE_GRAPHICS_COMPUTE
+                                                   : VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT),
+                               VK_PIPELINE_STAGE_TRANSFER_BIT, 0, {}, buffer_barrier, pre_copy);
+
+        const VkBufferImageCopy copy{
+            .bufferOffset = out_offset,
+            .bufferRowLength = 0,
+            .bufferImageHeight = 0,
+            .imageSubresource{
+                .aspectMask = aspect,
+                .mipLevel = 0,
+                .baseArrayLayer = 0,
+                .layerCount = depth,
+            },
+            .imageOffset = {0, 0, 0},
+            .imageExtent = {width, height, 1},
+        };
+        cmdbuf.CopyBufferToImage(out_buffer, dst_image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copy);
+
+        const VkImageMemoryBarrier post_copy{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = dst_image,
+            .subresourceRange{
+                .aspectMask = aspect,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE,
+                               0, {}, {}, post_copy);
+    });
+}
+
+namespace {
+constexpr u32 BL3DB_BINDING_INPUT_BUFFER = 0;
+constexpr u32 BL3DB_BINDING_OUTPUT_BUFFER = 1;
+
+struct alignas(16) BlockLinearUnswizzle3DBufferPushConstants {
+    std::array<u32, 3> dim;
+    u32 bytes_per_block_log2;
+    std::array<u32, 3> origin;
+    u32 slice_size;
+    u32 block_size;
+    u32 x_shift;
+    u32 block_height;
+    u32 block_height_mask;
+    u32 block_depth;
+    u32 block_depth_mask;
+};
+static_assert(sizeof(BlockLinearUnswizzle3DBufferPushConstants) <= 128);
+
+constexpr std::array<VkDescriptorSetLayoutBinding, 2> BL3DB_BINDINGS{{
+    {
+        .binding = BL3DB_BINDING_INPUT_BUFFER,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .descriptorCount = 1,
+        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+        .pImmutableSamplers = nullptr,
+    },
+    {
+        .binding = BL3DB_BINDING_OUTPUT_BUFFER,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .descriptorCount = 1,
+        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+        .pImmutableSamplers = nullptr,
+    },
+}};
+
+constexpr std::array<VkDescriptorUpdateTemplateEntry, 2> BL3DB_TEMPLATE{{
+    {
+        .dstBinding = BL3DB_BINDING_INPUT_BUFFER,
+        .dstArrayElement = 0,
+        .descriptorCount = 1,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .offset = BL3DB_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry),
+        .stride = sizeof(DescriptorUpdateEntry),
+    },
+    {
+        .dstBinding = BL3DB_BINDING_OUTPUT_BUFFER,
+        .dstArrayElement = 0,
+        .descriptorCount = 1,
+        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+        .offset = BL3DB_BINDING_OUTPUT_BUFFER * sizeof(DescriptorUpdateEntry),
+        .stride = sizeof(DescriptorUpdateEntry),
+    },
+}};
+
+constexpr DescriptorBankInfo BL3DB_BANK_INFO{
+    .uniform_buffers = 0,
+    .storage_buffers = 2,
+    .texture_buffers = 0,
+    .image_buffers = 0,
+    .textures = 0,
+    .images = 0,
+    .score = 2,
+};
+} // Anonymous namespace
+
+BlockLinearUnswizzle3DBufferPass::BlockLinearUnswizzle3DBufferPass(
+    const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+    StagingBufferPool& staging_buffer_pool_,
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
+    : ComputePass(device_, scheduler_, descriptor_pool_, BL3DB_BINDINGS, BL3DB_TEMPLATE,
+                  BL3DB_BANK_INFO,
+                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(BlockLinearUnswizzle3DBufferPushConstants)>,
+                  BLOCK_LINEAR_UNSWIZZLE_3D_BUFFER_COMP_SPV),
+      scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
+      compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
+
+BlockLinearUnswizzle3DBufferPass::~BlockLinearUnswizzle3DBufferPass() = default;
+
+bool BlockLinearUnswizzle3DBufferPass::IsSupported(const Device& device,
+                                                   const VideoCommon::ImageInfo& info) {
+    if (info.type != VideoCommon::ImageType::e3D) {
+        return false;
+    }
+    if (info.resources.levels != 1 || info.resources.layers != 1) {
+        return false;
+    }
+    if (info.num_samples > 1) {
+        return false;
+    }
+    if (info.size.depth <= 1) {
+        return false;
+    }
+    if (VideoCore::Surface::IsPixelFormatASTC(info.format)) {
+        return false;
+    }
+    if (VideoCore::Surface::IsPixelFormatBCn(info.format) && !device.IsOptimalBcnSupported()) {
+        return false;
+    }
+    const u32 bytes_per_block = VideoCore::Surface::BytesPerBlock(info.format);
+    return bytes_per_block == 4 || bytes_per_block == 8 || bytes_per_block == 16;
+}
+
+void BlockLinearUnswizzle3DBufferPass::Unswizzle(
+    Image& image, const StagingBufferRef& swizzled,
+    std::span<const VideoCommon::SwizzleParameters> swizzles) {
+    if (swizzles.empty()) {
+        return;
+    }
+
+    const VideoCommon::SwizzleParameters& sw = swizzles.front();
+    const auto params = VideoCommon::Accelerated::MakeBlockLinearSwizzle3DParams(sw, image.info);
+
+    const u32 blocks_x = sw.num_tiles.width;
+    const u32 blocks_y = sw.num_tiles.height;
+    const u32 blocks_z = sw.num_tiles.depth;
+    const u32 bytes_per_block = 1u << params.bytes_per_block_log2;
+    const VkDeviceSize output_size =
+        static_cast<VkDeviceSize>(blocks_x) * blocks_y * blocks_z * bytes_per_block;
+
+    const StagingBufferRef output =
+        staging_buffer_pool.Request(static_cast<size_t>(output_size), MemoryUsage::DeviceLocal);
+
+    BlockLinearUnswizzle3DBufferPushConstants pc{};
+    pc.dim = {blocks_x, blocks_y, blocks_z};
+    pc.bytes_per_block_log2 = params.bytes_per_block_log2;
+    pc.origin = params.origin;
+    pc.slice_size = params.slice_size;
+    pc.block_size = params.block_size;
+    pc.x_shift = params.x_shift;
+    pc.block_height = params.block_height;
+    pc.block_height_mask = params.block_height_mask;
+    pc.block_depth = params.block_depth;
+    pc.block_depth_mask = params.block_depth_mask;
+
+    scheduler.RequestOutsideRenderPassOperationContext();
+
+    compute_pass_descriptor_queue.Acquire(scheduler, 2);
+    compute_pass_descriptor_queue.AddBuffer(swizzled.buffer, sw.buffer_offset + swizzled.offset,
+                                            image.guest_size_bytes - sw.buffer_offset);
+    compute_pass_descriptor_queue.AddBuffer(output.buffer, output.offset, output_size);
+
+    const void* descriptor_data = compute_pass_descriptor_queue.UpdateData();
+    const VkDescriptorSet set = descriptor_allocator.Commit();
+
+    const u32 gx = Common::DivCeil(blocks_x, 8u);
+    const u32 gy = Common::DivCeil(blocks_y, 8u);
+    const u32 gz = Common::DivCeil(blocks_z, 4u);
+    const bool is_initialized = image.ExchangeInitialization();
+
+    const VkBuffer out_buffer = output.buffer;
+    const VkDeviceSize out_offset = output.offset;
+    const VkImage dst_image = image.Handle();
+    const VkImageAspectFlags aspect = image.AspectMask();
+    const VkExtent3D extent{
+        .width = image.info.size.width,
+        .height = image.info.size.height,
+        .depth = image.info.size.depth,
+    };
+
+    scheduler.Record([this, set, descriptor_data, pc, gx, gy, gz, output_size, out_buffer,
+                      out_offset, dst_image, aspect, extent,
+                      is_initialized](vk::CommandBuffer cmdbuf) {
+        if (dst_image == VK_NULL_HANDLE || out_buffer == VK_NULL_HANDLE) {
+            return;
+        }
+        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, {});
+        cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(pc), &pc);
+        cmdbuf.Dispatch(gx, gy, gz);
+
+        const VkBufferMemoryBarrier buffer_barrier{
+            .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .buffer = out_buffer,
+            .offset = out_offset,
+            .size = output_size,
+        };
+        const VkImageMemoryBarrier pre_copy{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = static_cast<VkAccessFlags>(is_initialized ? VK_ACCESS_SHADER_READ_BIT
+                                                                      : VK_ACCESS_NONE),
+            .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
+            .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = dst_image,
+            .subresourceRange{
+                .aspectMask = aspect,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
+                                   (is_initialized ? vk::PIPELINE_STAGE_GRAPHICS_COMPUTE
+                                                   : VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT),
+                               VK_PIPELINE_STAGE_TRANSFER_BIT, 0, {}, buffer_barrier, pre_copy);
+
+        const VkBufferImageCopy copy{
+            .bufferOffset = out_offset,
+            .bufferRowLength = 0,
+            .bufferImageHeight = 0,
+            .imageSubresource{
+                .aspectMask = aspect,
+                .mipLevel = 0,
+                .baseArrayLayer = 0,
+                .layerCount = 1,
+            },
+            .imageOffset = {0, 0, 0},
+            .imageExtent = extent,
+        };
+        cmdbuf.CopyBufferToImage(out_buffer, dst_image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copy);
+
+        const VkImageMemoryBarrier post_copy{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = dst_image,
+            .subresourceRange{
+                .aspectMask = aspect,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE,
+                               0, {}, {}, post_copy);
+    });
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_compute_pass.h

@@ -164,4 +164,42 @@ private:
     ComputePassDescriptorQueue& compute_pass_descriptor_queue;
 };
 
+class BlockLinearUnswizzle2DPass final : public ComputePass {
+public:
+    explicit BlockLinearUnswizzle2DPass(const Device& device_, Scheduler& scheduler_,
+                                        DescriptorPool& descriptor_pool_,
+                                        StagingBufferPool& staging_buffer_pool_,
+                                        ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
+    ~BlockLinearUnswizzle2DPass();
+
+    [[nodiscard]] static bool IsSupported(const VideoCommon::ImageInfo& info);
+
+    void Unswizzle(Image& image, const StagingBufferRef& swizzled,
+                   std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+private:
+    Scheduler& scheduler;
+    StagingBufferPool& staging_buffer_pool;
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
+};
+
+class BlockLinearUnswizzle3DBufferPass final : public ComputePass {
+public:
+    explicit BlockLinearUnswizzle3DBufferPass(
+        const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+        StagingBufferPool& staging_buffer_pool_,
+        ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
+    ~BlockLinearUnswizzle3DBufferPass();
+
+    [[nodiscard]] static bool IsSupported(const Device& device, const VideoCommon::ImageInfo& info);
+
+    void Unswizzle(Image& image, const StagingBufferRef& swizzled,
+                   std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+private:
+    Scheduler& scheduler;
+    StagingBufferPool& staging_buffer_pool;
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
+};
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -63,6 +63,8 @@ using VideoCommon::GenericEnvironment;
 using VideoCommon::GraphicsEnvironment;
 
 constexpr u32 CACHE_VERSION = 18;
+constexpr size_t VULKAN_CACHE_FLUSH_PIPELINES = 128;
+constexpr size_t VULKAN_CACHE_FLUSH_MIN_SECONDS = 30;
 constexpr std::array<char, 8> VULKAN_CACHE_MAGIC_NUMBER{'y', 'u', 'z', 'u', 'v', 'k', 'c', 'h'};
 
 template <typename Container>
@@ -699,6 +701,10 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
     if (use_vulkan_pipeline_cache) {
         SerializeVulkanPipelineCache(vulkan_pipeline_cache_filename, vulkan_pipeline_cache,
                                      CACHE_VERSION);
+        size_t size = 0;
+        vulkan_pipeline_cache.Read(&size, nullptr);
+        last_cache_size.store(size, std::memory_order_relaxed);
+        last_flush = std::chrono::steady_clock::now();
     }
 
     if (state.statistics) {
@@ -706,6 +712,35 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
     }
 }
 
+void PipelineCache::QueueVulkanPipelineCacheFlush() {
+    if (!use_vulkan_pipeline_cache || vulkan_pipeline_cache_filename.empty()) {
+        return;
+    }
+    if (++pipelines_since_flush < VULKAN_CACHE_FLUSH_PIPELINES) {
+        return;
+    }
+    const auto now = std::chrono::steady_clock::now();
+    const auto megabytes = last_cache_size.load(std::memory_order_relaxed) / (1024 * 1024);
+    const std::chrono::seconds interval{
+        std::max<size_t>(VULKAN_CACHE_FLUSH_MIN_SECONDS, megabytes)};
+    if (last_flush.time_since_epoch().count() != 0 && now - last_flush < interval) {
+        return;
+    }
+    if (flush_in_flight.exchange(true, std::memory_order_acq_rel)) {
+        return;
+    }
+    pipelines_since_flush = 0;
+    last_flush = now;
+    serialization_thread.QueueWork([this] {
+        SerializeVulkanPipelineCache(vulkan_pipeline_cache_filename, vulkan_pipeline_cache,
+                                     CACHE_VERSION);
+        size_t size = 0;
+        vulkan_pipeline_cache.Read(&size, nullptr);
+        last_cache_size.store(size, std::memory_order_relaxed);
+        flush_in_flight.store(false, std::memory_order_release);
+    });
+}
+
 GraphicsPipeline* PipelineCache::CurrentGraphicsPipelineSlowPath() {
     const auto [pair, is_new]{graphics_cache.try_emplace(graphics_key)};
     auto& pipeline{pair->second};
@@ -744,7 +779,7 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
     std::span<Shader::Environment* const> envs, PipelineStatistics* statistics,
     bool build_in_parallel) try {
     auto hash = key.Hash();
-    LOG_INFO(Render_Vulkan, "{:#016x}", hash);
+    LOG_DEBUG(Render_Vulkan, "{:#016x}", hash);
     size_t env_index{0};
     std::array<Shader::IR::Program, Maxwell::MaxShaderProgram> programs;
     const bool uses_vertex_a{key.unique_hashes[0] != 0};
@@ -880,6 +915,7 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline() {
         }
         SerializePipeline(key, env_ptrs, pipeline_cache_filename, CACHE_VERSION);
     });
+    QueueVulkanPipelineCacheFlush();
     return pipeline;
 }
 
@@ -899,6 +935,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
         SerializePipeline(key, std::array<const GenericEnvironment*, 1>{&env_},
                           pipeline_cache_filename, CACHE_VERSION);
     });
+    QueueVulkanPipelineCacheFlush();
     return pipeline;
 }
 
@@ -911,7 +948,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
         return nullptr;
     }
 
-    LOG_INFO(Render_Vulkan, "{:#016x}", hash);
+    LOG_DEBUG(Render_Vulkan, "{:#016x}", hash);
 
     Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()};
 

src/video_core/renderer_vulkan/vk_pipeline_cache.h

@@ -7,6 +7,8 @@
 #pragma once
 
 #include <array>
+#include <atomic>
+#include <chrono>
 #include <cstddef>
 #include <filesystem>
 #include <memory>
@@ -144,6 +146,8 @@ private:
     vk::PipelineCache LoadVulkanPipelineCache(const std::filesystem::path& filename,
                                               u32 expected_cache_version);
 
+    void QueueVulkanPipelineCacheFlush();
+
     const Device& device;
     Scheduler& scheduler;
     DescriptorPool& descriptor_pool;
@@ -171,6 +175,10 @@ private:
 
     std::filesystem::path vulkan_pipeline_cache_filename;
     vk::PipelineCache vulkan_pipeline_cache;
+    size_t pipelines_since_flush{};
+    std::chrono::steady_clock::time_point last_flush{};
+    std::atomic<size_t> last_cache_size{};
+    std::atomic_bool flush_in_flight{};
 
     Common::ThreadWorker workers;
     Common::ThreadWorker serialization_thread;

src/video_core/renderer_vulkan/vk_render_pass_cache.cpp

@@ -145,18 +145,6 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) {
         .preserveAttachmentCount = 0,
         .pPreserveAttachments = nullptr,
     };
-    const VkSubpassDependency dependency{
-            .srcSubpass = 0,  // Current subpass
-            .dstSubpass = 0,  // Same subpass (self-dependency)
-            .srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                            VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-                            VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
-            .dstStageMask = VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
-            .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
-                             VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
-            .dependencyFlags = VK_DEPENDENCY_BY_REGION_BIT
-    };
     pair->second = device->GetLogical().CreateRenderPass({
         .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
         .pNext = nullptr,
@@ -165,8 +153,8 @@ VkRenderPass RenderPassCache::Get(const RenderPassKey& key) {
         .pAttachments = descriptions.empty() ? nullptr : descriptions.data(),
         .subpassCount = 1,
         .pSubpasses = &subpass,
-        .dependencyCount = 1,
+        .dependencyCount = 0,
-        .pDependencies = &dependency,
+        .pDependencies = nullptr,
     });
     return *pair->second;
 }

src/video_core/renderer_vulkan/vk_render_pass_cache.h

@@ -9,6 +9,7 @@
 #include <mutex>
 #include <ankerl/unordered_dense.h>
 
+#include "common/container_hash.h"
 #include "video_core/surface.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
@@ -31,17 +32,26 @@ struct RenderPassKey {
 namespace std {
 template <>
 struct hash<Vulkan::RenderPassKey> {
+    static_assert(std::tuple_size_v<decltype(Vulkan::RenderPassKey::color_formats)> <= 8);
+    static_assert(static_cast<u32>(VideoCore::Surface::PixelFormat::Invalid) <= 0xFF);
+    static_assert(static_cast<u32>(VideoCore::Surface::PixelFormat::Max) <= 0xFF);
+    static_assert(VK_SAMPLE_COUNT_64_BIT <= 0xFF);
+
     [[nodiscard]] size_t operator()(const Vulkan::RenderPassKey& key) const noexcept {
-        size_t value = static_cast<size_t>(key.depth_format) << 48;
+        u64 formats = 0;
-        value ^= static_cast<size_t>(key.samples) << 52;
+        for (size_t index = 0; index < key.color_formats.size(); ++index) {
-        value ^= static_cast<size_t>(key.resolve_color) << 63;
+            formats |= static_cast<u64>(key.color_formats[index]) << (index * 8);
-        value ^= static_cast<size_t>(key.color_clear_mask) << 54;
-        value ^= static_cast<size_t>(key.depth_stencil_clear) << 62;
-        value ^= static_cast<size_t>(key.color_discard_mask) << 24;
-        for (size_t i = 0; i < key.color_formats.size(); ++i) {
-            value ^= static_cast<size_t>(key.color_formats[i]) << (i * 6);
         }
-        return value;
+        const u64 state = static_cast<u64>(key.depth_format) |
+                          (static_cast<u64>(key.samples) << 8) |
+                          (static_cast<u64>(key.color_clear_mask) << 16) |
+                          (static_cast<u64>(key.color_discard_mask) << 24) |
+                          (static_cast<u64>(key.resolve_color) << 32) |
+                          (static_cast<u64>(key.depth_stencil_clear) << 33);
+        size_t seed = 0;
+        Common::HashCombine(seed, formats);
+        Common::HashCombine(seed, state);
+        return seed;
     }
 };
 } // namespace std

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -628,18 +628,12 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im
             .subresourceRange = subresource_range,
     };
 
-    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
+    cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
-                           VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                           VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
                            read_barrier);
     cmdbuf.CopyBufferToImage(src_buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copies);
     // TODO: Move this to another API
-    cmdbuf.PipelineBarrier(
+    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, 0,
-            VK_PIPELINE_STAGE_TRANSFER_BIT,
+                           nullptr, nullptr, write_barrier);
-            VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
-            0, nullptr, nullptr, write_barrier);
 }
 
 [[nodiscard]] VkImageBlit MakeImageBlit(const Region2D& dst_region, const Region2D& src_region,
@@ -933,6 +927,10 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched
         bl3d_unswizzle_pass.emplace(device, scheduler, descriptor_pool,
                                    staging_buffer_pool, compute_pass_descriptor_queue);
     }
+    bl2d_unswizzle_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool,
+                                compute_pass_descriptor_queue);
+    bl3db_unswizzle_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool,
+                                 compute_pass_descriptor_queue);
 }
 
 void TextureCacheRuntime::Finish() {
@@ -1796,6 +1794,13 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
         }
         flags |= VideoCommon::ImageFlagBits::Converted;
         flags |= VideoCommon::ImageFlagBits::CostlyLoad;
+    } else if (runtime->bl2d_unswizzle_pass && BlockLinearUnswizzle2DPass::IsSupported(info)) {
+        flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
+        flags |= VideoCommon::ImageFlagBits::CostlyLoad;
+    } else if (runtime->bl3db_unswizzle_pass &&
+               BlockLinearUnswizzle3DBufferPass::IsSupported(runtime->device, info)) {
+        flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
+        flags |= VideoCommon::ImageFlagBits::CostlyLoad;
     }
     if (IsPixelFormatBCn(info.format) && !runtime->device.IsOptimalBcnSupported()) {
         flags |= VideoCommon::ImageFlagBits::Converted;
@@ -2827,11 +2832,25 @@ VkRenderPass Framebuffer::RenderPassVariant(u32 color_clear_mask, bool depth_ste
     if (color_clear_mask == 0 && !depth_stencil_clear && color_discard_mask == 0) {
         return renderpass;
     }
+    static_assert(NUM_RT <= 8);
+    const u32 variant_key = color_clear_mask | (color_discard_mask << 8) |
+                            (static_cast<u32>(depth_stencil_clear) << 16);
+    for (u32 index = 0; index < num_memoized_variants; ++index) {
+        if (variant_keys[index] == variant_key) {
+            return variant_render_passes[index];
+        }
+    }
     RenderPassKey key = render_pass_key;
     key.color_clear_mask = color_clear_mask;
     key.depth_stencil_clear = depth_stencil_clear;
     key.color_discard_mask = color_discard_mask;
-    return render_pass_cache->Get(key);
+    const VkRenderPass variant = render_pass_cache->Get(key);
+    if (num_memoized_variants < variant_keys.size()) {
+        variant_keys[num_memoized_variants] = variant_key;
+        variant_render_passes[num_memoized_variants] = variant;
+        ++num_memoized_variants;
+    }
+    return variant;
 }
 
 void TextureCacheRuntime::AccelerateImageUpload(
@@ -2843,6 +2862,15 @@ void TextureCacheRuntime::AccelerateImageUpload(
         return astc_decoder_pass->Assemble(image, map, swizzles);
     }
 
+    if (bl2d_unswizzle_pass && BlockLinearUnswizzle2DPass::IsSupported(image.info)) {
+        return bl2d_unswizzle_pass->Unswizzle(image, map, swizzles);
+    }
+
+    if (bl3db_unswizzle_pass &&
+        BlockLinearUnswizzle3DBufferPass::IsSupported(device, image.info)) {
+        return bl3db_unswizzle_pass->Unswizzle(image, map, swizzles);
+    }
+
     if (!Settings::values.gpu_unswizzle_enabled.GetValue() || !bl3d_unswizzle_pass) {
         if (IsPixelFormatBCn(image.info.format) && image.info.type == ImageType::e3D) {
             ASSERT(false && "GPU unswizzle is disabled for BCn 3D texture");

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -149,6 +149,8 @@ public:
     std::optional<ASTCDecoderPass> astc_decoder_pass;
 
     std::optional<BlockLinearUnswizzle3DPass> bl3d_unswizzle_pass;
+    std::optional<BlockLinearUnswizzle2DPass> bl2d_unswizzle_pass;
+    std::optional<BlockLinearUnswizzle3DBufferPass> bl3db_unswizzle_pass;
     const Settings::ResolutionScalingInfo& resolution;
     std::array<std::vector<VkFormat>, VideoCore::Surface::MaxPixelFormat> view_formats;
 
@@ -246,6 +248,8 @@ public:
     }
 
 private:
+    static constexpr size_t NUM_MEMOIZED_RENDER_PASS_VARIANTS = 8;
+
     vk::Framebuffer framebuffer;
     VkRenderPass renderpass{};
     VkExtent2D render_area{};
@@ -263,6 +267,9 @@ private:
     RenderPassKey render_pass_key{};
     RenderPassCache* render_pass_cache{nullptr};
     bool discard_msaa_color{};
+    mutable std::array<u32, NUM_MEMOIZED_RENDER_PASS_VARIANTS> variant_keys{};
+    mutable std::array<VkRenderPass, NUM_MEMOIZED_RENDER_PASS_VARIANTS> variant_render_passes{};
+    mutable u32 num_memoized_variants{};
 };
 
 class Image : public VideoCommon::ImageBase {

src/video_core/textures/decoders.cpp

@@ -1,9 +1,10 @@
-// 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 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <algorithm>
 #include <array>
 #include <cmath>
 #include <cstring>
@@ -30,12 +31,62 @@ constexpr u32 pdep(u32 value) {
     return result;
 }
 
+constexpr u32 SWIZZLE_RUN_BYTES = 16;
+constexpr u32 SWIZZLE_RUN_SHIFT = 4;
+constexpr u32 SWIZZLE_RUN_MASK = SWIZZLE_RUN_BYTES - 1;
+constexpr u32 SWIZZLE_RUN_INDEX_MASK = GOB_SIZE_X / SWIZZLE_RUN_BYTES - 1;
+
+static_assert((SWIZZLE_X_BITS & SWIZZLE_RUN_MASK) == SWIZZLE_RUN_MASK);
+
+constexpr std::array<u32, GOB_SIZE_X / SWIZZLE_RUN_BYTES> SWIZZLE_X_RUN_TABLE = [] {
+    std::array<u32, GOB_SIZE_X / SWIZZLE_RUN_BYTES> table{};
+    for (u32 index = 0; index < static_cast<u32>(table.size()); ++index) {
+        table[index] = pdep<SWIZZLE_X_BITS>(index << SWIZZLE_RUN_SHIFT);
+    }
+    return table;
+}();
+
 template <u32 mask, u32 incr_amount>
 void incrpdep(u32& value) {
     static constexpr u32 swizzled_incr = pdep<mask>(incr_amount);
     value = ((value | ~mask) + swizzled_incr) & mask;
 }
 
+template <bool TO_LINEAR>
+void SwizzleRow(std::span<u8> output, std::span<const u8> input, u32 offset_zy, u32 swizzled_y,
+                u32 x_shift, u32 x, u32 num_bytes, u32 linear) {
+    const auto copy = [&](u32 swizzled_x, u32 count) {
+        const u32 swizzled =
+            offset_zy + ((x >> GOB_SIZE_X_SHIFT) << x_shift) + (swizzled_x | swizzled_y);
+        u8* const dst = &output[TO_LINEAR ? swizzled : linear];
+        const u8* const src = &input[TO_LINEAR ? linear : swizzled];
+        std::memcpy(dst, src, count);
+        x += count;
+        linear += count;
+    };
+
+    u32 swizzled_run = SWIZZLE_X_RUN_TABLE[(x >> SWIZZLE_RUN_SHIFT) & SWIZZLE_RUN_INDEX_MASK];
+    u32 remaining = num_bytes;
+
+    const u32 head =
+        (std::min)(SWIZZLE_RUN_BYTES - (x & SWIZZLE_RUN_MASK), remaining) & SWIZZLE_RUN_MASK;
+    if (head != 0) {
+        copy(swizzled_run | (x & SWIZZLE_RUN_MASK), head);
+        remaining -= head;
+        incrpdep<SWIZZLE_X_BITS, SWIZZLE_RUN_BYTES>(swizzled_run);
+    }
+
+    while (remaining >= SWIZZLE_RUN_BYTES) {
+        copy(swizzled_run, SWIZZLE_RUN_BYTES);
+        remaining -= SWIZZLE_RUN_BYTES;
+        incrpdep<SWIZZLE_X_BITS, SWIZZLE_RUN_BYTES>(swizzled_run);
+    }
+
+    if (remaining != 0) {
+        copy(swizzled_run, remaining);
+    }
+}
+
 template <bool TO_LINEAR, u32 BYTES_PER_PIXEL>
 void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 height, u32 depth,
                  u32 block_height, u32 block_depth, u32 stride) {
@@ -70,23 +121,9 @@ void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32
             const u32 offset_y = (block_y >> block_height) * block_size +
                                  ((block_y & block_height_mask) << GOB_SIZE_SHIFT);
 
-            u32 swizzled_x = pdep<SWIZZLE_X_BITS>(origin_x * BYTES_PER_PIXEL);
+            SwizzleRow<TO_LINEAR>(output, input, offset_z + offset_y, swizzled_y, x_shift,
-            for (u32 column = 0; column < width;
+                                  origin_x * BYTES_PER_PIXEL, width * BYTES_PER_PIXEL,
-                 ++column, incrpdep<SWIZZLE_X_BITS, BYTES_PER_PIXEL>(swizzled_x)) {
+                                  slice * pitch * height + line * pitch);
-                const u32 x = (column + origin_x) * BYTES_PER_PIXEL;
-                const u32 offset_x = (x >> GOB_SIZE_X_SHIFT) << x_shift;
-
-                const u32 base_swizzled_offset = offset_z + offset_y + offset_x;
-                const u32 swizzled_offset = base_swizzled_offset + (swizzled_x | swizzled_y);
-
-                const u32 unswizzled_offset =
-                    slice * pitch * height + line * pitch + column * BYTES_PER_PIXEL;
-
-                u8* const dst = &output[TO_LINEAR ? swizzled_offset : unswizzled_offset];
-                const u8* const src = &input[TO_LINEAR ? unswizzled_offset : swizzled_offset];
-
-                std::memcpy(dst, src, BYTES_PER_PIXEL);
-            }
         }
     }
 }
@@ -129,23 +166,9 @@ void SwizzleSubrectImpl(std::span<u8> output, std::span<const u8> input, u32 wid
             const u32 offset_y = (block_y >> block_height) * block_size +
                                  ((block_y & block_height_mask) << GOB_SIZE_SHIFT);
 
-            u32 swizzled_x = pdep<SWIZZLE_X_BITS>(origin_x * BYTES_PER_PIXEL);
+            SwizzleRow<TO_LINEAR>(output, input, offset_z + offset_y, swizzled_y, x_shift,
-            for (u32 column = 0; column < extent_x;
+                                  origin_x * BYTES_PER_PIXEL, extent_x * BYTES_PER_PIXEL,
-                 ++column, incrpdep<SWIZZLE_X_BITS, BYTES_PER_PIXEL>(swizzled_x)) {
+                                  slice * pitch * height + line * pitch);
-                const u32 x = (column + origin_x) * BYTES_PER_PIXEL;
-                const u32 offset_x = (x >> GOB_SIZE_X_SHIFT) << x_shift;
-
-                const u32 base_swizzled_offset = offset_z + offset_y + offset_x;
-                const u32 swizzled_offset = base_swizzled_offset + (swizzled_x | swizzled_y);
-
-                const u32 unswizzled_offset =
-                    slice * pitch * height + line * pitch + column * BYTES_PER_PIXEL;
-
-                u8* const dst = &output[TO_LINEAR ? swizzled_offset : unswizzled_offset];
-                const u8* const src = &input[TO_LINEAR ? unswizzled_offset : swizzled_offset];
-
-                std::memcpy(dst, src, BYTES_PER_PIXEL);
-            }
         }
         unprocessed_lines -= lines_in_y;
         if (unprocessed_lines == 0) {

src/video_core/vulkan_common/vulkan_device.cpp

@@ -7,6 +7,8 @@
 #include <algorithm>
 #include <bitset>
 #include <chrono>
+#include <filesystem>
+#include <fstream>
 #include <optional>
 #include <thread>
 #include <ankerl/unordered_dense.h>
@@ -16,6 +18,8 @@
 #include <fmt/format.h>
 
 #include "common/assert.h"
+#include "common/fs/fs.h"
+#include "common/fs/path_util.h"
 #include "common/literals.h"
 #include <ranges>
 #include "common/settings.h"
@@ -393,6 +397,17 @@ std::vector<const char*> ExtensionListForVulkan(
     return output;
 }
 
+constexpr std::array<char, 8> STATIC_CACHE_MAGIC_NUMBER{'e', 'd', 'e', 'n', 's', 't', 'p', 'c'};
+constexpr u32 STATIC_CACHE_VERSION = 1;
+
+std::filesystem::path StaticPipelineCacheFilename() {
+    const auto shader_dir = Common::FS::GetEdenPath(Common::FS::EdenPath::ShaderDir);
+    if (!Common::FS::CreateDir(shader_dir)) {
+        return {};
+    }
+    return shader_dir / "vulkan_static_pipelines.bin";
+}
+
 } // Anonymous namespace
 
 void Device::RemoveExtension(bool& extension, const std::string& extension_name) {
@@ -780,15 +795,100 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
 
     vk::Check(vmaCreateAllocator(&allocator_info, &allocator));
 
+    owns_static_pipeline_cache = surface != VkSurfaceKHR{};
+    LoadStaticPipelineCache();
+
     // Initialize GPU logging if enabled
     InitializeGPULogging();
 }
 
 Device::~Device() {
+    SaveStaticPipelineCache();
     ShutdownGPULogging();
     vmaDestroyAllocator(allocator);
 }
 
+void Device::LoadStaticPipelineCache() {
+    const auto create = [this](size_t size, const void* data) {
+        static_pipeline_cache = logical.CreatePipelineCache({
+            .sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO,
+            .pNext = nullptr,
+            .flags = 0,
+            .initialDataSize = size,
+            .pInitialData = data,
+        });
+    };
+    if (!owns_static_pipeline_cache) {
+        create(0, nullptr);
+        return;
+    }
+    const auto filename = StaticPipelineCacheFilename();
+    if (filename.empty()) {
+        create(0, nullptr);
+        return;
+    }
+    std::vector<char> data;
+    try {
+        std::ifstream file(filename, std::ios::binary | std::ios::ate);
+        if (!file.is_open()) {
+            create(0, nullptr);
+            return;
+        }
+        file.exceptions(std::ifstream::failbit | std::ifstream::badbit);
+        const size_t total = static_cast<size_t>(file.tellg());
+        file.seekg(0, std::ios::beg);
+        std::array<char, 8> magic{};
+        u32 version{};
+        if (total < magic.size() + sizeof(version)) {
+            create(0, nullptr);
+            return;
+        }
+        file.read(magic.data(), magic.size())
+            .read(reinterpret_cast<char*>(&version), sizeof(version));
+        if (magic != STATIC_CACHE_MAGIC_NUMBER || version != STATIC_CACHE_VERSION) {
+            create(0, nullptr);
+            return;
+        }
+        data.resize(total - magic.size() - sizeof(version));
+        file.read(data.data(), static_cast<std::streamsize>(data.size()));
+    } catch (const std::ios_base::failure& e) {
+        create(0, nullptr);
+        return;
+    }
+    create(data.size(), data.empty() ? nullptr : data.data());
+}
+
+void Device::SaveStaticPipelineCache() const {
+    if (!owns_static_pipeline_cache || !static_pipeline_cache) {
+        return;
+    }
+    const auto filename = StaticPipelineCacheFilename();
+    if (filename.empty()) {
+        return;
+    }
+    size_t size = 0;
+    std::vector<char> data;
+    static_pipeline_cache.Read(&size, nullptr);
+    if (size == 0) {
+        return;
+    }
+    data.resize(size);
+    static_pipeline_cache.Read(&size, data.data());
+    try {
+        std::ofstream file(filename, std::ios::binary | std::ios::trunc);
+        file.exceptions(std::ofstream::failbit);
+        if (!file.is_open()) {
+            return;
+        }
+        file.write(STATIC_CACHE_MAGIC_NUMBER.data(), STATIC_CACHE_MAGIC_NUMBER.size())
+            .write(reinterpret_cast<const char*>(&STATIC_CACHE_VERSION),
+                   sizeof(STATIC_CACHE_VERSION))
+            .write(data.data(), static_cast<std::streamsize>(size));
+    } catch (const std::ios_base::failure& e) {
+        Common::FS::RemoveFile(filename);
+    }
+}
+
 VkFormat Device::GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage,
                                     FormatType format_type) const {
     if (IsFormatSupported(wanted_format, wanted_usage, format_type)) {

src/video_core/vulkan_common/vulkan_device.h

@@ -268,6 +268,10 @@ public:
         return physical;
     }
 
+    VkPipelineCache StaticPipelineCache() const noexcept {
+        return *static_pipeline_cache;
+    }
+
     /// Returns the main graphics queue.
     vk::Queue GetGraphicsQueue() const {
         return graphics_queue;
@@ -1088,6 +1092,9 @@ private:
     /// Returns true if the device natively supports blitting depth stencil images.
     bool TestDepthStencilBlits(VkFormat format) const;
 
+    void LoadStaticPipelineCache();
+    void SaveStaticPipelineCache() const;
+
 private:
     VkInstance instance;         ///< Vulkan instance.
     VmaAllocator allocator;      ///< VMA allocator.
@@ -1096,6 +1103,8 @@ private:
     vk::Device logical;          ///< Logical device.
     vk::Queue graphics_queue;    ///< Main graphics queue.
     vk::Queue present_queue;     ///< Main present queue.
+    vk::PipelineCache static_pipeline_cache;
+    bool owns_static_pipeline_cache{};
     u32 instance_version{};      ///< Vulkan instance version.
     u32 graphics_family{};       ///< Main graphics queue family index.
     u32 present_family{};        ///< Main present queue family index.