Merge pull request #4967 from ReinUsesLisp/new-texcache

video_core/texture_cache: Rewrite the texture cache

src/video_core/CMakeLists.txt

@@ -25,6 +25,7 @@ add_library(video_core STATIC
     command_classes/vic.h
     compatible_formats.cpp
     compatible_formats.h
+    delayed_destruction_ring.h
     dirty_flags.cpp
     dirty_flags.h
     dma_pusher.cpp
@@ -84,14 +85,10 @@ add_library(video_core STATIC
     renderer_opengl/gl_device.h
     renderer_opengl/gl_fence_manager.cpp
     renderer_opengl/gl_fence_manager.h
-    renderer_opengl/gl_framebuffer_cache.cpp
-    renderer_opengl/gl_framebuffer_cache.h
     renderer_opengl/gl_rasterizer.cpp
     renderer_opengl/gl_rasterizer.h
     renderer_opengl/gl_resource_manager.cpp
     renderer_opengl/gl_resource_manager.h
-    renderer_opengl/gl_sampler_cache.cpp
-    renderer_opengl/gl_sampler_cache.h
     renderer_opengl/gl_shader_cache.cpp
     renderer_opengl/gl_shader_cache.h
     renderer_opengl/gl_shader_decompiler.cpp
@@ -113,8 +110,10 @@ add_library(video_core STATIC
     renderer_opengl/maxwell_to_gl.h
     renderer_opengl/renderer_opengl.cpp
     renderer_opengl/renderer_opengl.h
-    renderer_opengl/utils.cpp
+    renderer_opengl/util_shaders.cpp
-    renderer_opengl/utils.h
+    renderer_opengl/util_shaders.h
+    renderer_vulkan/blit_image.cpp
+    renderer_vulkan/blit_image.h
     renderer_vulkan/fixed_pipeline_state.cpp
     renderer_vulkan/fixed_pipeline_state.h
     renderer_vulkan/maxwell_to_vk.cpp
@@ -141,8 +140,6 @@ add_library(video_core STATIC
     renderer_vulkan/vk_fence_manager.h
     renderer_vulkan/vk_graphics_pipeline.cpp
     renderer_vulkan/vk_graphics_pipeline.h
-    renderer_vulkan/vk_image.cpp
-    renderer_vulkan/vk_image.h
     renderer_vulkan/vk_master_semaphore.cpp
     renderer_vulkan/vk_master_semaphore.h
     renderer_vulkan/vk_memory_manager.cpp
@@ -153,12 +150,8 @@ add_library(video_core STATIC
     renderer_vulkan/vk_query_cache.h
     renderer_vulkan/vk_rasterizer.cpp
     renderer_vulkan/vk_rasterizer.h
-    renderer_vulkan/vk_renderpass_cache.cpp
-    renderer_vulkan/vk_renderpass_cache.h
     renderer_vulkan/vk_resource_pool.cpp
     renderer_vulkan/vk_resource_pool.h
-    renderer_vulkan/vk_sampler_cache.cpp
-    renderer_vulkan/vk_sampler_cache.h
     renderer_vulkan/vk_scheduler.cpp
     renderer_vulkan/vk_scheduler.h
     renderer_vulkan/vk_shader_decompiler.cpp
@@ -179,8 +172,6 @@ add_library(video_core STATIC
     renderer_vulkan/vk_update_descriptor.h
     renderer_vulkan/wrapper.cpp
     renderer_vulkan/wrapper.h
-    sampler_cache.cpp
-    sampler_cache.h
     shader_cache.h
     shader_notify.cpp
     shader_notify.h
@@ -237,19 +228,32 @@ add_library(video_core STATIC
     shader/transform_feedback.h
     surface.cpp
     surface.h
+    texture_cache/accelerated_swizzle.cpp
+    texture_cache/accelerated_swizzle.h
+    texture_cache/decode_bc4.cpp
+    texture_cache/decode_bc4.h
+    texture_cache/descriptor_table.h
+    texture_cache/formatter.cpp
+    texture_cache/formatter.h
     texture_cache/format_lookup_table.cpp
     texture_cache/format_lookup_table.h
-    texture_cache/surface_base.cpp
+    texture_cache/image_base.cpp
-    texture_cache/surface_base.h
+    texture_cache/image_base.h
-    texture_cache/surface_params.cpp
+    texture_cache/image_info.cpp
-    texture_cache/surface_params.h
+    texture_cache/image_info.h
-    texture_cache/surface_view.cpp
+    texture_cache/image_view_base.cpp
-    texture_cache/surface_view.h
+    texture_cache/image_view_base.h
+    texture_cache/image_view_info.cpp
+    texture_cache/image_view_info.h
+    texture_cache/render_targets.h
+    texture_cache/samples_helper.h
+    texture_cache/slot_vector.h
     texture_cache/texture_cache.h
+    texture_cache/types.h
+    texture_cache/util.cpp
+    texture_cache/util.h
     textures/astc.cpp
     textures/astc.h
-    textures/convert.cpp
-    textures/convert.h
     textures/decoders.cpp
     textures/decoders.h
     textures/texture.cpp

src/video_core/buffer_cache/buffer_cache.h

@@ -118,20 +118,17 @@ public:
     /// Prepares the buffer cache for data uploading
     /// @param max_size Maximum number of bytes that will be uploaded
     /// @return True when a stream buffer invalidation was required, false otherwise
-    bool Map(std::size_t max_size) {
+    void Map(std::size_t max_size) {
         std::lock_guard lock{mutex};
 
-        bool invalidated;
+        std::tie(buffer_ptr, buffer_offset_base) = stream_buffer.Map(max_size, 4);
-        std::tie(buffer_ptr, buffer_offset_base, invalidated) = stream_buffer->Map(max_size, 4);
         buffer_offset = buffer_offset_base;
-
-        return invalidated;
     }
 
     /// Finishes the upload stream
     void Unmap() {
         std::lock_guard lock{mutex};
-        stream_buffer->Unmap(buffer_offset - buffer_offset_base);
+        stream_buffer.Unmap(buffer_offset - buffer_offset_base);
     }
 
     /// Function called at the end of each frame, inteded for deferred operations
@@ -261,9 +258,9 @@ public:
 protected:
     explicit BufferCache(VideoCore::RasterizerInterface& rasterizer_,
                          Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_,
-                         std::unique_ptr<StreamBuffer> stream_buffer_)
+                         StreamBuffer& stream_buffer_)
         : rasterizer{rasterizer_}, gpu_memory{gpu_memory_}, cpu_memory{cpu_memory_},
-          stream_buffer{std::move(stream_buffer_)}, stream_buffer_handle{stream_buffer->Handle()} {}
+          stream_buffer{stream_buffer_} {}
 
     ~BufferCache() = default;
 
@@ -441,7 +438,7 @@ private:
 
         buffer_ptr += size;
         buffer_offset += size;
-        return BufferInfo{stream_buffer->Handle(), uploaded_offset, stream_buffer->Address()};
+        return BufferInfo{stream_buffer.Handle(), uploaded_offset, stream_buffer.Address()};
     }
 
     void AlignBuffer(std::size_t alignment) {
@@ -567,9 +564,7 @@ private:
     VideoCore::RasterizerInterface& rasterizer;
     Tegra::MemoryManager& gpu_memory;
     Core::Memory::Memory& cpu_memory;
-
+    StreamBuffer& stream_buffer;
-    std::unique_ptr<StreamBuffer> stream_buffer;
-    BufferType stream_buffer_handle;
 
     u8* buffer_ptr = nullptr;
     u64 buffer_offset = 0;

src/video_core/command_classes/vic.cpp

@@ -9,7 +9,7 @@
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
-#include "video_core/texture_cache/surface_params.h"
+#include "video_core/textures/decoders.h"
 
 extern "C" {
 #include <libswscale/swscale.h>
@@ -105,9 +105,9 @@ void Vic::Execute() {
             const auto size = Tegra::Texture::CalculateSize(true, 4, frame->width, frame->height, 1,
                                                             block_height, 0);
             std::vector<u8> swizzled_data(size);
-            Tegra::Texture::CopySwizzledData(frame->width, frame->height, 1, 4, 4,
+            Tegra::Texture::SwizzleSubrect(frame->width, frame->height, frame->width * 4,
-                                             swizzled_data.data(), converted_frame_buffer.get(),
+                                           frame->width, 4, swizzled_data.data(),
-                                             false, block_height, 0, 1);
+                                           converted_frame_buffer.get(), block_height, 0, 0);
 
             gpu.MemoryManager().WriteBlock(output_surface_luma_address, swizzled_data.data(), size);
             gpu.Maxwell3D().OnMemoryWrite();

src/video_core/compatible_formats.cpp

@@ -3,9 +3,9 @@
 // Refer to the license.txt file included.
 
 #include <array>
-#include <bitset>
 #include <cstddef>
 
+#include "common/common_types.h"
 #include "video_core/compatible_formats.h"
 #include "video_core/surface.h"
 
@@ -13,23 +13,25 @@ namespace VideoCore::Surface {
 
 namespace {
 
+using Table = std::array<std::array<u64, 2>, MaxPixelFormat>;
+
 // Compatibility table taken from Table 3.X.2 in:
 // https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_texture_view.txt
 
-constexpr std::array VIEW_CLASS_128_BITS = {
+constexpr std::array VIEW_CLASS_128_BITS{
     PixelFormat::R32G32B32A32_FLOAT,
     PixelFormat::R32G32B32A32_UINT,
     PixelFormat::R32G32B32A32_SINT,
 };
 
-constexpr std::array VIEW_CLASS_96_BITS = {
+constexpr std::array VIEW_CLASS_96_BITS{
     PixelFormat::R32G32B32_FLOAT,
 };
 // Missing formats:
 // PixelFormat::RGB32UI,
 // PixelFormat::RGB32I,
 
-constexpr std::array VIEW_CLASS_64_BITS = {
+constexpr std::array VIEW_CLASS_64_BITS{
     PixelFormat::R32G32_FLOAT,       PixelFormat::R32G32_UINT,
     PixelFormat::R32G32_SINT,        PixelFormat::R16G16B16A16_FLOAT,
     PixelFormat::R16G16B16A16_UNORM, PixelFormat::R16G16B16A16_SNORM,
@@ -38,7 +40,7 @@ constexpr std::array VIEW_CLASS_64_BITS = {
 
 // TODO: How should we handle 48 bits?
 
-constexpr std::array VIEW_CLASS_32_BITS = {
+constexpr std::array VIEW_CLASS_32_BITS{
     PixelFormat::R16G16_FLOAT,      PixelFormat::B10G11R11_FLOAT, PixelFormat::R32_FLOAT,
     PixelFormat::A2B10G10R10_UNORM, PixelFormat::R16G16_UINT,     PixelFormat::R32_UINT,
     PixelFormat::R16G16_SINT,       PixelFormat::R32_SINT,        PixelFormat::A8B8G8R8_UNORM,
@@ -50,43 +52,105 @@ constexpr std::array VIEW_CLASS_32_BITS = {
 
 // TODO: How should we handle 24 bits?
 
-constexpr std::array VIEW_CLASS_16_BITS = {
+constexpr std::array VIEW_CLASS_16_BITS{
     PixelFormat::R16_FLOAT,  PixelFormat::R8G8_UINT,  PixelFormat::R16_UINT,
     PixelFormat::R16_SINT,   PixelFormat::R8G8_UNORM, PixelFormat::R16_UNORM,
     PixelFormat::R8G8_SNORM, PixelFormat::R16_SNORM,  PixelFormat::R8G8_SINT,
 };
 
-constexpr std::array VIEW_CLASS_8_BITS = {
+constexpr std::array VIEW_CLASS_8_BITS{
     PixelFormat::R8_UINT,
     PixelFormat::R8_UNORM,
     PixelFormat::R8_SINT,
     PixelFormat::R8_SNORM,
 };
 
-constexpr std::array VIEW_CLASS_RGTC1_RED = {
+constexpr std::array VIEW_CLASS_RGTC1_RED{
     PixelFormat::BC4_UNORM,
     PixelFormat::BC4_SNORM,
 };
 
-constexpr std::array VIEW_CLASS_RGTC2_RG = {
+constexpr std::array VIEW_CLASS_RGTC2_RG{
     PixelFormat::BC5_UNORM,
     PixelFormat::BC5_SNORM,
 };
 
-constexpr std::array VIEW_CLASS_BPTC_UNORM = {
+constexpr std::array VIEW_CLASS_BPTC_UNORM{
     PixelFormat::BC7_UNORM,
     PixelFormat::BC7_SRGB,
 };
 
-constexpr std::array VIEW_CLASS_BPTC_FLOAT = {
+constexpr std::array VIEW_CLASS_BPTC_FLOAT{
     PixelFormat::BC6H_SFLOAT,
     PixelFormat::BC6H_UFLOAT,
 };
 
+constexpr std::array VIEW_CLASS_ASTC_4x4_RGBA{
+    PixelFormat::ASTC_2D_4X4_UNORM,
+    PixelFormat::ASTC_2D_4X4_SRGB,
+};
+
+constexpr std::array VIEW_CLASS_ASTC_5x4_RGBA{
+    PixelFormat::ASTC_2D_5X4_UNORM,
+    PixelFormat::ASTC_2D_5X4_SRGB,
+};
+
+constexpr std::array VIEW_CLASS_ASTC_5x5_RGBA{
+    PixelFormat::ASTC_2D_5X5_UNORM,
+    PixelFormat::ASTC_2D_5X5_SRGB,
+};
+
+constexpr std::array VIEW_CLASS_ASTC_6x5_RGBA{
+    PixelFormat::ASTC_2D_6X5_UNORM,
+    PixelFormat::ASTC_2D_6X5_SRGB,
+};
+
+constexpr std::array VIEW_CLASS_ASTC_6x6_RGBA{
+    PixelFormat::ASTC_2D_6X6_UNORM,
+    PixelFormat::ASTC_2D_6X6_SRGB,
+};
+
+constexpr std::array VIEW_CLASS_ASTC_8x5_RGBA{
+    PixelFormat::ASTC_2D_8X5_UNORM,
+    PixelFormat::ASTC_2D_8X5_SRGB,
+};
+
+constexpr std::array VIEW_CLASS_ASTC_8x8_RGBA{
+    PixelFormat::ASTC_2D_8X8_UNORM,
+    PixelFormat::ASTC_2D_8X8_SRGB,
+};
+
+// Missing formats:
+// PixelFormat::ASTC_2D_10X5_UNORM
+// PixelFormat::ASTC_2D_10X5_SRGB
+
+// Missing formats:
+// PixelFormat::ASTC_2D_10X6_UNORM
+// PixelFormat::ASTC_2D_10X6_SRGB
+
+constexpr std::array VIEW_CLASS_ASTC_10x8_RGBA{
+    PixelFormat::ASTC_2D_10X8_UNORM,
+    PixelFormat::ASTC_2D_10X8_SRGB,
+};
+
+constexpr std::array VIEW_CLASS_ASTC_10x10_RGBA{
+    PixelFormat::ASTC_2D_10X10_UNORM,
+    PixelFormat::ASTC_2D_10X10_SRGB,
+};
+
+// Missing formats
+// ASTC_2D_12X10_UNORM,
+// ASTC_2D_12X10_SRGB,
+
+constexpr std::array VIEW_CLASS_ASTC_12x12_RGBA{
+    PixelFormat::ASTC_2D_12X12_UNORM,
+    PixelFormat::ASTC_2D_12X12_SRGB,
+};
+
 // Compatibility table taken from Table 4.X.1 in:
 // https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_copy_image.txt
 
-constexpr std::array COPY_CLASS_128_BITS = {
+constexpr std::array COPY_CLASS_128_BITS{
     PixelFormat::R32G32B32A32_UINT, PixelFormat::R32G32B32A32_FLOAT, PixelFormat::R32G32B32A32_SINT,
     PixelFormat::BC2_UNORM,         PixelFormat::BC2_SRGB,           PixelFormat::BC3_UNORM,
     PixelFormat::BC3_SRGB,          PixelFormat::BC5_UNORM,          PixelFormat::BC5_SNORM,
@@ -97,7 +161,7 @@ constexpr std::array COPY_CLASS_128_BITS = {
 // PixelFormat::RGBA32I
 // COMPRESSED_RG_RGTC2
 
-constexpr std::array COPY_CLASS_64_BITS = {
+constexpr std::array COPY_CLASS_64_BITS{
     PixelFormat::R16G16B16A16_FLOAT, PixelFormat::R16G16B16A16_UINT,
     PixelFormat::R16G16B16A16_UNORM, PixelFormat::R16G16B16A16_SNORM,
     PixelFormat::R16G16B16A16_SINT,  PixelFormat::R32G32_UINT,
@@ -110,32 +174,36 @@ constexpr std::array COPY_CLASS_64_BITS = {
 // COMPRESSED_RGBA_S3TC_DXT1_EXT
 // COMPRESSED_SIGNED_RED_RGTC1
 
-void Enable(FormatCompatibility::Table& compatiblity, size_t format_a, size_t format_b) {
+constexpr void Enable(Table& table, size_t format_a, size_t format_b) {
-    compatiblity[format_a][format_b] = true;
+    table[format_a][format_b / 64] |= u64(1) << (format_b % 64);
-    compatiblity[format_b][format_a] = true;
+    table[format_b][format_a / 64] |= u64(1) << (format_a % 64);
 }
 
-void Enable(FormatCompatibility::Table& compatibility, PixelFormat format_a, PixelFormat format_b) {
+constexpr void Enable(Table& table, PixelFormat format_a, PixelFormat format_b) {
-    Enable(compatibility, static_cast<size_t>(format_a), static_cast<size_t>(format_b));
+    Enable(table, static_cast<size_t>(format_a), static_cast<size_t>(format_b));
 }
 
 template <typename Range>
-void EnableRange(FormatCompatibility::Table& compatibility, const Range& range) {
+constexpr void EnableRange(Table& table, const Range& range) {
     for (auto it_a = range.begin(); it_a != range.end(); ++it_a) {
         for (auto it_b = it_a; it_b != range.end(); ++it_b) {
-            Enable(compatibility, *it_a, *it_b);
+            Enable(table, *it_a, *it_b);
         }
     }
 }
 
-} // Anonymous namespace
+constexpr bool IsSupported(const Table& table, PixelFormat format_a, PixelFormat format_b) {
+    const size_t a = static_cast<size_t>(format_a);
+    const size_t b = static_cast<size_t>(format_b);
+    return ((table[a][b / 64] >> (b % 64)) & 1) != 0;
+}
 
-FormatCompatibility::FormatCompatibility() {
+constexpr Table MakeViewTable() {
+    Table view{};
     for (size_t i = 0; i < MaxPixelFormat; ++i) {
         // Identity is allowed
         Enable(view, i, i);
     }
-
     EnableRange(view, VIEW_CLASS_128_BITS);
     EnableRange(view, VIEW_CLASS_96_BITS);
     EnableRange(view, VIEW_CLASS_64_BITS);
@@ -146,10 +214,36 @@ FormatCompatibility::FormatCompatibility() {
     EnableRange(view, VIEW_CLASS_RGTC2_RG);
     EnableRange(view, VIEW_CLASS_BPTC_UNORM);
     EnableRange(view, VIEW_CLASS_BPTC_FLOAT);
+    EnableRange(view, VIEW_CLASS_ASTC_4x4_RGBA);
+    EnableRange(view, VIEW_CLASS_ASTC_5x4_RGBA);
+    EnableRange(view, VIEW_CLASS_ASTC_5x5_RGBA);
+    EnableRange(view, VIEW_CLASS_ASTC_6x5_RGBA);
+    EnableRange(view, VIEW_CLASS_ASTC_6x6_RGBA);
+    EnableRange(view, VIEW_CLASS_ASTC_8x5_RGBA);
+    EnableRange(view, VIEW_CLASS_ASTC_8x8_RGBA);
+    EnableRange(view, VIEW_CLASS_ASTC_10x8_RGBA);
+    EnableRange(view, VIEW_CLASS_ASTC_10x10_RGBA);
+    EnableRange(view, VIEW_CLASS_ASTC_12x12_RGBA);
+    return view;
+}
 
-    copy = view;
+constexpr Table MakeCopyTable() {
+    Table copy = MakeViewTable();
     EnableRange(copy, COPY_CLASS_128_BITS);
     EnableRange(copy, COPY_CLASS_64_BITS);
+    return copy;
+}
+
+} // Anonymous namespace
+
+bool IsViewCompatible(PixelFormat format_a, PixelFormat format_b) {
+    static constexpr Table TABLE = MakeViewTable();
+    return IsSupported(TABLE, format_a, format_b);
+}
+
+bool IsCopyCompatible(PixelFormat format_a, PixelFormat format_b) {
+    static constexpr Table TABLE = MakeCopyTable();
+    return IsSupported(TABLE, format_a, format_b);
 }
 
 } // namespace VideoCore::Surface

src/video_core/compatible_formats.h

@@ -4,31 +4,12 @@
 
 #pragma once
 
-#include <array>
-#include <bitset>
-#include <cstddef>
-
 #include "video_core/surface.h"
 
 namespace VideoCore::Surface {
 
-class FormatCompatibility {
+bool IsViewCompatible(PixelFormat format_a, PixelFormat format_b);
-public:
-    using Table = std::array<std::bitset<MaxPixelFormat>, MaxPixelFormat>;
 
-    explicit FormatCompatibility();
+bool IsCopyCompatible(PixelFormat format_a, PixelFormat format_b);
-
-    bool TestView(PixelFormat format_a, PixelFormat format_b) const noexcept {
-        return view[static_cast<size_t>(format_a)][static_cast<size_t>(format_b)];
-    }
-
-    bool TestCopy(PixelFormat format_a, PixelFormat format_b) const noexcept {
-        return copy[static_cast<size_t>(format_a)][static_cast<size_t>(format_b)];
-    }
-
-private:
-    Table view;
-    Table copy;
-};
 
 } // namespace VideoCore::Surface

src/video_core/delayed_destruction_ring.h

@@ -0,0 +1,32 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <cstddef>
+#include <utility>
+#include <vector>
+
+namespace VideoCommon {
+
+/// Container to push objects to be destroyed a few ticks in the future
+template <typename T, size_t TICKS_TO_DESTROY>
+class DelayedDestructionRing {
+public:
+    void Tick() {
+        index = (index + 1) % TICKS_TO_DESTROY;
+        elements[index].clear();
+    }
+
+    void Push(T&& object) {
+        elements[index].push_back(std::move(object));
+    }
+
+private:
+    size_t index = 0;
+    std::array<std::vector<T>, TICKS_TO_DESTROY> elements;
+};
+
+} // namespace VideoCommon

src/video_core/dirty_flags.cpp

@@ -16,6 +16,9 @@ namespace VideoCommon::Dirty {
 using Tegra::Engines::Maxwell3D;
 
 void SetupDirtyRenderTargets(Tegra::Engines::Maxwell3D::DirtyState::Tables& tables) {
+    FillBlock(tables[0], OFF(tic), NUM(tic), Descriptors);
+    FillBlock(tables[0], OFF(tsc), NUM(tsc), Descriptors);
+
     static constexpr std::size_t num_per_rt = NUM(rt[0]);
     static constexpr std::size_t begin = OFF(rt);
     static constexpr std::size_t num = num_per_rt * Maxwell3D::Regs::NumRenderTargets;
@@ -23,6 +26,10 @@ void SetupDirtyRenderTargets(Tegra::Engines::Maxwell3D::DirtyState::Tables& tabl
         FillBlock(tables[0], begin + rt * num_per_rt, num_per_rt, ColorBuffer0 + rt);
     }
     FillBlock(tables[1], begin, num, RenderTargets);
+    FillBlock(tables[0], OFF(render_area), NUM(render_area), RenderTargets);
+
+    tables[0][OFF(rt_control)] = RenderTargets;
+    tables[1][OFF(rt_control)] = RenderTargetControl;
 
     static constexpr std::array zeta_flags{ZetaBuffer, RenderTargets};
     for (std::size_t i = 0; i < std::size(zeta_flags); ++i) {

src/video_core/dirty_flags.h

@@ -16,7 +16,10 @@ namespace VideoCommon::Dirty {
 enum : u8 {
     NullEntry = 0,
 
+    Descriptors,
+
     RenderTargets,
+    RenderTargetControl,
     ColorBuffer0,
     ColorBuffer1,
     ColorBuffer2,

src/video_core/engines/fermi_2d.cpp

@@ -10,7 +10,11 @@
 
 namespace Tegra::Engines {
 
-Fermi2D::Fermi2D() = default;
+Fermi2D::Fermi2D() {
+    // Nvidia's OpenGL driver seems to assume these values
+    regs.src.depth = 1;
+    regs.dst.depth = 1;
+}
 
 Fermi2D::~Fermi2D() = default;
 
@@ -21,78 +25,43 @@ void Fermi2D::BindRasterizer(VideoCore::RasterizerInterface& rasterizer_) {
 void Fermi2D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
     ASSERT_MSG(method < Regs::NUM_REGS,
                "Invalid Fermi2D register, increase the size of the Regs structure");
-
     regs.reg_array[method] = method_argument;
 
-    switch (method) {
+    if (method == FERMI2D_REG_INDEX(pixels_from_memory.src_y0) + 1) {
-    // Trigger the surface copy on the last register write. This is blit_src_y, but this is 64-bit,
+        Blit();
-    // so trigger on the second 32-bit write.
-    case FERMI2D_REG_INDEX(blit_src_y) + 1: {
-        HandleSurfaceCopy();
-        break;
-    }
     }
 }
 
 void Fermi2D::CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending) {
-    for (std::size_t i = 0; i < amount; i++) {
+    for (u32 i = 0; i < amount; ++i) {
-        CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1);
+        CallMethod(method, base_start[i], methods_pending - i <= 1);
     }
 }
 
-static std::pair<u32, u32> DelimitLine(u32 src_1, u32 src_2, u32 dst_1, u32 dst_2, u32 src_line) {
+void Fermi2D::Blit() {
-    const u32 line_a = src_2 - src_1;
+    LOG_DEBUG(HW_GPU, "called. source address=0x{:x}, destination address=0x{:x}",
-    const u32 line_b = dst_2 - dst_1;
+              regs.src.Address(), regs.dst.Address());
-    const u32 excess = std::max<s32>(0, line_a - src_line + src_1);
-    return {line_b - (excess * line_b) / line_a, excess};
-}
 
-void Fermi2D::HandleSurfaceCopy() {
+    UNIMPLEMENTED_IF_MSG(regs.operation != Operation::SrcCopy, "Operation is not copy");
-    LOG_DEBUG(HW_GPU, "Requested a surface copy with operation {}", regs.operation);
+    UNIMPLEMENTED_IF_MSG(regs.src.layer != 0, "Source layer is not zero");
+    UNIMPLEMENTED_IF_MSG(regs.dst.layer != 0, "Destination layer is not zero");
+    UNIMPLEMENTED_IF_MSG(regs.src.depth != 1, "Source depth is not one");
+    UNIMPLEMENTED_IF_MSG(regs.clip_enable != 0, "Clipped blit enabled");
 
-    // TODO(Subv): Only raw copies are implemented.
+    const auto& args = regs.pixels_from_memory;
-    ASSERT(regs.operation == Operation::SrcCopy);
+    const Config config{
-
-    const u32 src_blit_x1{static_cast<u32>(regs.blit_src_x >> 32)};
-    const u32 src_blit_y1{static_cast<u32>(regs.blit_src_y >> 32)};
-    u32 src_blit_x2, src_blit_y2;
-    if (regs.blit_control.origin == Origin::Corner) {
-        src_blit_x2 =
-            static_cast<u32>((regs.blit_src_x + (regs.blit_du_dx * regs.blit_dst_width)) >> 32);
-        src_blit_y2 =
-            static_cast<u32>((regs.blit_src_y + (regs.blit_dv_dy * regs.blit_dst_height)) >> 32);
-    } else {
-        src_blit_x2 = static_cast<u32>((regs.blit_src_x >> 32) + regs.blit_dst_width);
-        src_blit_y2 = static_cast<u32>((regs.blit_src_y >> 32) + regs.blit_dst_height);
-    }
-    u32 dst_blit_x2 = regs.blit_dst_x + regs.blit_dst_width;
-    u32 dst_blit_y2 = regs.blit_dst_y + regs.blit_dst_height;
-    const auto [new_dst_w, src_excess_x] =
-        DelimitLine(src_blit_x1, src_blit_x2, regs.blit_dst_x, dst_blit_x2, regs.src.width);
-    const auto [new_dst_h, src_excess_y] =
-        DelimitLine(src_blit_y1, src_blit_y2, regs.blit_dst_y, dst_blit_y2, regs.src.height);
-    dst_blit_x2 = new_dst_w + regs.blit_dst_x;
-    src_blit_x2 = src_blit_x2 - src_excess_x;
-    dst_blit_y2 = new_dst_h + regs.blit_dst_y;
-    src_blit_y2 = src_blit_y2 - src_excess_y;
-    const auto [new_src_w, dst_excess_x] =
-        DelimitLine(regs.blit_dst_x, dst_blit_x2, src_blit_x1, src_blit_x2, regs.dst.width);
-    const auto [new_src_h, dst_excess_y] =
-        DelimitLine(regs.blit_dst_y, dst_blit_y2, src_blit_y1, src_blit_y2, regs.dst.height);
-    src_blit_x2 = new_src_w + src_blit_x1;
-    dst_blit_x2 = dst_blit_x2 - dst_excess_x;
-    src_blit_y2 = new_src_h + src_blit_y1;
-    dst_blit_y2 = dst_blit_y2 - dst_excess_y;
-    const Common::Rectangle<u32> src_rect{src_blit_x1, src_blit_y1, src_blit_x2, src_blit_y2};
-    const Common::Rectangle<u32> dst_rect{regs.blit_dst_x, regs.blit_dst_y, dst_blit_x2,
-                                          dst_blit_y2};
-    const Config copy_config{
         .operation = regs.operation,
-        .filter = regs.blit_control.filter,
+        .filter = args.sample_mode.filter,
-        .src_rect = src_rect,
+        .dst_x0 = args.dst_x0,
-        .dst_rect = dst_rect,
+        .dst_y0 = args.dst_y0,
+        .dst_x1 = args.dst_x0 + args.dst_width,
+        .dst_y1 = args.dst_y0 + args.dst_height,
+        .src_x0 = static_cast<s32>(args.src_x0 >> 32),
+        .src_y0 = static_cast<s32>(args.src_y0 >> 32),
+        .src_x1 = static_cast<s32>((args.du_dx * args.dst_width + args.src_x0) >> 32),
+        .src_y1 = static_cast<s32>((args.dv_dy * args.dst_height + args.src_y0) >> 32),
     };
-    if (!rasterizer->AccelerateSurfaceCopy(regs.src, regs.dst, copy_config)) {
+    if (!rasterizer->AccelerateSurfaceCopy(regs.src, regs.dst, config)) {
         UNIMPLEMENTED();
     }
 }

src/video_core/engines/fermi_2d.h

@@ -53,8 +53,8 @@ public:
     };
 
     enum class Filter : u32 {
-        PointSample = 0, // Nearest
+        Point = 0,
-        Linear = 1,
+        Bilinear = 1,
     };
 
     enum class Operation : u32 {
@@ -67,88 +67,235 @@ public:
         BlendPremult = 6,
     };
 
-    struct Regs {
+    enum class MemoryLayout : u32 {
-        static constexpr std::size_t NUM_REGS = 0x258;
+        BlockLinear = 0,
+        Pitch = 1,
+    };
 
-        struct Surface {
+    enum class CpuIndexWrap : u32 {
-            RenderTargetFormat format;
+        Wrap = 0,
-            BitField<0, 1, u32> linear;
+        NoWrap = 1,
-            union {
+    };
-                BitField<0, 4, u32> block_width;
-                BitField<4, 4, u32> block_height;
-                BitField<8, 4, u32> block_depth;
-            };
-            u32 depth;
-            u32 layer;
-            u32 pitch;
-            u32 width;
-            u32 height;
-            u32 address_high;
-            u32 address_low;
-
-            GPUVAddr Address() const {
-                return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
-                                             address_low);
-            }
-
-            u32 BlockWidth() const {
-                return block_width.Value();
-            }
-
-            u32 BlockHeight() const {
-                return block_height.Value();
-            }
-
-            u32 BlockDepth() const {
-                return block_depth.Value();
-            }
-        };
-        static_assert(sizeof(Surface) == 0x28, "Surface has incorrect size");
 
+    struct Surface {
+        RenderTargetFormat format;
+        MemoryLayout linear;
         union {
+            BitField<0, 4, u32> block_width;
+            BitField<4, 4, u32> block_height;
+            BitField<8, 4, u32> block_depth;
+        };
+        u32 depth;
+        u32 layer;
+        u32 pitch;
+        u32 width;
+        u32 height;
+        u32 addr_upper;
+        u32 addr_lower;
+
+        [[nodiscard]] constexpr GPUVAddr Address() const noexcept {
+            return (static_cast<GPUVAddr>(addr_upper) << 32) | static_cast<GPUVAddr>(addr_lower);
+        }
+    };
+    static_assert(sizeof(Surface) == 0x28, "Surface has incorrect size");
+
+    enum class SectorPromotion : u32 {
+        NoPromotion = 0,
+        PromoteTo2V = 1,
+        PromoteTo2H = 2,
+        PromoteTo4 = 3,
+    };
+
+    enum class NumTpcs : u32 {
+        All = 0,
+        One = 1,
+    };
+
+    enum class RenderEnableMode : u32 {
+        False = 0,
+        True = 1,
+        Conditional = 2,
+        RenderIfEqual = 3,
+        RenderIfNotEqual = 4,
+    };
+
+    enum class ColorKeyFormat : u32 {
+        A16R56G6B5 = 0,
+        A1R5G55B5 = 1,
+        A8R8G8B8 = 2,
+        A2R10G10B10 = 3,
+        Y8 = 4,
+        Y16 = 5,
+        Y32 = 6,
+    };
+
+    union Beta4 {
+        BitField<0, 8, u32> b;
+        BitField<8, 8, u32> g;
+        BitField<16, 8, u32> r;
+        BitField<24, 8, u32> a;
+    };
+
+    struct Point {
+        u32 x;
+        u32 y;
+    };
+
+    enum class PatternSelect : u32 {
+        MonoChrome8x8 = 0,
+        MonoChrome64x1 = 1,
+        MonoChrome1x64 = 2,
+        Color = 3,
+    };
+
+    enum class NotifyType : u32 {
+        WriteOnly = 0,
+        WriteThenAwaken = 1,
+    };
+
+    enum class MonochromePatternColorFormat : u32 {
+        A8X8R8G6B5 = 0,
+        A1R5G5B5 = 1,
+        A8R8G8B8 = 2,
+        A8Y8 = 3,
+        A8X8Y16 = 4,
+        Y32 = 5,
+    };
+
+    enum class MonochromePatternFormat : u32 {
+        CGA6_M1 = 0,
+        LE_M1 = 1,
+    };
+
+    union Regs {
+        static constexpr std::size_t NUM_REGS = 0x258;
+        struct {
+            u32 object;
+            INSERT_UNION_PADDING_WORDS(0x3F);
+            u32 no_operation;
+            NotifyType notify;
+            INSERT_UNION_PADDING_WORDS(0x2);
+            u32 wait_for_idle;
+            INSERT_UNION_PADDING_WORDS(0xB);
+            u32 pm_trigger;
+            INSERT_UNION_PADDING_WORDS(0xF);
+            u32 context_dma_notify;
+            u32 dst_context_dma;
+            u32 src_context_dma;
+            u32 semaphore_context_dma;
+            INSERT_UNION_PADDING_WORDS(0x1C);
+            Surface dst;
+            CpuIndexWrap pixels_from_cpu_index_wrap;
+            u32 kind2d_check_enable;
+            Surface src;
+            SectorPromotion pixels_from_memory_sector_promotion;
+            INSERT_UNION_PADDING_WORDS(0x1);
+            NumTpcs num_tpcs;
+            u32 render_enable_addr_upper;
+            u32 render_enable_addr_lower;
+            RenderEnableMode render_enable_mode;
+            INSERT_UNION_PADDING_WORDS(0x4);
+            u32 clip_x0;
+            u32 clip_y0;
+            u32 clip_width;
+            u32 clip_height;
+            BitField<0, 1, u32> clip_enable;
+            BitField<0, 3, ColorKeyFormat> color_key_format;
+            u32 color_key;
+            BitField<0, 1, u32> color_key_enable;
+            BitField<0, 8, u32> rop;
+            u32 beta1;
+            Beta4 beta4;
+            Operation operation;
+            union {
+                BitField<0, 6, u32> x;
+                BitField<8, 6, u32> y;
+            } pattern_offset;
+            BitField<0, 2, PatternSelect> pattern_select;
+            INSERT_UNION_PADDING_WORDS(0xC);
             struct {
-                INSERT_UNION_PADDING_WORDS(0x80);
+                BitField<0, 3, MonochromePatternColorFormat> color_format;
-
+                BitField<0, 1, MonochromePatternFormat> format;
-                Surface dst;
+                u32 color0;
-
+                u32 color1;
-                INSERT_UNION_PADDING_WORDS(2);
+                u32 pattern0;
-
+                u32 pattern1;
-                Surface src;
+            } monochrome_pattern;
-
+            struct {
-                INSERT_UNION_PADDING_WORDS(0x15);
+                std::array<u32, 0x40> X8R8G8B8;
-
+                std::array<u32, 0x20> R5G6B5;
-                Operation operation;
+                std::array<u32, 0x20> X1R5G5B5;
-
+                std::array<u32, 0x10> Y8;
-                INSERT_UNION_PADDING_WORDS(0x177);
+            } color_pattern;
-
+            INSERT_UNION_PADDING_WORDS(0x10);
+            struct {
+                u32 prim_mode;
+                u32 prim_color_format;
+                u32 prim_color;
+                u32 line_tie_break_bits;
+                INSERT_UNION_PADDING_WORDS(0x14);
+                u32 prim_point_xy;
+                INSERT_UNION_PADDING_WORDS(0x7);
+                std::array<Point, 0x40> prim_point;
+            } render_solid;
+            struct {
+                u32 data_type;
+                u32 color_format;
+                u32 index_format;
+                u32 mono_format;
+                u32 wrap;
+                u32 color0;
+                u32 color1;
+                u32 mono_opacity;
+                INSERT_UNION_PADDING_WORDS(0x6);
+                u32 src_width;
+                u32 src_height;
+                u32 dx_du_frac;
+                u32 dx_du_int;
+                u32 dx_dv_frac;
+                u32 dy_dv_int;
+                u32 dst_x0_frac;
+                u32 dst_x0_int;
+                u32 dst_y0_frac;
+                u32 dst_y0_int;
+                u32 data;
+            } pixels_from_cpu;
+            INSERT_UNION_PADDING_WORDS(0x3);
+            u32 big_endian_control;
+            INSERT_UNION_PADDING_WORDS(0x3);
+            struct {
+                BitField<0, 3, u32> block_shape;
+                BitField<0, 5, u32> corral_size;
+                BitField<0, 1, u32> safe_overlap;
                 union {
-                    u32 raw;
                     BitField<0, 1, Origin> origin;
                     BitField<4, 1, Filter> filter;
-                } blit_control;
+                } sample_mode;
-
                 INSERT_UNION_PADDING_WORDS(0x8);
-
+                s32 dst_x0;
-                u32 blit_dst_x;
+                s32 dst_y0;
-                u32 blit_dst_y;
+                s32 dst_width;
-                u32 blit_dst_width;
+                s32 dst_height;
-                u32 blit_dst_height;
+                s64 du_dx;
-                u64 blit_du_dx;
+                s64 dv_dy;
-                u64 blit_dv_dy;
+                s64 src_x0;
-                u64 blit_src_x;
+                s64 src_y0;
-                u64 blit_src_y;
+            } pixels_from_memory;
-
-                INSERT_UNION_PADDING_WORDS(0x21);
-            };
-            std::array<u32, NUM_REGS> reg_array;
         };
+        std::array<u32, NUM_REGS> reg_array;
     } regs{};
 
     struct Config {
-        Operation operation{};
+        Operation operation;
-        Filter filter{};
+        Filter filter;
-        Common::Rectangle<u32> src_rect;
+        s32 dst_x0;
-        Common::Rectangle<u32> dst_rect;
+        s32 dst_y0;
+        s32 dst_x1;
+        s32 dst_y1;
+        s32 src_x0;
+        s32 src_y0;
+        s32 src_x1;
+        s32 src_y1;
     };
 
 private:
@@ -156,25 +303,49 @@ private:
 
     /// Performs the copy from the source surface to the destination surface as configured in the
     /// registers.
-    void HandleSurfaceCopy();
+    void Blit();
 };
 
 #define ASSERT_REG_POSITION(field_name, position)                                                  \
-    static_assert(offsetof(Fermi2D::Regs, field_name) == position * 4,                             \
+    static_assert(offsetof(Fermi2D::Regs, field_name) == position,                                 \
                   "Field " #field_name " has invalid position")
 
-ASSERT_REG_POSITION(dst, 0x80);
+ASSERT_REG_POSITION(object, 0x0);
-ASSERT_REG_POSITION(src, 0x8C);
+ASSERT_REG_POSITION(no_operation, 0x100);
-ASSERT_REG_POSITION(operation, 0xAB);
+ASSERT_REG_POSITION(notify, 0x104);
-ASSERT_REG_POSITION(blit_control, 0x223);
+ASSERT_REG_POSITION(wait_for_idle, 0x110);
-ASSERT_REG_POSITION(blit_dst_x, 0x22c);
+ASSERT_REG_POSITION(pm_trigger, 0x140);
-ASSERT_REG_POSITION(blit_dst_y, 0x22d);
+ASSERT_REG_POSITION(context_dma_notify, 0x180);
-ASSERT_REG_POSITION(blit_dst_width, 0x22e);
+ASSERT_REG_POSITION(dst_context_dma, 0x184);
-ASSERT_REG_POSITION(blit_dst_height, 0x22f);
+ASSERT_REG_POSITION(src_context_dma, 0x188);
-ASSERT_REG_POSITION(blit_du_dx, 0x230);
+ASSERT_REG_POSITION(semaphore_context_dma, 0x18C);
-ASSERT_REG_POSITION(blit_dv_dy, 0x232);
+ASSERT_REG_POSITION(dst, 0x200);
-ASSERT_REG_POSITION(blit_src_x, 0x234);
+ASSERT_REG_POSITION(pixels_from_cpu_index_wrap, 0x228);
-ASSERT_REG_POSITION(blit_src_y, 0x236);
+ASSERT_REG_POSITION(kind2d_check_enable, 0x22C);
+ASSERT_REG_POSITION(src, 0x230);
+ASSERT_REG_POSITION(pixels_from_memory_sector_promotion, 0x258);
+ASSERT_REG_POSITION(num_tpcs, 0x260);
+ASSERT_REG_POSITION(render_enable_addr_upper, 0x264);
+ASSERT_REG_POSITION(render_enable_addr_lower, 0x268);
+ASSERT_REG_POSITION(clip_x0, 0x280);
+ASSERT_REG_POSITION(clip_y0, 0x284);
+ASSERT_REG_POSITION(clip_width, 0x288);
+ASSERT_REG_POSITION(clip_height, 0x28c);
+ASSERT_REG_POSITION(clip_enable, 0x290);
+ASSERT_REG_POSITION(color_key_format, 0x294);
+ASSERT_REG_POSITION(color_key, 0x298);
+ASSERT_REG_POSITION(rop, 0x2A0);
+ASSERT_REG_POSITION(beta1, 0x2A4);
+ASSERT_REG_POSITION(beta4, 0x2A8);
+ASSERT_REG_POSITION(operation, 0x2AC);
+ASSERT_REG_POSITION(pattern_offset, 0x2B0);
+ASSERT_REG_POSITION(pattern_select, 0x2B4);
+ASSERT_REG_POSITION(monochrome_pattern, 0x2E8);
+ASSERT_REG_POSITION(color_pattern, 0x300);
+ASSERT_REG_POSITION(render_solid, 0x580);
+ASSERT_REG_POSITION(pixels_from_cpu, 0x800);
+ASSERT_REG_POSITION(big_endian_control, 0x870);
+ASSERT_REG_POSITION(pixels_from_memory, 0x880);
 
 #undef ASSERT_REG_POSITION
 

src/video_core/engines/kepler_compute.cpp

@@ -58,24 +58,6 @@ void KeplerCompute::CallMultiMethod(u32 method, const u32* base_start, u32 amoun
     }
 }
 
-Texture::FullTextureInfo KeplerCompute::GetTexture(std::size_t offset) const {
-    const std::bitset<8> cbuf_mask = launch_description.const_buffer_enable_mask.Value();
-    ASSERT(cbuf_mask[regs.tex_cb_index]);
-
-    const auto& texinfo = launch_description.const_buffer_config[regs.tex_cb_index];
-    ASSERT(texinfo.Address() != 0);
-
-    const GPUVAddr address = texinfo.Address() + offset * sizeof(Texture::TextureHandle);
-    ASSERT(address < texinfo.Address() + texinfo.size);
-
-    const Texture::TextureHandle tex_handle{memory_manager.Read<u32>(address)};
-    return GetTextureInfo(tex_handle);
-}
-
-Texture::FullTextureInfo KeplerCompute::GetTextureInfo(Texture::TextureHandle tex_handle) const {
-    return Texture::FullTextureInfo{GetTICEntry(tex_handle.tic_id), GetTSCEntry(tex_handle.tsc_id)};
-}
-
 u32 KeplerCompute::AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const {
     ASSERT(stage == ShaderType::Compute);
     const auto& buffer = launch_description.const_buffer_config[const_buffer];
@@ -98,9 +80,11 @@ SamplerDescriptor KeplerCompute::AccessBindlessSampler(ShaderType stage, u64 con
 
 SamplerDescriptor KeplerCompute::AccessSampler(u32 handle) const {
     const Texture::TextureHandle tex_handle{handle};
-    const Texture::FullTextureInfo tex_info = GetTextureInfo(tex_handle);
+    const Texture::TICEntry tic = GetTICEntry(tex_handle.tic_id);
-    SamplerDescriptor result = SamplerDescriptor::FromTIC(tex_info.tic);
+    const Texture::TSCEntry tsc = GetTSCEntry(tex_handle.tsc_id);
-    result.is_shadow.Assign(tex_info.tsc.depth_compare_enabled.Value());
+
+    SamplerDescriptor result = SamplerDescriptor::FromTIC(tic);
+    result.is_shadow.Assign(tsc.depth_compare_enabled.Value());
     return result;
 }
 

src/video_core/engines/kepler_compute.h

@@ -209,11 +209,6 @@ public:
     void CallMultiMethod(u32 method, const u32* base_start, u32 amount,
                          u32 methods_pending) override;
 
-    Texture::FullTextureInfo GetTexture(std::size_t offset) const;
-
-    /// Given a texture handle, returns the TSC and TIC entries.
-    Texture::FullTextureInfo GetTextureInfo(Texture::TextureHandle tex_handle) const;
-
     u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const override;
 
     SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const override;

src/video_core/engines/maxwell_3d.cpp

@@ -2,7 +2,6 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <cinttypes>
 #include <cstring>
 #include <optional>
 #include "common/assert.h"
@@ -227,6 +226,10 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
             OnMemoryWrite();
         }
         return;
+    case MAXWELL3D_REG_INDEX(fragment_barrier):
+        return rasterizer->FragmentBarrier();
+    case MAXWELL3D_REG_INDEX(tiled_cache_barrier):
+        return rasterizer->TiledCacheBarrier();
     }
 }
 
@@ -639,7 +642,7 @@ void Maxwell3D::FinishCBData() {
 }
 
 Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {
-    const GPUVAddr tic_address_gpu{regs.tic.TICAddress() + tic_index * sizeof(Texture::TICEntry)};
+    const GPUVAddr tic_address_gpu{regs.tic.Address() + tic_index * sizeof(Texture::TICEntry)};
 
     Texture::TICEntry tic_entry;
     memory_manager.ReadBlockUnsafe(tic_address_gpu, &tic_entry, sizeof(Texture::TICEntry));
@@ -648,43 +651,19 @@ Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {
 }
 
 Texture::TSCEntry Maxwell3D::GetTSCEntry(u32 tsc_index) const {
-    const GPUVAddr tsc_address_gpu{regs.tsc.TSCAddress() + tsc_index * sizeof(Texture::TSCEntry)};
+    const GPUVAddr tsc_address_gpu{regs.tsc.Address() + tsc_index * sizeof(Texture::TSCEntry)};
 
     Texture::TSCEntry tsc_entry;
     memory_manager.ReadBlockUnsafe(tsc_address_gpu, &tsc_entry, sizeof(Texture::TSCEntry));
     return tsc_entry;
 }
 
-Texture::FullTextureInfo Maxwell3D::GetTextureInfo(Texture::TextureHandle tex_handle) const {
-    return Texture::FullTextureInfo{GetTICEntry(tex_handle.tic_id), GetTSCEntry(tex_handle.tsc_id)};
-}
-
-Texture::FullTextureInfo Maxwell3D::GetStageTexture(ShaderType stage, std::size_t offset) const {
-    const auto stage_index = static_cast<std::size_t>(stage);
-    const auto& shader = state.shader_stages[stage_index];
-    const auto& tex_info_buffer = shader.const_buffers[regs.tex_cb_index];
-    ASSERT(tex_info_buffer.enabled && tex_info_buffer.address != 0);
-
-    const GPUVAddr tex_info_address =
-        tex_info_buffer.address + offset * sizeof(Texture::TextureHandle);
-
-    ASSERT(tex_info_address < tex_info_buffer.address + tex_info_buffer.size);
-
-    const Texture::TextureHandle tex_handle{memory_manager.Read<u32>(tex_info_address)};
-
-    return GetTextureInfo(tex_handle);
-}
-
 u32 Maxwell3D::GetRegisterValue(u32 method) const {
     ASSERT_MSG(method < Regs::NUM_REGS, "Invalid Maxwell3D register");
     return regs.reg_array[method];
 }
 
 void Maxwell3D::ProcessClearBuffers() {
-    ASSERT(regs.clear_buffers.R == regs.clear_buffers.G &&
-           regs.clear_buffers.R == regs.clear_buffers.B &&
-           regs.clear_buffers.R == regs.clear_buffers.A);
-
     rasterizer->Clear();
 }
 
@@ -692,9 +671,7 @@ u32 Maxwell3D::AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offse
     ASSERT(stage != ShaderType::Compute);
     const auto& shader_stage = state.shader_stages[static_cast<std::size_t>(stage)];
     const auto& buffer = shader_stage.const_buffers[const_buffer];
-    u32 result;
+    return memory_manager.Read<u32>(buffer.address + offset);
-    std::memcpy(&result, memory_manager.GetPointer(buffer.address + offset), sizeof(u32));
-    return result;
 }
 
 SamplerDescriptor Maxwell3D::AccessBoundSampler(ShaderType stage, u64 offset) const {
@@ -712,9 +689,11 @@ SamplerDescriptor Maxwell3D::AccessBindlessSampler(ShaderType stage, u64 const_b
 
 SamplerDescriptor Maxwell3D::AccessSampler(u32 handle) const {
     const Texture::TextureHandle tex_handle{handle};
-    const Texture::FullTextureInfo tex_info = GetTextureInfo(tex_handle);
+    const Texture::TICEntry tic = GetTICEntry(tex_handle.tic_id);
-    SamplerDescriptor result = SamplerDescriptor::FromTIC(tex_info.tic);
+    const Texture::TSCEntry tsc = GetTSCEntry(tex_handle.tsc_id);
-    result.is_shadow.Assign(tex_info.tsc.depth_compare_enabled.Value());
+
+    SamplerDescriptor result = SamplerDescriptor::FromTIC(tic);
+    result.is_shadow.Assign(tsc.depth_compare_enabled.Value());
     return result;
 }
 

src/video_core/engines/maxwell_3d.h

@@ -438,16 +438,6 @@ public:
             DecrWrapOGL = 0x8508,
         };
 
-        enum class MemoryLayout : u32 {
-            Linear = 0,
-            BlockLinear = 1,
-        };
-
-        enum class InvMemoryLayout : u32 {
-            BlockLinear = 0,
-            Linear = 1,
-        };
-
         enum class CounterReset : u32 {
             SampleCnt = 0x01,
             Unk02 = 0x02,
@@ -589,21 +579,31 @@ public:
             NegativeW = 7,
         };
 
+        enum class SamplerIndex : u32 {
+            Independently = 0,
+            ViaHeaderIndex = 1,
+        };
+
+        struct TileMode {
+            union {
+                BitField<0, 4, u32> block_width;
+                BitField<4, 4, u32> block_height;
+                BitField<8, 4, u32> block_depth;
+                BitField<12, 1, u32> is_pitch_linear;
+                BitField<16, 1, u32> is_3d;
+            };
+        };
+        static_assert(sizeof(TileMode) == 4);
+
         struct RenderTargetConfig {
             u32 address_high;
             u32 address_low;
             u32 width;
             u32 height;
             Tegra::RenderTargetFormat format;
+            TileMode tile_mode;
             union {
-                BitField<0, 3, u32> block_width;
+                BitField<0, 16, u32> depth;
-                BitField<4, 3, u32> block_height;
-                BitField<8, 3, u32> block_depth;
-                BitField<12, 1, InvMemoryLayout> type;
-                BitField<16, 1, u32> is_3d;
-            } memory_layout;
-            union {
-                BitField<0, 16, u32> layers;
                 BitField<16, 1, u32> volume;
             };
             u32 layer_stride;
@@ -832,7 +832,11 @@ public:
 
                 u32 patch_vertices;
 
-                INSERT_UNION_PADDING_WORDS(0xC);
+                INSERT_UNION_PADDING_WORDS(0x4);
+
+                u32 fragment_barrier;
+
+                INSERT_UNION_PADDING_WORDS(0x7);
 
                 std::array<ScissorTest, NumViewports> scissor_test;
 
@@ -842,7 +846,15 @@ public:
                 u32 stencil_back_mask;
                 u32 stencil_back_func_mask;
 
-                INSERT_UNION_PADDING_WORDS(0xC);
+                INSERT_UNION_PADDING_WORDS(0x5);
+
+                u32 invalidate_texture_data_cache;
+
+                INSERT_UNION_PADDING_WORDS(0x1);
+
+                u32 tiled_cache_barrier;
+
+                INSERT_UNION_PADDING_WORDS(0x4);
 
                 u32 color_mask_common;
 
@@ -866,12 +878,7 @@ public:
                     u32 address_high;
                     u32 address_low;
                     Tegra::DepthFormat format;
-                    union {
+                    TileMode tile_mode;
-                        BitField<0, 4, u32> block_width;
-                        BitField<4, 4, u32> block_height;
-                        BitField<8, 4, u32> block_depth;
-                        BitField<20, 1, InvMemoryLayout> type;
-                    } memory_layout;
                     u32 layer_stride;
 
                     GPUVAddr Address() const {
@@ -880,7 +887,18 @@ public:
                     }
                 } zeta;
 
-                INSERT_UNION_PADDING_WORDS(0x41);
+                struct {
+                    union {
+                        BitField<0, 16, u32> x;
+                        BitField<16, 16, u32> width;
+                    };
+                    union {
+                        BitField<0, 16, u32> y;
+                        BitField<16, 16, u32> height;
+                    };
+                } render_area;
+
+                INSERT_UNION_PADDING_WORDS(0x3F);
 
                 union {
                     BitField<0, 4, u32> stencil;
@@ -921,7 +939,7 @@ public:
                         BitField<25, 3, u32> map_7;
                     };
 
-                    u32 GetMap(std::size_t index) const {
+                    u32 Map(std::size_t index) const {
                         const std::array<u32, NumRenderTargets> maps{map_0, map_1, map_2, map_3,
                                                                      map_4, map_5, map_6, map_7};
                         ASSERT(index < maps.size());
@@ -934,11 +952,13 @@ public:
                 u32 zeta_width;
                 u32 zeta_height;
                 union {
-                    BitField<0, 16, u32> zeta_layers;
+                    BitField<0, 16, u32> zeta_depth;
                     BitField<16, 1, u32> zeta_volume;
                 };
 
-                INSERT_UNION_PADDING_WORDS(0x26);
+                SamplerIndex sampler_index;
+
+                INSERT_UNION_PADDING_WORDS(0x25);
 
                 u32 depth_test_enable;
 
@@ -964,6 +984,7 @@ public:
                     float b;
                     float a;
                 } blend_color;
+
                 INSERT_UNION_PADDING_WORDS(0x4);
 
                 struct {
@@ -1001,7 +1022,12 @@ public:
                 float line_width_smooth;
                 float line_width_aliased;
 
-                INSERT_UNION_PADDING_WORDS(0x1F);
+                INSERT_UNION_PADDING_WORDS(0x1B);
+
+                u32 invalidate_sampler_cache_no_wfi;
+                u32 invalidate_texture_header_cache_no_wfi;
+
+                INSERT_UNION_PADDING_WORDS(0x2);
 
                 u32 vb_element_base;
                 u32 vb_base_instance;
@@ -1045,13 +1071,13 @@ public:
                 } condition;
 
                 struct {
-                    u32 tsc_address_high;
+                    u32 address_high;
-                    u32 tsc_address_low;
+                    u32 address_low;
-                    u32 tsc_limit;
+                    u32 limit;
 
-                    GPUVAddr TSCAddress() const {
+                    GPUVAddr Address() const {
-                        return static_cast<GPUVAddr>(
+                        return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
-                            (static_cast<GPUVAddr>(tsc_address_high) << 32) | tsc_address_low);
+                                                     address_low);
                     }
                 } tsc;
 
@@ -1062,13 +1088,13 @@ public:
                 u32 line_smooth_enable;
 
                 struct {
-                    u32 tic_address_high;
+                    u32 address_high;
-                    u32 tic_address_low;
+                    u32 address_low;
-                    u32 tic_limit;
+                    u32 limit;
 
-                    GPUVAddr TICAddress() const {
+                    GPUVAddr Address() const {
-                        return static_cast<GPUVAddr>(
+                        return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
-                            (static_cast<GPUVAddr>(tic_address_high) << 32) | tic_address_low);
+                                                     address_low);
                     }
                 } tic;
 
@@ -1397,12 +1423,6 @@ public:
 
     void FlushMMEInlineDraw();
 
-    /// Given a texture handle, returns the TSC and TIC entries.
-    Texture::FullTextureInfo GetTextureInfo(Texture::TextureHandle tex_handle) const;
-
-    /// Returns the texture information for a specific texture in a specific shader stage.
-    Texture::FullTextureInfo GetStageTexture(ShaderType stage, std::size_t offset) const;
-
     u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const override;
 
     SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const override;
@@ -1598,10 +1618,13 @@ ASSERT_REG_POSITION(polygon_offset_point_enable, 0x370);
 ASSERT_REG_POSITION(polygon_offset_line_enable, 0x371);
 ASSERT_REG_POSITION(polygon_offset_fill_enable, 0x372);
 ASSERT_REG_POSITION(patch_vertices, 0x373);
+ASSERT_REG_POSITION(fragment_barrier, 0x378);
 ASSERT_REG_POSITION(scissor_test, 0x380);
 ASSERT_REG_POSITION(stencil_back_func_ref, 0x3D5);
 ASSERT_REG_POSITION(stencil_back_mask, 0x3D6);
 ASSERT_REG_POSITION(stencil_back_func_mask, 0x3D7);
+ASSERT_REG_POSITION(invalidate_texture_data_cache, 0x3DD);
+ASSERT_REG_POSITION(tiled_cache_barrier, 0x3DF);
 ASSERT_REG_POSITION(color_mask_common, 0x3E4);
 ASSERT_REG_POSITION(depth_bounds, 0x3E7);
 ASSERT_REG_POSITION(rt_separate_frag_data, 0x3EB);
@@ -1609,6 +1632,7 @@ ASSERT_REG_POSITION(multisample_raster_enable, 0x3ED);
 ASSERT_REG_POSITION(multisample_raster_samples, 0x3EE);
 ASSERT_REG_POSITION(multisample_sample_mask, 0x3EF);
 ASSERT_REG_POSITION(zeta, 0x3F8);
+ASSERT_REG_POSITION(render_area, 0x3FD);
 ASSERT_REG_POSITION(clear_flags, 0x43E);
 ASSERT_REG_POSITION(fill_rectangle, 0x44F);
 ASSERT_REG_POSITION(vertex_attrib_format, 0x458);
@@ -1617,7 +1641,8 @@ ASSERT_REG_POSITION(multisample_coverage_to_color, 0x47E);
 ASSERT_REG_POSITION(rt_control, 0x487);
 ASSERT_REG_POSITION(zeta_width, 0x48a);
 ASSERT_REG_POSITION(zeta_height, 0x48b);
-ASSERT_REG_POSITION(zeta_layers, 0x48c);
+ASSERT_REG_POSITION(zeta_depth, 0x48c);
+ASSERT_REG_POSITION(sampler_index, 0x48D);
 ASSERT_REG_POSITION(depth_test_enable, 0x4B3);
 ASSERT_REG_POSITION(independent_blend_enable, 0x4B9);
 ASSERT_REG_POSITION(depth_write_enabled, 0x4BA);
@@ -1641,6 +1666,8 @@ ASSERT_REG_POSITION(frag_color_clamp, 0x4EA);
 ASSERT_REG_POSITION(screen_y_control, 0x4EB);
 ASSERT_REG_POSITION(line_width_smooth, 0x4EC);
 ASSERT_REG_POSITION(line_width_aliased, 0x4ED);
+ASSERT_REG_POSITION(invalidate_sampler_cache_no_wfi, 0x509);
+ASSERT_REG_POSITION(invalidate_texture_header_cache_no_wfi, 0x50A);
 ASSERT_REG_POSITION(vb_element_base, 0x50D);
 ASSERT_REG_POSITION(vb_base_instance, 0x50E);
 ASSERT_REG_POSITION(clip_distance_enabled, 0x544);

src/video_core/engines/maxwell_dma.cpp

@@ -96,6 +96,7 @@ void MaxwellDMA::CopyPitchToPitch() {
 }
 
 void MaxwellDMA::CopyBlockLinearToPitch() {
+    UNIMPLEMENTED_IF(regs.src_params.block_size.width != 0);
     UNIMPLEMENTED_IF(regs.src_params.block_size.depth != 0);
     UNIMPLEMENTED_IF(regs.src_params.layer != 0);
 
@@ -135,6 +136,8 @@ void MaxwellDMA::CopyBlockLinearToPitch() {
 }
 
 void MaxwellDMA::CopyPitchToBlockLinear() {
+    UNIMPLEMENTED_IF_MSG(regs.dst_params.block_size.width != 0, "Block width is not one");
+
     const auto& dst_params = regs.dst_params;
     const u32 bytes_per_pixel = regs.pitch_in / regs.line_length_in;
     const u32 width = dst_params.width;

src/video_core/fence_manager.h

@@ -9,6 +9,7 @@
 
 #include "common/common_types.h"
 #include "core/core.h"
+#include "video_core/delayed_destruction_ring.h"
 #include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"
@@ -47,6 +48,11 @@ protected:
 template <typename TFence, typename TTextureCache, typename TTBufferCache, typename TQueryCache>
 class FenceManager {
 public:
+    /// Notify the fence manager about a new frame
+    void TickFrame() {
+        delayed_destruction_ring.Tick();
+    }
+
     void SignalSemaphore(GPUVAddr addr, u32 value) {
         TryReleasePendingFences();
         const bool should_flush = ShouldFlush();
@@ -86,7 +92,7 @@ public:
             } else {
                 gpu.IncrementSyncPoint(current_fence->GetPayload());
             }
-            fences.pop();
+            PopFence();
         }
     }
 
@@ -132,7 +138,7 @@ private:
             } else {
                 gpu.IncrementSyncPoint(current_fence->GetPayload());
             }
-            fences.pop();
+            PopFence();
         }
     }
 
@@ -158,7 +164,14 @@ private:
         query_cache.CommitAsyncFlushes();
     }
 
+    void PopFence() {
+        delayed_destruction_ring.Push(std::move(fences.front()));
+        fences.pop();
+    }
+
     std::queue<TFence> fences;
+
+    DelayedDestructionRing<TFence, 6> delayed_destruction_ring;
 };
 
 } // namespace VideoCommon

src/video_core/host_shaders/CMakeLists.txt

@@ -1,8 +1,26 @@
-set(SHADER_SOURCES
+set(SHADER_FILES
+    block_linear_unswizzle_2d.comp
+    block_linear_unswizzle_3d.comp
+    convert_depth_to_float.frag
+    convert_float_to_depth.frag
+    full_screen_triangle.vert
+    opengl_copy_bc4.comp
     opengl_present.frag
     opengl_present.vert
+    pitch_unswizzle.comp
+    vulkan_blit_color_float.frag
+    vulkan_blit_depth_stencil.frag
+    vulkan_present.frag
+    vulkan_present.vert
+    vulkan_quad_array.comp
+    vulkan_quad_indexed.comp
+    vulkan_uint8.comp
 )
 
+find_program(GLSLANGVALIDATOR "glslangValidator" REQUIRED)
+
+set(GLSL_FLAGS "")
+
 set(SHADER_INCLUDE ${CMAKE_CURRENT_BINARY_DIR}/include)
 set(SHADER_DIR ${SHADER_INCLUDE}/video_core/host_shaders)
 set(HOST_SHADERS_INCLUDE ${SHADER_INCLUDE} PARENT_SCOPE)
@@ -10,27 +28,44 @@ set(HOST_SHADERS_INCLUDE ${SHADER_INCLUDE} PARENT_SCOPE)
 set(INPUT_FILE ${CMAKE_CURRENT_SOURCE_DIR}/source_shader.h.in)
 set(HEADER_GENERATOR ${CMAKE_CURRENT_SOURCE_DIR}/StringShaderHeader.cmake)
 
-foreach(FILENAME IN ITEMS ${SHADER_SOURCES})
+foreach(FILENAME IN ITEMS ${SHADER_FILES})
     string(REPLACE "." "_" SHADER_NAME ${FILENAME})
     set(SOURCE_FILE ${CMAKE_CURRENT_SOURCE_DIR}/${FILENAME})
-    set(HEADER_FILE ${SHADER_DIR}/${SHADER_NAME}.h)
+    # Skip generating source headers on Vulkan exclusive files
-    add_custom_command(
+    if (NOT ${FILENAME} MATCHES "vulkan.*")
-        OUTPUT
+        set(SOURCE_HEADER_FILE ${SHADER_DIR}/${SHADER_NAME}.h)
-            ${HEADER_FILE}
+        add_custom_command(
-        COMMAND
+            OUTPUT
-            ${CMAKE_COMMAND} -P ${HEADER_GENERATOR} ${SOURCE_FILE} ${HEADER_FILE} ${INPUT_FILE}
+                ${SOURCE_HEADER_FILE}
-        MAIN_DEPENDENCY
+            COMMAND
-            ${SOURCE_FILE}
+                ${CMAKE_COMMAND} -P ${HEADER_GENERATOR} ${SOURCE_FILE} ${SOURCE_HEADER_FILE} ${INPUT_FILE}
-        DEPENDS
+            MAIN_DEPENDENCY
-            ${INPUT_FILE}
+                ${SOURCE_FILE}
-            # HEADER_GENERATOR should be included here but msbuild seems to assume it's always modified
+            DEPENDS
-    )
+                ${INPUT_FILE}
-    set(SHADER_HEADERS ${SHADER_HEADERS} ${HEADER_FILE})
+                # HEADER_GENERATOR should be included here but msbuild seems to assume it's always modified
+        )
+        set(SHADER_HEADERS ${SHADER_HEADERS} ${SOURCE_HEADER_FILE})
+    endif()
+    # Skip compiling to SPIR-V OpenGL exclusive files
+    if (NOT ${FILENAME} MATCHES "opengl.*")
+        string(TOUPPER ${SHADER_NAME}_SPV SPIRV_VARIABLE_NAME)
+        set(SPIRV_HEADER_FILE ${SHADER_DIR}/${SHADER_NAME}_spv.h)
+        add_custom_command(
+            OUTPUT
+                ${SPIRV_HEADER_FILE}
+            COMMAND
+                ${GLSLANGVALIDATOR} -V ${GLSL_FLAGS} --variable-name ${SPIRV_VARIABLE_NAME} -o ${SPIRV_HEADER_FILE} ${SOURCE_FILE}
+            MAIN_DEPENDENCY
+                ${SOURCE_FILE}
+        )
+        set(SHADER_HEADERS ${SHADER_HEADERS} ${SPIRV_HEADER_FILE})
+    endif()
 endforeach()
 
 add_custom_target(host_shaders
     DEPENDS
         ${SHADER_HEADERS}
     SOURCES
-        ${SHADER_SOURCES}
+        ${SHADER_FILES}
 )

src/video_core/host_shaders/block_linear_unswizzle_2d.comp

@@ -0,0 +1,122 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 430
+
+#ifdef VULKAN
+
+#extension GL_EXT_shader_16bit_storage : require
+#extension GL_EXT_shader_8bit_storage : require
+#define HAS_EXTENDED_TYPES 1
+#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
+
+#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
+
+#extension GL_NV_gpu_shader5 : enable
+#ifdef GL_NV_gpu_shader5
+#define HAS_EXTENDED_TYPES 1
+#else
+#define HAS_EXTENDED_TYPES 0
+#endif
+#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_IMAGE 0
+
+#endif
+
+BEGIN_PUSH_CONSTANTS
+UNIFORM(0) uvec3 origin;
+UNIFORM(1) ivec3 destination;
+UNIFORM(2) uint bytes_per_block_log2;
+UNIFORM(3) uint layer_stride;
+UNIFORM(4) uint block_size;
+UNIFORM(5) uint x_shift;
+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[]; };
+#endif
+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_IMAGE) uniform writeonly uimage2DArray output_image;
+
+layout(local_size_x = 32, local_size_y = 32, local_size_z = 1) in;
+
+const uint GOB_SIZE_X = 64;
+const uint GOB_SIZE_Y = 8;
+const uint GOB_SIZE_Z = 1;
+const uint GOB_SIZE = GOB_SIZE_X * GOB_SIZE_Y * GOB_SIZE_Z;
+
+const uint GOB_SIZE_X_SHIFT = 6;
+const uint GOB_SIZE_Y_SHIFT = 3;
+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 - 1, GOB_SIZE_Y - 1);
+
+uint SwizzleOffset(uvec2 pos) {
+    pos = pos & SWIZZLE_MASK;
+    return swizzle_table[pos.y * 64 + pos.x];
+}
+
+uvec4 ReadTexel(uint offset) {
+    switch (bytes_per_block_log2) {
+#if HAS_EXTENDED_TYPES
+    case 0:
+        return uvec4(u8data[offset], 0, 0, 0);
+    case 1:
+        return uvec4(u16data[offset / 2], 0, 0, 0);
+#else
+    case 0:
+        return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 24), 8), 0, 0, 0);
+    case 1:
+        return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 16), 16), 0, 0, 0);
+#endif
+    case 2:
+        return uvec4(u32data[offset / 4], 0, 0, 0);
+    case 3:
+        return uvec4(u64data[offset / 8], 0, 0);
+    case 4:
+        return u128data[offset / 16];
+    }
+    return uvec4(0);
+}
+
+void main() {
+    uvec3 pos = gl_GlobalInvocationID + origin;
+    pos.x <<= bytes_per_block_log2;
+
+    // Read as soon as possible due to its latency
+    const uint swizzle = SwizzleOffset(pos.xy);
+
+    const uint block_y = pos.y >> GOB_SIZE_Y_SHIFT;
+
+    uint offset = 0;
+    offset += pos.z * layer_stride;
+    offset += (block_y >> block_height) * block_size;
+    offset += (block_y & block_height_mask) << GOB_SIZE_SHIFT;
+    offset += (pos.x >> GOB_SIZE_X_SHIFT) << x_shift;
+    offset += swizzle;
+
+    const uvec4 texel = ReadTexel(offset);
+    const ivec3 coord = ivec3(gl_GlobalInvocationID) + destination;
+    imageStore(output_image, coord, texel);
+}

src/video_core/host_shaders/block_linear_unswizzle_3d.comp

@@ -0,0 +1,125 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 430
+
+#ifdef VULKAN
+
+#extension GL_EXT_shader_16bit_storage : require
+#extension GL_EXT_shader_8bit_storage : require
+#define HAS_EXTENDED_TYPES 1
+#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
+
+#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
+
+#extension GL_NV_gpu_shader5 : enable
+#ifdef GL_NV_gpu_shader5
+#define HAS_EXTENDED_TYPES 1
+#else
+#define HAS_EXTENDED_TYPES 0
+#endif
+#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_IMAGE 0
+
+#endif
+
+BEGIN_PUSH_CONSTANTS
+UNIFORM(0) uvec3 origin;
+UNIFORM(1) ivec3 destination;
+UNIFORM(2) uint bytes_per_block_log2;
+UNIFORM(3) uint slice_size;
+UNIFORM(4) uint block_size;
+UNIFORM(5) uint x_shift;
+UNIFORM(6) uint block_height;
+UNIFORM(7) uint block_height_mask;
+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[]; };
+#endif
+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_IMAGE) uniform writeonly uimage3D output_image;
+
+layout(local_size_x = 16, local_size_y = 8, local_size_z = 8) in;
+
+const uint GOB_SIZE_X = 64;
+const uint GOB_SIZE_Y = 8;
+const uint GOB_SIZE_Z = 1;
+const uint GOB_SIZE = GOB_SIZE_X * GOB_SIZE_Y * GOB_SIZE_Z;
+
+const uint GOB_SIZE_X_SHIFT = 6;
+const uint GOB_SIZE_Y_SHIFT = 3;
+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 - 1, GOB_SIZE_Y - 1);
+
+uint SwizzleOffset(uvec2 pos) {
+    pos = pos & SWIZZLE_MASK;
+    return swizzle_table[pos.y * 64 + pos.x];
+}
+
+uvec4 ReadTexel(uint offset) {
+    switch (bytes_per_block_log2) {
+#if HAS_EXTENDED_TYPES
+    case 0:
+        return uvec4(u8data[offset], 0, 0, 0);
+    case 1:
+        return uvec4(u16data[offset / 2], 0, 0, 0);
+#else
+    case 0:
+        return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 24), 8), 0, 0, 0);
+    case 1:
+        return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 16), 16), 0, 0, 0);
+#endif
+    case 2:
+        return uvec4(u32data[offset / 4], 0, 0, 0);
+    case 3:
+        return uvec4(u64data[offset / 8], 0, 0);
+    case 4:
+        return u128data[offset / 16];
+    }
+    return uvec4(0);
+}
+
+void main() {
+    uvec3 pos = gl_GlobalInvocationID + origin;
+    pos.x <<= bytes_per_block_log2;
+
+    // Read as soon as possible due to its latency
+    const uint swizzle = SwizzleOffset(pos.xy);
+
+    const uint block_y = pos.y >> GOB_SIZE_Y_SHIFT;
+
+    uint offset = 0;
+    offset += (pos.z >> block_depth) * slice_size;
+    offset += (pos.z & block_depth_mask) << (GOB_SIZE_SHIFT + block_height);
+    offset += (block_y >> block_height) * block_size;
+    offset += (block_y & block_height_mask) << GOB_SIZE_SHIFT;
+    offset += (pos.x >> GOB_SIZE_X_SHIFT) << x_shift;
+    offset += swizzle;
+
+    const uvec4 texel = ReadTexel(offset);
+    const ivec3 coord = ivec3(gl_GlobalInvocationID) + destination;
+    imageStore(output_image, coord, texel);
+}

src/video_core/host_shaders/convert_depth_to_float.frag

@@ -0,0 +1,13 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 450
+
+layout(binding = 0) uniform sampler2D depth_texture;
+layout(location = 0) out float output_color;
+
+void main() {
+    ivec2 coord = ivec2(gl_FragCoord.xy);
+    output_color = texelFetch(depth_texture, coord, 0).r;
+}

src/video_core/host_shaders/convert_float_to_depth.frag

@@ -0,0 +1,13 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 450
+
+layout(binding = 0) uniform sampler2D color_texture;
+
+void main() {
+    ivec2 coord = ivec2(gl_FragCoord.xy);
+    float color = texelFetch(color_texture, coord, 0).r;
+    gl_FragDepth = color;
+}

src/video_core/host_shaders/full_screen_triangle.vert

@@ -0,0 +1,29 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 450
+
+#ifdef VULKAN
+#define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
+#define END_PUSH_CONSTANTS };
+#define UNIFORM(n)
+#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
+#define BEGIN_PUSH_CONSTANTS
+#define END_PUSH_CONSTANTS
+#define UNIFORM(n) layout (location = n) uniform
+#endif
+
+BEGIN_PUSH_CONSTANTS
+UNIFORM(0) vec2 tex_scale;
+UNIFORM(1) vec2 tex_offset;
+END_PUSH_CONSTANTS
+
+layout(location = 0) out vec2 texcoord;
+
+void main() {
+    float x = float((gl_VertexIndex & 1) << 2);
+    float y = float((gl_VertexIndex & 2) << 1);
+    gl_Position = vec4(x - 1.0, y - 1.0, 0.0, 1.0);
+    texcoord = fma(vec2(x, y) / 2.0, tex_scale, tex_offset);
+}

src/video_core/host_shaders/opengl_copy_bc4.comp

@@ -0,0 +1,70 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 430 core
+#extension GL_ARB_gpu_shader_int64 : require
+
+layout (local_size_x = 4, local_size_y = 4) in;
+
+layout(binding = 0, rg32ui) readonly uniform uimage3D bc4_input;
+layout(binding = 1, rgba8ui) writeonly uniform uimage3D bc4_output;
+
+layout(location = 0) uniform uvec3 src_offset;
+layout(location = 1) uniform uvec3 dst_offset;
+
+// https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_texture_compression_rgtc.txt
+uint DecompressBlock(uint64_t bits, uvec2 coord) {
+    const uint code_offset = 16 + 3 * (4 * coord.y + coord.x);
+    const uint code = uint(bits >> code_offset) & 7;
+    const uint red0 = uint(bits >> 0) & 0xff;
+    const uint red1 = uint(bits >> 8) & 0xff;
+    if (red0 > red1) {
+        switch (code) {
+        case 0:
+            return red0;
+        case 1:
+            return red1;
+        case 2:
+            return (6 * red0 + 1 * red1) / 7;
+        case 3:
+            return (5 * red0 + 2 * red1) / 7;
+        case 4:
+            return (4 * red0 + 3 * red1) / 7;
+        case 5:
+            return (3 * red0 + 4 * red1) / 7;
+        case 6:
+            return (2 * red0 + 5 * red1) / 7;
+        case 7:
+            return (1 * red0 + 6 * red1) / 7;
+        }
+    } else {
+        switch (code) {
+        case 0:
+            return red0;
+        case 1:
+            return red1;
+        case 2:
+            return (4 * red0 + 1 * red1) / 5;
+        case 3:
+            return (3 * red0 + 2 * red1) / 5;
+        case 4:
+            return (2 * red0 + 3 * red1) / 5;
+        case 5:
+            return (1 * red0 + 4 * red1) / 5;
+        case 6:
+            return 0;
+        case 7:
+            return 0xff;
+        }
+    }
+    return 0;
+}
+
+void main() {
+    uvec2 packed_bits = imageLoad(bc4_input, ivec3(gl_WorkGroupID + src_offset)).rg;
+    uint64_t bits = packUint2x32(packed_bits);
+    uint red = DecompressBlock(bits, gl_LocalInvocationID.xy);
+    uvec4 color = uvec4(red & 0xff, 0, 0, 0xff);
+    imageStore(bc4_output, ivec3(gl_GlobalInvocationID + dst_offset), color);
+}

src/video_core/host_shaders/opengl_present.frag

@@ -1,3 +1,7 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
 #version 430 core
 
 layout (location = 0) in vec2 frag_tex_coord;

src/video_core/host_shaders/opengl_present.vert

@@ -1,3 +1,7 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
 #version 430 core
 
 out gl_PerVertex {

src/video_core/host_shaders/pitch_unswizzle.comp

@@ -0,0 +1,86 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 430
+
+#ifdef VULKAN
+
+#extension GL_EXT_shader_16bit_storage : require
+#extension GL_EXT_shader_8bit_storage : require
+#define HAS_EXTENDED_TYPES 1
+#define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
+#define END_PUSH_CONSTANTS };
+#define UNIFORM(n)
+#define BINDING_INPUT_BUFFER 0
+#define BINDING_OUTPUT_IMAGE 1
+
+#else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
+
+#extension GL_NV_gpu_shader5 : enable
+#ifdef GL_NV_gpu_shader5
+#define HAS_EXTENDED_TYPES 1
+#else
+#define HAS_EXTENDED_TYPES 0
+#endif
+#define BEGIN_PUSH_CONSTANTS
+#define END_PUSH_CONSTANTS
+#define UNIFORM(n) layout (location = n) uniform
+#define BINDING_INPUT_BUFFER 0
+#define BINDING_OUTPUT_IMAGE 0
+
+#endif
+
+BEGIN_PUSH_CONSTANTS
+UNIFORM(0) uvec2 origin;
+UNIFORM(1) ivec2 destination;
+UNIFORM(2) uint bytes_per_block;
+UNIFORM(3) uint pitch;
+END_PUSH_CONSTANTS
+
+#if HAS_EXTENDED_TYPES
+layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU8 { uint8_t u8data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU16 { uint16_t u16data[]; };
+#endif
+layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU32 { uint u32data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU64 { uvec2 u64data[]; };
+layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU128 { uvec4 u128data[]; };
+
+layout(binding = BINDING_OUTPUT_IMAGE) writeonly uniform uimage2D output_image;
+
+layout(local_size_x = 32, local_size_y = 32, local_size_z = 1) in;
+
+uvec4 ReadTexel(uint offset) {
+    switch (bytes_per_block) {
+#if HAS_EXTENDED_TYPES
+    case 1:
+        return uvec4(u8data[offset], 0, 0, 0);
+    case 2:
+        return uvec4(u16data[offset / 2], 0, 0, 0);
+#else
+    case 1:
+        return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 24), 8), 0, 0, 0);
+    case 2:
+        return uvec4(bitfieldExtract(u32data[offset / 4], int((offset * 8) & 16), 16), 0, 0, 0);
+#endif
+    case 4:
+        return uvec4(u32data[offset / 4], 0, 0, 0);
+    case 8:
+        return uvec4(u64data[offset / 8], 0, 0);
+    case 16:
+        return u128data[offset / 16];
+    }
+    return uvec4(0);
+}
+
+void main() {
+    uvec2 pos = gl_GlobalInvocationID.xy + origin;
+
+    uint offset = 0;
+    offset += pos.x * bytes_per_block;
+    offset += pos.y * pitch;
+
+    const uvec4 texel = ReadTexel(offset);
+    const ivec2 coord = ivec2(gl_GlobalInvocationID.xy) + destination;
+    imageStore(output_image, coord, texel);
+}

src/video_core/host_shaders/vulkan_blit_color_float.frag

@@ -0,0 +1,14 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 450
+
+layout(binding = 0) uniform sampler2D tex;
+
+layout(location = 0) in vec2 texcoord;
+layout(location = 0) out vec4 color;
+
+void main() {
+    color = textureLod(tex, texcoord, 0);
+}

src/video_core/host_shaders/vulkan_blit_depth_stencil.frag

@@ -0,0 +1,16 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#version 450
+#extension GL_ARB_shader_stencil_export : require
+
+layout(binding = 0) uniform sampler2D depth_tex;
+layout(binding = 1) uniform isampler2D stencil_tex;
+
+layout(location = 0) in vec2 texcoord;
+
+void main() {
+    gl_FragDepth = textureLod(depth_tex, texcoord, 0).r;
+    gl_FragStencilRefARB = textureLod(stencil_tex, texcoord, 0).r;
+}

src/video_core/host_shaders/vulkan_present.frag

@@ -2,15 +2,6 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-/*
- * Build instructions:
- * $ glslangValidator -V $THIS_FILE -o output.spv
- * $ spirv-opt -O --strip-debug output.spv -o optimized.spv
- * $ xxd -i optimized.spv
- *
- * Then copy that bytecode to the C++ file
- */
-
 #version 460 core
 
 layout (location = 0) in vec2 frag_tex_coord;

src/video_core/host_shaders/vulkan_present.vert

@@ -2,15 +2,6 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-/*
- * Build instructions:
- * $ glslangValidator -V $THIS_FILE -o output.spv
- * $ spirv-opt -O --strip-debug output.spv -o optimized.spv
- * $ xxd -i optimized.spv
- *
- * Then copy that bytecode to the C++ file
- */
-
 #version 460 core
 
 layout (location = 0) in vec2 vert_position;

src/video_core/host_shaders/vulkan_quad_array.comp

@@ -2,15 +2,6 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-/*
- * Build instructions:
- * $ glslangValidator -V $THIS_FILE -o output.spv
- * $ spirv-opt -O --strip-debug output.spv -o optimized.spv
- * $ xxd -i optimized.spv
- *
- * Then copy that bytecode to the C++ file
- */
-
 #version 460 core
 
 layout (local_size_x = 1024) in;

src/video_core/host_shaders/vulkan_quad_indexed.comp

@@ -2,15 +2,6 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-/*
- * Build instructions:
- * $ glslangValidator -V quad_indexed.comp -o output.spv
- * $ spirv-opt -O --strip-debug output.spv -o optimized.spv
- * $ xxd -i optimized.spv
- *
- * Then copy that bytecode to the C++ file
- */
-
 #version 460 core
 
 layout (local_size_x = 1024) in;

src/video_core/host_shaders/vulkan_uint8.comp

@@ -2,15 +2,6 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-/*
- * Build instructions:
- * $ glslangValidator -V $THIS_FILE -o output.spv
- * $ spirv-opt -O --strip-debug output.spv -o optimized.spv
- * $ xxd -i optimized.spv
- *
- * Then copy that bytecode to the C++ file
- */
-
 #version 460 core
 #extension GL_EXT_shader_16bit_storage : require
 #extension GL_EXT_shader_8bit_storage : require

src/video_core/memory_manager.cpp

@@ -57,7 +57,10 @@ void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) {
     }
 
     // Flush and invalidate through the GPU interface, to be asynchronous if possible.
-    system.GPU().FlushAndInvalidateRegion(*GpuToCpuAddress(gpu_addr), size);
+    const std::optional<VAddr> cpu_addr = GpuToCpuAddress(gpu_addr);
+    ASSERT(cpu_addr);
+
+    rasterizer->UnmapMemory(*cpu_addr, size);
 
     UpdateRange(gpu_addr, PageEntry::State::Unmapped, size);
 }

src/video_core/morton.cpp

@@ -1,250 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <array>
-#include <cstring>
-#include "common/assert.h"
-#include "common/common_types.h"
-#include "video_core/morton.h"
-#include "video_core/surface.h"
-#include "video_core/textures/decoders.h"
-
-namespace VideoCore {
-
-using Surface::GetBytesPerPixel;
-using Surface::PixelFormat;
-
-using MortonCopyFn = void (*)(u32, u32, u32, u32, u32, u32, u8*, u8*);
-using ConversionArray = std::array<MortonCopyFn, Surface::MaxPixelFormat>;
-
-template <bool morton_to_linear, PixelFormat format>
-static void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth, u32 depth,
-                       u32 tile_width_spacing, u8* buffer, u8* addr) {
-    constexpr u32 bytes_per_pixel = GetBytesPerPixel(format);
-
-    // With the BCn formats (DXT and DXN), each 4x4 tile is swizzled instead of just individual
-    // pixel values.
-    constexpr u32 tile_size_x{GetDefaultBlockWidth(format)};
-    constexpr u32 tile_size_y{GetDefaultBlockHeight(format)};
-
-    if constexpr (morton_to_linear) {
-        Tegra::Texture::UnswizzleTexture(buffer, addr, tile_size_x, tile_size_y, bytes_per_pixel,
-                                         stride, height, depth, block_height, block_depth,
-                                         tile_width_spacing);
-    } else {
-        Tegra::Texture::CopySwizzledData((stride + tile_size_x - 1) / tile_size_x,
-                                         (height + tile_size_y - 1) / tile_size_y, depth,
-                                         bytes_per_pixel, bytes_per_pixel, addr, buffer, false,
-                                         block_height, block_depth, tile_width_spacing);
-    }
-}
-
-static constexpr ConversionArray morton_to_linear_fns = {
-    MortonCopy<true, PixelFormat::A8B8G8R8_UNORM>,
-    MortonCopy<true, PixelFormat::A8B8G8R8_SNORM>,
-    MortonCopy<true, PixelFormat::A8B8G8R8_SINT>,
-    MortonCopy<true, PixelFormat::A8B8G8R8_UINT>,
-    MortonCopy<true, PixelFormat::R5G6B5_UNORM>,
-    MortonCopy<true, PixelFormat::B5G6R5_UNORM>,
-    MortonCopy<true, PixelFormat::A1R5G5B5_UNORM>,
-    MortonCopy<true, PixelFormat::A2B10G10R10_UNORM>,
-    MortonCopy<true, PixelFormat::A2B10G10R10_UINT>,
-    MortonCopy<true, PixelFormat::A1B5G5R5_UNORM>,
-    MortonCopy<true, PixelFormat::R8_UNORM>,
-    MortonCopy<true, PixelFormat::R8_SNORM>,
-    MortonCopy<true, PixelFormat::R8_SINT>,
-    MortonCopy<true, PixelFormat::R8_UINT>,
-    MortonCopy<true, PixelFormat::R16G16B16A16_FLOAT>,
-    MortonCopy<true, PixelFormat::R16G16B16A16_UNORM>,
-    MortonCopy<true, PixelFormat::R16G16B16A16_SNORM>,
-    MortonCopy<true, PixelFormat::R16G16B16A16_SINT>,
-    MortonCopy<true, PixelFormat::R16G16B16A16_UINT>,
-    MortonCopy<true, PixelFormat::B10G11R11_FLOAT>,
-    MortonCopy<true, PixelFormat::R32G32B32A32_UINT>,
-    MortonCopy<true, PixelFormat::BC1_RGBA_UNORM>,
-    MortonCopy<true, PixelFormat::BC2_UNORM>,
-    MortonCopy<true, PixelFormat::BC3_UNORM>,
-    MortonCopy<true, PixelFormat::BC4_UNORM>,
-    MortonCopy<true, PixelFormat::BC4_SNORM>,
-    MortonCopy<true, PixelFormat::BC5_UNORM>,
-    MortonCopy<true, PixelFormat::BC5_SNORM>,
-    MortonCopy<true, PixelFormat::BC7_UNORM>,
-    MortonCopy<true, PixelFormat::BC6H_UFLOAT>,
-    MortonCopy<true, PixelFormat::BC6H_SFLOAT>,
-    MortonCopy<true, PixelFormat::ASTC_2D_4X4_UNORM>,
-    MortonCopy<true, PixelFormat::B8G8R8A8_UNORM>,
-    MortonCopy<true, PixelFormat::R32G32B32A32_FLOAT>,
-    MortonCopy<true, PixelFormat::R32G32B32A32_SINT>,
-    MortonCopy<true, PixelFormat::R32G32_FLOAT>,
-    MortonCopy<true, PixelFormat::R32G32_SINT>,
-    MortonCopy<true, PixelFormat::R32_FLOAT>,
-    MortonCopy<true, PixelFormat::R16_FLOAT>,
-    MortonCopy<true, PixelFormat::R16_UNORM>,
-    MortonCopy<true, PixelFormat::R16_SNORM>,
-    MortonCopy<true, PixelFormat::R16_UINT>,
-    MortonCopy<true, PixelFormat::R16_SINT>,
-    MortonCopy<true, PixelFormat::R16G16_UNORM>,
-    MortonCopy<true, PixelFormat::R16G16_FLOAT>,
-    MortonCopy<true, PixelFormat::R16G16_UINT>,
-    MortonCopy<true, PixelFormat::R16G16_SINT>,
-    MortonCopy<true, PixelFormat::R16G16_SNORM>,
-    MortonCopy<true, PixelFormat::R32G32B32_FLOAT>,
-    MortonCopy<true, PixelFormat::A8B8G8R8_SRGB>,
-    MortonCopy<true, PixelFormat::R8G8_UNORM>,
-    MortonCopy<true, PixelFormat::R8G8_SNORM>,
-    MortonCopy<true, PixelFormat::R8G8_SINT>,
-    MortonCopy<true, PixelFormat::R8G8_UINT>,
-    MortonCopy<true, PixelFormat::R32G32_UINT>,
-    MortonCopy<true, PixelFormat::R16G16B16X16_FLOAT>,
-    MortonCopy<true, PixelFormat::R32_UINT>,
-    MortonCopy<true, PixelFormat::R32_SINT>,
-    MortonCopy<true, PixelFormat::ASTC_2D_8X8_UNORM>,
-    MortonCopy<true, PixelFormat::ASTC_2D_8X5_UNORM>,
-    MortonCopy<true, PixelFormat::ASTC_2D_5X4_UNORM>,
-    MortonCopy<true, PixelFormat::B8G8R8A8_SRGB>,
-    MortonCopy<true, PixelFormat::BC1_RGBA_SRGB>,
-    MortonCopy<true, PixelFormat::BC2_SRGB>,
-    MortonCopy<true, PixelFormat::BC3_SRGB>,
-    MortonCopy<true, PixelFormat::BC7_SRGB>,
-    MortonCopy<true, PixelFormat::A4B4G4R4_UNORM>,
-    MortonCopy<true, PixelFormat::ASTC_2D_4X4_SRGB>,
-    MortonCopy<true, PixelFormat::ASTC_2D_8X8_SRGB>,
-    MortonCopy<true, PixelFormat::ASTC_2D_8X5_SRGB>,
-    MortonCopy<true, PixelFormat::ASTC_2D_5X4_SRGB>,
-    MortonCopy<true, PixelFormat::ASTC_2D_5X5_UNORM>,
-    MortonCopy<true, PixelFormat::ASTC_2D_5X5_SRGB>,
-    MortonCopy<true, PixelFormat::ASTC_2D_10X8_UNORM>,
-    MortonCopy<true, PixelFormat::ASTC_2D_10X8_SRGB>,
-    MortonCopy<true, PixelFormat::ASTC_2D_6X6_UNORM>,
-    MortonCopy<true, PixelFormat::ASTC_2D_6X6_SRGB>,
-    MortonCopy<true, PixelFormat::ASTC_2D_10X10_UNORM>,
-    MortonCopy<true, PixelFormat::ASTC_2D_10X10_SRGB>,
-    MortonCopy<true, PixelFormat::ASTC_2D_12X12_UNORM>,
-    MortonCopy<true, PixelFormat::ASTC_2D_12X12_SRGB>,
-    MortonCopy<true, PixelFormat::ASTC_2D_8X6_UNORM>,
-    MortonCopy<true, PixelFormat::ASTC_2D_8X6_SRGB>,
-    MortonCopy<true, PixelFormat::ASTC_2D_6X5_UNORM>,
-    MortonCopy<true, PixelFormat::ASTC_2D_6X5_SRGB>,
-    MortonCopy<true, PixelFormat::E5B9G9R9_FLOAT>,
-    MortonCopy<true, PixelFormat::D32_FLOAT>,
-    MortonCopy<true, PixelFormat::D16_UNORM>,
-    MortonCopy<true, PixelFormat::D24_UNORM_S8_UINT>,
-    MortonCopy<true, PixelFormat::S8_UINT_D24_UNORM>,
-    MortonCopy<true, PixelFormat::D32_FLOAT_S8_UINT>,
-};
-
-static constexpr ConversionArray linear_to_morton_fns = {
-    MortonCopy<false, PixelFormat::A8B8G8R8_UNORM>,
-    MortonCopy<false, PixelFormat::A8B8G8R8_SNORM>,
-    MortonCopy<false, PixelFormat::A8B8G8R8_SINT>,
-    MortonCopy<false, PixelFormat::A8B8G8R8_UINT>,
-    MortonCopy<false, PixelFormat::R5G6B5_UNORM>,
-    MortonCopy<false, PixelFormat::B5G6R5_UNORM>,
-    MortonCopy<false, PixelFormat::A1R5G5B5_UNORM>,
-    MortonCopy<false, PixelFormat::A2B10G10R10_UNORM>,
-    MortonCopy<false, PixelFormat::A2B10G10R10_UINT>,
-    MortonCopy<false, PixelFormat::A1B5G5R5_UNORM>,
-    MortonCopy<false, PixelFormat::R8_UNORM>,
-    MortonCopy<false, PixelFormat::R8_SNORM>,
-    MortonCopy<false, PixelFormat::R8_SINT>,
-    MortonCopy<false, PixelFormat::R8_UINT>,
-    MortonCopy<false, PixelFormat::R16G16B16A16_FLOAT>,
-    MortonCopy<false, PixelFormat::R16G16B16A16_SNORM>,
-    MortonCopy<false, PixelFormat::R16G16B16A16_SINT>,
-    MortonCopy<false, PixelFormat::R16G16B16A16_UNORM>,
-    MortonCopy<false, PixelFormat::R16G16B16A16_UINT>,
-    MortonCopy<false, PixelFormat::B10G11R11_FLOAT>,
-    MortonCopy<false, PixelFormat::R32G32B32A32_UINT>,
-    MortonCopy<false, PixelFormat::BC1_RGBA_UNORM>,
-    MortonCopy<false, PixelFormat::BC2_UNORM>,
-    MortonCopy<false, PixelFormat::BC3_UNORM>,
-    MortonCopy<false, PixelFormat::BC4_UNORM>,
-    MortonCopy<false, PixelFormat::BC4_SNORM>,
-    MortonCopy<false, PixelFormat::BC5_UNORM>,
-    MortonCopy<false, PixelFormat::BC5_SNORM>,
-    MortonCopy<false, PixelFormat::BC7_UNORM>,
-    MortonCopy<false, PixelFormat::BC6H_UFLOAT>,
-    MortonCopy<false, PixelFormat::BC6H_SFLOAT>,
-    // TODO(Subv): Swizzling ASTC formats are not supported
-    nullptr,
-    MortonCopy<false, PixelFormat::B8G8R8A8_UNORM>,
-    MortonCopy<false, PixelFormat::R32G32B32A32_FLOAT>,
-    MortonCopy<false, PixelFormat::R32G32B32A32_SINT>,
-    MortonCopy<false, PixelFormat::R32G32_FLOAT>,
-    MortonCopy<false, PixelFormat::R32G32_SINT>,
-    MortonCopy<false, PixelFormat::R32_FLOAT>,
-    MortonCopy<false, PixelFormat::R16_FLOAT>,
-    MortonCopy<false, PixelFormat::R16_UNORM>,
-    MortonCopy<false, PixelFormat::R16_SNORM>,
-    MortonCopy<false, PixelFormat::R16_UINT>,
-    MortonCopy<false, PixelFormat::R16_SINT>,
-    MortonCopy<false, PixelFormat::R16G16_UNORM>,
-    MortonCopy<false, PixelFormat::R16G16_FLOAT>,
-    MortonCopy<false, PixelFormat::R16G16_UINT>,
-    MortonCopy<false, PixelFormat::R16G16_SINT>,
-    MortonCopy<false, PixelFormat::R16G16_SNORM>,
-    MortonCopy<false, PixelFormat::R32G32B32_FLOAT>,
-    MortonCopy<false, PixelFormat::A8B8G8R8_SRGB>,
-    MortonCopy<false, PixelFormat::R8G8_UNORM>,
-    MortonCopy<false, PixelFormat::R8G8_SNORM>,
-    MortonCopy<false, PixelFormat::R8G8_SINT>,
-    MortonCopy<false, PixelFormat::R8G8_UINT>,
-    MortonCopy<false, PixelFormat::R32G32_UINT>,
-    MortonCopy<false, PixelFormat::R16G16B16X16_FLOAT>,
-    MortonCopy<false, PixelFormat::R32_UINT>,
-    MortonCopy<false, PixelFormat::R32_SINT>,
-    nullptr,
-    nullptr,
-    nullptr,
-    MortonCopy<false, PixelFormat::B8G8R8A8_SRGB>,
-    MortonCopy<false, PixelFormat::BC1_RGBA_SRGB>,
-    MortonCopy<false, PixelFormat::BC2_SRGB>,
-    MortonCopy<false, PixelFormat::BC3_SRGB>,
-    MortonCopy<false, PixelFormat::BC7_SRGB>,
-    MortonCopy<false, PixelFormat::A4B4G4R4_UNORM>,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    nullptr,
-    MortonCopy<false, PixelFormat::E5B9G9R9_FLOAT>,
-    MortonCopy<false, PixelFormat::D32_FLOAT>,
-    MortonCopy<false, PixelFormat::D16_UNORM>,
-    MortonCopy<false, PixelFormat::D24_UNORM_S8_UINT>,
-    MortonCopy<false, PixelFormat::S8_UINT_D24_UNORM>,
-    MortonCopy<false, PixelFormat::D32_FLOAT_S8_UINT>,
-};
-
-static MortonCopyFn GetSwizzleFunction(MortonSwizzleMode mode, Surface::PixelFormat format) {
-    switch (mode) {
-    case MortonSwizzleMode::MortonToLinear:
-        return morton_to_linear_fns[static_cast<std::size_t>(format)];
-    case MortonSwizzleMode::LinearToMorton:
-        return linear_to_morton_fns[static_cast<std::size_t>(format)];
-    }
-    UNREACHABLE();
-    return morton_to_linear_fns[static_cast<std::size_t>(format)];
-}
-
-void MortonSwizzle(MortonSwizzleMode mode, Surface::PixelFormat format, u32 stride,
-                   u32 block_height, u32 height, u32 block_depth, u32 depth, u32 tile_width_spacing,
-                   u8* buffer, u8* addr) {
-    GetSwizzleFunction(mode, format)(stride, block_height, height, block_depth, depth,
-                                     tile_width_spacing, buffer, addr);
-}
-
-} // namespace VideoCore

src/video_core/morton.h

@@ -1,18 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include "common/common_types.h"
-#include "video_core/surface.h"
-
-namespace VideoCore {
-
-enum class MortonSwizzleMode { MortonToLinear, LinearToMorton };
-
-void MortonSwizzle(MortonSwizzleMode mode, VideoCore::Surface::PixelFormat format, u32 stride,
-                   u32 block_height, u32 height, u32 block_depth, u32 depth, u32 tile_width_spacing,
-                   u8* buffer, u8* addr);
-
-} // namespace VideoCore

src/video_core/rasterizer_interface.h

@@ -76,6 +76,9 @@ public:
     /// Sync memory between guest and host.
     virtual void SyncGuestHost() = 0;
 
+    /// Unmap memory range
+    virtual void UnmapMemory(VAddr addr, u64 size) = 0;
+
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
     /// and invalidated
     virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0;
@@ -83,6 +86,12 @@ public:
     /// Notify the host renderer to wait for previous primitive and compute operations.
     virtual void WaitForIdle() = 0;
 
+    /// Notify the host renderer to wait for reads and writes to render targets and flush caches.
+    virtual void FragmentBarrier() = 0;
+
+    /// Notify the host renderer to make available previous render target writes.
+    virtual void TiledCacheBarrier() = 0;
+
     /// Notify the rasterizer to send all written commands to the host GPU.
     virtual void FlushCommands() = 0;
 
@@ -91,8 +100,7 @@ public:
 
     /// Attempt to use a faster method to perform a surface copy
     [[nodiscard]] virtual bool AccelerateSurfaceCopy(
-        const Tegra::Engines::Fermi2D::Regs::Surface& src,
+        const Tegra::Engines::Fermi2D::Surface& src, const Tegra::Engines::Fermi2D::Surface& dst,
-        const Tegra::Engines::Fermi2D::Regs::Surface& dst,
         const Tegra::Engines::Fermi2D::Config& copy_config) {
         return false;
     }

src/video_core/renderer_opengl/gl_buffer_cache.cpp

@@ -61,10 +61,9 @@ void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst
 
 OGLBufferCache::OGLBufferCache(VideoCore::RasterizerInterface& rasterizer_,
                                Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_,
-                               const Device& device_, std::size_t stream_size_)
+                               const Device& device_, OGLStreamBuffer& stream_buffer_,
-    : GenericBufferCache{rasterizer_, gpu_memory_, cpu_memory_,
+                               StateTracker& state_tracker)
-                         std::make_unique<OGLStreamBuffer>(device_, stream_size_, true)},
+    : GenericBufferCache{rasterizer_, gpu_memory_, cpu_memory_, stream_buffer_}, device{device_} {
-      device{device_} {
     if (!device.HasFastBufferSubData()) {
         return;
     }

src/video_core/renderer_opengl/gl_buffer_cache.h

@@ -22,6 +22,7 @@ namespace OpenGL {
 class Device;
 class OGLStreamBuffer;
 class RasterizerOpenGL;
+class StateTracker;
 
 class Buffer : public VideoCommon::BufferBlock {
 public:
@@ -52,9 +53,10 @@ private:
 using GenericBufferCache = VideoCommon::BufferCache<Buffer, GLuint, OGLStreamBuffer>;
 class OGLBufferCache final : public GenericBufferCache {
 public:
-    explicit OGLBufferCache(VideoCore::RasterizerInterface& rasterizer_,
+    explicit OGLBufferCache(VideoCore::RasterizerInterface& rasterizer,
-                            Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_,
+                            Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory,
-                            const Device& device_, std::size_t stream_size_);
+                            const Device& device, OGLStreamBuffer& stream_buffer,
+                            StateTracker& state_tracker);
     ~OGLBufferCache();
 
     BufferInfo GetEmptyBuffer(std::size_t) override;

src/video_core/renderer_opengl/gl_device.cpp

@@ -5,9 +5,11 @@
 #include <algorithm>
 #include <array>
 #include <cstddef>
+#include <cstdlib>
 #include <cstring>
 #include <limits>
 #include <optional>
+#include <span>
 #include <vector>
 
 #include <glad/glad.h>
@@ -27,27 +29,29 @@ constexpr u32 ReservedUniformBlocks = 1;
 
 constexpr u32 NumStages = 5;
 
-constexpr std::array LimitUBOs = {
+constexpr std::array LIMIT_UBOS = {
     GL_MAX_VERTEX_UNIFORM_BLOCKS,          GL_MAX_TESS_CONTROL_UNIFORM_BLOCKS,
     GL_MAX_TESS_EVALUATION_UNIFORM_BLOCKS, GL_MAX_GEOMETRY_UNIFORM_BLOCKS,
-    GL_MAX_FRAGMENT_UNIFORM_BLOCKS,        GL_MAX_COMPUTE_UNIFORM_BLOCKS};
+    GL_MAX_FRAGMENT_UNIFORM_BLOCKS,        GL_MAX_COMPUTE_UNIFORM_BLOCKS,
-
+};
-constexpr std::array LimitSSBOs = {
+constexpr std::array LIMIT_SSBOS = {
     GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS,          GL_MAX_TESS_CONTROL_SHADER_STORAGE_BLOCKS,
     GL_MAX_TESS_EVALUATION_SHADER_STORAGE_BLOCKS, GL_MAX_GEOMETRY_SHADER_STORAGE_BLOCKS,
-    GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS,        GL_MAX_COMPUTE_SHADER_STORAGE_BLOCKS};
+    GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS,        GL_MAX_COMPUTE_SHADER_STORAGE_BLOCKS,
-
+};
-constexpr std::array LimitSamplers = {GL_MAX_VERTEX_TEXTURE_IMAGE_UNITS,
+constexpr std::array LIMIT_SAMPLERS = {
-                                      GL_MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS,
+    GL_MAX_VERTEX_TEXTURE_IMAGE_UNITS,
-                                      GL_MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS,
+    GL_MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS,
-                                      GL_MAX_GEOMETRY_TEXTURE_IMAGE_UNITS,
+    GL_MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS,
-                                      GL_MAX_TEXTURE_IMAGE_UNITS,
+    GL_MAX_GEOMETRY_TEXTURE_IMAGE_UNITS,
-                                      GL_MAX_COMPUTE_TEXTURE_IMAGE_UNITS};
+    GL_MAX_TEXTURE_IMAGE_UNITS,
-
+    GL_MAX_COMPUTE_TEXTURE_IMAGE_UNITS,
-constexpr std::array LimitImages = {
+};
+constexpr std::array LIMIT_IMAGES = {
     GL_MAX_VERTEX_IMAGE_UNIFORMS,          GL_MAX_TESS_CONTROL_IMAGE_UNIFORMS,
     GL_MAX_TESS_EVALUATION_IMAGE_UNIFORMS, GL_MAX_GEOMETRY_IMAGE_UNIFORMS,
-    GL_MAX_FRAGMENT_IMAGE_UNIFORMS,        GL_MAX_COMPUTE_IMAGE_UNIFORMS};
+    GL_MAX_FRAGMENT_IMAGE_UNIFORMS,        GL_MAX_COMPUTE_IMAGE_UNIFORMS,
+};
 
 template <typename T>
 T GetInteger(GLenum pname) {
@@ -76,8 +80,8 @@ std::vector<std::string_view> GetExtensions() {
     return extensions;
 }
 
-bool HasExtension(const std::vector<std::string_view>& images, std::string_view extension) {
+bool HasExtension(std::span<const std::string_view> extensions, std::string_view extension) {
-    return std::find(images.begin(), images.end(), extension) != images.end();
+    return std::ranges::find(extensions, extension) != extensions.end();
 }
 
 u32 Extract(u32& base, u32& num, u32 amount, std::optional<GLenum> limit = {}) {
@@ -91,8 +95,8 @@ u32 Extract(u32& base, u32& num, u32 amount, std::optional<GLenum> limit = {}) {
 
 std::array<u32, Tegra::Engines::MaxShaderTypes> BuildMaxUniformBuffers() noexcept {
     std::array<u32, Tegra::Engines::MaxShaderTypes> max;
-    std::transform(LimitUBOs.begin(), LimitUBOs.end(), max.begin(),
+    std::ranges::transform(LIMIT_UBOS, max.begin(),
-                   [](GLenum pname) { return GetInteger<u32>(pname); });
+                           [](GLenum pname) { return GetInteger<u32>(pname); });
     return max;
 }
 
@@ -115,9 +119,10 @@ std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> BuildBaseBindin
     for (std::size_t i = 0; i < NumStages; ++i) {
         const std::size_t stage = stage_swizzle[i];
         bindings[stage] = {
-            Extract(base_ubo, num_ubos, total_ubos / NumStages, LimitUBOs[stage]),
+            Extract(base_ubo, num_ubos, total_ubos / NumStages, LIMIT_UBOS[stage]),
-            Extract(base_ssbo, num_ssbos, total_ssbos / NumStages, LimitSSBOs[stage]),
+            Extract(base_ssbo, num_ssbos, total_ssbos / NumStages, LIMIT_SSBOS[stage]),
-            Extract(base_samplers, num_samplers, total_samplers / NumStages, LimitSamplers[stage])};
+            Extract(base_samplers, num_samplers, total_samplers / NumStages,
+                    LIMIT_SAMPLERS[stage])};
     }
 
     u32 num_images = GetInteger<u32>(GL_MAX_IMAGE_UNITS);
@@ -130,7 +135,7 @@ std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> BuildBaseBindin
 
     // Reserve at least 4 image bindings on the fragment stage.
     bindings[4].image =
-        Extract(base_images, num_images, std::max(4U, num_images / NumStages), LimitImages[4]);
+        Extract(base_images, num_images, std::max(4U, num_images / NumStages), LIMIT_IMAGES[4]);
 
     // This is guaranteed to be at least 1.
     const u32 total_extracted_images = num_images / (NumStages - 1);
@@ -142,7 +147,7 @@ std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> BuildBaseBindin
             continue;
         }
         bindings[stage].image =
-            Extract(base_images, num_images, total_extracted_images, LimitImages[stage]);
+            Extract(base_images, num_images, total_extracted_images, LIMIT_IMAGES[stage]);
     }
 
     // Compute doesn't care about any of this.
@@ -188,6 +193,11 @@ bool IsASTCSupported() {
     return true;
 }
 
+[[nodiscard]] bool IsDebugToolAttached(std::span<const std::string_view> extensions) {
+    const bool nsight = std::getenv("NVTX_INJECTION64_PATH") || std::getenv("NSIGHT_LAUNCHED");
+    return nsight || HasExtension(extensions, "GL_EXT_debug_tool");
+}
+
 } // Anonymous namespace
 
 Device::Device()
@@ -206,9 +216,8 @@ Device::Device()
             "Beta driver 443.24 is known to have issues. There might be performance issues.");
         disable_fast_buffer_sub_data = true;
     }
-
+    uniform_buffer_alignment = GetInteger<size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT);
-    uniform_buffer_alignment = GetInteger<std::size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT);
+    shader_storage_alignment = GetInteger<size_t>(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT);
-    shader_storage_alignment = GetInteger<std::size_t>(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT);
     max_vertex_attributes = GetInteger<u32>(GL_MAX_VERTEX_ATTRIBS);
     max_varyings = GetInteger<u32>(GL_MAX_VARYING_VECTORS);
     max_compute_shared_memory_size = GetInteger<u32>(GL_MAX_COMPUTE_SHARED_MEMORY_SIZE);
@@ -224,6 +233,7 @@ Device::Device()
     has_precise_bug = TestPreciseBug();
     has_nv_viewport_array2 = GLAD_GL_NV_viewport_array2;
     has_vertex_buffer_unified_memory = GLAD_GL_NV_vertex_buffer_unified_memory;
+    has_debugging_tool_attached = IsDebugToolAttached(extensions);
 
     // At the moment of writing this, only Nvidia's driver optimizes BufferSubData on exclusive
     // uniform buffers as "push constants"

src/video_core/renderer_opengl/gl_device.h

@@ -36,11 +36,11 @@ public:
         return GetBaseBindings(static_cast<std::size_t>(shader_type));
     }
 
-    std::size_t GetUniformBufferAlignment() const {
+    size_t GetUniformBufferAlignment() const {
         return uniform_buffer_alignment;
     }
 
-    std::size_t GetShaderStorageBufferAlignment() const {
+    size_t GetShaderStorageBufferAlignment() const {
         return shader_storage_alignment;
     }
 
@@ -104,6 +104,10 @@ public:
         return has_nv_viewport_array2;
     }
 
+    bool HasDebuggingToolAttached() const {
+        return has_debugging_tool_attached;
+    }
+
     bool UseAssemblyShaders() const {
         return use_assembly_shaders;
     }
@@ -118,8 +122,8 @@ private:
 
     std::array<u32, Tegra::Engines::MaxShaderTypes> max_uniform_buffers{};
     std::array<BaseBindings, Tegra::Engines::MaxShaderTypes> base_bindings{};
-    std::size_t uniform_buffer_alignment{};
+    size_t uniform_buffer_alignment{};
-    std::size_t shader_storage_alignment{};
+    size_t shader_storage_alignment{};
     u32 max_vertex_attributes{};
     u32 max_varyings{};
     u32 max_compute_shared_memory_size{};
@@ -135,6 +139,7 @@ private:
     bool has_precise_bug{};
     bool has_fast_buffer_sub_data{};
     bool has_nv_viewport_array2{};
+    bool has_debugging_tool_attached{};
     bool use_assembly_shaders{};
     bool use_asynchronous_shaders{};
 };

src/video_core/renderer_opengl/gl_fence_manager.cpp

@@ -46,7 +46,7 @@ void GLInnerFence::Wait() {
 }
 
 FenceManagerOpenGL::FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer_,
-                                       Tegra::GPU& gpu_, TextureCacheOpenGL& texture_cache_,
+                                       Tegra::GPU& gpu_, TextureCache& texture_cache_,
                                        OGLBufferCache& buffer_cache_, QueryCache& query_cache_)
     : GenericFenceManager{rasterizer_, gpu_, texture_cache_, buffer_cache_, query_cache_} {}
 

src/video_core/renderer_opengl/gl_fence_manager.h

@@ -33,12 +33,12 @@ private:
 
 using Fence = std::shared_ptr<GLInnerFence>;
 using GenericFenceManager =
-    VideoCommon::FenceManager<Fence, TextureCacheOpenGL, OGLBufferCache, QueryCache>;
+    VideoCommon::FenceManager<Fence, TextureCache, OGLBufferCache, QueryCache>;
 
 class FenceManagerOpenGL final : public GenericFenceManager {
 public:
     explicit FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_,
-                                TextureCacheOpenGL& texture_cache_, OGLBufferCache& buffer_cache_,
+                                TextureCache& texture_cache_, OGLBufferCache& buffer_cache_,
                                 QueryCache& query_cache_);
 
 protected:

src/video_core/renderer_opengl/gl_framebuffer_cache.cpp

@@ -1,85 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <tuple>
-#include <unordered_map>
-#include <utility>
-
-#include <glad/glad.h>
-
-#include "common/common_types.h"
-#include "video_core/engines/maxwell_3d.h"
-#include "video_core/renderer_opengl/gl_framebuffer_cache.h"
-
-namespace OpenGL {
-
-using Maxwell = Tegra::Engines::Maxwell3D::Regs;
-using VideoCore::Surface::SurfaceType;
-
-FramebufferCacheOpenGL::FramebufferCacheOpenGL() = default;
-
-FramebufferCacheOpenGL::~FramebufferCacheOpenGL() = default;
-
-GLuint FramebufferCacheOpenGL::GetFramebuffer(const FramebufferCacheKey& key) {
-    const auto [entry, is_cache_miss] = cache.try_emplace(key);
-    auto& framebuffer{entry->second};
-    if (is_cache_miss) {
-        framebuffer = CreateFramebuffer(key);
-    }
-    return framebuffer.handle;
-}
-
-OGLFramebuffer FramebufferCacheOpenGL::CreateFramebuffer(const FramebufferCacheKey& key) {
-    OGLFramebuffer framebuffer;
-    framebuffer.Create();
-
-    // TODO(Rodrigo): Use DSA here after Nvidia fixes their framebuffer DSA bugs.
-    glBindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer.handle);
-
-    if (key.zeta) {
-        const bool stencil = key.zeta->GetSurfaceParams().type == SurfaceType::DepthStencil;
-        const GLenum attach_target = stencil ? GL_DEPTH_STENCIL_ATTACHMENT : GL_DEPTH_ATTACHMENT;
-        key.zeta->Attach(attach_target, GL_DRAW_FRAMEBUFFER);
-    }
-
-    std::size_t num_buffers = 0;
-    std::array<GLenum, Maxwell::NumRenderTargets> targets;
-
-    for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) {
-        if (!key.colors[index]) {
-            targets[index] = GL_NONE;
-            continue;
-        }
-        const GLenum attach_target = GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index);
-        key.colors[index]->Attach(attach_target, GL_DRAW_FRAMEBUFFER);
-
-        const u32 attachment = (key.color_attachments >> (BitsPerAttachment * index)) & 0b1111;
-        targets[index] = GL_COLOR_ATTACHMENT0 + attachment;
-        num_buffers = index + 1;
-    }
-
-    if (num_buffers > 0) {
-        glDrawBuffers(static_cast<GLsizei>(num_buffers), std::data(targets));
-    } else {
-        glDrawBuffer(GL_NONE);
-    }
-
-    return framebuffer;
-}
-
-std::size_t FramebufferCacheKey::Hash() const noexcept {
-    std::size_t hash = std::hash<View>{}(zeta);
-    for (const auto& color : colors) {
-        hash ^= std::hash<View>{}(color);
-    }
-    hash ^= static_cast<std::size_t>(color_attachments) << 16;
-    return hash;
-}
-
-bool FramebufferCacheKey::operator==(const FramebufferCacheKey& rhs) const noexcept {
-    return std::tie(colors, zeta, color_attachments) ==
-           std::tie(rhs.colors, rhs.zeta, rhs.color_attachments);
-}
-
-} // namespace OpenGL

src/video_core/renderer_opengl/gl_framebuffer_cache.h

@@ -1,68 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <array>
-#include <cstddef>
-#include <unordered_map>
-
-#include <glad/glad.h>
-
-#include "common/common_types.h"
-#include "video_core/engines/maxwell_3d.h"
-#include "video_core/renderer_opengl/gl_resource_manager.h"
-#include "video_core/renderer_opengl/gl_texture_cache.h"
-
-namespace OpenGL {
-
-constexpr std::size_t BitsPerAttachment = 4;
-
-struct FramebufferCacheKey {
-    View zeta;
-    std::array<View, Tegra::Engines::Maxwell3D::Regs::NumRenderTargets> colors;
-    u32 color_attachments = 0;
-
-    std::size_t Hash() const noexcept;
-
-    bool operator==(const FramebufferCacheKey& rhs) const noexcept;
-
-    bool operator!=(const FramebufferCacheKey& rhs) const noexcept {
-        return !operator==(rhs);
-    }
-
-    void SetAttachment(std::size_t index, u32 attachment) {
-        color_attachments |= attachment << (BitsPerAttachment * index);
-    }
-};
-
-} // namespace OpenGL
-
-namespace std {
-
-template <>
-struct hash<OpenGL::FramebufferCacheKey> {
-    std::size_t operator()(const OpenGL::FramebufferCacheKey& k) const noexcept {
-        return k.Hash();
-    }
-};
-
-} // namespace std
-
-namespace OpenGL {
-
-class FramebufferCacheOpenGL {
-public:
-    FramebufferCacheOpenGL();
-    ~FramebufferCacheOpenGL();
-
-    GLuint GetFramebuffer(const FramebufferCacheKey& key);
-
-private:
-    OGLFramebuffer CreateFramebuffer(const FramebufferCacheKey& key);
-
-    std::unordered_map<FramebufferCacheKey, OGLFramebuffer> cache;
-};
-
-} // namespace OpenGL

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -25,12 +25,15 @@
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/shader_type.h"
 #include "video_core/memory_manager.h"
+#include "video_core/renderer_opengl/gl_device.h"
 #include "video_core/renderer_opengl/gl_query_cache.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
 #include "video_core/renderer_opengl/gl_shader_cache.h"
+#include "video_core/renderer_opengl/gl_texture_cache.h"
 #include "video_core/renderer_opengl/maxwell_to_gl.h"
 #include "video_core/renderer_opengl/renderer_opengl.h"
 #include "video_core/shader_cache.h"
+#include "video_core/texture_cache/texture_cache.h"
 
 namespace OpenGL {
 
@@ -55,18 +58,32 @@ MICROPROFILE_DEFINE(OpenGL_PrimitiveAssembly, "OpenGL", "Prim Asmbl", MP_RGB(255
 
 namespace {
 
-constexpr std::size_t NUM_CONST_BUFFERS_PER_STAGE = 18;
+constexpr size_t NUM_CONST_BUFFERS_PER_STAGE = 18;
-constexpr std::size_t NUM_CONST_BUFFERS_BYTES_PER_STAGE =
+constexpr size_t NUM_CONST_BUFFERS_BYTES_PER_STAGE =
     NUM_CONST_BUFFERS_PER_STAGE * Maxwell::MaxConstBufferSize;
-constexpr std::size_t TOTAL_CONST_BUFFER_BYTES =
+constexpr size_t TOTAL_CONST_BUFFER_BYTES =
     NUM_CONST_BUFFERS_BYTES_PER_STAGE * Maxwell::MaxShaderStage;
 
-constexpr std::size_t NUM_SUPPORTED_VERTEX_ATTRIBUTES = 16;
+constexpr size_t NUM_SUPPORTED_VERTEX_ATTRIBUTES = 16;
-constexpr std::size_t NUM_SUPPORTED_VERTEX_BINDINGS = 16;
+constexpr size_t NUM_SUPPORTED_VERTEX_BINDINGS = 16;
+
+constexpr size_t MAX_TEXTURES = 192;
+constexpr size_t MAX_IMAGES = 48;
+
+struct TextureHandle {
+    constexpr TextureHandle(u32 data, bool via_header_index) {
+        const Tegra::Texture::TextureHandle handle{data};
+        image = handle.tic_id;
+        sampler = via_header_index ? image : handle.tsc_id.Value();
+    }
+
+    u32 image;
+    u32 sampler;
+};
 
 template <typename Engine, typename Entry>
-Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry& entry,
+TextureHandle GetTextureInfo(const Engine& engine, bool via_header_index, const Entry& entry,
-                                               ShaderType shader_type, std::size_t index = 0) {
+                             ShaderType shader_type, size_t index = 0) {
     if constexpr (std::is_same_v<Entry, SamplerEntry>) {
         if (entry.is_separated) {
             const u32 buffer_1 = entry.buffer;
@@ -75,21 +92,16 @@ Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry
             const u32 offset_2 = entry.secondary_offset;
             const u32 handle_1 = engine.AccessConstBuffer32(shader_type, buffer_1, offset_1);
             const u32 handle_2 = engine.AccessConstBuffer32(shader_type, buffer_2, offset_2);
-            return engine.GetTextureInfo(handle_1 | handle_2);
+            return TextureHandle(handle_1 | handle_2, via_header_index);
         }
     }
     if (entry.is_bindless) {
-        const u32 handle = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset);
+        const u32 raw = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset);
-        return engine.GetTextureInfo(handle);
+        return TextureHandle(raw, via_header_index);
-    }
-
-    const auto& gpu_profile = engine.AccessGuestDriverProfile();
-    const u32 offset = entry.offset + static_cast<u32>(index * gpu_profile.GetTextureHandlerSize());
-    if constexpr (std::is_same_v<Engine, Tegra::Engines::Maxwell3D>) {
-        return engine.GetStageTexture(shader_type, offset);
-    } else {
-        return engine.GetTexture(offset);
     }
+    const u32 buffer = engine.GetBoundBuffer();
+    const u64 offset = (entry.offset + index) * sizeof(u32);
+    return TextureHandle(engine.AccessConstBuffer32(shader_type, buffer, offset), via_header_index);
 }
 
 std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer,
@@ -97,7 +109,6 @@ std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer,
     if (!entry.IsIndirect()) {
         return entry.GetSize();
     }
-
     if (buffer.size > Maxwell::MaxConstBufferSize) {
         LOG_WARNING(Render_OpenGL, "Indirect constbuffer size {} exceeds maximum {}", buffer.size,
                     Maxwell::MaxConstBufferSize);
@@ -147,23 +158,60 @@ void UpdateBindlessSSBOs(GLenum target, const BindlessSSBO* ssbos, size_t num_ss
                                     reinterpret_cast<const GLuint*>(ssbos));
 }
 
+ImageViewType ImageViewTypeFromEntry(const SamplerEntry& entry) {
+    if (entry.is_buffer) {
+        return ImageViewType::Buffer;
+    }
+    switch (entry.type) {
+    case Tegra::Shader::TextureType::Texture1D:
+        return entry.is_array ? ImageViewType::e1DArray : ImageViewType::e1D;
+    case Tegra::Shader::TextureType::Texture2D:
+        return entry.is_array ? ImageViewType::e2DArray : ImageViewType::e2D;
+    case Tegra::Shader::TextureType::Texture3D:
+        return ImageViewType::e3D;
+    case Tegra::Shader::TextureType::TextureCube:
+        return entry.is_array ? ImageViewType::CubeArray : ImageViewType::Cube;
+    }
+    UNREACHABLE();
+    return ImageViewType::e2D;
+}
+
+ImageViewType ImageViewTypeFromEntry(const ImageEntry& entry) {
+    switch (entry.type) {
+    case Tegra::Shader::ImageType::Texture1D:
+        return ImageViewType::e1D;
+    case Tegra::Shader::ImageType::Texture1DArray:
+        return ImageViewType::e1DArray;
+    case Tegra::Shader::ImageType::Texture2D:
+        return ImageViewType::e2D;
+    case Tegra::Shader::ImageType::Texture2DArray:
+        return ImageViewType::e2DArray;
+    case Tegra::Shader::ImageType::Texture3D:
+        return ImageViewType::e3D;
+    case Tegra::Shader::ImageType::TextureBuffer:
+        return ImageViewType::Buffer;
+    }
+    UNREACHABLE();
+    return ImageViewType::e2D;
+}
+
 } // Anonymous namespace
 
 RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
                                    Core::Memory::Memory& cpu_memory_, const Device& device_,
                                    ScreenInfo& screen_info_, ProgramManager& program_manager_,
                                    StateTracker& state_tracker_)
-    : RasterizerAccelerated{cpu_memory_}, gpu(gpu_), maxwell3d(gpu.Maxwell3D()),
+    : RasterizerAccelerated(cpu_memory_), gpu(gpu_), maxwell3d(gpu.Maxwell3D()),
       kepler_compute(gpu.KeplerCompute()), gpu_memory(gpu.MemoryManager()), device(device_),
       screen_info(screen_info_), program_manager(program_manager_), state_tracker(state_tracker_),
-      texture_cache(*this, maxwell3d, gpu_memory, device, state_tracker),
+      stream_buffer(device, state_tracker),
+      texture_cache_runtime(device, program_manager, state_tracker),
+      texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory),
       shader_cache(*this, emu_window_, gpu, maxwell3d, kepler_compute, gpu_memory, device),
       query_cache(*this, maxwell3d, gpu_memory),
-      buffer_cache(*this, gpu_memory, cpu_memory_, device, STREAM_BUFFER_SIZE),
+      buffer_cache(*this, gpu_memory, cpu_memory_, device, stream_buffer, state_tracker),
       fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache),
       async_shaders(emu_window_) {
-    CheckExtensions();
-
     unified_uniform_buffer.Create();
     glNamedBufferStorage(unified_uniform_buffer.handle, TOTAL_CONST_BUFFER_BYTES, nullptr, 0);
 
@@ -174,7 +222,6 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra
                                  nullptr, 0);
         }
     }
-
     if (device.UseAsynchronousShaders()) {
         async_shaders.AllocateWorkers();
     }
@@ -186,14 +233,6 @@ RasterizerOpenGL::~RasterizerOpenGL() {
     }
 }
 
-void RasterizerOpenGL::CheckExtensions() {
-    if (!GLAD_GL_ARB_texture_filter_anisotropic && !GLAD_GL_EXT_texture_filter_anisotropic) {
-        LOG_WARNING(
-            Render_OpenGL,
-            "Anisotropic filter is not supported! This can cause graphical issues in some games.");
-    }
-}
-
 void RasterizerOpenGL::SetupVertexFormat() {
     auto& flags = maxwell3d.dirty.flags;
     if (!flags[Dirty::VertexFormats]) {
@@ -316,10 +355,16 @@ GLintptr RasterizerOpenGL::SetupIndexBuffer() {
     return info.offset;
 }
 
-void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
+void RasterizerOpenGL::SetupShaders() {
     MICROPROFILE_SCOPE(OpenGL_Shader);
     u32 clip_distances = 0;
 
+    std::array<Shader*, Maxwell::MaxShaderStage> shaders{};
+    image_view_indices.clear();
+    sampler_handles.clear();
+
+    texture_cache.SynchronizeGraphicsDescriptors();
+
     for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
         const auto& shader_config = maxwell3d.regs.shader_config[index];
         const auto program{static_cast<Maxwell::ShaderProgram>(index)};
@@ -338,7 +383,6 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
             }
             continue;
         }
-
         // Currently this stages are not supported in the OpenGL backend.
         // TODO(Blinkhawk): Port tesselation shaders from Vulkan to OpenGL
         if (program == Maxwell::ShaderProgram::TesselationControl ||
@@ -347,7 +391,6 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
         }
 
         Shader* const shader = shader_cache.GetStageProgram(program, async_shaders);
-
         const GLuint program_handle = shader->IsBuilt() ? shader->GetHandle() : 0;
         switch (program) {
         case Maxwell::ShaderProgram::VertexA:
@@ -363,14 +406,17 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
         default:
             UNIMPLEMENTED_MSG("Unimplemented shader index={}, enable={}, offset=0x{:08X}", index,
                               shader_config.enable.Value(), shader_config.offset);
+            break;
         }
 
         // Stage indices are 0 - 5
-        const std::size_t stage = index == 0 ? 0 : index - 1;
+        const size_t stage = index == 0 ? 0 : index - 1;
+        shaders[stage] = shader;
+
         SetupDrawConstBuffers(stage, shader);
         SetupDrawGlobalMemory(stage, shader);
-        SetupDrawTextures(stage, shader);
+        SetupDrawTextures(shader, stage);
-        SetupDrawImages(stage, shader);
+        SetupDrawImages(shader, stage);
 
         // Workaround for Intel drivers.
         // When a clip distance is enabled but not set in the shader it crops parts of the screen
@@ -384,9 +430,23 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
             ++index;
         }
     }
-
     SyncClipEnabled(clip_distances);
     maxwell3d.dirty.flags[Dirty::Shaders] = false;
+
+    const std::span indices_span(image_view_indices.data(), image_view_indices.size());
+    texture_cache.FillGraphicsImageViews(indices_span, image_view_ids);
+
+    size_t image_view_index = 0;
+    size_t texture_index = 0;
+    size_t image_index = 0;
+    for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
+        const Shader* const shader = shaders[stage];
+        if (shader) {
+            const auto base = device.GetBaseBindings(stage);
+            BindTextures(shader->GetEntries(), base.sampler, base.image, image_view_index,
+                         texture_index, image_index);
+        }
+    }
 }
 
 std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
@@ -417,98 +477,6 @@ void RasterizerOpenGL::LoadDiskResources(u64 title_id, const std::atomic_bool& s
     shader_cache.LoadDiskCache(title_id, stop_loading, callback);
 }
 
-void RasterizerOpenGL::ConfigureFramebuffers() {
-    MICROPROFILE_SCOPE(OpenGL_Framebuffer);
-    if (!maxwell3d.dirty.flags[VideoCommon::Dirty::RenderTargets]) {
-        return;
-    }
-    maxwell3d.dirty.flags[VideoCommon::Dirty::RenderTargets] = false;
-
-    texture_cache.GuardRenderTargets(true);
-
-    View depth_surface = texture_cache.GetDepthBufferSurface(true);
-
-    const auto& regs = maxwell3d.regs;
-    UNIMPLEMENTED_IF(regs.rt_separate_frag_data == 0);
-
-    // Bind the framebuffer surfaces
-    FramebufferCacheKey key;
-    const auto colors_count = static_cast<std::size_t>(regs.rt_control.count);
-    for (std::size_t index = 0; index < colors_count; ++index) {
-        View color_surface{texture_cache.GetColorBufferSurface(index, true)};
-        if (!color_surface) {
-            continue;
-        }
-        // Assume that a surface will be written to if it is used as a framebuffer, even
-        // if the shader doesn't actually write to it.
-        texture_cache.MarkColorBufferInUse(index);
-
-        key.SetAttachment(index, regs.rt_control.GetMap(index));
-        key.colors[index] = std::move(color_surface);
-    }
-
-    if (depth_surface) {
-        // Assume that a surface will be written to if it is used as a framebuffer, even if
-        // the shader doesn't actually write to it.
-        texture_cache.MarkDepthBufferInUse();
-        key.zeta = std::move(depth_surface);
-    }
-
-    texture_cache.GuardRenderTargets(false);
-
-    glBindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer_cache.GetFramebuffer(key));
-}
-
-void RasterizerOpenGL::ConfigureClearFramebuffer(bool using_color, bool using_depth_stencil) {
-    const auto& regs = maxwell3d.regs;
-
-    texture_cache.GuardRenderTargets(true);
-    View color_surface;
-
-    if (using_color) {
-        // Determine if we have to preserve the contents.
-        // First we have to make sure all clear masks are enabled.
-        bool preserve_contents = !regs.clear_buffers.R || !regs.clear_buffers.G ||
-                                 !regs.clear_buffers.B || !regs.clear_buffers.A;
-        const std::size_t index = regs.clear_buffers.RT;
-        if (regs.clear_flags.scissor) {
-            // Then we have to confirm scissor testing clears the whole image.
-            const auto& scissor = regs.scissor_test[0];
-            preserve_contents |= scissor.min_x > 0;
-            preserve_contents |= scissor.min_y > 0;
-            preserve_contents |= scissor.max_x < regs.rt[index].width;
-            preserve_contents |= scissor.max_y < regs.rt[index].height;
-        }
-
-        color_surface = texture_cache.GetColorBufferSurface(index, preserve_contents);
-        texture_cache.MarkColorBufferInUse(index);
-    }
-
-    View depth_surface;
-    if (using_depth_stencil) {
-        bool preserve_contents = false;
-        if (regs.clear_flags.scissor) {
-            // For depth stencil clears we only have to confirm scissor test covers the whole image.
-            const auto& scissor = regs.scissor_test[0];
-            preserve_contents |= scissor.min_x > 0;
-            preserve_contents |= scissor.min_y > 0;
-            preserve_contents |= scissor.max_x < regs.zeta_width;
-            preserve_contents |= scissor.max_y < regs.zeta_height;
-        }
-
-        depth_surface = texture_cache.GetDepthBufferSurface(preserve_contents);
-        texture_cache.MarkDepthBufferInUse();
-    }
-    texture_cache.GuardRenderTargets(false);
-
-    FramebufferCacheKey key;
-    key.colors[0] = std::move(color_surface);
-    key.zeta = std::move(depth_surface);
-
-    state_tracker.NotifyFramebuffer();
-    glBindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer_cache.GetFramebuffer(key));
-}
-
 void RasterizerOpenGL::Clear() {
     if (!maxwell3d.ShouldExecute()) {
         return;
@@ -523,8 +491,9 @@ void RasterizerOpenGL::Clear() {
         regs.clear_buffers.A) {
         use_color = true;
 
-        state_tracker.NotifyColorMask0();
+        const GLuint index = regs.clear_buffers.RT;
-        glColorMaski(0, regs.clear_buffers.R != 0, regs.clear_buffers.G != 0,
+        state_tracker.NotifyColorMask(index);
+        glColorMaski(index, regs.clear_buffers.R != 0, regs.clear_buffers.G != 0,
                      regs.clear_buffers.B != 0, regs.clear_buffers.A != 0);
 
         // TODO(Rodrigo): Determine if clamping is used on clears
@@ -557,15 +526,17 @@ void RasterizerOpenGL::Clear() {
         state_tracker.NotifyScissor0();
         glDisablei(GL_SCISSOR_TEST, 0);
     }
-
     UNIMPLEMENTED_IF(regs.clear_flags.viewport);
 
-    ConfigureClearFramebuffer(use_color, use_depth || use_stencil);
+    {
-
+        auto lock = texture_cache.AcquireLock();
-    if (use_color) {
+        texture_cache.UpdateRenderTargets(true);
-        glClearBufferfv(GL_COLOR, 0, regs.clear_color);
+        state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle());
     }
 
+    if (use_color) {
+        glClearBufferfv(GL_COLOR, regs.clear_buffers.RT, regs.clear_color);
+    }
     if (use_depth && use_stencil) {
         glClearBufferfi(GL_DEPTH_STENCIL, 0, regs.clear_depth, regs.clear_stencil);
     } else if (use_depth) {
@@ -622,16 +593,7 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
                    (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment());
 
     // Prepare the vertex array.
-    const bool invalidated = buffer_cache.Map(buffer_size);
+    buffer_cache.Map(buffer_size);
-
-    if (invalidated) {
-        // When the stream buffer has been invalidated, we have to consider vertex buffers as dirty
-        auto& dirty = maxwell3d.dirty.flags;
-        dirty[Dirty::VertexBuffers] = true;
-        for (int index = Dirty::VertexBuffer0; index <= Dirty::VertexBuffer31; ++index) {
-            dirty[index] = true;
-        }
-    }
 
     // Prepare vertex array format.
     SetupVertexFormat();
@@ -655,22 +617,16 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
     }
 
     // Setup shaders and their used resources.
-    texture_cache.GuardSamplers(true);
+    auto lock = texture_cache.AcquireLock();
-    const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d.regs.draw.topology);
+    SetupShaders();
-    SetupShaders(primitive_mode);
-    texture_cache.GuardSamplers(false);
-
-    ConfigureFramebuffers();
 
     // Signal the buffer cache that we are not going to upload more things.
     buffer_cache.Unmap();
-
+    texture_cache.UpdateRenderTargets(false);
+    state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle());
     program_manager.BindGraphicsPipeline();
 
-    if (texture_cache.TextureBarrier()) {
+    const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d.regs.draw.topology);
-        glTextureBarrier();
-    }
-
     BeginTransformFeedback(primitive_mode);
 
     const GLuint base_instance = static_cast<GLuint>(maxwell3d.regs.vb_base_instance);
@@ -722,15 +678,13 @@ void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
     buffer_cache.Acquire();
     current_cbuf = 0;
 
-    auto kernel = shader_cache.GetComputeKernel(code_addr);
+    Shader* const kernel = shader_cache.GetComputeKernel(code_addr);
-    program_manager.BindCompute(kernel->GetHandle());
 
-    SetupComputeTextures(kernel);
+    auto lock = texture_cache.AcquireLock();
-    SetupComputeImages(kernel);
+    BindComputeTextures(kernel);
 
-    const std::size_t buffer_size =
+    const size_t buffer_size = Tegra::Engines::KeplerCompute::NumConstBuffers *
-        Tegra::Engines::KeplerCompute::NumConstBuffers *
+                               (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment());
-        (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment());
     buffer_cache.Map(buffer_size);
 
     SetupComputeConstBuffers(kernel);
@@ -739,7 +693,6 @@ void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
     buffer_cache.Unmap();
 
     const auto& launch_desc = kepler_compute.launch_description;
-    program_manager.BindCompute(kernel->GetHandle());
     glDispatchCompute(launch_desc.grid_dim_x, launch_desc.grid_dim_y, launch_desc.grid_dim_z);
     ++num_queued_commands;
 }
@@ -760,7 +713,10 @@ void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) {
     if (addr == 0 || size == 0) {
         return;
     }
-    texture_cache.FlushRegion(addr, size);
+    {
+        auto lock = texture_cache.AcquireLock();
+        texture_cache.DownloadMemory(addr, size);
+    }
     buffer_cache.FlushRegion(addr, size);
     query_cache.FlushRegion(addr, size);
 }
@@ -769,7 +725,8 @@ bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size) {
     if (!Settings::IsGPULevelHigh()) {
         return buffer_cache.MustFlushRegion(addr, size);
     }
-    return texture_cache.MustFlushRegion(addr, size) || buffer_cache.MustFlushRegion(addr, size);
+    return texture_cache.IsRegionGpuModified(addr, size) ||
+           buffer_cache.MustFlushRegion(addr, size);
 }
 
 void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
@@ -777,7 +734,10 @@ void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
     if (addr == 0 || size == 0) {
         return;
     }
-    texture_cache.InvalidateRegion(addr, size);
+    {
+        auto lock = texture_cache.AcquireLock();
+        texture_cache.WriteMemory(addr, size);
+    }
     shader_cache.InvalidateRegion(addr, size);
     buffer_cache.InvalidateRegion(addr, size);
     query_cache.InvalidateRegion(addr, size);
@@ -788,18 +748,29 @@ void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) {
     if (addr == 0 || size == 0) {
         return;
     }
-    texture_cache.OnCPUWrite(addr, size);
+    {
+        auto lock = texture_cache.AcquireLock();
+        texture_cache.WriteMemory(addr, size);
+    }
     shader_cache.OnCPUWrite(addr, size);
     buffer_cache.OnCPUWrite(addr, size);
 }
 
 void RasterizerOpenGL::SyncGuestHost() {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
-    texture_cache.SyncGuestHost();
     buffer_cache.SyncGuestHost();
     shader_cache.SyncGuestHost();
 }
 
+void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) {
+    {
+        auto lock = texture_cache.AcquireLock();
+        texture_cache.UnmapMemory(addr, size);
+    }
+    buffer_cache.OnCPUWrite(addr, size);
+    shader_cache.OnCPUWrite(addr, size);
+}
+
 void RasterizerOpenGL::SignalSemaphore(GPUVAddr addr, u32 value) {
     if (!gpu.IsAsync()) {
         gpu_memory.Write<u32>(addr, value);
@@ -841,6 +812,14 @@ void RasterizerOpenGL::WaitForIdle() {
                     GL_SHADER_STORAGE_BARRIER_BIT | GL_QUERY_BUFFER_BARRIER_BIT);
 }
 
+void RasterizerOpenGL::FragmentBarrier() {
+    glMemoryBarrier(GL_FRAMEBUFFER_BARRIER_BIT);
+}
+
+void RasterizerOpenGL::TiledCacheBarrier() {
+    glTextureBarrier();
+}
+
 void RasterizerOpenGL::FlushCommands() {
     // Only flush when we have commands queued to OpenGL.
     if (num_queued_commands == 0) {
@@ -854,47 +833,97 @@ void RasterizerOpenGL::TickFrame() {
     // Ticking a frame means that buffers will be swapped, calling glFlush implicitly.
     num_queued_commands = 0;
 
+    fence_manager.TickFrame();
     buffer_cache.TickFrame();
+    {
+        auto lock = texture_cache.AcquireLock();
+        texture_cache.TickFrame();
+    }
 }
 
-bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
+bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surface& src,
-                                             const Tegra::Engines::Fermi2D::Regs::Surface& dst,
+                                             const Tegra::Engines::Fermi2D::Surface& dst,
                                              const Tegra::Engines::Fermi2D::Config& copy_config) {
     MICROPROFILE_SCOPE(OpenGL_Blits);
-    texture_cache.DoFermiCopy(src, dst, copy_config);
+    auto lock = texture_cache.AcquireLock();
+    texture_cache.BlitImage(dst, src, copy_config);
     return true;
 }
 
 bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
                                          VAddr framebuffer_addr, u32 pixel_stride) {
-    if (!framebuffer_addr) {
+    if (framebuffer_addr == 0) {
-        return {};
+        return false;
     }
-
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
 
-    const auto surface{texture_cache.TryFindFramebufferSurface(framebuffer_addr)};
+    auto lock = texture_cache.AcquireLock();
-    if (!surface) {
+    ImageView* const image_view{texture_cache.TryFindFramebufferImageView(framebuffer_addr)};
-        return {};
+    if (!image_view) {
+        return false;
     }
-
     // Verify that the cached surface is the same size and format as the requested framebuffer
-    const auto& params{surface->GetSurfaceParams()};
+    // ASSERT_MSG(image_view->size.width == config.width, "Framebuffer width is different");
-    const auto& pixel_format{
+    // ASSERT_MSG(image_view->size.height == config.height, "Framebuffer height is different");
-        VideoCore::Surface::PixelFormatFromGPUPixelFormat(config.pixel_format)};
-    ASSERT_MSG(params.width == config.width, "Framebuffer width is different");
-    ASSERT_MSG(params.height == config.height, "Framebuffer height is different");
-
-    if (params.pixel_format != pixel_format) {
-        LOG_DEBUG(Render_OpenGL, "Framebuffer pixel_format is different");
-    }
-
-    screen_info.display_texture = surface->GetTexture();
-    screen_info.display_srgb = surface->GetSurfaceParams().srgb_conversion;
 
+    screen_info.display_texture = image_view->Handle(ImageViewType::e2D);
+    screen_info.display_srgb = VideoCore::Surface::IsPixelFormatSRGB(image_view->format);
     return true;
 }
 
+void RasterizerOpenGL::BindComputeTextures(Shader* kernel) {
+    image_view_indices.clear();
+    sampler_handles.clear();
+
+    texture_cache.SynchronizeComputeDescriptors();
+
+    SetupComputeTextures(kernel);
+    SetupComputeImages(kernel);
+
+    const std::span indices_span(image_view_indices.data(), image_view_indices.size());
+    texture_cache.FillComputeImageViews(indices_span, image_view_ids);
+
+    program_manager.BindCompute(kernel->GetHandle());
+    size_t image_view_index = 0;
+    size_t texture_index = 0;
+    size_t image_index = 0;
+    BindTextures(kernel->GetEntries(), 0, 0, image_view_index, texture_index, image_index);
+}
+
+void RasterizerOpenGL::BindTextures(const ShaderEntries& entries, GLuint base_texture,
+                                    GLuint base_image, size_t& image_view_index,
+                                    size_t& texture_index, size_t& image_index) {
+    const GLuint* const samplers = sampler_handles.data() + texture_index;
+    const GLuint* const textures = texture_handles.data() + texture_index;
+    const GLuint* const images = image_handles.data() + image_index;
+
+    const size_t num_samplers = entries.samplers.size();
+    for (const auto& sampler : entries.samplers) {
+        for (size_t i = 0; i < sampler.size; ++i) {
+            const ImageViewId image_view_id = image_view_ids[image_view_index++];
+            const ImageView& image_view = texture_cache.GetImageView(image_view_id);
+            const GLuint handle = image_view.Handle(ImageViewTypeFromEntry(sampler));
+            texture_handles[texture_index++] = handle;
+        }
+    }
+    const size_t num_images = entries.images.size();
+    for (size_t unit = 0; unit < num_images; ++unit) {
+        // TODO: Mark as modified
+        const ImageViewId image_view_id = image_view_ids[image_view_index++];
+        const ImageView& image_view = texture_cache.GetImageView(image_view_id);
+        const GLuint handle = image_view.Handle(ImageViewTypeFromEntry(entries.images[unit]));
+        image_handles[image_index] = handle;
+        ++image_index;
+    }
+    if (num_samplers > 0) {
+        glBindSamplers(base_texture, static_cast<GLsizei>(num_samplers), samplers);
+        glBindTextures(base_texture, static_cast<GLsizei>(num_samplers), textures);
+    }
+    if (num_images > 0) {
+        glBindImageTextures(base_image, static_cast<GLsizei>(num_images), images);
+    }
+}
+
 void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, Shader* shader) {
     static constexpr std::array PARAMETER_LUT{
         GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV,          GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV,
@@ -999,7 +1028,6 @@ void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, Shader* sh
         GL_VERTEX_PROGRAM_NV,   GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV,
         GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV,
     };
-
     const auto& cbufs{maxwell3d.state.shader_stages[stage_index]};
     const auto& entries{shader->GetEntries().global_memory_entries};
 
@@ -1056,77 +1084,53 @@ void RasterizerOpenGL::SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& e
     }
 }
 
-void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, Shader* shader) {
+void RasterizerOpenGL::SetupDrawTextures(const Shader* shader, size_t stage_index) {
-    MICROPROFILE_SCOPE(OpenGL_Texture);
+    const bool via_header_index =
-    u32 binding = device.GetBaseBindings(stage_index).sampler;
+        maxwell3d.regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
     for (const auto& entry : shader->GetEntries().samplers) {
         const auto shader_type = static_cast<ShaderType>(stage_index);
-        for (std::size_t i = 0; i < entry.size; ++i) {
+        for (size_t index = 0; index < entry.size; ++index) {
-            const auto texture = GetTextureInfo(maxwell3d, entry, shader_type, i);
+            const auto handle =
-            SetupTexture(binding++, texture, entry);
+                GetTextureInfo(maxwell3d, via_header_index, entry, shader_type, index);
+            const Sampler* const sampler = texture_cache.GetGraphicsSampler(handle.sampler);
+            sampler_handles.push_back(sampler->Handle());
+            image_view_indices.push_back(handle.image);
         }
     }
 }
 
-void RasterizerOpenGL::SetupComputeTextures(Shader* kernel) {
+void RasterizerOpenGL::SetupComputeTextures(const Shader* kernel) {
-    MICROPROFILE_SCOPE(OpenGL_Texture);
+    const bool via_header_index = kepler_compute.launch_description.linked_tsc;
-    u32 binding = 0;
     for (const auto& entry : kernel->GetEntries().samplers) {
-        for (std::size_t i = 0; i < entry.size; ++i) {
+        for (size_t i = 0; i < entry.size; ++i) {
-            const auto texture = GetTextureInfo(kepler_compute, entry, ShaderType::Compute, i);
+            const auto handle =
-            SetupTexture(binding++, texture, entry);
+                GetTextureInfo(kepler_compute, via_header_index, entry, ShaderType::Compute, i);
+            const Sampler* const sampler = texture_cache.GetComputeSampler(handle.sampler);
+            sampler_handles.push_back(sampler->Handle());
+            image_view_indices.push_back(handle.image);
         }
     }
 }
 
-void RasterizerOpenGL::SetupTexture(u32 binding, const Tegra::Texture::FullTextureInfo& texture,
+void RasterizerOpenGL::SetupDrawImages(const Shader* shader, size_t stage_index) {
-                                    const SamplerEntry& entry) {
+    const bool via_header_index =
-    const auto view = texture_cache.GetTextureSurface(texture.tic, entry);
+        maxwell3d.regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
-    if (!view) {
-        // Can occur when texture addr is null or its memory is unmapped/invalid
-        glBindSampler(binding, 0);
-        glBindTextureUnit(binding, 0);
-        return;
-    }
-    const GLuint handle = view->GetTexture(texture.tic.x_source, texture.tic.y_source,
-                                           texture.tic.z_source, texture.tic.w_source);
-    glBindTextureUnit(binding, handle);
-    if (!view->GetSurfaceParams().IsBuffer()) {
-        glBindSampler(binding, sampler_cache.GetSampler(texture.tsc));
-    }
-}
-
-void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, Shader* shader) {
-    u32 binding = device.GetBaseBindings(stage_index).image;
     for (const auto& entry : shader->GetEntries().images) {
         const auto shader_type = static_cast<ShaderType>(stage_index);
-        const auto tic = GetTextureInfo(maxwell3d, entry, shader_type).tic;
+        const auto handle = GetTextureInfo(maxwell3d, via_header_index, entry, shader_type);
-        SetupImage(binding++, tic, entry);
+        image_view_indices.push_back(handle.image);
     }
 }
 
-void RasterizerOpenGL::SetupComputeImages(Shader* shader) {
+void RasterizerOpenGL::SetupComputeImages(const Shader* shader) {
-    u32 binding = 0;
+    const bool via_header_index = kepler_compute.launch_description.linked_tsc;
     for (const auto& entry : shader->GetEntries().images) {
-        const auto tic = GetTextureInfo(kepler_compute, entry, ShaderType::Compute).tic;
+        const auto handle =
-        SetupImage(binding++, tic, entry);
+            GetTextureInfo(kepler_compute, via_header_index, entry, ShaderType::Compute);
+        image_view_indices.push_back(handle.image);
     }
 }
 
-void RasterizerOpenGL::SetupImage(u32 binding, const Tegra::Texture::TICEntry& tic,
-                                  const ImageEntry& entry) {
-    const auto view = texture_cache.GetImageSurface(tic, entry);
-    if (!view) {
-        glBindImageTexture(binding, 0, 0, GL_FALSE, 0, GL_READ_ONLY, GL_R8);
-        return;
-    }
-    if (entry.is_written) {
-        view->MarkAsModified(texture_cache.Tick());
-    }
-    const GLuint handle = view->GetTexture(tic.x_source, tic.y_source, tic.z_source, tic.w_source);
-    glBindImageTexture(binding, handle, 0, GL_TRUE, 0, GL_READ_WRITE, view->GetFormat());
-}
-
 void RasterizerOpenGL::SyncViewport() {
     auto& flags = maxwell3d.dirty.flags;
     const auto& regs = maxwell3d.regs;
@@ -1526,17 +1530,9 @@ void RasterizerOpenGL::SyncPointState() {
     flags[Dirty::PointSize] = false;
 
     oglEnable(GL_POINT_SPRITE, maxwell3d.regs.point_sprite_enable);
+    oglEnable(GL_PROGRAM_POINT_SIZE, maxwell3d.regs.vp_point_size.enable);
 
-    if (maxwell3d.regs.vp_point_size.enable) {
-        // By definition of GL_POINT_SIZE, it only matters if GL_PROGRAM_POINT_SIZE is disabled.
-        glEnable(GL_PROGRAM_POINT_SIZE);
-        return;
-    }
-
-    // Limit the point size to 1 since nouveau sometimes sets a point size of 0 (and that's invalid
-    // in OpenGL).
     glPointSize(std::max(1.0f, maxwell3d.regs.point_size));
-    glDisable(GL_PROGRAM_POINT_SIZE);
 }
 
 void RasterizerOpenGL::SyncLineState() {

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -7,12 +7,13 @@
 #include <array>
 #include <atomic>
 #include <cstddef>
-#include <map>
 #include <memory>
 #include <optional>
 #include <tuple>
 #include <utility>
 
+#include <boost/container/static_vector.hpp>
+
 #include <glad/glad.h>
 
 #include "common/common_types.h"
@@ -23,16 +24,14 @@
 #include "video_core/renderer_opengl/gl_buffer_cache.h"
 #include "video_core/renderer_opengl/gl_device.h"
 #include "video_core/renderer_opengl/gl_fence_manager.h"
-#include "video_core/renderer_opengl/gl_framebuffer_cache.h"
 #include "video_core/renderer_opengl/gl_query_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
-#include "video_core/renderer_opengl/gl_sampler_cache.h"
 #include "video_core/renderer_opengl/gl_shader_cache.h"
 #include "video_core/renderer_opengl/gl_shader_decompiler.h"
 #include "video_core/renderer_opengl/gl_shader_manager.h"
 #include "video_core/renderer_opengl/gl_state_tracker.h"
+#include "video_core/renderer_opengl/gl_stream_buffer.h"
 #include "video_core/renderer_opengl/gl_texture_cache.h"
-#include "video_core/renderer_opengl/utils.h"
 #include "video_core/shader/async_shaders.h"
 #include "video_core/textures/texture.h"
 
@@ -51,7 +50,7 @@ class MemoryManager;
 namespace OpenGL {
 
 struct ScreenInfo;
-struct DrawParameters;
+struct ShaderEntries;
 
 struct BindlessSSBO {
     GLuint64EXT address;
@@ -79,15 +78,18 @@ public:
     void InvalidateRegion(VAddr addr, u64 size) override;
     void OnCPUWrite(VAddr addr, u64 size) override;
     void SyncGuestHost() override;
+    void UnmapMemory(VAddr addr, u64 size) override;
     void SignalSemaphore(GPUVAddr addr, u32 value) override;
     void SignalSyncPoint(u32 value) override;
     void ReleaseFences() override;
     void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
     void WaitForIdle() override;
+    void FragmentBarrier() override;
+    void TiledCacheBarrier() override;
     void FlushCommands() override;
     void TickFrame() override;
-    bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
+    bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surface& src,
-                               const Tegra::Engines::Fermi2D::Regs::Surface& dst,
+                               const Tegra::Engines::Fermi2D::Surface& dst,
                                const Tegra::Engines::Fermi2D::Config& copy_config) override;
     bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
                            u32 pixel_stride) override;
@@ -108,11 +110,14 @@ public:
     }
 
 private:
-    /// Configures the color and depth framebuffer states.
+    static constexpr size_t MAX_TEXTURES = 192;
-    void ConfigureFramebuffers();
+    static constexpr size_t MAX_IMAGES = 48;
+    static constexpr size_t MAX_IMAGE_VIEWS = MAX_TEXTURES + MAX_IMAGES;
 
-    /// Configures the color and depth framebuffer for clearing.
+    void BindComputeTextures(Shader* kernel);
-    void ConfigureClearFramebuffer(bool using_color, bool using_depth_stencil);
+
+    void BindTextures(const ShaderEntries& entries, GLuint base_texture, GLuint base_image,
+                      size_t& image_view_index, size_t& texture_index, size_t& image_index);
 
     /// Configures the current constbuffers to use for the draw command.
     void SetupDrawConstBuffers(std::size_t stage_index, Shader* shader);
@@ -136,23 +141,16 @@ private:
                            size_t size, BindlessSSBO* ssbo);
 
     /// Configures the current textures to use for the draw command.
-    void SetupDrawTextures(std::size_t stage_index, Shader* shader);
+    void SetupDrawTextures(const Shader* shader, size_t stage_index);
 
     /// Configures the textures used in a compute shader.
-    void SetupComputeTextures(Shader* kernel);
+    void SetupComputeTextures(const Shader* kernel);
-
-    /// Configures a texture.
-    void SetupTexture(u32 binding, const Tegra::Texture::FullTextureInfo& texture,
-                      const SamplerEntry& entry);
 
     /// Configures images in a graphics shader.
-    void SetupDrawImages(std::size_t stage_index, Shader* shader);
+    void SetupDrawImages(const Shader* shader, size_t stage_index);
 
     /// Configures images in a compute shader.
-    void SetupComputeImages(Shader* shader);
+    void SetupComputeImages(const Shader* shader);
-
-    /// Configures an image.
-    void SetupImage(u32 binding, const Tegra::Texture::TICEntry& tic, const ImageEntry& entry);
 
     /// Syncs the viewport and depth range to match the guest state
     void SyncViewport();
@@ -227,9 +225,6 @@ private:
     /// End a transform feedback
     void EndTransformFeedback();
 
-    /// Check for extension that are not strictly required but are needed for correct emulation
-    void CheckExtensions();
-
     std::size_t CalculateVertexArraysSize() const;
 
     std::size_t CalculateIndexBufferSize() const;
@@ -242,7 +237,7 @@ private:
 
     GLintptr SetupIndexBuffer();
 
-    void SetupShaders(GLenum primitive_mode);
+    void SetupShaders();
 
     Tegra::GPU& gpu;
     Tegra::Engines::Maxwell3D& maxwell3d;
@@ -254,19 +249,21 @@ private:
     ProgramManager& program_manager;
     StateTracker& state_tracker;
 
-    TextureCacheOpenGL texture_cache;
+    OGLStreamBuffer stream_buffer;
+    TextureCacheRuntime texture_cache_runtime;
+    TextureCache texture_cache;
     ShaderCacheOpenGL shader_cache;
-    SamplerCacheOpenGL sampler_cache;
-    FramebufferCacheOpenGL framebuffer_cache;
     QueryCache query_cache;
     OGLBufferCache buffer_cache;
     FenceManagerOpenGL fence_manager;
 
     VideoCommon::Shader::AsyncShaders async_shaders;
 
-    static constexpr std::size_t STREAM_BUFFER_SIZE = 128 * 1024 * 1024;
+    boost::container::static_vector<u32, MAX_IMAGE_VIEWS> image_view_indices;
-
+    std::array<ImageViewId, MAX_IMAGE_VIEWS> image_view_ids;
-    GLint vertex_binding = 0;
+    boost::container::static_vector<GLuint, MAX_TEXTURES> sampler_handles;
+    std::array<GLuint, MAX_TEXTURES> texture_handles;
+    std::array<GLuint, MAX_IMAGES> image_handles;
 
     std::array<OGLBuffer, Tegra::Engines::Maxwell3D::Regs::NumTransformFeedbackBuffers>
         transform_feedback_buffers;
@@ -280,7 +277,7 @@ private:
     std::size_t current_cbuf = 0;
     OGLBuffer unified_uniform_buffer;
 
-    /// Number of commands queued to the OpenGL driver. Reseted on flush.
+    /// Number of commands queued to the OpenGL driver. Resetted on flush.
     std::size_t num_queued_commands = 0;
 
     u32 last_clip_distance_mask = 0;

src/video_core/renderer_opengl/gl_resource_manager.cpp

@@ -71,7 +71,7 @@ void OGLSampler::Create() {
         return;
 
     MICROPROFILE_SCOPE(OpenGL_ResourceCreation);
-    glGenSamplers(1, &handle);
+    glCreateSamplers(1, &handle);
 }
 
 void OGLSampler::Release() {

src/video_core/renderer_opengl/gl_sampler_cache.cpp

@@ -1,52 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include "common/logging/log.h"
-#include "video_core/renderer_opengl/gl_resource_manager.h"
-#include "video_core/renderer_opengl/gl_sampler_cache.h"
-#include "video_core/renderer_opengl/maxwell_to_gl.h"
-
-namespace OpenGL {
-
-SamplerCacheOpenGL::SamplerCacheOpenGL() = default;
-
-SamplerCacheOpenGL::~SamplerCacheOpenGL() = default;
-
-OGLSampler SamplerCacheOpenGL::CreateSampler(const Tegra::Texture::TSCEntry& tsc) const {
-    OGLSampler sampler;
-    sampler.Create();
-
-    const GLuint sampler_id{sampler.handle};
-    glSamplerParameteri(
-        sampler_id, GL_TEXTURE_MAG_FILTER,
-        MaxwellToGL::TextureFilterMode(tsc.mag_filter, Tegra::Texture::TextureMipmapFilter::None));
-    glSamplerParameteri(sampler_id, GL_TEXTURE_MIN_FILTER,
-                        MaxwellToGL::TextureFilterMode(tsc.min_filter, tsc.mipmap_filter));
-    glSamplerParameteri(sampler_id, GL_TEXTURE_WRAP_S, MaxwellToGL::WrapMode(tsc.wrap_u));
-    glSamplerParameteri(sampler_id, GL_TEXTURE_WRAP_T, MaxwellToGL::WrapMode(tsc.wrap_v));
-    glSamplerParameteri(sampler_id, GL_TEXTURE_WRAP_R, MaxwellToGL::WrapMode(tsc.wrap_p));
-    glSamplerParameteri(sampler_id, GL_TEXTURE_COMPARE_MODE,
-                        tsc.depth_compare_enabled == 1 ? GL_COMPARE_REF_TO_TEXTURE : GL_NONE);
-    glSamplerParameteri(sampler_id, GL_TEXTURE_COMPARE_FUNC,
-                        MaxwellToGL::DepthCompareFunc(tsc.depth_compare_func));
-    glSamplerParameterfv(sampler_id, GL_TEXTURE_BORDER_COLOR, tsc.GetBorderColor().data());
-    glSamplerParameterf(sampler_id, GL_TEXTURE_MIN_LOD, tsc.GetMinLod());
-    glSamplerParameterf(sampler_id, GL_TEXTURE_MAX_LOD, tsc.GetMaxLod());
-    glSamplerParameterf(sampler_id, GL_TEXTURE_LOD_BIAS, tsc.GetLodBias());
-    if (GLAD_GL_ARB_texture_filter_anisotropic) {
-        glSamplerParameterf(sampler_id, GL_TEXTURE_MAX_ANISOTROPY, tsc.GetMaxAnisotropy());
-    } else if (GLAD_GL_EXT_texture_filter_anisotropic) {
-        glSamplerParameterf(sampler_id, GL_TEXTURE_MAX_ANISOTROPY_EXT, tsc.GetMaxAnisotropy());
-    } else {
-        LOG_WARNING(Render_OpenGL, "Anisotropy not supported by host GPU driver");
-    }
-
-    return sampler;
-}
-
-GLuint SamplerCacheOpenGL::ToSamplerType(const OGLSampler& sampler) const {
-    return sampler.handle;
-}
-
-} // namespace OpenGL

src/video_core/renderer_opengl/gl_sampler_cache.h

@@ -1,25 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <glad/glad.h>
-
-#include "video_core/renderer_opengl/gl_resource_manager.h"
-#include "video_core/sampler_cache.h"
-
-namespace OpenGL {
-
-class SamplerCacheOpenGL final : public VideoCommon::SamplerCache<GLuint, OGLSampler> {
-public:
-    explicit SamplerCacheOpenGL();
-    ~SamplerCacheOpenGL();
-
-protected:
-    OGLSampler CreateSampler(const Tegra::Texture::TSCEntry& tsc) const override;
-
-    GLuint ToSamplerType(const OGLSampler& sampler) const override;
-};
-
-} // namespace OpenGL

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -27,7 +27,6 @@
 #include "video_core/renderer_opengl/gl_shader_decompiler.h"
 #include "video_core/renderer_opengl/gl_shader_disk_cache.h"
 #include "video_core/renderer_opengl/gl_state_tracker.h"
-#include "video_core/renderer_opengl/utils.h"
 #include "video_core/shader/memory_util.h"
 #include "video_core/shader/registry.h"
 #include "video_core/shader/shader_ir.h"

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -38,11 +38,9 @@ using Tegra::Shader::IpaSampleMode;
 using Tegra::Shader::PixelImap;
 using Tegra::Shader::Register;
 using Tegra::Shader::TextureType;
-using VideoCommon::Shader::BuildTransformFeedback;
-using VideoCommon::Shader::Registry;
 
-using namespace std::string_literals;
 using namespace VideoCommon::Shader;
+using namespace std::string_literals;
 
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 using Operation = const OperationNode&;
@@ -2753,11 +2751,11 @@ private:
         }
     }
 
-    std::string GetSampler(const Sampler& sampler) const {
+    std::string GetSampler(const SamplerEntry& sampler) const {
         return AppendSuffix(sampler.index, "sampler");
     }
 
-    std::string GetImage(const Image& image) const {
+    std::string GetImage(const ImageEntry& image) const {
         return AppendSuffix(image.index, "image");
     }
 

src/video_core/renderer_opengl/gl_shader_decompiler.h

@@ -20,8 +20,8 @@ namespace OpenGL {
 class Device;
 
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
-using SamplerEntry = VideoCommon::Shader::Sampler;
+using SamplerEntry = VideoCommon::Shader::SamplerEntry;
-using ImageEntry = VideoCommon::Shader::Image;
+using ImageEntry = VideoCommon::Shader::ImageEntry;
 
 class ConstBufferEntry : public VideoCommon::Shader::ConstBuffer {
 public:

src/video_core/renderer_opengl/gl_shader_manager.cpp

@@ -83,6 +83,21 @@ void ProgramManager::RestoreGuestPipeline() {
     }
 }
 
+void ProgramManager::BindHostCompute(GLuint program) {
+    if (use_assembly_programs) {
+        glDisable(GL_COMPUTE_PROGRAM_NV);
+    }
+    glUseProgram(program);
+    is_graphics_bound = false;
+}
+
+void ProgramManager::RestoreGuestCompute() {
+    if (use_assembly_programs) {
+        glEnable(GL_COMPUTE_PROGRAM_NV);
+        glUseProgram(0);
+    }
+}
+
 void ProgramManager::UseVertexShader(GLuint program) {
     if (use_assembly_programs) {
         BindProgram(GL_VERTEX_PROGRAM_NV, program, current_state.vertex, vertex_enabled);

src/video_core/renderer_opengl/gl_shader_manager.h

@@ -45,6 +45,12 @@ public:
     /// Rewinds BindHostPipeline state changes.
     void RestoreGuestPipeline();
 
+    /// Binds an OpenGL GLSL program object unsynchronized with the guest state.
+    void BindHostCompute(GLuint program);
+
+    /// Rewinds BindHostCompute state changes.
+    void RestoreGuestCompute();
+
     void UseVertexShader(GLuint program);
     void UseGeometryShader(GLuint program);
     void UseFragmentShader(GLuint program);

src/video_core/renderer_opengl/gl_state_tracker.cpp

@@ -249,4 +249,11 @@ StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags}
     }
 }
 
+void StateTracker::InvalidateStreamBuffer() {
+    flags[Dirty::VertexBuffers] = true;
+    for (int index = Dirty::VertexBuffer0; index <= Dirty::VertexBuffer31; ++index) {
+        flags[index] = true;
+    }
+}
+
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_state_tracker.h

@@ -92,6 +92,8 @@ class StateTracker {
 public:
     explicit StateTracker(Tegra::GPU& gpu);
 
+    void InvalidateStreamBuffer();
+
     void BindIndexBuffer(GLuint new_index_buffer) {
         if (index_buffer == new_index_buffer) {
             return;
@@ -100,6 +102,14 @@ public:
         glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, new_index_buffer);
     }
 
+    void BindFramebuffer(GLuint new_framebuffer) {
+        if (framebuffer == new_framebuffer) {
+            return;
+        }
+        framebuffer = new_framebuffer;
+        glBindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer);
+    }
+
     void NotifyScreenDrawVertexArray() {
         flags[OpenGL::Dirty::VertexFormats] = true;
         flags[OpenGL::Dirty::VertexFormat0 + 0] = true;
@@ -129,9 +139,9 @@ public:
         flags[OpenGL::Dirty::Scissor0] = true;
     }
 
-    void NotifyColorMask0() {
+    void NotifyColorMask(size_t index) {
         flags[OpenGL::Dirty::ColorMasks] = true;
-        flags[OpenGL::Dirty::ColorMask0] = true;
+        flags[OpenGL::Dirty::ColorMask0 + index] = true;
     }
 
     void NotifyBlend0() {
@@ -190,6 +200,7 @@ public:
 private:
     Tegra::Engines::Maxwell3D::DirtyState::Flags& flags;
 
+    GLuint framebuffer = 0;
     GLuint index_buffer = 0;
 };
 

src/video_core/renderer_opengl/gl_stream_buffer.cpp

@@ -9,6 +9,7 @@
 #include "common/assert.h"
 #include "common/microprofile.h"
 #include "video_core/renderer_opengl/gl_device.h"
+#include "video_core/renderer_opengl/gl_state_tracker.h"
 #include "video_core/renderer_opengl/gl_stream_buffer.h"
 
 MICROPROFILE_DEFINE(OpenGL_StreamBuffer, "OpenGL", "Stream Buffer Orphaning",
@@ -16,24 +17,14 @@ MICROPROFILE_DEFINE(OpenGL_StreamBuffer, "OpenGL", "Stream Buffer Orphaning",
 
 namespace OpenGL {
 
-OGLStreamBuffer::OGLStreamBuffer(const Device& device, GLsizeiptr size, bool vertex_data_usage)
+OGLStreamBuffer::OGLStreamBuffer(const Device& device, StateTracker& state_tracker_)
-    : buffer_size(size) {
+    : state_tracker{state_tracker_} {
     gl_buffer.Create();
 
-    GLsizeiptr allocate_size = size;
-    if (vertex_data_usage) {
-        // On AMD GPU there is a strange crash in indexed drawing. The crash happens when the buffer
-        // read position is near the end and is an out-of-bound access to the vertex buffer. This is
-        // probably a bug in the driver and is related to the usage of vec3<byte> attributes in the
-        // vertex array. Doubling the allocation size for the vertex buffer seems to avoid the
-        // crash.
-        allocate_size *= 2;
-    }
-
     static constexpr GLbitfield flags = GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT;
-    glNamedBufferStorage(gl_buffer.handle, allocate_size, nullptr, flags);
+    glNamedBufferStorage(gl_buffer.handle, BUFFER_SIZE, nullptr, flags);
     mapped_ptr = static_cast<u8*>(
-        glMapNamedBufferRange(gl_buffer.handle, 0, buffer_size, flags | GL_MAP_FLUSH_EXPLICIT_BIT));
+        glMapNamedBufferRange(gl_buffer.handle, 0, BUFFER_SIZE, flags | GL_MAP_FLUSH_EXPLICIT_BIT));
 
     if (device.UseAssemblyShaders() || device.HasVertexBufferUnifiedMemory()) {
         glMakeNamedBufferResidentNV(gl_buffer.handle, GL_READ_ONLY);
@@ -46,25 +37,24 @@ OGLStreamBuffer::~OGLStreamBuffer() {
     gl_buffer.Release();
 }
 
-std::tuple<u8*, GLintptr, bool> OGLStreamBuffer::Map(GLsizeiptr size, GLintptr alignment) {
+std::pair<u8*, GLintptr> OGLStreamBuffer::Map(GLsizeiptr size, GLintptr alignment) {
-    ASSERT(size <= buffer_size);
+    ASSERT(size <= BUFFER_SIZE);
-    ASSERT(alignment <= buffer_size);
+    ASSERT(alignment <= BUFFER_SIZE);
     mapped_size = size;
 
     if (alignment > 0) {
         buffer_pos = Common::AlignUp<std::size_t>(buffer_pos, alignment);
     }
 
-    bool invalidate = false;
+    if (buffer_pos + size > BUFFER_SIZE) {
-    if (buffer_pos + size > buffer_size) {
         MICROPROFILE_SCOPE(OpenGL_StreamBuffer);
         glInvalidateBufferData(gl_buffer.handle);
+        state_tracker.InvalidateStreamBuffer();
 
         buffer_pos = 0;
-        invalidate = true;
     }
 
-    return std::make_tuple(mapped_ptr + buffer_pos, buffer_pos, invalidate);
+    return std::make_pair(mapped_ptr + buffer_pos, buffer_pos);
 }
 
 void OGLStreamBuffer::Unmap(GLsizeiptr size) {

src/video_core/renderer_opengl/gl_stream_buffer.h

@@ -4,29 +4,31 @@
 
 #pragma once
 
-#include <tuple>
+#include <utility>
+
 #include <glad/glad.h>
+
 #include "common/common_types.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 
 namespace OpenGL {
 
 class Device;
+class StateTracker;
 
 class OGLStreamBuffer : private NonCopyable {
 public:
-    explicit OGLStreamBuffer(const Device& device, GLsizeiptr size, bool vertex_data_usage);
+    explicit OGLStreamBuffer(const Device& device, StateTracker& state_tracker_);
     ~OGLStreamBuffer();
 
     /*
      * Allocates a linear chunk of memory in the GPU buffer with at least "size" bytes
      * and the optional alignment requirement.
      * If the buffer is full, the whole buffer is reallocated which invalidates old chunks.
-     * The return values are the pointer to the new chunk, the offset within the buffer,
+     * The return values are the pointer to the new chunk, and the offset within the buffer.
-     * and the invalidation flag for previous chunks.
      * The actual used size must be specified on unmapping the chunk.
      */
-    std::tuple<u8*, GLintptr, bool> Map(GLsizeiptr size, GLintptr alignment = 0);
+    std::pair<u8*, GLintptr> Map(GLsizeiptr size, GLintptr alignment = 0);
 
     void Unmap(GLsizeiptr size);
 
@@ -39,15 +41,18 @@ public:
     }
 
     GLsizeiptr Size() const noexcept {
-        return buffer_size;
+        return BUFFER_SIZE;
     }
 
 private:
+    static constexpr GLsizeiptr BUFFER_SIZE = 256 * 1024 * 1024;
+
+    StateTracker& state_tracker;
+
     OGLBuffer gl_buffer;
 
     GLuint64EXT gpu_address = 0;
     GLintptr buffer_pos = 0;
-    GLsizeiptr buffer_size = 0;
     GLsizeiptr mapped_size = 0;
     u8* mapped_ptr = nullptr;
 };

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -4,157 +4,247 @@
 
 #pragma once
 
-#include <array>
-#include <functional>
 #include <memory>
-#include <unordered_map>
+#include <span>
-#include <utility>
-#include <vector>
 
 #include <glad/glad.h>
 
-#include "common/common_types.h"
-#include "video_core/engines/shader_bytecode.h"
-#include "video_core/renderer_opengl/gl_device.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
+#include "video_core/renderer_opengl/util_shaders.h"
 #include "video_core/texture_cache/texture_cache.h"
 
 namespace OpenGL {
 
-using VideoCommon::SurfaceParams;
+class Device;
-using VideoCommon::ViewParams;
+class ProgramManager;
-
-class CachedSurfaceView;
-class CachedSurface;
-class TextureCacheOpenGL;
 class StateTracker;
 
-using Surface = std::shared_ptr<CachedSurface>;
+class Framebuffer;
-using View = std::shared_ptr<CachedSurfaceView>;
+class Image;
-using TextureCacheBase = VideoCommon::TextureCache<Surface, View>;
+class ImageView;
+class Sampler;
 
-class CachedSurface final : public VideoCommon::SurfaceBase<View> {
+using VideoCommon::ImageId;
-    friend CachedSurfaceView;
+using VideoCommon::ImageViewId;
+using VideoCommon::ImageViewType;
+using VideoCommon::NUM_RT;
+using VideoCommon::Offset2D;
+using VideoCommon::RenderTargets;
 
+class ImageBufferMap {
 public:
-    explicit CachedSurface(GPUVAddr gpu_addr_, const SurfaceParams& params_,
+    explicit ImageBufferMap(GLuint handle, u8* map, size_t size, OGLSync* sync);
-                           bool is_astc_supported_);
+    ~ImageBufferMap();
-    ~CachedSurface();
 
-    void UploadTexture(const std::vector<u8>& staging_buffer) override;
+    GLuint Handle() const noexcept {
-    void DownloadTexture(std::vector<u8>& staging_buffer) override;
+        return handle;
-
-    GLenum GetTarget() const {
-        return target;
     }
 
-    GLuint GetTexture() const {
+    std::span<u8> Span() const noexcept {
+        return span;
+    }
+
+private:
+    std::span<u8> span;
+    OGLSync* sync;
+    GLuint handle;
+};
+
+struct FormatProperties {
+    GLenum compatibility_class;
+    bool compatibility_by_size;
+    bool is_compressed;
+};
+
+class TextureCacheRuntime {
+    friend Framebuffer;
+    friend Image;
+    friend ImageView;
+    friend Sampler;
+
+public:
+    explicit TextureCacheRuntime(const Device& device, ProgramManager& program_manager,
+                                 StateTracker& state_tracker);
+    ~TextureCacheRuntime();
+
+    void Finish();
+
+    ImageBufferMap MapUploadBuffer(size_t size);
+
+    ImageBufferMap MapDownloadBuffer(size_t size);
+
+    void CopyImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies);
+
+    void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view) {
+        UNIMPLEMENTED();
+    }
+
+    bool CanImageBeCopied(const Image& dst, const Image& src);
+
+    void EmulateCopyImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies);
+
+    void BlitFramebuffer(Framebuffer* dst, Framebuffer* src,
+                         const std::array<Offset2D, 2>& dst_region,
+                         const std::array<Offset2D, 2>& src_region,
+                         Tegra::Engines::Fermi2D::Filter filter,
+                         Tegra::Engines::Fermi2D::Operation operation);
+
+    void AccelerateImageUpload(Image& image, const ImageBufferMap& map, size_t buffer_offset,
+                               std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+    void InsertUploadMemoryBarrier();
+
+    FormatProperties FormatInfo(VideoCommon::ImageType type, GLenum internal_format) const;
+
+private:
+    struct StagingBuffers {
+        explicit StagingBuffers(GLenum storage_flags_, GLenum map_flags_);
+        ~StagingBuffers();
+
+        ImageBufferMap RequestMap(size_t requested_size, bool insert_fence);
+
+        size_t RequestBuffer(size_t requested_size);
+
+        std::optional<size_t> FindBuffer(size_t requested_size);
+
+        std::vector<OGLSync> syncs;
+        std::vector<OGLBuffer> buffers;
+        std::vector<u8*> maps;
+        std::vector<size_t> sizes;
+        GLenum storage_flags;
+        GLenum map_flags;
+    };
+
+    const Device& device;
+    StateTracker& state_tracker;
+    UtilShaders util_shaders;
+
+    std::array<std::unordered_map<GLenum, FormatProperties>, 3> format_properties;
+
+    StagingBuffers upload_buffers{GL_MAP_WRITE_BIT, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT};
+    StagingBuffers download_buffers{GL_MAP_READ_BIT, GL_MAP_READ_BIT};
+
+    OGLTexture null_image_1d_array;
+    OGLTexture null_image_cube_array;
+    OGLTexture null_image_3d;
+    OGLTexture null_image_rect;
+    OGLTextureView null_image_view_1d;
+    OGLTextureView null_image_view_2d;
+    OGLTextureView null_image_view_2d_array;
+    OGLTextureView null_image_view_cube;
+
+    std::array<GLuint, VideoCommon::NUM_IMAGE_VIEW_TYPES> null_image_views;
+};
+
+class Image : public VideoCommon::ImageBase {
+    friend ImageView;
+
+public:
+    explicit Image(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, GPUVAddr gpu_addr,
+                   VAddr cpu_addr);
+
+    void UploadMemory(const ImageBufferMap& map, size_t buffer_offset,
+                      std::span<const VideoCommon::BufferImageCopy> copies);
+
+    void UploadMemory(const ImageBufferMap& map, size_t buffer_offset,
+                      std::span<const VideoCommon::BufferCopy> copies);
+
+    void DownloadMemory(ImageBufferMap& map, size_t buffer_offset,
+                        std::span<const VideoCommon::BufferImageCopy> copies);
+
+    GLuint Handle() const noexcept {
         return texture.handle;
     }
 
-    bool IsCompressed() const {
-        return is_compressed;
-    }
-
-protected:
-    void DecorateSurfaceName() override;
-
-    View CreateView(const ViewParams& view_key) override;
-    View CreateViewInner(const ViewParams& view_key, bool is_proxy);
-
 private:
-    void UploadTextureMipmap(u32 level, const std::vector<u8>& staging_buffer);
+    void CopyBufferToImage(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset);
 
-    GLenum internal_format{};
+    void CopyImageToBuffer(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset);
-    GLenum format{};
-    GLenum type{};
-    bool is_compressed{};
-    GLenum target{};
-    u32 view_count{};
 
     OGLTexture texture;
-    OGLBuffer texture_buffer;
+    OGLTextureView store_view;
+    OGLBuffer buffer;
+    GLenum gl_internal_format = GL_NONE;
+    GLenum gl_store_format = GL_NONE;
+    GLenum gl_format = GL_NONE;
+    GLenum gl_type = GL_NONE;
 };
 
-class CachedSurfaceView final : public VideoCommon::ViewBase {
+class ImageView : public VideoCommon::ImageViewBase {
+    friend Image;
+
 public:
-    explicit CachedSurfaceView(CachedSurface& surface_, const ViewParams& params_, bool is_proxy_);
+    explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageViewInfo&, ImageId, Image&);
-    ~CachedSurfaceView();
+    explicit ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams&);
 
-    /// @brief Attaches this texture view to the currently bound fb_target framebuffer
+    [[nodiscard]] GLuint Handle(ImageViewType query_type) const noexcept {
-    /// @param attachment   Attachment to bind textures to
+        return views[static_cast<size_t>(query_type)];
-    /// @param fb_target    Framebuffer target to attach to (e.g. DRAW_FRAMEBUFFER)
-    void Attach(GLenum attachment, GLenum fb_target) const;
-
-    GLuint GetTexture(Tegra::Texture::SwizzleSource x_source,
-                      Tegra::Texture::SwizzleSource y_source,
-                      Tegra::Texture::SwizzleSource z_source,
-                      Tegra::Texture::SwizzleSource w_source);
-
-    void DecorateViewName(GPUVAddr gpu_addr, const std::string& prefix);
-
-    void MarkAsModified(u64 tick) {
-        surface.MarkAsModified(true, tick);
     }
 
-    GLuint GetTexture() const {
+    [[nodiscard]] GLuint DefaultHandle() const noexcept {
-        if (is_proxy) {
+        return default_handle;
-            return surface.GetTexture();
-        }
-        return main_view.handle;
     }
 
-    GLenum GetFormat() const {
+    [[nodiscard]] GLenum Format() const noexcept {
-        return format;
+        return internal_format;
-    }
-
-    const SurfaceParams& GetSurfaceParams() const {
-        return surface.GetSurfaceParams();
     }
 
 private:
-    OGLTextureView CreateTextureView() const;
+    void SetupView(const Device& device, Image& image, ImageViewType view_type, GLuint handle,
+                   const VideoCommon::ImageViewInfo& info,
+                   VideoCommon::SubresourceRange view_range);
 
-    CachedSurface& surface;
+    std::array<GLuint, VideoCommon::NUM_IMAGE_VIEW_TYPES> views{};
-    const GLenum format;
+    std::vector<OGLTextureView> stored_views;
-    const GLenum target;
+    GLuint default_handle = 0;
-    const bool is_proxy;
+    GLenum internal_format = GL_NONE;
-
-    std::unordered_map<u32, OGLTextureView> view_cache;
-    OGLTextureView main_view;
-
-    // Use an invalid default so it always fails the comparison test
-    u32 current_swizzle = 0xffffffff;
-    GLuint current_view = 0;
 };
 
-class TextureCacheOpenGL final : public TextureCacheBase {
+class ImageAlloc : public VideoCommon::ImageAllocBase {};
+
+class Sampler {
 public:
-    explicit TextureCacheOpenGL(VideoCore::RasterizerInterface& rasterizer_,
+    explicit Sampler(TextureCacheRuntime&, const Tegra::Texture::TSCEntry&);
-                                Tegra::Engines::Maxwell3D& maxwell3d_,
-                                Tegra::MemoryManager& gpu_memory_, const Device& device_,
-                                StateTracker& state_tracker);
-    ~TextureCacheOpenGL();
 
-protected:
+    GLuint Handle() const noexcept {
-    Surface CreateSurface(GPUVAddr gpu_addr, const SurfaceParams& params) override;
+        return sampler.handle;
-
+    }
-    void ImageCopy(Surface& src_surface, Surface& dst_surface,
-                   const VideoCommon::CopyParams& copy_params) override;
-
-    void ImageBlit(View& src_view, View& dst_view,
-                   const Tegra::Engines::Fermi2D::Config& copy_config) override;
-
-    void BufferCopy(Surface& src_surface, Surface& dst_surface) override;
 
 private:
-    GLuint FetchPBO(std::size_t buffer_size);
+    OGLSampler sampler;
-
-    StateTracker& state_tracker;
-
-    OGLFramebuffer src_framebuffer;
-    OGLFramebuffer dst_framebuffer;
-    std::unordered_map<u32, OGLBuffer> copy_pbo_cache;
 };
 
+class Framebuffer {
+public:
+    explicit Framebuffer(TextureCacheRuntime&, std::span<ImageView*, NUM_RT> color_buffers,
+                         ImageView* depth_buffer, const VideoCommon::RenderTargets& key);
+
+    [[nodiscard]] GLuint Handle() const noexcept {
+        return framebuffer.handle;
+    }
+
+    [[nodiscard]] GLbitfield BufferBits() const noexcept {
+        return buffer_bits;
+    }
+
+private:
+    OGLFramebuffer framebuffer;
+    GLbitfield buffer_bits = GL_NONE;
+};
+
+struct TextureCacheParams {
+    static constexpr bool ENABLE_VALIDATION = true;
+    static constexpr bool FRAMEBUFFER_BLITS = true;
+    static constexpr bool HAS_EMULATED_COPIES = true;
+
+    using Runtime = OpenGL::TextureCacheRuntime;
+    using Image = OpenGL::Image;
+    using ImageAlloc = OpenGL::ImageAlloc;
+    using ImageView = OpenGL::ImageView;
+    using Sampler = OpenGL::Sampler;
+    using Framebuffer = OpenGL::Framebuffer;
+};
+
+using TextureCache = VideoCommon::TextureCache<TextureCacheParams>;
+
 } // namespace OpenGL

src/video_core/renderer_opengl/maxwell_to_gl.h

@@ -475,6 +475,19 @@ inline GLenum PolygonMode(Maxwell::PolygonMode polygon_mode) {
     return GL_FILL;
 }
 
+inline GLenum ReductionFilter(Tegra::Texture::SamplerReduction filter) {
+    switch (filter) {
+    case Tegra::Texture::SamplerReduction::WeightedAverage:
+        return GL_WEIGHTED_AVERAGE_ARB;
+    case Tegra::Texture::SamplerReduction::Min:
+        return GL_MIN;
+    case Tegra::Texture::SamplerReduction::Max:
+        return GL_MAX;
+    }
+    UNREACHABLE_MSG("Invalid reduction filter={}", static_cast<int>(filter));
+    return GL_WEIGHTED_AVERAGE_ARB;
+}
+
 inline GLenum ViewportSwizzle(Maxwell::ViewportSwizzle swizzle) {
     // Enumeration order matches register order. We can convert it arithmetically.
     return GL_VIEWPORT_SWIZZLE_POSITIVE_X_NV + static_cast<GLenum>(swizzle);

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -23,10 +23,10 @@
 #include "core/telemetry_session.h"
 #include "video_core/host_shaders/opengl_present_frag.h"
 #include "video_core/host_shaders/opengl_present_vert.h"
-#include "video_core/morton.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
 #include "video_core/renderer_opengl/gl_shader_manager.h"
 #include "video_core/renderer_opengl/renderer_opengl.h"
+#include "video_core/textures/decoders.h"
 
 namespace OpenGL {
 
@@ -140,11 +140,10 @@ void RendererOpenGL::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
     if (!framebuffer) {
         return;
     }
-
     PrepareRendertarget(framebuffer);
     RenderScreenshot();
 
-    glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
+    state_tracker.BindFramebuffer(0);
     DrawScreen(emu_window.GetFramebufferLayout());
 
     ++m_current_frame;
@@ -187,19 +186,20 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf
     // Reset the screen info's display texture to its own permanent texture
     screen_info.display_texture = screen_info.texture.resource.handle;
 
-    const auto pixel_format{
-        VideoCore::Surface::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format)};
-    const u32 bytes_per_pixel{VideoCore::Surface::GetBytesPerPixel(pixel_format)};
-    const u64 size_in_bytes{framebuffer.stride * framebuffer.height * bytes_per_pixel};
-    u8* const host_ptr{cpu_memory.GetPointer(framebuffer_addr)};
-    rasterizer->FlushRegion(ToCacheAddr(host_ptr), size_in_bytes);
-
     // TODO(Rodrigo): Read this from HLE
     constexpr u32 block_height_log2 = 4;
-    VideoCore::MortonSwizzle(VideoCore::MortonSwizzleMode::MortonToLinear, pixel_format,
+    const auto pixel_format{
-                             framebuffer.stride, block_height_log2, framebuffer.height, 0, 1, 1,
+        VideoCore::Surface::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format)};
-                             gl_framebuffer_data.data(), host_ptr);
+    const u32 bytes_per_pixel{VideoCore::Surface::BytesPerBlock(pixel_format)};
+    const u64 size_in_bytes{Tegra::Texture::CalculateSize(
+        true, bytes_per_pixel, framebuffer.stride, framebuffer.height, 1, block_height_log2, 0)};
+    const u8* const host_ptr{cpu_memory.GetPointer(framebuffer_addr)};
+    const std::span<const u8> input_data(host_ptr, size_in_bytes);
+    Tegra::Texture::UnswizzleTexture(gl_framebuffer_data, input_data, bytes_per_pixel,
+                                     framebuffer.width, framebuffer.height, 1, block_height_log2,
+                                     0);
 
+    glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
     glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast<GLint>(framebuffer.stride));
 
     // Update existing texture
@@ -238,6 +238,10 @@ void RendererOpenGL::InitOpenGLObjects() {
     glUseProgramStages(pipeline.handle, GL_VERTEX_SHADER_BIT, vertex_program.handle);
     glUseProgramStages(pipeline.handle, GL_FRAGMENT_SHADER_BIT, fragment_program.handle);
 
+    // Generate presentation sampler
+    present_sampler.Create();
+    glSamplerParameteri(present_sampler.handle, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+
     // Generate VBO handle for drawing
     vertex_buffer.Create();
 
@@ -255,6 +259,11 @@ void RendererOpenGL::InitOpenGLObjects() {
     // Clear screen to black
     LoadColorToActiveGLTexture(0, 0, 0, 0, screen_info.texture);
 
+    // Enable seamless cubemaps when per texture parameters are not available
+    if (!GLAD_GL_ARB_seamless_cubemap_per_texture && !GLAD_GL_AMD_seamless_cubemap_per_texture) {
+        glEnable(GL_TEXTURE_CUBE_MAP_SEAMLESS);
+    }
+
     // Enable unified vertex attributes and query vertex buffer address when the driver supports it
     if (device.HasVertexBufferUnifiedMemory()) {
         glEnableClientState(GL_VERTEX_ATTRIB_ARRAY_UNIFIED_NV);
@@ -296,7 +305,7 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture,
 
     const auto pixel_format{
         VideoCore::Surface::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format)};
-    const u32 bytes_per_pixel{VideoCore::Surface::GetBytesPerPixel(pixel_format)};
+    const u32 bytes_per_pixel{VideoCore::Surface::BytesPerBlock(pixel_format)};
     gl_framebuffer_data.resize(texture.width * texture.height * bytes_per_pixel);
 
     GLint internal_format;
@@ -315,8 +324,8 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture,
         internal_format = GL_RGBA8;
         texture.gl_format = GL_RGBA;
         texture.gl_type = GL_UNSIGNED_INT_8_8_8_8_REV;
-        UNIMPLEMENTED_MSG("Unknown framebuffer pixel format: {}",
+        // UNIMPLEMENTED_MSG("Unknown framebuffer pixel format: {}",
-                          static_cast<u32>(framebuffer.pixel_format));
+        //                   static_cast<u32>(framebuffer.pixel_format));
     }
 
     texture.resource.Release();
@@ -382,7 +391,7 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) {
     state_tracker.NotifyPolygonModes();
     state_tracker.NotifyViewport0();
     state_tracker.NotifyScissor0();
-    state_tracker.NotifyColorMask0();
+    state_tracker.NotifyColorMask(0);
     state_tracker.NotifyBlend0();
     state_tracker.NotifyFramebuffer();
     state_tracker.NotifyFrontFace();
@@ -440,7 +449,7 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) {
     }
 
     glBindTextureUnit(0, screen_info.display_texture);
-    glBindSampler(0, 0);
+    glBindSampler(0, present_sampler.handle);
 
     glClear(GL_COLOR_BUFFER_BIT);
     glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
@@ -473,6 +482,8 @@ void RendererOpenGL::RenderScreenshot() {
 
     DrawScreen(layout);
 
+    glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
+    glPixelStorei(GL_PACK_ROW_LENGTH, 0);
     glReadPixels(0, 0, layout.width, layout.height, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV,
                  renderer_settings.screenshot_bits);
 

src/video_core/renderer_opengl/renderer_opengl.h

@@ -102,6 +102,7 @@ private:
     StateTracker state_tracker{gpu};
 
     // OpenGL object IDs
+    OGLSampler present_sampler;
     OGLBuffer vertex_buffer;
     OGLProgram vertex_program;
     OGLProgram fragment_program;

src/video_core/renderer_opengl/util_shaders.cpp

@@ -0,0 +1,224 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <bit>
+#include <span>
+#include <string_view>
+
+#include <glad/glad.h>
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "common/div_ceil.h"
+#include "video_core/host_shaders/block_linear_unswizzle_2d_comp.h"
+#include "video_core/host_shaders/block_linear_unswizzle_3d_comp.h"
+#include "video_core/host_shaders/opengl_copy_bc4_comp.h"
+#include "video_core/host_shaders/pitch_unswizzle_comp.h"
+#include "video_core/renderer_opengl/gl_resource_manager.h"
+#include "video_core/renderer_opengl/gl_shader_manager.h"
+#include "video_core/renderer_opengl/gl_texture_cache.h"
+#include "video_core/renderer_opengl/util_shaders.h"
+#include "video_core/surface.h"
+#include "video_core/texture_cache/accelerated_swizzle.h"
+#include "video_core/texture_cache/types.h"
+#include "video_core/texture_cache/util.h"
+#include "video_core/textures/decoders.h"
+
+namespace OpenGL {
+
+using namespace HostShaders;
+
+using VideoCommon::Extent3D;
+using VideoCommon::ImageCopy;
+using VideoCommon::ImageType;
+using VideoCommon::SwizzleParameters;
+using VideoCommon::Accelerated::MakeBlockLinearSwizzle2DParams;
+using VideoCommon::Accelerated::MakeBlockLinearSwizzle3DParams;
+using VideoCore::Surface::BytesPerBlock;
+
+namespace {
+
+OGLProgram MakeProgram(std::string_view source) {
+    OGLShader shader;
+    shader.Create(source, GL_COMPUTE_SHADER);
+
+    OGLProgram program;
+    program.Create(true, false, shader.handle);
+    return program;
+}
+
+} // Anonymous namespace
+
+UtilShaders::UtilShaders(ProgramManager& program_manager_)
+    : program_manager{program_manager_},
+      block_linear_unswizzle_2d_program(MakeProgram(BLOCK_LINEAR_UNSWIZZLE_2D_COMP)),
+      block_linear_unswizzle_3d_program(MakeProgram(BLOCK_LINEAR_UNSWIZZLE_3D_COMP)),
+      pitch_unswizzle_program(MakeProgram(PITCH_UNSWIZZLE_COMP)),
+      copy_bc4_program(MakeProgram(OPENGL_COPY_BC4_COMP)) {
+    const auto swizzle_table = Tegra::Texture::MakeSwizzleTable();
+    swizzle_table_buffer.Create();
+    glNamedBufferStorage(swizzle_table_buffer.handle, sizeof(swizzle_table), &swizzle_table, 0);
+}
+
+UtilShaders::~UtilShaders() = default;
+
+void UtilShaders::BlockLinearUpload2D(Image& image, const ImageBufferMap& map, size_t buffer_offset,
+                                      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_OUTPUT_IMAGE = 0;
+
+    program_manager.BindHostCompute(block_linear_unswizzle_2d_program.handle);
+    glFlushMappedNamedBufferRange(map.Handle(), buffer_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) {
+        const Extent3D num_tiles = swizzle.num_tiles;
+        const size_t input_offset = swizzle.buffer_offset + buffer_offset;
+
+        const u32 num_dispatches_x = Common::DivCeil(num_tiles.width, WORKGROUP_SIZE.width);
+        const u32 num_dispatches_y = Common::DivCeil(num_tiles.height, WORKGROUP_SIZE.height);
+
+        const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
+        glUniform3uiv(0, 1, params.origin.data());
+        glUniform3iv(1, 1, params.destination.data());
+        glUniform1ui(2, params.bytes_per_block_log2);
+        glUniform1ui(3, params.layer_stride);
+        glUniform1ui(4, params.block_size);
+        glUniform1ui(5, params.x_shift);
+        glUniform1ui(6, params.block_height);
+        glUniform1ui(7, params.block_height_mask);
+        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.Handle(),
+                          input_offset, image.guest_size_bytes - swizzle.buffer_offset);
+        glBindImageTexture(BINDING_OUTPUT_IMAGE, image.Handle(), swizzle.level, GL_TRUE, 0,
+                           GL_WRITE_ONLY, store_format);
+        glDispatchCompute(num_dispatches_x, num_dispatches_y, image.info.resources.layers);
+    }
+    program_manager.RestoreGuestCompute();
+}
+
+void UtilShaders::BlockLinearUpload3D(Image& image, const ImageBufferMap& map, size_t buffer_offset,
+                                      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_OUTPUT_IMAGE = 0;
+
+    glFlushMappedNamedBufferRange(map.Handle(), buffer_offset, image.guest_size_bytes);
+    program_manager.BindHostCompute(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) {
+        const Extent3D num_tiles = swizzle.num_tiles;
+        const size_t input_offset = swizzle.buffer_offset + buffer_offset;
+
+        const u32 num_dispatches_x = Common::DivCeil(num_tiles.width, WORKGROUP_SIZE.width);
+        const u32 num_dispatches_y = Common::DivCeil(num_tiles.height, WORKGROUP_SIZE.height);
+        const u32 num_dispatches_z = Common::DivCeil(num_tiles.depth, WORKGROUP_SIZE.depth);
+
+        const auto params = MakeBlockLinearSwizzle3DParams(swizzle, image.info);
+        glUniform3uiv(0, 1, params.origin.data());
+        glUniform3iv(1, 1, params.destination.data());
+        glUniform1ui(2, params.bytes_per_block_log2);
+        glUniform1ui(3, params.slice_size);
+        glUniform1ui(4, params.block_size);
+        glUniform1ui(5, params.x_shift);
+        glUniform1ui(6, params.block_height);
+        glUniform1ui(7, params.block_height_mask);
+        glUniform1ui(8, params.block_depth);
+        glUniform1ui(9, params.block_depth_mask);
+        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.Handle(),
+                          input_offset, image.guest_size_bytes - swizzle.buffer_offset);
+        glBindImageTexture(BINDING_OUTPUT_IMAGE, image.Handle(), swizzle.level, GL_TRUE, 0,
+                           GL_WRITE_ONLY, store_format);
+        glDispatchCompute(num_dispatches_x, num_dispatches_y, num_dispatches_z);
+    }
+    program_manager.RestoreGuestCompute();
+}
+
+void UtilShaders::PitchUpload(Image& image, const ImageBufferMap& map, size_t buffer_offset,
+                              std::span<const SwizzleParameters> swizzles) {
+    static constexpr Extent3D WORKGROUP_SIZE{32, 32, 1};
+    static constexpr GLuint BINDING_INPUT_BUFFER = 0;
+    static constexpr GLuint BINDING_OUTPUT_IMAGE = 0;
+    static constexpr GLuint LOC_ORIGIN = 0;
+    static constexpr GLuint LOC_DESTINATION = 1;
+    static constexpr GLuint LOC_BYTES_PER_BLOCK = 2;
+    static constexpr GLuint LOC_PITCH = 3;
+
+    const u32 bytes_per_block = BytesPerBlock(image.info.format);
+    const GLenum format = StoreFormat(bytes_per_block);
+    const u32 pitch = image.info.pitch;
+
+    UNIMPLEMENTED_IF_MSG(!std::has_single_bit(bytes_per_block),
+                         "Non-power of two images are not implemented");
+
+    program_manager.BindHostCompute(pitch_unswizzle_program.handle);
+    glFlushMappedNamedBufferRange(map.Handle(), buffer_offset, image.guest_size_bytes);
+    glUniform2ui(LOC_ORIGIN, 0, 0);
+    glUniform2i(LOC_DESTINATION, 0, 0);
+    glUniform1ui(LOC_BYTES_PER_BLOCK, bytes_per_block);
+    glUniform1ui(LOC_PITCH, pitch);
+    glBindImageTexture(BINDING_OUTPUT_IMAGE, image.Handle(), 0, GL_FALSE, 0, GL_WRITE_ONLY, format);
+    for (const SwizzleParameters& swizzle : swizzles) {
+        const Extent3D num_tiles = swizzle.num_tiles;
+        const size_t input_offset = swizzle.buffer_offset + buffer_offset;
+
+        const u32 num_dispatches_x = Common::DivCeil(num_tiles.width, WORKGROUP_SIZE.width);
+        const u32 num_dispatches_y = Common::DivCeil(num_tiles.height, WORKGROUP_SIZE.height);
+
+        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.Handle(),
+                          input_offset, image.guest_size_bytes - swizzle.buffer_offset);
+        glDispatchCompute(num_dispatches_x, num_dispatches_y, 1);
+    }
+    program_manager.RestoreGuestCompute();
+}
+
+void UtilShaders::CopyBC4(Image& dst_image, Image& src_image, std::span<const ImageCopy> copies) {
+    static constexpr GLuint BINDING_INPUT_IMAGE = 0;
+    static constexpr GLuint BINDING_OUTPUT_IMAGE = 1;
+    static constexpr GLuint LOC_SRC_OFFSET = 0;
+    static constexpr GLuint LOC_DST_OFFSET = 1;
+
+    program_manager.BindHostCompute(copy_bc4_program.handle);
+
+    for (const ImageCopy& copy : copies) {
+        ASSERT(copy.src_subresource.base_layer == 0);
+        ASSERT(copy.src_subresource.num_layers == 1);
+        ASSERT(copy.dst_subresource.base_layer == 0);
+        ASSERT(copy.dst_subresource.num_layers == 1);
+
+        glUniform3ui(LOC_SRC_OFFSET, copy.src_offset.x, copy.src_offset.y, copy.src_offset.z);
+        glUniform3ui(LOC_DST_OFFSET, copy.dst_offset.x, copy.dst_offset.y, copy.dst_offset.z);
+        glBindImageTexture(BINDING_INPUT_IMAGE, src_image.Handle(), copy.src_subresource.base_level,
+                           GL_FALSE, 0, GL_READ_ONLY, GL_RG32UI);
+        glBindImageTexture(BINDING_OUTPUT_IMAGE, dst_image.Handle(),
+                           copy.dst_subresource.base_level, GL_FALSE, 0, GL_WRITE_ONLY, GL_RGBA8UI);
+        glDispatchCompute(copy.extent.width, copy.extent.height, copy.extent.depth);
+    }
+    program_manager.RestoreGuestCompute();
+}
+
+GLenum StoreFormat(u32 bytes_per_block) {
+    switch (bytes_per_block) {
+    case 1:
+        return GL_R8UI;
+    case 2:
+        return GL_R16UI;
+    case 4:
+        return GL_R32UI;
+    case 8:
+        return GL_RG32UI;
+    case 16:
+        return GL_RGBA32UI;
+    }
+    UNREACHABLE();
+    return GL_R8UI;
+}
+
+} // namespace OpenGL

src/video_core/renderer_opengl/util_shaders.h

@@ -0,0 +1,51 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <span>
+
+#include <glad/glad.h>
+
+#include "common/common_types.h"
+#include "video_core/renderer_opengl/gl_resource_manager.h"
+#include "video_core/texture_cache/types.h"
+
+namespace OpenGL {
+
+class Image;
+class ImageBufferMap;
+class ProgramManager;
+
+class UtilShaders {
+public:
+    explicit UtilShaders(ProgramManager& program_manager);
+    ~UtilShaders();
+
+    void BlockLinearUpload2D(Image& image, const ImageBufferMap& map, size_t buffer_offset,
+                             std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+    void BlockLinearUpload3D(Image& image, const ImageBufferMap& map, size_t buffer_offset,
+                             std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+    void PitchUpload(Image& image, const ImageBufferMap& map, size_t buffer_offset,
+                     std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+    void CopyBC4(Image& dst_image, Image& src_image,
+                 std::span<const VideoCommon::ImageCopy> copies);
+
+private:
+    ProgramManager& program_manager;
+
+    OGLBuffer swizzle_table_buffer;
+
+    OGLProgram block_linear_unswizzle_2d_program;
+    OGLProgram block_linear_unswizzle_3d_program;
+    OGLProgram pitch_unswizzle_program;
+    OGLProgram copy_bc4_program;
+};
+
+GLenum StoreFormat(u32 bytes_per_block);
+
+} // namespace OpenGL

src/video_core/renderer_opengl/utils.cpp

@@ -1,42 +0,0 @@
-// Copyright 2014 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <string>
-#include <vector>
-
-#include <fmt/format.h>
-#include <glad/glad.h>
-
-#include "common/common_types.h"
-#include "video_core/renderer_opengl/gl_state_tracker.h"
-#include "video_core/renderer_opengl/utils.h"
-
-namespace OpenGL {
-
-void LabelGLObject(GLenum identifier, GLuint handle, VAddr addr, std::string_view extra_info) {
-    if (!GLAD_GL_KHR_debug) {
-        // We don't need to throw an error as this is just for debugging
-        return;
-    }
-
-    std::string object_label;
-    if (extra_info.empty()) {
-        switch (identifier) {
-        case GL_TEXTURE:
-            object_label = fmt::format("Texture@0x{:016X}", addr);
-            break;
-        case GL_PROGRAM:
-            object_label = fmt::format("Shader@0x{:016X}", addr);
-            break;
-        default:
-            object_label = fmt::format("Object(0x{:X})@0x{:016X}", identifier, addr);
-            break;
-        }
-    } else {
-        object_label = fmt::format("{}@0x{:016X}", extra_info, addr);
-    }
-    glObjectLabel(identifier, handle, -1, static_cast<const GLchar*>(object_label.c_str()));
-}
-
-} // namespace OpenGL

src/video_core/renderer_opengl/utils.h

@@ -1,16 +0,0 @@
-// Copyright 2014 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <string_view>
-#include <vector>
-#include <glad/glad.h>
-#include "common/common_types.h"
-
-namespace OpenGL {
-
-void LabelGLObject(GLenum identifier, GLuint handle, VAddr addr, std::string_view extra_info = {});
-
-} // namespace OpenGL

src/video_core/renderer_vulkan/blit_image.cpp

@@ -0,0 +1,624 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+
+#include "video_core/host_shaders/convert_depth_to_float_frag_spv.h"
+#include "video_core/host_shaders/convert_float_to_depth_frag_spv.h"
+#include "video_core/host_shaders/full_screen_triangle_vert_spv.h"
+#include "video_core/host_shaders/vulkan_blit_color_float_frag_spv.h"
+#include "video_core/host_shaders/vulkan_blit_depth_stencil_frag_spv.h"
+#include "video_core/renderer_vulkan/blit_image.h"
+#include "video_core/renderer_vulkan/maxwell_to_vk.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_shader_util.h"
+#include "video_core/renderer_vulkan/vk_state_tracker.h"
+#include "video_core/renderer_vulkan/vk_texture_cache.h"
+#include "video_core/renderer_vulkan/vk_update_descriptor.h"
+#include "video_core/renderer_vulkan/wrapper.h"
+#include "video_core/surface.h"
+
+namespace Vulkan {
+
+using VideoCommon::ImageViewType;
+
+namespace {
+struct PushConstants {
+    std::array<float, 2> tex_scale;
+    std::array<float, 2> tex_offset;
+};
+
+template <u32 binding>
+inline constexpr VkDescriptorSetLayoutBinding TEXTURE_DESCRIPTOR_SET_LAYOUT_BINDING{
+    .binding = binding,
+    .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+    .descriptorCount = 1,
+    .stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT,
+    .pImmutableSamplers = nullptr,
+};
+constexpr std::array TWO_TEXTURES_DESCRIPTOR_SET_LAYOUT_BINDINGS{
+    TEXTURE_DESCRIPTOR_SET_LAYOUT_BINDING<0>,
+    TEXTURE_DESCRIPTOR_SET_LAYOUT_BINDING<1>,
+};
+constexpr VkDescriptorSetLayoutCreateInfo ONE_TEXTURE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO{
+    .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
+    .pNext = nullptr,
+    .flags = 0,
+    .bindingCount = 1,
+    .pBindings = &TEXTURE_DESCRIPTOR_SET_LAYOUT_BINDING<0>,
+};
+constexpr VkDescriptorSetLayoutCreateInfo TWO_TEXTURES_DESCRIPTOR_SET_LAYOUT_CREATE_INFO{
+    .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
+    .pNext = nullptr,
+    .flags = 0,
+    .bindingCount = static_cast<u32>(TWO_TEXTURES_DESCRIPTOR_SET_LAYOUT_BINDINGS.size()),
+    .pBindings = TWO_TEXTURES_DESCRIPTOR_SET_LAYOUT_BINDINGS.data(),
+};
+constexpr VkPushConstantRange PUSH_CONSTANT_RANGE{
+    .stageFlags = VK_SHADER_STAGE_VERTEX_BIT,
+    .offset = 0,
+    .size = sizeof(PushConstants),
+};
+constexpr VkPipelineVertexInputStateCreateInfo PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO{
+    .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
+    .pNext = nullptr,
+    .flags = 0,
+    .vertexBindingDescriptionCount = 0,
+    .pVertexBindingDescriptions = nullptr,
+    .vertexAttributeDescriptionCount = 0,
+    .pVertexAttributeDescriptions = nullptr,
+};
+constexpr VkPipelineInputAssemblyStateCreateInfo PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO{
+    .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
+    .pNext = nullptr,
+    .flags = 0,
+    .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST,
+    .primitiveRestartEnable = VK_FALSE,
+};
+constexpr VkPipelineViewportStateCreateInfo PIPELINE_VIEWPORT_STATE_CREATE_INFO{
+    .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
+    .pNext = nullptr,
+    .flags = 0,
+    .viewportCount = 1,
+    .pViewports = nullptr,
+    .scissorCount = 1,
+    .pScissors = nullptr,
+};
+constexpr VkPipelineRasterizationStateCreateInfo PIPELINE_RASTERIZATION_STATE_CREATE_INFO{
+    .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
+    .pNext = nullptr,
+    .flags = 0,
+    .depthClampEnable = VK_FALSE,
+    .rasterizerDiscardEnable = VK_FALSE,
+    .polygonMode = VK_POLYGON_MODE_FILL,
+    .cullMode = VK_CULL_MODE_BACK_BIT,
+    .frontFace = VK_FRONT_FACE_CLOCKWISE,
+    .depthBiasEnable = VK_FALSE,
+    .depthBiasConstantFactor = 0.0f,
+    .depthBiasClamp = 0.0f,
+    .depthBiasSlopeFactor = 0.0f,
+    .lineWidth = 1.0f,
+};
+constexpr VkPipelineMultisampleStateCreateInfo PIPELINE_MULTISAMPLE_STATE_CREATE_INFO{
+    .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
+    .pNext = nullptr,
+    .flags = 0,
+    .rasterizationSamples = VK_SAMPLE_COUNT_1_BIT,
+    .sampleShadingEnable = VK_FALSE,
+    .minSampleShading = 0.0f,
+    .pSampleMask = nullptr,
+    .alphaToCoverageEnable = VK_FALSE,
+    .alphaToOneEnable = VK_FALSE,
+};
+constexpr std::array DYNAMIC_STATES{
+    VK_DYNAMIC_STATE_VIEWPORT,
+    VK_DYNAMIC_STATE_SCISSOR,
+};
+constexpr VkPipelineDynamicStateCreateInfo PIPELINE_DYNAMIC_STATE_CREATE_INFO{
+    .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
+    .pNext = nullptr,
+    .flags = 0,
+    .dynamicStateCount = static_cast<u32>(DYNAMIC_STATES.size()),
+    .pDynamicStates = DYNAMIC_STATES.data(),
+};
+constexpr VkPipelineColorBlendStateCreateInfo PIPELINE_COLOR_BLEND_STATE_EMPTY_CREATE_INFO{
+    .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
+    .pNext = nullptr,
+    .flags = 0,
+    .logicOpEnable = VK_FALSE,
+    .logicOp = VK_LOGIC_OP_CLEAR,
+    .attachmentCount = 0,
+    .pAttachments = nullptr,
+    .blendConstants = {0.0f, 0.0f, 0.0f, 0.0f},
+};
+constexpr VkPipelineColorBlendAttachmentState PIPELINE_COLOR_BLEND_ATTACHMENT_STATE{
+    .blendEnable = VK_FALSE,
+    .srcColorBlendFactor = VK_BLEND_FACTOR_ZERO,
+    .dstColorBlendFactor = VK_BLEND_FACTOR_ZERO,
+    .colorBlendOp = VK_BLEND_OP_ADD,
+    .srcAlphaBlendFactor = VK_BLEND_FACTOR_ZERO,
+    .dstAlphaBlendFactor = VK_BLEND_FACTOR_ZERO,
+    .alphaBlendOp = VK_BLEND_OP_ADD,
+    .colorWriteMask = VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT |
+                      VK_COLOR_COMPONENT_B_BIT | VK_COLOR_COMPONENT_A_BIT,
+};
+constexpr VkPipelineColorBlendStateCreateInfo PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO{
+    .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
+    .pNext = nullptr,
+    .flags = 0,
+    .logicOpEnable = VK_FALSE,
+    .logicOp = VK_LOGIC_OP_CLEAR,
+    .attachmentCount = 1,
+    .pAttachments = &PIPELINE_COLOR_BLEND_ATTACHMENT_STATE,
+    .blendConstants = {0.0f, 0.0f, 0.0f, 0.0f},
+};
+constexpr VkPipelineDepthStencilStateCreateInfo PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO{
+    .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
+    .pNext = nullptr,
+    .flags = 0,
+    .depthTestEnable = VK_TRUE,
+    .depthWriteEnable = VK_TRUE,
+    .depthCompareOp = VK_COMPARE_OP_ALWAYS,
+    .depthBoundsTestEnable = VK_FALSE,
+    .stencilTestEnable = VK_FALSE,
+    .front = VkStencilOpState{},
+    .back = VkStencilOpState{},
+    .minDepthBounds = 0.0f,
+    .maxDepthBounds = 0.0f,
+};
+
+template <VkFilter filter>
+inline constexpr VkSamplerCreateInfo SAMPLER_CREATE_INFO{
+    .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
+    .pNext = nullptr,
+    .flags = 0,
+    .magFilter = filter,
+    .minFilter = filter,
+    .mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST,
+    .addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER,
+    .addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER,
+    .addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER,
+    .mipLodBias = 0.0f,
+    .anisotropyEnable = VK_FALSE,
+    .maxAnisotropy = 0.0f,
+    .compareEnable = VK_FALSE,
+    .compareOp = VK_COMPARE_OP_NEVER,
+    .minLod = 0.0f,
+    .maxLod = 0.0f,
+    .borderColor = VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE,
+    .unnormalizedCoordinates = VK_TRUE,
+};
+
+constexpr VkPipelineLayoutCreateInfo PipelineLayoutCreateInfo(
+    const VkDescriptorSetLayout* set_layout) {
+    return VkPipelineLayoutCreateInfo{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .setLayoutCount = 1,
+        .pSetLayouts = set_layout,
+        .pushConstantRangeCount = 1,
+        .pPushConstantRanges = &PUSH_CONSTANT_RANGE,
+    };
+}
+
+constexpr VkPipelineShaderStageCreateInfo PipelineShaderStageCreateInfo(VkShaderStageFlagBits stage,
+                                                                        VkShaderModule shader) {
+    return VkPipelineShaderStageCreateInfo{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .stage = stage,
+        .module = shader,
+        .pName = "main",
+        .pSpecializationInfo = nullptr,
+    };
+}
+
+constexpr std::array<VkPipelineShaderStageCreateInfo, 2> MakeStages(
+    VkShaderModule vertex_shader, VkShaderModule fragment_shader) {
+    return std::array{
+        PipelineShaderStageCreateInfo(VK_SHADER_STAGE_VERTEX_BIT, vertex_shader),
+        PipelineShaderStageCreateInfo(VK_SHADER_STAGE_FRAGMENT_BIT, fragment_shader),
+    };
+}
+
+void UpdateOneTextureDescriptorSet(const VKDevice& device, VkDescriptorSet descriptor_set,
+                                   VkSampler sampler, VkImageView image_view) {
+    const VkDescriptorImageInfo image_info{
+        .sampler = sampler,
+        .imageView = image_view,
+        .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
+    };
+    const VkWriteDescriptorSet write_descriptor_set{
+        .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
+        .pNext = nullptr,
+        .dstSet = descriptor_set,
+        .dstBinding = 0,
+        .dstArrayElement = 0,
+        .descriptorCount = 1,
+        .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+        .pImageInfo = &image_info,
+        .pBufferInfo = nullptr,
+        .pTexelBufferView = nullptr,
+    };
+    device.GetLogical().UpdateDescriptorSets(write_descriptor_set, nullptr);
+}
+
+void UpdateTwoTexturesDescriptorSet(const VKDevice& device, VkDescriptorSet descriptor_set,
+                                    VkSampler sampler, VkImageView image_view_0,
+                                    VkImageView image_view_1) {
+    const VkDescriptorImageInfo image_info_0{
+        .sampler = sampler,
+        .imageView = image_view_0,
+        .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
+    };
+    const VkDescriptorImageInfo image_info_1{
+        .sampler = sampler,
+        .imageView = image_view_1,
+        .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
+    };
+    const std::array write_descriptor_sets{
+        VkWriteDescriptorSet{
+            .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
+            .pNext = nullptr,
+            .dstSet = descriptor_set,
+            .dstBinding = 0,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+            .pImageInfo = &image_info_0,
+            .pBufferInfo = nullptr,
+            .pTexelBufferView = nullptr,
+        },
+        VkWriteDescriptorSet{
+            .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
+            .pNext = nullptr,
+            .dstSet = descriptor_set,
+            .dstBinding = 1,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+            .pImageInfo = &image_info_1,
+            .pBufferInfo = nullptr,
+            .pTexelBufferView = nullptr,
+        },
+    };
+    device.GetLogical().UpdateDescriptorSets(write_descriptor_sets, nullptr);
+}
+
+void BindBlitState(vk::CommandBuffer cmdbuf, VkPipelineLayout layout,
+                   const std::array<Offset2D, 2>& dst_region,
+                   const std::array<Offset2D, 2>& src_region) {
+    const VkOffset2D offset{
+        .x = std::min(dst_region[0].x, dst_region[1].x),
+        .y = std::min(dst_region[0].y, dst_region[1].y),
+    };
+    const VkExtent2D extent{
+        .width = static_cast<u32>(std::abs(dst_region[1].x - dst_region[0].x)),
+        .height = static_cast<u32>(std::abs(dst_region[1].y - dst_region[0].y)),
+    };
+    const VkViewport viewport{
+        .x = static_cast<float>(offset.x),
+        .y = static_cast<float>(offset.y),
+        .width = static_cast<float>(extent.width),
+        .height = static_cast<float>(extent.height),
+        .minDepth = 0.0f,
+        .maxDepth = 1.0f,
+    };
+    // TODO: Support scissored blits
+    const VkRect2D scissor{
+        .offset = offset,
+        .extent = extent,
+    };
+    const float scale_x = static_cast<float>(src_region[1].x - src_region[0].x);
+    const float scale_y = static_cast<float>(src_region[1].y - src_region[0].y);
+    const PushConstants push_constants{
+        .tex_scale = {scale_x, scale_y},
+        .tex_offset = {static_cast<float>(src_region[0].x), static_cast<float>(src_region[0].y)},
+    };
+    cmdbuf.SetViewport(0, viewport);
+    cmdbuf.SetScissor(0, scissor);
+    cmdbuf.PushConstants(layout, VK_SHADER_STAGE_VERTEX_BIT, push_constants);
+}
+
+} // Anonymous namespace
+
+BlitImageHelper::BlitImageHelper(const VKDevice& device_, VKScheduler& scheduler_,
+                                 StateTracker& state_tracker_, VKDescriptorPool& descriptor_pool)
+    : device{device_}, scheduler{scheduler_}, state_tracker{state_tracker_},
+      one_texture_set_layout(device.GetLogical().CreateDescriptorSetLayout(
+          ONE_TEXTURE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO)),
+      two_textures_set_layout(device.GetLogical().CreateDescriptorSetLayout(
+          TWO_TEXTURES_DESCRIPTOR_SET_LAYOUT_CREATE_INFO)),
+      one_texture_descriptor_allocator(descriptor_pool, *one_texture_set_layout),
+      two_textures_descriptor_allocator(descriptor_pool, *two_textures_set_layout),
+      one_texture_pipeline_layout(device.GetLogical().CreatePipelineLayout(
+          PipelineLayoutCreateInfo(one_texture_set_layout.address()))),
+      two_textures_pipeline_layout(device.GetLogical().CreatePipelineLayout(
+          PipelineLayoutCreateInfo(two_textures_set_layout.address()))),
+      full_screen_vert(BuildShader(device, FULL_SCREEN_TRIANGLE_VERT_SPV)),
+      blit_color_to_color_frag(BuildShader(device, VULKAN_BLIT_COLOR_FLOAT_FRAG_SPV)),
+      convert_depth_to_float_frag(BuildShader(device, CONVERT_DEPTH_TO_FLOAT_FRAG_SPV)),
+      convert_float_to_depth_frag(BuildShader(device, CONVERT_FLOAT_TO_DEPTH_FRAG_SPV)),
+      linear_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_LINEAR>)),
+      nearest_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_NEAREST>)) {
+    if (device.IsExtShaderStencilExportSupported()) {
+        blit_depth_stencil_frag = BuildShader(device, VULKAN_BLIT_DEPTH_STENCIL_FRAG_SPV);
+    }
+}
+
+BlitImageHelper::~BlitImageHelper() = default;
+
+void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, const ImageView& src_image_view,
+                                const std::array<Offset2D, 2>& dst_region,
+                                const std::array<Offset2D, 2>& src_region,
+                                Tegra::Engines::Fermi2D::Filter filter,
+                                Tegra::Engines::Fermi2D::Operation operation) {
+    const bool is_linear = filter == Tegra::Engines::Fermi2D::Filter::Bilinear;
+    const BlitImagePipelineKey key{
+        .renderpass = dst_framebuffer->RenderPass(),
+        .operation = operation,
+    };
+    const VkPipelineLayout layout = *one_texture_pipeline_layout;
+    const VkImageView src_view = src_image_view.Handle(ImageViewType::e2D);
+    const VkSampler sampler = is_linear ? *linear_sampler : *nearest_sampler;
+    const VkPipeline pipeline = FindOrEmplacePipeline(key);
+    const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit();
+    scheduler.RequestRenderpass(dst_framebuffer);
+    scheduler.Record([dst_region, src_region, pipeline, layout, sampler, src_view, descriptor_set,
+                      &device = device](vk::CommandBuffer cmdbuf) {
+        // TODO: Barriers
+        UpdateOneTextureDescriptorSet(device, descriptor_set, sampler, src_view);
+        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
+        cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set,
+                                  nullptr);
+        BindBlitState(cmdbuf, layout, dst_region, src_region);
+        cmdbuf.Draw(3, 1, 0, 0);
+    });
+    scheduler.InvalidateState();
+}
+
+void BlitImageHelper::BlitDepthStencil(const Framebuffer* dst_framebuffer,
+                                       VkImageView src_depth_view, VkImageView src_stencil_view,
+                                       const std::array<Offset2D, 2>& dst_region,
+                                       const std::array<Offset2D, 2>& src_region,
+                                       Tegra::Engines::Fermi2D::Filter filter,
+                                       Tegra::Engines::Fermi2D::Operation operation) {
+    ASSERT(filter == Tegra::Engines::Fermi2D::Filter::Point);
+    ASSERT(operation == Tegra::Engines::Fermi2D::Operation::SrcCopy);
+
+    const VkPipelineLayout layout = *two_textures_pipeline_layout;
+    const VkSampler sampler = *nearest_sampler;
+    const VkPipeline pipeline = BlitDepthStencilPipeline(dst_framebuffer->RenderPass());
+    const VkDescriptorSet descriptor_set = two_textures_descriptor_allocator.Commit();
+    scheduler.RequestRenderpass(dst_framebuffer);
+    scheduler.Record([dst_region, src_region, pipeline, layout, sampler, src_depth_view,
+                      src_stencil_view, descriptor_set,
+                      &device = device](vk::CommandBuffer cmdbuf) {
+        // TODO: Barriers
+        UpdateTwoTexturesDescriptorSet(device, descriptor_set, sampler, src_depth_view,
+                                       src_stencil_view);
+        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
+        cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set,
+                                  nullptr);
+        BindBlitState(cmdbuf, layout, dst_region, src_region);
+        cmdbuf.Draw(3, 1, 0, 0);
+    });
+    scheduler.InvalidateState();
+}
+
+void BlitImageHelper::ConvertD32ToR32(const Framebuffer* dst_framebuffer,
+                                      const ImageView& src_image_view) {
+    ConvertDepthToColorPipeline(convert_d32_to_r32_pipeline, dst_framebuffer->RenderPass());
+    Convert(*convert_d32_to_r32_pipeline, dst_framebuffer, src_image_view);
+}
+
+void BlitImageHelper::ConvertR32ToD32(const Framebuffer* dst_framebuffer,
+                                      const ImageView& src_image_view) {
+
+    ConvertColorToDepthPipeline(convert_r32_to_d32_pipeline, dst_framebuffer->RenderPass());
+    Convert(*convert_r32_to_d32_pipeline, dst_framebuffer, src_image_view);
+}
+
+void BlitImageHelper::ConvertD16ToR16(const Framebuffer* dst_framebuffer,
+                                      const ImageView& src_image_view) {
+    ConvertDepthToColorPipeline(convert_d16_to_r16_pipeline, dst_framebuffer->RenderPass());
+    Convert(*convert_d16_to_r16_pipeline, dst_framebuffer, src_image_view);
+}
+
+void BlitImageHelper::ConvertR16ToD16(const Framebuffer* dst_framebuffer,
+                                      const ImageView& src_image_view) {
+    ConvertColorToDepthPipeline(convert_r16_to_d16_pipeline, dst_framebuffer->RenderPass());
+    Convert(*convert_r16_to_d16_pipeline, dst_framebuffer, src_image_view);
+}
+
+void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
+                              const ImageView& src_image_view) {
+    const VkPipelineLayout layout = *one_texture_pipeline_layout;
+    const VkImageView src_view = src_image_view.Handle(ImageViewType::e2D);
+    const VkSampler sampler = *nearest_sampler;
+    const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit();
+    const VkExtent2D extent{
+        .width = src_image_view.size.width,
+        .height = src_image_view.size.height,
+    };
+    scheduler.RequestRenderpass(dst_framebuffer);
+    scheduler.Record([pipeline, layout, sampler, src_view, descriptor_set, extent,
+                      &device = device](vk::CommandBuffer cmdbuf) {
+        const VkOffset2D offset{
+            .x = 0,
+            .y = 0,
+        };
+        const VkViewport viewport{
+            .x = 0.0f,
+            .y = 0.0f,
+            .width = static_cast<float>(extent.width),
+            .height = static_cast<float>(extent.height),
+            .minDepth = 0.0f,
+            .maxDepth = 0.0f,
+        };
+        const VkRect2D scissor{
+            .offset = offset,
+            .extent = extent,
+        };
+        const PushConstants push_constants{
+            .tex_scale = {viewport.width, viewport.height},
+            .tex_offset = {0.0f, 0.0f},
+        };
+        UpdateOneTextureDescriptorSet(device, descriptor_set, sampler, src_view);
+
+        // TODO: Barriers
+        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
+        cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set,
+                                  nullptr);
+        cmdbuf.SetViewport(0, viewport);
+        cmdbuf.SetScissor(0, scissor);
+        cmdbuf.PushConstants(layout, VK_SHADER_STAGE_VERTEX_BIT, push_constants);
+        cmdbuf.Draw(3, 1, 0, 0);
+    });
+    scheduler.InvalidateState();
+}
+
+VkPipeline BlitImageHelper::FindOrEmplacePipeline(const BlitImagePipelineKey& key) {
+    const auto it = std::ranges::find(blit_color_keys, key);
+    if (it != blit_color_keys.end()) {
+        return *blit_color_pipelines[std::distance(blit_color_keys.begin(), it)];
+    }
+    blit_color_keys.push_back(key);
+
+    const std::array stages = MakeStages(*full_screen_vert, *blit_color_to_color_frag);
+    const VkPipelineColorBlendAttachmentState blend_attachment{
+        .blendEnable = VK_FALSE,
+        .srcColorBlendFactor = VK_BLEND_FACTOR_ZERO,
+        .dstColorBlendFactor = VK_BLEND_FACTOR_ZERO,
+        .colorBlendOp = VK_BLEND_OP_ADD,
+        .srcAlphaBlendFactor = VK_BLEND_FACTOR_ZERO,
+        .dstAlphaBlendFactor = VK_BLEND_FACTOR_ZERO,
+        .alphaBlendOp = VK_BLEND_OP_ADD,
+        .colorWriteMask = VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT |
+                          VK_COLOR_COMPONENT_B_BIT | VK_COLOR_COMPONENT_A_BIT,
+    };
+    // TODO: programmable blending
+    const VkPipelineColorBlendStateCreateInfo color_blend_create_info{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .logicOpEnable = VK_FALSE,
+        .logicOp = VK_LOGIC_OP_CLEAR,
+        .attachmentCount = 1,
+        .pAttachments = &blend_attachment,
+        .blendConstants = {0.0f, 0.0f, 0.0f, 0.0f},
+    };
+    blit_color_pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({
+        .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .stageCount = static_cast<u32>(stages.size()),
+        .pStages = stages.data(),
+        .pVertexInputState = &PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
+        .pInputAssemblyState = &PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
+        .pTessellationState = nullptr,
+        .pViewportState = &PIPELINE_VIEWPORT_STATE_CREATE_INFO,
+        .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
+        .pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
+        .pDepthStencilState = nullptr,
+        .pColorBlendState = &color_blend_create_info,
+        .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
+        .layout = *one_texture_pipeline_layout,
+        .renderPass = key.renderpass,
+        .subpass = 0,
+        .basePipelineHandle = VK_NULL_HANDLE,
+        .basePipelineIndex = 0,
+    }));
+    return *blit_color_pipelines.back();
+}
+
+VkPipeline BlitImageHelper::BlitDepthStencilPipeline(VkRenderPass renderpass) {
+    if (blit_depth_stencil_pipeline) {
+        return *blit_depth_stencil_pipeline;
+    }
+    const std::array stages = MakeStages(*full_screen_vert, *blit_depth_stencil_frag);
+    blit_depth_stencil_pipeline = device.GetLogical().CreateGraphicsPipeline({
+        .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .stageCount = static_cast<u32>(stages.size()),
+        .pStages = stages.data(),
+        .pVertexInputState = &PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
+        .pInputAssemblyState = &PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
+        .pTessellationState = nullptr,
+        .pViewportState = &PIPELINE_VIEWPORT_STATE_CREATE_INFO,
+        .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
+        .pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
+        .pDepthStencilState = &PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
+        .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_EMPTY_CREATE_INFO,
+        .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
+        .layout = *two_textures_pipeline_layout,
+        .renderPass = renderpass,
+        .subpass = 0,
+        .basePipelineHandle = VK_NULL_HANDLE,
+        .basePipelineIndex = 0,
+    });
+    return *blit_depth_stencil_pipeline;
+}
+
+void BlitImageHelper::ConvertDepthToColorPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass) {
+    if (pipeline) {
+        return;
+    }
+    const std::array stages = MakeStages(*full_screen_vert, *convert_depth_to_float_frag);
+    pipeline = device.GetLogical().CreateGraphicsPipeline({
+        .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .stageCount = static_cast<u32>(stages.size()),
+        .pStages = stages.data(),
+        .pVertexInputState = &PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
+        .pInputAssemblyState = &PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
+        .pTessellationState = nullptr,
+        .pViewportState = &PIPELINE_VIEWPORT_STATE_CREATE_INFO,
+        .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
+        .pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
+        .pDepthStencilState = nullptr,
+        .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO,
+        .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
+        .layout = *one_texture_pipeline_layout,
+        .renderPass = renderpass,
+        .subpass = 0,
+        .basePipelineHandle = VK_NULL_HANDLE,
+        .basePipelineIndex = 0,
+    });
+}
+
+void BlitImageHelper::ConvertColorToDepthPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass) {
+    if (pipeline) {
+        return;
+    }
+    const std::array stages = MakeStages(*full_screen_vert, *convert_float_to_depth_frag);
+    pipeline = device.GetLogical().CreateGraphicsPipeline({
+        .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .stageCount = static_cast<u32>(stages.size()),
+        .pStages = stages.data(),
+        .pVertexInputState = &PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
+        .pInputAssemblyState = &PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
+        .pTessellationState = nullptr,
+        .pViewportState = &PIPELINE_VIEWPORT_STATE_CREATE_INFO,
+        .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
+        .pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
+        .pDepthStencilState = &PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
+        .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_EMPTY_CREATE_INFO,
+        .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
+        .layout = *one_texture_pipeline_layout,
+        .renderPass = renderpass,
+        .subpass = 0,
+        .basePipelineHandle = VK_NULL_HANDLE,
+        .basePipelineIndex = 0,
+    });
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/blit_image.h

@@ -0,0 +1,97 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <compare>
+
+#include "video_core/engines/fermi_2d.h"
+#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
+#include "video_core/renderer_vulkan/wrapper.h"
+#include "video_core/texture_cache/types.h"
+
+namespace Vulkan {
+
+using VideoCommon::Offset2D;
+
+class VKDevice;
+class VKScheduler;
+class StateTracker;
+
+class Framebuffer;
+class ImageView;
+
+struct BlitImagePipelineKey {
+    constexpr auto operator<=>(const BlitImagePipelineKey&) const noexcept = default;
+
+    VkRenderPass renderpass;
+    Tegra::Engines::Fermi2D::Operation operation;
+};
+
+class BlitImageHelper {
+public:
+    explicit BlitImageHelper(const VKDevice& device, VKScheduler& scheduler,
+                             StateTracker& state_tracker, VKDescriptorPool& descriptor_pool);
+    ~BlitImageHelper();
+
+    void BlitColor(const Framebuffer* dst_framebuffer, const ImageView& src_image_view,
+                   const std::array<Offset2D, 2>& dst_region,
+                   const std::array<Offset2D, 2>& src_region,
+                   Tegra::Engines::Fermi2D::Filter filter,
+                   Tegra::Engines::Fermi2D::Operation operation);
+
+    void BlitDepthStencil(const Framebuffer* dst_framebuffer, VkImageView src_depth_view,
+                          VkImageView src_stencil_view, const std::array<Offset2D, 2>& dst_region,
+                          const std::array<Offset2D, 2>& src_region,
+                          Tegra::Engines::Fermi2D::Filter filter,
+                          Tegra::Engines::Fermi2D::Operation operation);
+
+    void ConvertD32ToR32(const Framebuffer* dst_framebuffer, const ImageView& src_image_view);
+
+    void ConvertR32ToD32(const Framebuffer* dst_framebuffer, const ImageView& src_image_view);
+
+    void ConvertD16ToR16(const Framebuffer* dst_framebuffer, const ImageView& src_image_view);
+
+    void ConvertR16ToD16(const Framebuffer* dst_framebuffer, const ImageView& src_image_view);
+
+private:
+    void Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
+                 const ImageView& src_image_view);
+
+    [[nodiscard]] VkPipeline FindOrEmplacePipeline(const BlitImagePipelineKey& key);
+
+    [[nodiscard]] VkPipeline BlitDepthStencilPipeline(VkRenderPass renderpass);
+
+    void ConvertDepthToColorPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass);
+
+    void ConvertColorToDepthPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass);
+
+    const VKDevice& device;
+    VKScheduler& scheduler;
+    StateTracker& state_tracker;
+
+    vk::DescriptorSetLayout one_texture_set_layout;
+    vk::DescriptorSetLayout two_textures_set_layout;
+    DescriptorAllocator one_texture_descriptor_allocator;
+    DescriptorAllocator two_textures_descriptor_allocator;
+    vk::PipelineLayout one_texture_pipeline_layout;
+    vk::PipelineLayout two_textures_pipeline_layout;
+    vk::ShaderModule full_screen_vert;
+    vk::ShaderModule blit_color_to_color_frag;
+    vk::ShaderModule blit_depth_stencil_frag;
+    vk::ShaderModule convert_depth_to_float_frag;
+    vk::ShaderModule convert_float_to_depth_frag;
+    vk::Sampler linear_sampler;
+    vk::Sampler nearest_sampler;
+
+    std::vector<BlitImagePipelineKey> blit_color_keys;
+    std::vector<vk::Pipeline> blit_color_pipelines;
+    vk::Pipeline blit_depth_stencil_pipeline;
+    vk::Pipeline convert_d32_to_r32_pipeline;
+    vk::Pipeline convert_r32_to_d32_pipeline;
+    vk::Pipeline convert_d16_to_r16_pipeline;
+    vk::Pipeline convert_r16_to_d16_pipeline;
+};
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/fixed_pipeline_state.cpp

@@ -60,6 +60,7 @@ void FixedPipelineState::Fill(const Maxwell& regs, bool has_extended_dynamic_sta
     logic_op.Assign(PackLogicOp(regs.logic_op.operation));
     rasterize_enable.Assign(regs.rasterize_enable != 0 ? 1 : 0);
     topology.Assign(regs.draw.topology);
+    msaa_mode.Assign(regs.multisample_mode);
 
     raw2 = 0;
     const auto test_func =

src/video_core/renderer_vulkan/fixed_pipeline_state.h

@@ -186,6 +186,7 @@ struct FixedPipelineState {
         BitField<19, 4, u32> logic_op;
         BitField<23, 1, u32> rasterize_enable;
         BitField<24, 4, Maxwell::PrimitiveTopology> topology;
+        BitField<28, 4, Tegra::Texture::MsaaMode> msaa_mode;
     };
     union {
         u32 raw2;

src/video_core/renderer_vulkan/maxwell_to_vk.cpp

@@ -122,7 +122,7 @@ struct FormatTuple {
     {VK_FORMAT_A8B8G8R8_SINT_PACK32, Attachable | Storage},     // A8B8G8R8_SINT
     {VK_FORMAT_A8B8G8R8_UINT_PACK32, Attachable | Storage},     // A8B8G8R8_UINT
     {VK_FORMAT_R5G6B5_UNORM_PACK16, Attachable},                // R5G6B5_UNORM
-    {VK_FORMAT_B5G6R5_UNORM_PACK16, Attachable},                // B5G6R5_UNORM
+    {VK_FORMAT_B5G6R5_UNORM_PACK16},                            // B5G6R5_UNORM
     {VK_FORMAT_A1R5G5B5_UNORM_PACK16, Attachable},              // A1R5G5B5_UNORM
     {VK_FORMAT_A2B10G10R10_UNORM_PACK32, Attachable | Storage}, // A2B10G10R10_UNORM
     {VK_FORMAT_A2B10G10R10_UINT_PACK32, Attachable | Storage},  // A2B10G10R10_UINT
@@ -163,7 +163,7 @@ struct FormatTuple {
     {VK_FORMAT_R16G16_UNORM, Attachable | Storage},            // R16G16_UNORM
     {VK_FORMAT_R16G16_SFLOAT, Attachable | Storage},           // R16G16_FLOAT
     {VK_FORMAT_UNDEFINED},                                     // R16G16_UINT
-    {VK_FORMAT_UNDEFINED},                                     // R16G16_SINT
+    {VK_FORMAT_R16G16_SINT, Attachable | Storage},             // R16G16_SINT
     {VK_FORMAT_R16G16_SNORM, Attachable | Storage},            // R16G16_SNORM
     {VK_FORMAT_UNDEFINED},                                     // R32G32B32_FLOAT
     {VK_FORMAT_R8G8B8A8_SRGB, Attachable},                     // A8B8G8R8_SRGB
@@ -233,18 +233,20 @@ FormatInfo SurfaceFormat(const VKDevice& device, FormatType format_type, PixelFo
 
     // Use A8B8G8R8_UNORM on hardware that doesn't support ASTC natively
     if (!device.IsOptimalAstcSupported() && VideoCore::Surface::IsPixelFormatASTC(pixel_format)) {
-        tuple.format = VideoCore::Surface::IsPixelFormatSRGB(pixel_format)
+        const bool is_srgb = VideoCore::Surface::IsPixelFormatSRGB(pixel_format);
-                           ? VK_FORMAT_A8B8G8R8_SRGB_PACK32
+        tuple.format = is_srgb ? VK_FORMAT_A8B8G8R8_SRGB_PACK32 : VK_FORMAT_A8B8G8R8_UNORM_PACK32;
-                           : VK_FORMAT_A8B8G8R8_UNORM_PACK32;
     }
     const bool attachable = tuple.usage & Attachable;
     const bool storage = tuple.usage & Storage;
 
     VkFormatFeatureFlags usage;
-    if (format_type == FormatType::Buffer) {
+    switch (format_type) {
+    case FormatType::Buffer:
         usage =
             VK_FORMAT_FEATURE_STORAGE_TEXEL_BUFFER_BIT | VK_FORMAT_FEATURE_UNIFORM_TEXEL_BUFFER_BIT;
-    } else {
+        break;
+    case FormatType::Linear:
+    case FormatType::Optimal:
         usage = VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT | VK_FORMAT_FEATURE_TRANSFER_DST_BIT |
                 VK_FORMAT_FEATURE_TRANSFER_SRC_BIT;
         if (attachable) {
@@ -254,6 +256,7 @@ FormatInfo SurfaceFormat(const VKDevice& device, FormatType format_type, PixelFo
         if (storage) {
             usage |= VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT;
         }
+        break;
     }
     return {device.GetSupportedFormat(tuple.format, usage, format_type), attachable, storage};
 }
@@ -724,4 +727,17 @@ VkViewportCoordinateSwizzleNV ViewportSwizzle(Maxwell::ViewportSwizzle swizzle)
     return {};
 }
 
+VkSamplerReductionMode SamplerReduction(Tegra::Texture::SamplerReduction reduction) {
+    switch (reduction) {
+    case Tegra::Texture::SamplerReduction::WeightedAverage:
+        return VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT;
+    case Tegra::Texture::SamplerReduction::Min:
+        return VK_SAMPLER_REDUCTION_MODE_MIN_EXT;
+    case Tegra::Texture::SamplerReduction::Max:
+        return VK_SAMPLER_REDUCTION_MODE_MAX_EXT;
+    }
+    UNREACHABLE_MSG("Invalid sampler mode={}", static_cast<int>(reduction));
+    return VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT;
+}
+
 } // namespace Vulkan::MaxwellToVK

src/video_core/renderer_vulkan/maxwell_to_vk.h

@@ -61,4 +61,6 @@ VkComponentSwizzle SwizzleSource(Tegra::Texture::SwizzleSource swizzle);
 
 VkViewportCoordinateSwizzleNV ViewportSwizzle(Maxwell::ViewportSwizzle swizzle);
 
+VkSamplerReductionMode SamplerReduction(Tegra::Texture::SamplerReduction reduction);
+
 } // namespace Vulkan::MaxwellToVK

src/video_core/renderer_vulkan/renderer_vulkan.cpp

@@ -92,9 +92,9 @@ Common::DynamicLibrary OpenVulkanLibrary() {
     return library;
 }
 
-std::pair<vk::Instance, u32> CreateInstance(
+std::pair<vk::Instance, u32> CreateInstance(Common::DynamicLibrary& library,
-    Common::DynamicLibrary& library, vk::InstanceDispatch& dld,
+                                            vk::InstanceDispatch& dld, WindowSystemType window_type,
-    WindowSystemType window_type = WindowSystemType::Headless, bool enable_layers = false) {
+                                            bool enable_debug_utils, bool enable_layers) {
     if (!library.IsOpen()) {
         LOG_ERROR(Render_Vulkan, "Vulkan library not available");
         return {};
@@ -133,7 +133,7 @@ std::pair<vk::Instance, u32> CreateInstance(
     if (window_type != Core::Frontend::WindowSystemType::Headless) {
         extensions.push_back(VK_KHR_SURFACE_EXTENSION_NAME);
     }
-    if (enable_layers) {
+    if (enable_debug_utils) {
         extensions.push_back(VK_EXT_DEBUG_UTILS_EXTENSION_NAME);
     }
     extensions.push_back(VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME);
@@ -287,7 +287,7 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
 bool RendererVulkan::Init() {
     library = OpenVulkanLibrary();
     std::tie(instance, instance_version) = CreateInstance(
-        library, dld, render_window.GetWindowInfo().type, Settings::values.renderer_debug);
+        library, dld, render_window.GetWindowInfo().type, true, Settings::values.renderer_debug);
     if (!instance || !CreateDebugCallback() || !CreateSurface() || !PickDevices()) {
         return false;
     }
@@ -447,7 +447,8 @@ void RendererVulkan::Report() const {
 std::vector<std::string> RendererVulkan::EnumerateDevices() {
     vk::InstanceDispatch dld;
     Common::DynamicLibrary library = OpenVulkanLibrary();
-    vk::Instance instance = CreateInstance(library, dld).first;
+    vk::Instance instance =
+        CreateInstance(library, dld, WindowSystemType::Headless, false, false).first;
     if (!instance) {
         return {};
     }

src/video_core/renderer_vulkan/renderer_vulkan.h

@@ -33,10 +33,9 @@ class VKDevice;
 class VKMemoryManager;
 class VKSwapchain;
 class VKScheduler;
-class VKImage;
 
 struct VKScreenInfo {
-    VKImage* image{};
+    VkImageView image_view{};
     u32 width{};
     u32 height{};
     bool is_srgb{};

src/video_core/renderer_vulkan/vk_blit_screen.cpp

@@ -16,12 +16,12 @@
 #include "core/frontend/emu_window.h"
 #include "core/memory.h"
 #include "video_core/gpu.h"
-#include "video_core/morton.h"
+#include "video_core/host_shaders/vulkan_present_frag_spv.h"
+#include "video_core/host_shaders/vulkan_present_vert_spv.h"
 #include "video_core/rasterizer_interface.h"
 #include "video_core/renderer_vulkan/renderer_vulkan.h"
 #include "video_core/renderer_vulkan/vk_blit_screen.h"
 #include "video_core/renderer_vulkan/vk_device.h"
-#include "video_core/renderer_vulkan/vk_image.h"
 #include "video_core/renderer_vulkan/vk_master_semaphore.h"
 #include "video_core/renderer_vulkan/vk_memory_manager.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
@@ -29,108 +29,12 @@
 #include "video_core/renderer_vulkan/vk_swapchain.h"
 #include "video_core/renderer_vulkan/wrapper.h"
 #include "video_core/surface.h"
+#include "video_core/textures/decoders.h"
 
 namespace Vulkan {
 
 namespace {
 
-// Generated from the "shaders/" directory, read the instructions there.
-constexpr u8 blit_vertex_code[] = {
-    0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x27, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c, 0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30,
-    0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
-    0x0f, 0x00, 0x09, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e,
-    0x00, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
-    0x25, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x0b, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x0b, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00,
-    0x0b, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
-    0x48, 0x00, 0x05, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00,
-    0x04, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x48, 0x00, 0x04, 0x00, 0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00,
-    0x48, 0x00, 0x05, 0x00, 0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x07, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x11, 0x00, 0x00, 0x00,
-    0x02, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x13, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x13, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x19, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x24, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x25, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x03, 0x00,
-    0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x16, 0x00, 0x03, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x20, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x04, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x08, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x08, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x04, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x06, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
-    0x0c, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
-    0x0c, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
-    0x0e, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00,
-    0x0e, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x18, 0x00, 0x04, 0x00,
-    0x10, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00,
-    0x11, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x12, 0x00, 0x00, 0x00,
-    0x02, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x12, 0x00, 0x00, 0x00,
-    0x13, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x14, 0x00, 0x00, 0x00,
-    0x02, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x17, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00,
-    0x19, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x1b, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x1c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x80, 0x3f, 0x20, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00,
-    0x03, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
-    0x03, 0x00, 0x00, 0x00, 0x17, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
-    0x24, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00,
-    0x25, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00,
-    0x05, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x14, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
-    0x13, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x10, 0x00, 0x00, 0x00,
-    0x16, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00,
-    0x1a, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x51, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x1d, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x51, 0x00, 0x05, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
-    0x50, 0x00, 0x07, 0x00, 0x07, 0x00, 0x00, 0x00, 0x1f, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00,
-    0x1e, 0x00, 0x00, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x91, 0x00, 0x05, 0x00,
-    0x07, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x1f, 0x00, 0x00, 0x00,
-    0x41, 0x00, 0x05, 0x00, 0x21, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00,
-    0x0f, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00, 0x22, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
-    0x3d, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x26, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
-    0x3e, 0x00, 0x03, 0x00, 0x24, 0x00, 0x00, 0x00, 0x26, 0x00, 0x00, 0x00, 0xfd, 0x00, 0x01, 0x00,
-    0x38, 0x00, 0x01, 0x00};
-
-constexpr u8 blit_fragment_code[] = {
-    0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x14, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c, 0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30,
-    0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
-    0x0f, 0x00, 0x07, 0x00, 0x04, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e,
-    0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x10, 0x00, 0x03, 0x00,
-    0x04, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
-    0x1e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x0d, 0x00, 0x00, 0x00,
-    0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x0d, 0x00, 0x00, 0x00,
-    0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x11, 0x00, 0x00, 0x00,
-    0x1e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x21, 0x00, 0x03, 0x00, 0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x16, 0x00, 0x03, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x08, 0x00, 0x00, 0x00,
-    0x03, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x08, 0x00, 0x00, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x19, 0x00, 0x09, 0x00, 0x0a, 0x00, 0x00, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1b, 0x00, 0x03, 0x00,
-    0x0b, 0x00, 0x00, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0c, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x0c, 0x00, 0x00, 0x00,
-    0x0d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x0f, 0x00, 0x00, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x10, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x10, 0x00, 0x00, 0x00,
-    0x11, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00,
-    0x05, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x00, 0x00,
-    0x0d, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x12, 0x00, 0x00, 0x00,
-    0x11, 0x00, 0x00, 0x00, 0x57, 0x00, 0x05, 0x00, 0x07, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00,
-    0x0e, 0x00, 0x00, 0x00, 0x12, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00, 0x09, 0x00, 0x00, 0x00,
-    0x13, 0x00, 0x00, 0x00, 0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00};
-
 struct ScreenRectVertex {
     ScreenRectVertex() = default;
     explicit ScreenRectVertex(f32 x, f32 y, f32 u, f32 v) : position{{x, y}}, tex_coord{{u, v}} {}
@@ -173,9 +77,9 @@ constexpr std::array<f32, 4 * 4> MakeOrthographicMatrix(f32 width, f32 height) {
     // clang-format on
 }
 
-std::size_t GetBytesPerPixel(const Tegra::FramebufferConfig& framebuffer) {
+u32 GetBytesPerPixel(const Tegra::FramebufferConfig& framebuffer) {
     using namespace VideoCore::Surface;
-    return GetBytesPerPixel(PixelFormatFromGPUPixelFormat(framebuffer.pixel_format));
+    return BytesPerBlock(PixelFormatFromGPUPixelFormat(framebuffer.pixel_format));
 }
 
 std::size_t GetSizeInBytes(const Tegra::FramebufferConfig& framebuffer) {
@@ -239,34 +143,30 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
     scheduler.Wait(resource_ticks[image_index]);
     resource_ticks[image_index] = scheduler.CurrentTick();
 
-    VKImage* blit_image = use_accelerated ? screen_info.image : raw_images[image_index].get();
+    UpdateDescriptorSet(image_index,
-
+                        use_accelerated ? screen_info.image_view : *raw_image_views[image_index]);
-    UpdateDescriptorSet(image_index, blit_image->GetPresentView());
 
     BufferData data;
     SetUniformData(data, framebuffer);
     SetVertexData(data, framebuffer);
 
     auto map = buffer_commit->Map();
-    std::memcpy(map.GetAddress(), &data, sizeof(data));
+    std::memcpy(map.Address(), &data, sizeof(data));
 
     if (!use_accelerated) {
         const u64 image_offset = GetRawImageOffset(framebuffer, image_index);
 
-        const auto pixel_format =
-            VideoCore::Surface::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format);
         const VAddr framebuffer_addr = framebuffer.address + framebuffer.offset;
-        const auto host_ptr = cpu_memory.GetPointer(framebuffer_addr);
+        const u8* const host_ptr = cpu_memory.GetPointer(framebuffer_addr);
-        rasterizer.FlushRegion(ToCacheAddr(host_ptr), GetSizeInBytes(framebuffer));
+        const size_t size_bytes = GetSizeInBytes(framebuffer);
+        rasterizer.FlushRegion(ToCacheAddr(host_ptr), size_bytes);
 
         // TODO(Rodrigo): Read this from HLE
         constexpr u32 block_height_log2 = 4;
-        VideoCore::MortonSwizzle(VideoCore::MortonSwizzleMode::MortonToLinear, pixel_format,
+        const u32 bytes_per_pixel = GetBytesPerPixel(framebuffer);
-                                 framebuffer.stride, block_height_log2, framebuffer.height, 0, 1, 1,
+        Tegra::Texture::UnswizzleTexture(
-                                 map.GetAddress() + image_offset, host_ptr);
+            std::span(map.Address() + image_offset, size_bytes), std::span(host_ptr, size_bytes),
-
+            bytes_per_pixel, framebuffer.width, framebuffer.height, 1, block_height_log2, 0);
-        blit_image->Transition(0, 1, 0, 1, VK_PIPELINE_STAGE_TRANSFER_BIT,
-                               VK_ACCESS_TRANSFER_WRITE_BIT, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
 
         const VkBufferImageCopy copy{
             .bufferOffset = image_offset,
@@ -288,15 +188,44 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
                 },
         };
         scheduler.Record(
-            [buffer = *buffer, image = *blit_image->GetHandle(), copy](vk::CommandBuffer cmdbuf) {
+            [buffer = *buffer, image = *raw_images[image_index], copy](vk::CommandBuffer cmdbuf) {
-                cmdbuf.CopyBufferToImage(buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copy);
+                const VkImageMemoryBarrier base_barrier{
+                    .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+                    .pNext = nullptr,
+                    .srcAccessMask = 0,
+                    .dstAccessMask = 0,
+                    .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
+                    .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+                    .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                    .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                    .image = image,
+                    .subresourceRange =
+                        {
+                            .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                            .baseMipLevel = 0,
+                            .levelCount = 1,
+                            .baseArrayLayer = 0,
+                            .layerCount = 1,
+                        },
+                };
+                VkImageMemoryBarrier read_barrier = base_barrier;
+                read_barrier.srcAccessMask = VK_ACCESS_HOST_WRITE_BIT;
+                read_barrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
+                read_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED;
+
+                VkImageMemoryBarrier write_barrier = base_barrier;
+                write_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
+                write_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
+
+                cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
+                                       0, read_barrier);
+                cmdbuf.CopyBufferToImage(buffer, image, VK_IMAGE_LAYOUT_GENERAL, copy);
+                cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
+                                       VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier);
             });
     }
     map.Release();
 
-    blit_image->Transition(0, 1, 0, 1, VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
-                           VK_ACCESS_SHADER_READ_BIT, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
-
     scheduler.Record([renderpass = *renderpass, framebuffer = *framebuffers[image_index],
                       descriptor_set = descriptor_sets[image_index], buffer = *buffer,
                       size = swapchain.GetSize(), pipeline = *pipeline,
@@ -304,31 +233,31 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
         const VkClearValue clear_color{
             .color = {.float32 = {0.0f, 0.0f, 0.0f, 0.0f}},
         };
-
+        const VkRenderPassBeginInfo renderpass_bi{
-        VkRenderPassBeginInfo renderpass_bi;
+            .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
-        renderpass_bi.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO;
+            .pNext = nullptr,
-        renderpass_bi.pNext = nullptr;
+            .renderPass = renderpass,
-        renderpass_bi.renderPass = renderpass;
+            .framebuffer = framebuffer,
-        renderpass_bi.framebuffer = framebuffer;
+            .renderArea =
-        renderpass_bi.renderArea.offset.x = 0;
+                {
-        renderpass_bi.renderArea.offset.y = 0;
+                    .offset = {0, 0},
-        renderpass_bi.renderArea.extent = size;
+                    .extent = size,
-        renderpass_bi.clearValueCount = 1;
+                },
-        renderpass_bi.pClearValues = &clear_color;
+            .clearValueCount = 1,
-
+            .pClearValues = &clear_color,
-        VkViewport viewport;
+        };
-        viewport.x = 0.0f;
+        const VkViewport viewport{
-        viewport.y = 0.0f;
+            .x = 0.0f,
-        viewport.width = static_cast<float>(size.width);
+            .y = 0.0f,
-        viewport.height = static_cast<float>(size.height);
+            .width = static_cast<float>(size.width),
-        viewport.minDepth = 0.0f;
+            .height = static_cast<float>(size.height),
-        viewport.maxDepth = 1.0f;
+            .minDepth = 0.0f,
-
+            .maxDepth = 1.0f,
-        VkRect2D scissor;
+        };
-        scissor.offset.x = 0;
+        const VkRect2D scissor{
-        scissor.offset.y = 0;
+            .offset = {0, 0},
-        scissor.extent = size;
+            .extent = size,
-
+        };
         cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE);
         cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
         cmdbuf.SetViewport(0, viewport);
@@ -372,8 +301,8 @@ void VKBlitScreen::RefreshResources(const Tegra::FramebufferConfig& framebuffer)
 }
 
 void VKBlitScreen::CreateShaders() {
-    vertex_shader = BuildShader(device, sizeof(blit_vertex_code), blit_vertex_code);
+    vertex_shader = BuildShader(device, VULKAN_PRESENT_VERT_SPV);
-    fragment_shader = BuildShader(device, sizeof(blit_fragment_code), blit_fragment_code);
+    fragment_shader = BuildShader(device, VULKAN_PRESENT_FRAG_SPV);
 }
 
 void VKBlitScreen::CreateSemaphores() {
@@ -420,7 +349,7 @@ void VKBlitScreen::CreateRenderPass() {
 
     const VkAttachmentReference color_attachment_ref{
         .attachment = 0,
-        .layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
+        .layout = VK_IMAGE_LAYOUT_GENERAL,
     };
 
     const VkSubpassDescription subpass_description{
@@ -735,34 +664,56 @@ void VKBlitScreen::CreateStagingBuffer(const Tegra::FramebufferConfig& framebuff
 
 void VKBlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) {
     raw_images.resize(image_count);
+    raw_image_views.resize(image_count);
     raw_buffer_commits.resize(image_count);
 
-    const VkImageCreateInfo ci{
+    for (size_t i = 0; i < image_count; ++i) {
-        .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
+        raw_images[i] = device.GetLogical().CreateImage(VkImageCreateInfo{
-        .pNext = nullptr,
+            .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
-        .flags = 0,
+            .pNext = nullptr,
-        .imageType = VK_IMAGE_TYPE_2D,
+            .flags = 0,
-        .format = GetFormat(framebuffer),
+            .imageType = VK_IMAGE_TYPE_2D,
-        .extent =
+            .format = GetFormat(framebuffer),
-            {
+            .extent =
-                .width = framebuffer.width,
+                {
-                .height = framebuffer.height,
+                    .width = framebuffer.width,
-                .depth = 1,
+                    .height = framebuffer.height,
-            },
+                    .depth = 1,
-        .mipLevels = 1,
+                },
-        .arrayLayers = 1,
+            .mipLevels = 1,
-        .samples = VK_SAMPLE_COUNT_1_BIT,
+            .arrayLayers = 1,
-        .tiling = VK_IMAGE_TILING_LINEAR,
+            .samples = VK_SAMPLE_COUNT_1_BIT,
-        .usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT,
+            .tiling = VK_IMAGE_TILING_LINEAR,
-        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+            .usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT,
-        .queueFamilyIndexCount = 0,
+            .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
-        .pQueueFamilyIndices = nullptr,
+            .queueFamilyIndexCount = 0,
-        .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+            .pQueueFamilyIndices = nullptr,
-    };
+            .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
-
+        });
-    for (std::size_t i = 0; i < image_count; ++i) {
+        raw_buffer_commits[i] = memory_manager.Commit(raw_images[i], false);
-        raw_images[i] = std::make_unique<VKImage>(device, scheduler, ci, VK_IMAGE_ASPECT_COLOR_BIT);
+        raw_image_views[i] = device.GetLogical().CreateImageView(VkImageViewCreateInfo{
-        raw_buffer_commits[i] = memory_manager.Commit(raw_images[i]->GetHandle(), false);
+            .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
+            .pNext = nullptr,
+            .flags = 0,
+            .image = *raw_images[i],
+            .viewType = VK_IMAGE_VIEW_TYPE_2D,
+            .format = GetFormat(framebuffer),
+            .components =
+                {
+                    .r = VK_COMPONENT_SWIZZLE_IDENTITY,
+                    .g = VK_COMPONENT_SWIZZLE_IDENTITY,
+                    .b = VK_COMPONENT_SWIZZLE_IDENTITY,
+                    .a = VK_COMPONENT_SWIZZLE_IDENTITY,
+                },
+            .subresourceRange =
+                {
+                    .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                    .baseMipLevel = 0,
+                    .levelCount = 1,
+                    .baseArrayLayer = 0,
+                    .layerCount = 1,
+                },
+        });
     }
 }
 
@@ -789,7 +740,7 @@ void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView imag
     const VkDescriptorImageInfo image_info{
         .sampler = *sampler,
         .imageView = image_view,
-        .imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL,
+        .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
     };
 
     const VkWriteDescriptorSet sampler_write{

src/video_core/renderer_vulkan/vk_blit_screen.h

@@ -35,7 +35,6 @@ struct ScreenInfo;
 
 class RasterizerVulkan;
 class VKDevice;
-class VKImage;
 class VKScheduler;
 class VKSwapchain;
 
@@ -110,7 +109,8 @@ private:
     std::vector<u64> resource_ticks;
 
     std::vector<vk::Semaphore> semaphores;
-    std::vector<std::unique_ptr<VKImage>> raw_images;
+    std::vector<vk::Image> raw_images;
+    std::vector<vk::ImageView> raw_image_views;
     std::vector<VKMemoryCommit> raw_buffer_commits;
     u32 raw_width = 0;
     u32 raw_height = 0;

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -31,15 +31,19 @@ constexpr VkAccessFlags UPLOAD_ACCESS_BARRIERS =
     VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_UNIFORM_READ_BIT |
     VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDEX_READ_BIT;
 
+constexpr VkAccessFlags TRANSFORM_FEEDBACK_WRITE_ACCESS =
+    VK_ACCESS_TRANSFORM_FEEDBACK_WRITE_BIT_EXT | VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_WRITE_BIT_EXT;
+
 std::unique_ptr<VKStreamBuffer> CreateStreamBuffer(const VKDevice& device, VKScheduler& scheduler) {
-    return std::make_unique<VKStreamBuffer>(device, scheduler, BUFFER_USAGE);
+    return std::make_unique<VKStreamBuffer>(device, scheduler);
 }
 
 } // Anonymous namespace
 
-Buffer::Buffer(const VKDevice& device, VKMemoryManager& memory_manager, VKScheduler& scheduler_,
+Buffer::Buffer(const VKDevice& device_, VKMemoryManager& memory_manager, VKScheduler& scheduler_,
                VKStagingBufferPool& staging_pool_, VAddr cpu_addr_, std::size_t size_)
-    : BufferBlock{cpu_addr_, size_}, scheduler{scheduler_}, staging_pool{staging_pool_} {
+    : BufferBlock{cpu_addr_, size_}, device{device_}, scheduler{scheduler_}, staging_pool{
+                                                                                 staging_pool_} {
     const VkBufferCreateInfo ci{
         .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
         .pNext = nullptr,
@@ -64,24 +68,39 @@ void Buffer::Upload(std::size_t offset, std::size_t data_size, const u8* data) {
     scheduler.RequestOutsideRenderPassOperationContext();
 
     const VkBuffer handle = Handle();
-    scheduler.Record(
+    scheduler.Record([staging = *staging.handle, handle, offset, data_size,
-        [staging = *staging.handle, handle, offset, data_size](vk::CommandBuffer cmdbuf) {
+                      &device = device](vk::CommandBuffer cmdbuf) {
-            cmdbuf.CopyBuffer(staging, handle, VkBufferCopy{0, offset, data_size});
+        const VkBufferMemoryBarrier read_barrier{
-
+            .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
-            const VkBufferMemoryBarrier barrier{
+            .pNext = nullptr,
-                .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
+            .srcAccessMask =
-                .pNext = nullptr,
+                VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT |
-                .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+                VK_ACCESS_HOST_WRITE_BIT |
-                .dstAccessMask = UPLOAD_ACCESS_BARRIERS,
+                (device.IsExtTransformFeedbackSupported() ? TRANSFORM_FEEDBACK_WRITE_ACCESS : 0),
-                .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
-                .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .buffer = handle,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .offset = offset,
+            .buffer = handle,
-                .size = data_size,
+            .offset = offset,
-            };
+            .size = data_size,
-            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0, {},
+        };
-                                   barrier, {});
+        const VkBufferMemoryBarrier write_barrier{
-        });
+            .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .dstAccessMask = UPLOAD_ACCESS_BARRIERS,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .buffer = handle,
+            .offset = offset,
+            .size = data_size,
+        };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
+                               0, read_barrier);
+        cmdbuf.CopyBuffer(staging, handle, VkBufferCopy{0, offset, data_size});
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0,
+                               write_barrier);
+    });
 }
 
 void Buffer::Download(std::size_t offset, std::size_t data_size, u8* data) {
@@ -150,8 +169,10 @@ void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst
 VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer_,
                              Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_,
                              const VKDevice& device_, VKMemoryManager& memory_manager_,
-                             VKScheduler& scheduler_, VKStagingBufferPool& staging_pool_)
+                             VKScheduler& scheduler_, VKStreamBuffer& stream_buffer_,
-    : BufferCache{rasterizer_, gpu_memory_, cpu_memory_, CreateStreamBuffer(device_, scheduler_)},
+                             VKStagingBufferPool& staging_pool_)
+    : VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer>{rasterizer_, gpu_memory_,
+                                                                 cpu_memory_, stream_buffer_},
       device{device_}, memory_manager{memory_manager_}, scheduler{scheduler_}, staging_pool{
                                                                                    staging_pool_} {}
 

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -41,6 +41,7 @@ public:
     }
 
 private:
+    const VKDevice& device;
     VKScheduler& scheduler;
     VKStagingBufferPool& staging_pool;
 
@@ -49,10 +50,11 @@ private:
 
 class VKBufferCache final : public VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer> {
 public:
-    explicit VKBufferCache(VideoCore::RasterizerInterface& rasterizer_,
+    explicit VKBufferCache(VideoCore::RasterizerInterface& rasterizer,
-                           Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_,
+                           Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory,
-                           const VKDevice& device_, VKMemoryManager& memory_manager_,
+                           const VKDevice& device, VKMemoryManager& memory_manager,
-                           VKScheduler& scheduler_, VKStagingBufferPool& staging_pool_);
+                           VKScheduler& scheduler, VKStreamBuffer& stream_buffer,
+                           VKStagingBufferPool& staging_pool);
     ~VKBufferCache();
 
     BufferInfo GetEmptyBuffer(std::size_t size) override;

src/video_core/renderer_vulkan/vk_compute_pass.cpp

@@ -10,6 +10,9 @@
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/common_types.h"
+#include "video_core/host_shaders/vulkan_quad_array_comp_spv.h"
+#include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h"
+#include "video_core/host_shaders/vulkan_uint8_comp_spv.h"
 #include "video_core/renderer_vulkan/vk_compute_pass.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
 #include "video_core/renderer_vulkan/vk_device.h"
@@ -22,99 +25,6 @@ namespace Vulkan {
 
 namespace {
 
-// Quad array SPIR-V module. Generated from the "shaders/" directory, read the instructions there.
-constexpr u8 quad_array[] = {
-    0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x54, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c, 0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30,
-    0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
-    0x0f, 0x00, 0x06, 0x00, 0x05, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e,
-    0x00, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x10, 0x00, 0x06, 0x00, 0x04, 0x00, 0x00, 0x00,
-    0x11, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
-    0x47, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
-    0x47, 0x00, 0x04, 0x00, 0x13, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
-    0x48, 0x00, 0x05, 0x00, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x14, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
-    0x47, 0x00, 0x04, 0x00, 0x16, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x47, 0x00, 0x04, 0x00, 0x16, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x48, 0x00, 0x05, 0x00, 0x29, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x29, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x47, 0x00, 0x04, 0x00, 0x4a, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
-    0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x03, 0x00, 0x03, 0x00, 0x00, 0x00,
-    0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x03, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0d, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00, 0x13, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x1e, 0x00, 0x03, 0x00, 0x14, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
-    0x15, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
-    0x15, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
-    0x18, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x14, 0x00, 0x02, 0x00,
-    0x1b, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x29, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x20, 0x00, 0x04, 0x00, 0x2a, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00,
-    0x3b, 0x00, 0x04, 0x00, 0x2a, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
-    0x2b, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x20, 0x00, 0x04, 0x00, 0x2d, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x31, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
-    0x1c, 0x00, 0x04, 0x00, 0x34, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00,
-    0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x35, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
-    0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x37, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
-    0x2c, 0x00, 0x09, 0x00, 0x34, 0x00, 0x00, 0x00, 0x38, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00,
-    0x35, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00,
-    0x37, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x3a, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
-    0x34, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x44, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x47, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x49, 0x00, 0x00, 0x00,
-    0x00, 0x04, 0x00, 0x00, 0x2c, 0x00, 0x06, 0x00, 0x09, 0x00, 0x00, 0x00, 0x4a, 0x00, 0x00, 0x00,
-    0x49, 0x00, 0x00, 0x00, 0x35, 0x00, 0x00, 0x00, 0x35, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00,
-    0x02, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
-    0xf8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x3a, 0x00, 0x00, 0x00,
-    0x3b, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x4c, 0x00, 0x00, 0x00,
-    0xf8, 0x00, 0x02, 0x00, 0x4c, 0x00, 0x00, 0x00, 0xf6, 0x00, 0x04, 0x00, 0x4b, 0x00, 0x00, 0x00,
-    0x4e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x4d, 0x00, 0x00, 0x00,
-    0xf8, 0x00, 0x02, 0x00, 0x4d, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x0d, 0x00, 0x00, 0x00,
-    0x0e, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x12, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00,
-    0x44, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x17, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
-    0x17, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00,
-    0x19, 0x00, 0x00, 0x00, 0xae, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
-    0x12, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00, 0xf7, 0x00, 0x03, 0x00, 0x1e, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0xfa, 0x00, 0x04, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00,
-    0x1e, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00,
-    0x4b, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1e, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00,
-    0x21, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x21, 0x00, 0x00, 0x00, 0xf5, 0x00, 0x07, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00,
-    0x48, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0xb0, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00,
-    0x27, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0xf6, 0x00, 0x04, 0x00,
-    0x23, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xfa, 0x00, 0x04, 0x00,
-    0x27, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00,
-    0x22, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x2d, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00,
-    0x2b, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x2f, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x32, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x31, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x33, 0x00, 0x00, 0x00, 0x2f, 0x00, 0x00, 0x00, 0x32, 0x00, 0x00, 0x00,
-    0x3e, 0x00, 0x03, 0x00, 0x3b, 0x00, 0x00, 0x00, 0x38, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00,
-    0x07, 0x00, 0x00, 0x00, 0x3c, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00,
-    0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x00, 0x00, 0x3c, 0x00, 0x00, 0x00,
-    0x80, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x00, 0x00, 0x33, 0x00, 0x00, 0x00,
-    0x3d, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00,
-    0x12, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x44, 0x00, 0x00, 0x00,
-    0x45, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00,
-    0x3e, 0x00, 0x03, 0x00, 0x45, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x47, 0x00, 0x00, 0x00,
-    0xf9, 0x00, 0x02, 0x00, 0x21, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x23, 0x00, 0x00, 0x00,
-    0xf9, 0x00, 0x02, 0x00, 0x4b, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x4e, 0x00, 0x00, 0x00,
-    0xf9, 0x00, 0x02, 0x00, 0x4c, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x4b, 0x00, 0x00, 0x00,
-    0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00,
-};
-
 VkDescriptorSetLayoutBinding BuildQuadArrayPassDescriptorSetLayoutBinding() {
     return {
         .binding = 0,
@@ -144,208 +54,6 @@ VkPushConstantRange BuildComputePushConstantRange(std::size_t size) {
     };
 }
 
-// Uint8 SPIR-V module. Generated from the "shaders/" directory.
-constexpr u8 uint8_pass[] = {
-    0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x2f, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00,
-    0x51, 0x11, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x61, 0x11, 0x00, 0x00, 0x0a, 0x00, 0x07, 0x00,
-    0x53, 0x50, 0x56, 0x5f, 0x4b, 0x48, 0x52, 0x5f, 0x31, 0x36, 0x62, 0x69, 0x74, 0x5f, 0x73, 0x74,
-    0x6f, 0x72, 0x61, 0x67, 0x65, 0x00, 0x00, 0x00, 0x0a, 0x00, 0x07, 0x00, 0x53, 0x50, 0x56, 0x5f,
-    0x4b, 0x48, 0x52, 0x5f, 0x38, 0x62, 0x69, 0x74, 0x5f, 0x73, 0x74, 0x6f, 0x72, 0x61, 0x67, 0x65,
-    0x00, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c,
-    0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30, 0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x06, 0x00, 0x05, 0x00, 0x00, 0x00,
-    0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e, 0x00, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00,
-    0x10, 0x00, 0x06, 0x00, 0x04, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
-    0x0b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x12, 0x00, 0x00, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x48, 0x00, 0x04, 0x00, 0x13, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x13, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00,
-    0x13, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x15, 0x00, 0x00, 0x00,
-    0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x15, 0x00, 0x00, 0x00,
-    0x21, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x1f, 0x00, 0x00, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x48, 0x00, 0x04, 0x00, 0x20, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x20, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00,
-    0x20, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00,
-    0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00,
-    0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x2e, 0x00, 0x00, 0x00,
-    0x0b, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x21, 0x00, 0x03, 0x00, 0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
-    0x07, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
-    0x0a, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
-    0x0a, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
-    0x0d, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
-    0x11, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00,
-    0x12, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x13, 0x00, 0x00, 0x00,
-    0x12, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x14, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x13, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x14, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
-    0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x14, 0x00, 0x02, 0x00, 0x1a, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
-    0x1e, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00,
-    0x1f, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x20, 0x00, 0x00, 0x00,
-    0x1f, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x20, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
-    0x02, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x26, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x11, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x2a, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x1e, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x00, 0x00,
-    0x00, 0x04, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2d, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x06, 0x00, 0x09, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00,
-    0x2c, 0x00, 0x00, 0x00, 0x2d, 0x00, 0x00, 0x00, 0x2d, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00,
-    0x02, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
-    0xf8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00,
-    0x08, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x0d, 0x00, 0x00, 0x00,
-    0x0e, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00,
-    0x08, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x10, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x44, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x16, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
-    0x17, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0xb0, 0x00, 0x05, 0x00,
-    0x1a, 0x00, 0x00, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
-    0xf7, 0x00, 0x03, 0x00, 0x1d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xfa, 0x00, 0x04, 0x00,
-    0x1b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00,
-    0x1c, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
-    0x08, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
-    0x08, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x26, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00,
-    0x15, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
-    0x11, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x71, 0x00, 0x04, 0x00,
-    0x1e, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00,
-    0x2a, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
-    0x24, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00,
-    0xf9, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00,
-    0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00,
-};
-
-// Quad indexed SPIR-V module. Generated from the "shaders/" directory.
-constexpr u8 QUAD_INDEXED_SPV[] = {
-    0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x7c, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c, 0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30,
-    0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
-    0x0f, 0x00, 0x06, 0x00, 0x05, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e,
-    0x00, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x10, 0x00, 0x06, 0x00, 0x04, 0x00, 0x00, 0x00,
-    0x11, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
-    0x47, 0x00, 0x04, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
-    0x47, 0x00, 0x04, 0x00, 0x15, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
-    0x48, 0x00, 0x04, 0x00, 0x16, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
-    0x48, 0x00, 0x05, 0x00, 0x16, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x16, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
-    0x47, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x47, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
-    0x48, 0x00, 0x05, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x22, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
-    0x23, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x22, 0x00, 0x00, 0x00,
-    0x02, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x56, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x04, 0x00, 0x00, 0x00, 0x48, 0x00, 0x04, 0x00, 0x57, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x18, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x57, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x57, 0x00, 0x00, 0x00,
-    0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x59, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x59, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x72, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00,
-    0x19, 0x00, 0x00, 0x00, 0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x03, 0x00,
-    0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x20, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00,
-    0x07, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
-    0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x0a, 0x00, 0x00, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
-    0x0c, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
-    0x0d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0e, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x13, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00, 0x15, 0x00, 0x00, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x16, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
-    0x20, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00,
-    0x3b, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x14, 0x00, 0x02, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x21, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
-    0x24, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x25, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x26, 0x00, 0x00, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x2b, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
-    0x3b, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x3f, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x04, 0x00, 0x41, 0x00, 0x00, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x42, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x43, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x09, 0x00, 0x41, 0x00, 0x00, 0x00,
-    0x44, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00,
-    0x42, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x43, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
-    0x46, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x41, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00,
-    0x56, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x57, 0x00, 0x00, 0x00,
-    0x56, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x58, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x57, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x58, 0x00, 0x00, 0x00, 0x59, 0x00, 0x00, 0x00,
-    0x02, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x5b, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x69, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x70, 0x00, 0x00, 0x00,
-    0x00, 0x04, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x71, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x06, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x72, 0x00, 0x00, 0x00,
-    0x70, 0x00, 0x00, 0x00, 0x71, 0x00, 0x00, 0x00, 0x71, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00,
-    0x02, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
-    0xf8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x46, 0x00, 0x00, 0x00,
-    0x47, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x74, 0x00, 0x00, 0x00,
-    0xf8, 0x00, 0x02, 0x00, 0x74, 0x00, 0x00, 0x00, 0xf6, 0x00, 0x04, 0x00, 0x73, 0x00, 0x00, 0x00,
-    0x76, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x75, 0x00, 0x00, 0x00,
-    0xf8, 0x00, 0x02, 0x00, 0x75, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x0e, 0x00, 0x00, 0x00,
-    0x0f, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00,
-    0x44, 0x00, 0x05, 0x00, 0x09, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00,
-    0x19, 0x00, 0x00, 0x00, 0xaf, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
-    0x14, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00, 0xf7, 0x00, 0x03, 0x00, 0x1e, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0xfa, 0x00, 0x04, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00,
-    0x1e, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00,
-    0x73, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00,
-    0x26, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
-    0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00,
-    0xc4, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
-    0x28, 0x00, 0x00, 0x00, 0x82, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00,
-    0x2b, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0xc4, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x31, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00, 0x82, 0x00, 0x05, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x32, 0x00, 0x00, 0x00, 0x31, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
-    0xf9, 0x00, 0x02, 0x00, 0x35, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x35, 0x00, 0x00, 0x00,
-    0xf5, 0x00, 0x07, 0x00, 0x09, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00,
-    0x1e, 0x00, 0x00, 0x00, 0x6f, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0xb0, 0x00, 0x05, 0x00,
-    0x1b, 0x00, 0x00, 0x00, 0x3c, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x00, 0x00,
-    0xf6, 0x00, 0x04, 0x00, 0x37, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0xfa, 0x00, 0x04, 0x00, 0x3c, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x37, 0x00, 0x00, 0x00,
-    0xf8, 0x00, 0x02, 0x00, 0x36, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x40, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x3f, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00,
-    0x47, 0x00, 0x00, 0x00, 0x44, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x07, 0x00, 0x00, 0x00,
-    0x48, 0x00, 0x00, 0x00, 0x47, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x49, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
-    0x06, 0x00, 0x00, 0x00, 0x4a, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x49, 0x00, 0x00, 0x00,
-    0xc3, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x4e, 0x00, 0x00, 0x00, 0x4a, 0x00, 0x00, 0x00,
-    0x2e, 0x00, 0x00, 0x00, 0xc7, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x52, 0x00, 0x00, 0x00,
-    0x4a, 0x00, 0x00, 0x00, 0x32, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
-    0x54, 0x00, 0x00, 0x00, 0x52, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00,
-    0x5b, 0x00, 0x00, 0x00, 0x5c, 0x00, 0x00, 0x00, 0x59, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00,
-    0x4e, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x5d, 0x00, 0x00, 0x00,
-    0x5c, 0x00, 0x00, 0x00, 0xcb, 0x00, 0x06, 0x00, 0x09, 0x00, 0x00, 0x00, 0x62, 0x00, 0x00, 0x00,
-    0x5d, 0x00, 0x00, 0x00, 0x54, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x67, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00,
-    0x41, 0x00, 0x05, 0x00, 0x69, 0x00, 0x00, 0x00, 0x6a, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
-    0x42, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x6b, 0x00, 0x00, 0x00,
-    0x6a, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00, 0x09, 0x00, 0x00, 0x00, 0x6c, 0x00, 0x00, 0x00,
-    0x62, 0x00, 0x00, 0x00, 0x6b, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x5b, 0x00, 0x00, 0x00,
-    0x6d, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x67, 0x00, 0x00, 0x00,
-    0x3e, 0x00, 0x03, 0x00, 0x6d, 0x00, 0x00, 0x00, 0x6c, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
-    0x09, 0x00, 0x00, 0x00, 0x6f, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
-    0xf9, 0x00, 0x02, 0x00, 0x35, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x37, 0x00, 0x00, 0x00,
-    0xf9, 0x00, 0x02, 0x00, 0x73, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x76, 0x00, 0x00, 0x00,
-    0xf9, 0x00, 0x02, 0x00, 0x74, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x73, 0x00, 0x00, 0x00,
-    0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00,
-};
-
 std::array<VkDescriptorSetLayoutBinding, 2> BuildInputOutputDescriptorSetBindings() {
     return {{
         {
@@ -381,8 +89,8 @@ VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() {
 VKComputePass::VKComputePass(const VKDevice& device, VKDescriptorPool& descriptor_pool,
                              vk::Span<VkDescriptorSetLayoutBinding> bindings,
                              vk::Span<VkDescriptorUpdateTemplateEntryKHR> templates,
-                             vk::Span<VkPushConstantRange> push_constants, std::size_t code_size,
+                             vk::Span<VkPushConstantRange> push_constants,
-                             const u8* code) {
+                             std::span<const u32> code) {
     descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout({
         .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
         .pNext = nullptr,
@@ -390,7 +98,6 @@ VKComputePass::VKComputePass(const VKDevice& device, VKDescriptorPool& descripto
         .bindingCount = bindings.size(),
         .pBindings = bindings.data(),
     });
-
     layout = device.GetLogical().CreatePipelineLayout({
         .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
         .pNext = nullptr,
@@ -400,7 +107,6 @@ VKComputePass::VKComputePass(const VKDevice& device, VKDescriptorPool& descripto
         .pushConstantRangeCount = push_constants.size(),
         .pPushConstantRanges = push_constants.data(),
     });
-
     if (!templates.empty()) {
         descriptor_template = device.GetLogical().CreateDescriptorUpdateTemplateKHR({
             .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_UPDATE_TEMPLATE_CREATE_INFO_KHR,
@@ -417,18 +123,13 @@ VKComputePass::VKComputePass(const VKDevice& device, VKDescriptorPool& descripto
 
         descriptor_allocator.emplace(descriptor_pool, *descriptor_set_layout);
     }
-
-    auto code_copy = std::make_unique<u32[]>(code_size / sizeof(u32) + 1);
-    std::memcpy(code_copy.get(), code, code_size);
-
     module = device.GetLogical().CreateShaderModule({
         .sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
         .pNext = nullptr,
         .flags = 0,
-        .codeSize = code_size,
+        .codeSize = static_cast<u32>(code.size_bytes()),
-        .pCode = code_copy.get(),
+        .pCode = code.data(),
     });
-
     pipeline = device.GetLogical().CreateComputePipeline({
         .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
         .pNext = nullptr,
@@ -467,7 +168,7 @@ QuadArrayPass::QuadArrayPass(const VKDevice& device_, VKScheduler& scheduler_,
                              VKUpdateDescriptorQueue& update_descriptor_queue_)
     : VKComputePass(device_, descriptor_pool_, BuildQuadArrayPassDescriptorSetLayoutBinding(),
                     BuildQuadArrayPassDescriptorUpdateTemplateEntry(),
-                    BuildComputePushConstantRange(sizeof(u32)), std::size(quad_array), quad_array),
+                    BuildComputePushConstantRange(sizeof(u32)), VULKAN_QUAD_ARRAY_COMP_SPV),
       scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
       update_descriptor_queue{update_descriptor_queue_} {}
 
@@ -510,12 +211,11 @@ std::pair<VkBuffer, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32
     return {*buffer.handle, 0};
 }
 
-Uint8Pass::Uint8Pass(const VKDevice& device_, VKScheduler& scheduler_,
+Uint8Pass::Uint8Pass(const VKDevice& device, VKScheduler& scheduler_,
-                     VKDescriptorPool& descriptor_pool_, VKStagingBufferPool& staging_buffer_pool_,
+                     VKDescriptorPool& descriptor_pool, VKStagingBufferPool& staging_buffer_pool_,
                      VKUpdateDescriptorQueue& update_descriptor_queue_)
-    : VKComputePass(device_, descriptor_pool_, BuildInputOutputDescriptorSetBindings(),
+    : VKComputePass(device, descriptor_pool, BuildInputOutputDescriptorSetBindings(),
-                    BuildInputOutputDescriptorUpdateTemplate(), {}, std::size(uint8_pass),
+                    BuildInputOutputDescriptorUpdateTemplate(), {}, VULKAN_UINT8_COMP_SPV),
-                    uint8_pass),
       scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
       update_descriptor_queue{update_descriptor_queue_} {}
 
@@ -561,8 +261,7 @@ QuadIndexedPass::QuadIndexedPass(const VKDevice& device_, VKScheduler& scheduler
                                  VKUpdateDescriptorQueue& update_descriptor_queue_)
     : VKComputePass(device_, descriptor_pool_, BuildInputOutputDescriptorSetBindings(),
                     BuildInputOutputDescriptorUpdateTemplate(),
-                    BuildComputePushConstantRange(sizeof(u32) * 2), std::size(QUAD_INDEXED_SPV),
+                    BuildComputePushConstantRange(sizeof(u32) * 2), VULKAN_QUAD_INDEXED_COMP_SPV),
-                    QUAD_INDEXED_SPV),
       scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
       update_descriptor_queue{update_descriptor_queue_} {}
 

src/video_core/renderer_vulkan/vk_compute_pass.h

@@ -5,6 +5,7 @@
 #pragma once
 
 #include <optional>
+#include <span>
 #include <utility>
 
 #include "common/common_types.h"
@@ -24,8 +25,7 @@ public:
     explicit VKComputePass(const VKDevice& device, VKDescriptorPool& descriptor_pool,
                            vk::Span<VkDescriptorSetLayoutBinding> bindings,
                            vk::Span<VkDescriptorUpdateTemplateEntryKHR> templates,
-                           vk::Span<VkPushConstantRange> push_constants, std::size_t code_size,
+                           vk::Span<VkPushConstantRange> push_constants, std::span<const u32> code);
-                           const u8* code);
     ~VKComputePass();
 
 protected:

src/video_core/renderer_vulkan/vk_device.cpp

@@ -46,6 +46,7 @@ constexpr std::array REQUIRED_EXTENSIONS{
     VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME,
     VK_KHR_DESCRIPTOR_UPDATE_TEMPLATE_EXTENSION_NAME,
     VK_KHR_TIMELINE_SEMAPHORE_EXTENSION_NAME,
+    VK_KHR_SAMPLER_MIRROR_CLAMP_TO_EDGE_EXTENSION_NAME,
     VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME,
     VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME,
     VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME,
@@ -122,6 +123,7 @@ std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties(
         VK_FORMAT_R16G16_UNORM,
         VK_FORMAT_R16G16_SNORM,
         VK_FORMAT_R16G16_SFLOAT,
+        VK_FORMAT_R16G16_SINT,
         VK_FORMAT_R16_UNORM,
         VK_FORMAT_R16_UINT,
         VK_FORMAT_R8G8B8A8_SRGB,
@@ -161,18 +163,32 @@ std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties(
         VK_FORMAT_BC2_SRGB_BLOCK,
         VK_FORMAT_BC3_SRGB_BLOCK,
         VK_FORMAT_BC7_SRGB_BLOCK,
+        VK_FORMAT_ASTC_4x4_UNORM_BLOCK,
         VK_FORMAT_ASTC_4x4_SRGB_BLOCK,
-        VK_FORMAT_ASTC_8x8_SRGB_BLOCK,
+        VK_FORMAT_ASTC_5x4_UNORM_BLOCK,
-        VK_FORMAT_ASTC_8x5_SRGB_BLOCK,
         VK_FORMAT_ASTC_5x4_SRGB_BLOCK,
         VK_FORMAT_ASTC_5x5_UNORM_BLOCK,
         VK_FORMAT_ASTC_5x5_SRGB_BLOCK,
-        VK_FORMAT_ASTC_10x8_UNORM_BLOCK,
+        VK_FORMAT_ASTC_6x5_UNORM_BLOCK,
-        VK_FORMAT_ASTC_10x8_SRGB_BLOCK,
+        VK_FORMAT_ASTC_6x5_SRGB_BLOCK,
         VK_FORMAT_ASTC_6x6_UNORM_BLOCK,
         VK_FORMAT_ASTC_6x6_SRGB_BLOCK,
+        VK_FORMAT_ASTC_8x5_UNORM_BLOCK,
+        VK_FORMAT_ASTC_8x5_SRGB_BLOCK,
+        VK_FORMAT_ASTC_8x6_UNORM_BLOCK,
+        VK_FORMAT_ASTC_8x6_SRGB_BLOCK,
+        VK_FORMAT_ASTC_8x8_UNORM_BLOCK,
+        VK_FORMAT_ASTC_8x8_SRGB_BLOCK,
+        VK_FORMAT_ASTC_10x5_UNORM_BLOCK,
+        VK_FORMAT_ASTC_10x5_SRGB_BLOCK,
+        VK_FORMAT_ASTC_10x6_UNORM_BLOCK,
+        VK_FORMAT_ASTC_10x6_SRGB_BLOCK,
+        VK_FORMAT_ASTC_10x8_UNORM_BLOCK,
+        VK_FORMAT_ASTC_10x8_SRGB_BLOCK,
         VK_FORMAT_ASTC_10x10_UNORM_BLOCK,
         VK_FORMAT_ASTC_10x10_SRGB_BLOCK,
+        VK_FORMAT_ASTC_12x10_UNORM_BLOCK,
+        VK_FORMAT_ASTC_12x10_SRGB_BLOCK,
         VK_FORMAT_ASTC_12x12_UNORM_BLOCK,
         VK_FORMAT_ASTC_12x12_SRGB_BLOCK,
         VK_FORMAT_ASTC_8x6_UNORM_BLOCK,
@@ -192,7 +208,7 @@ std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties(
 
 VKDevice::VKDevice(VkInstance instance_, u32 instance_version_, vk::PhysicalDevice physical_,
                    VkSurfaceKHR surface, const vk::InstanceDispatch& dld_)
-    : dld{dld_}, physical{physical_}, properties{physical.GetProperties()},
+    : instance{instance_}, dld{dld_}, physical{physical_}, properties{physical.GetProperties()},
       instance_version{instance_version_}, format_properties{GetFormatProperties(physical, dld)} {
     SetupFamilies(surface);
     SetupFeatures();
@@ -214,7 +230,7 @@ bool VKDevice::Create() {
     features2.features = {
         .robustBufferAccess = false,
         .fullDrawIndexUint32 = false,
-        .imageCubeArray = false,
+        .imageCubeArray = true,
         .independentBlend = true,
         .geometryShader = true,
         .tessellationShader = true,
@@ -242,7 +258,7 @@ bool VKDevice::Create() {
         .shaderTessellationAndGeometryPointSize = false,
         .shaderImageGatherExtended = true,
         .shaderStorageImageExtendedFormats = false,
-        .shaderStorageImageMultisample = false,
+        .shaderStorageImageMultisample = true,
         .shaderStorageImageReadWithoutFormat = is_formatless_image_load_supported,
         .shaderStorageImageWriteWithoutFormat = true,
         .shaderUniformBufferArrayDynamicIndexing = false,
@@ -268,7 +284,6 @@ bool VKDevice::Create() {
         .variableMultisampleRate = false,
         .inheritedQueries = false,
     };
-
     VkPhysicalDeviceTimelineSemaphoreFeaturesKHR timeline_semaphore{
         .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES_KHR,
         .pNext = nullptr,
@@ -380,6 +395,20 @@ bool VKDevice::Create() {
         LOG_INFO(Render_Vulkan, "Device doesn't support extended dynamic state");
     }
 
+    VkPhysicalDeviceRobustness2FeaturesEXT robustness2;
+    if (ext_robustness2) {
+        robustness2 = {
+            .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_FEATURES_EXT,
+            .pNext = nullptr,
+            .robustBufferAccess2 = false,
+            .robustImageAccess2 = true,
+            .nullDescriptor = true,
+        };
+        SetNext(next, robustness2);
+    } else {
+        LOG_INFO(Render_Vulkan, "Device doesn't support robustness2");
+    }
+
     if (!ext_depth_range_unrestricted) {
         LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted");
     }
@@ -405,7 +434,14 @@ bool VKDevice::Create() {
     }
 
     CollectTelemetryParameters();
+    CollectToolingInfo();
 
+    if (ext_extended_dynamic_state && driver_id == VK_DRIVER_ID_MESA_RADV) {
+        LOG_WARNING(
+            Render_Vulkan,
+            "Blacklisting RADV for VK_EXT_extended_dynamic state, likely due to a bug in yuzu");
+        ext_extended_dynamic_state = false;
+    }
     if (ext_extended_dynamic_state && IsRDNA(properties.deviceName, driver_id)) {
         // AMD's proprietary driver supports VK_EXT_extended_dynamic_state but on RDNA devices it
         // seems to cause stability issues
@@ -458,7 +494,7 @@ void VKDevice::ReportLoss() const {
     LOG_CRITICAL(Render_Vulkan, "Device loss occured!");
 
     // Wait for the log to flush and for Nsight Aftermath to dump the results
-    std::this_thread::sleep_for(std::chrono::seconds{3});
+    std::this_thread::sleep_for(std::chrono::seconds{15});
 }
 
 void VKDevice::SaveShader(const std::vector<u32>& spirv) const {
@@ -499,6 +535,16 @@ bool VKDevice::IsOptimalAstcSupported(const VkPhysicalDeviceFeatures& features)
     return true;
 }
 
+bool VKDevice::TestDepthStencilBlits() const {
+    static constexpr VkFormatFeatureFlags required_features =
+        VK_FORMAT_FEATURE_BLIT_SRC_BIT | VK_FORMAT_FEATURE_BLIT_DST_BIT;
+    const auto test_features = [](VkFormatProperties props) {
+        return (props.optimalTilingFeatures & required_features) == required_features;
+    };
+    return test_features(format_properties.at(VK_FORMAT_D32_SFLOAT_S8_UINT)) &&
+           test_features(format_properties.at(VK_FORMAT_D24_UNORM_S8_UINT));
+}
+
 bool VKDevice::IsFormatSupported(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage,
                                  FormatType format_type) const {
     const auto it = format_properties.find(wanted_format);
@@ -569,6 +615,7 @@ bool VKDevice::IsSuitable(vk::PhysicalDevice physical, VkSurfaceKHR surface) {
     const auto features{physical.GetFeatures()};
     const std::array feature_report = {
         std::make_pair(features.vertexPipelineStoresAndAtomics, "vertexPipelineStoresAndAtomics"),
+        std::make_pair(features.imageCubeArray, "imageCubeArray"),
         std::make_pair(features.independentBlend, "independentBlend"),
         std::make_pair(features.depthClamp, "depthClamp"),
         std::make_pair(features.samplerAnisotropy, "samplerAnisotropy"),
@@ -580,6 +627,7 @@ bool VKDevice::IsSuitable(vk::PhysicalDevice physical, VkSurfaceKHR surface) {
         std::make_pair(features.occlusionQueryPrecise, "occlusionQueryPrecise"),
         std::make_pair(features.fragmentStoresAndAtomics, "fragmentStoresAndAtomics"),
         std::make_pair(features.shaderImageGatherExtended, "shaderImageGatherExtended"),
+        std::make_pair(features.shaderStorageImageMultisample, "shaderStorageImageMultisample"),
         std::make_pair(features.shaderStorageImageWriteWithoutFormat,
                        "shaderStorageImageWriteWithoutFormat"),
     };
@@ -608,6 +656,7 @@ std::vector<const char*> VKDevice::LoadExtensions() {
     bool has_ext_transform_feedback{};
     bool has_ext_custom_border_color{};
     bool has_ext_extended_dynamic_state{};
+    bool has_ext_robustness2{};
     for (const VkExtensionProperties& extension : physical.EnumerateDeviceExtensionProperties()) {
         const auto test = [&](std::optional<std::reference_wrapper<bool>> status, const char* name,
                               bool push) {
@@ -627,11 +676,15 @@ std::vector<const char*> VKDevice::LoadExtensions() {
         test(has_khr_shader_float16_int8, VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME, false);
         test(ext_depth_range_unrestricted, VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME, true);
         test(ext_index_type_uint8, VK_EXT_INDEX_TYPE_UINT8_EXTENSION_NAME, true);
+        test(ext_sampler_filter_minmax, VK_EXT_SAMPLER_FILTER_MINMAX_EXTENSION_NAME, true);
         test(ext_shader_viewport_index_layer, VK_EXT_SHADER_VIEWPORT_INDEX_LAYER_EXTENSION_NAME,
              true);
+        test(ext_tooling_info, VK_EXT_TOOLING_INFO_EXTENSION_NAME, true);
+        test(ext_shader_stencil_export, VK_EXT_SHADER_STENCIL_EXPORT_EXTENSION_NAME, true);
         test(has_ext_transform_feedback, VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME, false);
         test(has_ext_custom_border_color, VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME, false);
         test(has_ext_extended_dynamic_state, VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME, false);
+        test(has_ext_robustness2, VK_EXT_ROBUSTNESS_2_EXTENSION_NAME, false);
         if (instance_version >= VK_API_VERSION_1_1) {
             test(has_ext_subgroup_size_control, VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME, false);
         }
@@ -733,6 +786,18 @@ std::vector<const char*> VKDevice::LoadExtensions() {
         }
     }
 
+    if (has_ext_robustness2) {
+        VkPhysicalDeviceRobustness2FeaturesEXT robustness2;
+        robustness2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_FEATURES_EXT;
+        robustness2.pNext = nullptr;
+        features.pNext = &robustness2;
+        physical.GetFeatures2KHR(features);
+        if (robustness2.nullDescriptor && robustness2.robustImageAccess2) {
+            extensions.push_back(VK_EXT_ROBUSTNESS_2_EXTENSION_NAME);
+            ext_robustness2 = true;
+        }
+    }
+
     return extensions;
 }
 
@@ -764,6 +829,7 @@ void VKDevice::SetupFamilies(VkSurfaceKHR surface) {
 void VKDevice::SetupFeatures() {
     const auto supported_features{physical.GetFeatures()};
     is_formatless_image_load_supported = supported_features.shaderStorageImageReadWithoutFormat;
+    is_blit_depth_stencil_supported = TestDepthStencilBlits();
     is_optimal_astc_supported = IsOptimalAstcSupported(supported_features);
 }
 
@@ -794,6 +860,32 @@ void VKDevice::CollectTelemetryParameters() {
     }
 }
 
+void VKDevice::CollectToolingInfo() {
+    if (!ext_tooling_info) {
+        return;
+    }
+    const auto vkGetPhysicalDeviceToolPropertiesEXT =
+        reinterpret_cast<PFN_vkGetPhysicalDeviceToolPropertiesEXT>(
+            dld.vkGetInstanceProcAddr(instance, "vkGetPhysicalDeviceToolPropertiesEXT"));
+    if (!vkGetPhysicalDeviceToolPropertiesEXT) {
+        return;
+    }
+    u32 tool_count = 0;
+    if (vkGetPhysicalDeviceToolPropertiesEXT(physical, &tool_count, nullptr) != VK_SUCCESS) {
+        return;
+    }
+    std::vector<VkPhysicalDeviceToolPropertiesEXT> tools(tool_count);
+    if (vkGetPhysicalDeviceToolPropertiesEXT(physical, &tool_count, tools.data()) != VK_SUCCESS) {
+        return;
+    }
+    for (const VkPhysicalDeviceToolPropertiesEXT& tool : tools) {
+        const std::string_view name = tool.name;
+        LOG_INFO(Render_Vulkan, "{}", name);
+        has_renderdoc = has_renderdoc || name == "RenderDoc";
+        has_nsight_graphics = has_nsight_graphics || name == "NVIDIA Nsight Graphics";
+    }
+}
+
 std::vector<VkDeviceQueueCreateInfo> VKDevice::GetDeviceQueueCreateInfos() const {
     static constexpr float QUEUE_PRIORITY = 1.0f;
 

src/video_core/renderer_vulkan/vk_device.h

@@ -157,6 +157,11 @@ public:
         return is_formatless_image_load_supported;
     }
 
+    /// Returns true when blitting from and to depth stencil images is supported.
+    bool IsBlitDepthStencilSupported() const {
+        return is_blit_depth_stencil_supported;
+    }
+
     /// Returns true if the device supports VK_NV_viewport_swizzle.
     bool IsNvViewportSwizzleSupported() const {
         return nv_viewport_swizzle;
@@ -172,6 +177,11 @@ public:
         return ext_index_type_uint8;
     }
 
+    /// Returns true if the device supports VK_EXT_sampler_filter_minmax.
+    bool IsExtSamplerFilterMinmaxSupported() const {
+        return ext_sampler_filter_minmax;
+    }
+
     /// Returns true if the device supports VK_EXT_depth_range_unrestricted.
     bool IsExtDepthRangeUnrestrictedSupported() const {
         return ext_depth_range_unrestricted;
@@ -197,6 +207,16 @@ public:
         return ext_extended_dynamic_state;
     }
 
+    /// Returns true if the device supports VK_EXT_shader_stencil_export.
+    bool IsExtShaderStencilExportSupported() const {
+        return ext_shader_stencil_export;
+    }
+
+    /// Returns true when a known debugging tool is attached.
+    bool HasDebuggingToolAttached() const {
+        return has_renderdoc || has_nsight_graphics;
+    }
+
     /// Returns the vendor name reported from Vulkan.
     std::string_view GetVendorName() const {
         return vendor_name;
@@ -228,16 +248,23 @@ private:
     /// Collects telemetry information from the device.
     void CollectTelemetryParameters();
 
+    /// Collects information about attached tools.
+    void CollectToolingInfo();
+
     /// Returns a list of queue initialization descriptors.
     std::vector<VkDeviceQueueCreateInfo> GetDeviceQueueCreateInfos() const;
 
     /// Returns true if ASTC textures are natively supported.
     bool IsOptimalAstcSupported(const VkPhysicalDeviceFeatures& features) const;
 
+    /// Returns true if the device natively supports blitting depth stencil images.
+    bool TestDepthStencilBlits() const;
+
     /// Returns true if a format is supported.
     bool IsFormatSupported(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage,
                            FormatType format_type) const;
 
+    VkInstance instance;                    ///< Vulkan instance.
     vk::DeviceDispatch dld;                 ///< Device function pointers.
     vk::PhysicalDevice physical;            ///< Physical device.
     VkPhysicalDeviceProperties properties;  ///< Device properties.
@@ -253,15 +280,22 @@ private:
     bool is_float16_supported{};            ///< Support for float16 arithmetics.
     bool is_warp_potentially_bigger{};      ///< Host warp size can be bigger than guest.
     bool is_formatless_image_load_supported{}; ///< Support for shader image read without format.
+    bool is_blit_depth_stencil_supported{};    ///< Support for blitting from and to depth stencil.
     bool nv_viewport_swizzle{};                ///< Support for VK_NV_viewport_swizzle.
     bool khr_uniform_buffer_standard_layout{}; ///< Support for std430 on UBOs.
     bool ext_index_type_uint8{};               ///< Support for VK_EXT_index_type_uint8.
+    bool ext_sampler_filter_minmax{};          ///< Support for VK_EXT_sampler_filter_minmax.
     bool ext_depth_range_unrestricted{};       ///< Support for VK_EXT_depth_range_unrestricted.
     bool ext_shader_viewport_index_layer{};    ///< Support for VK_EXT_shader_viewport_index_layer.
+    bool ext_tooling_info{};                   ///< Support for VK_EXT_tooling_info.
     bool ext_transform_feedback{};             ///< Support for VK_EXT_transform_feedback.
     bool ext_custom_border_color{};            ///< Support for VK_EXT_custom_border_color.
     bool ext_extended_dynamic_state{};         ///< Support for VK_EXT_extended_dynamic_state.
+    bool ext_robustness2{};                    ///< Support for VK_EXT_robustness2.
+    bool ext_shader_stencil_export{};          ///< Support for VK_EXT_shader_stencil_export.
     bool nv_device_diagnostics_config{};       ///< Support for VK_NV_device_diagnostics_config.
+    bool has_renderdoc{};                      ///< Has RenderDoc attached
+    bool has_nsight_graphics{};                ///< Has Nsight Graphics attached
 
     // Asynchronous Graphics Pipeline setting
     bool use_asynchronous_shaders{}; ///< Setting to use asynchronous shaders/graphics pipeline

src/video_core/renderer_vulkan/vk_fence_manager.cpp

@@ -73,10 +73,9 @@ bool InnerFence::IsEventSignalled() const {
 }
 
 VKFenceManager::VKFenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_,
-                               Tegra::MemoryManager& memory_manager_,
+                               Tegra::MemoryManager& memory_manager_, TextureCache& texture_cache_,
-                               VKTextureCache& texture_cache_, VKBufferCache& buffer_cache_,
+                               VKBufferCache& buffer_cache_, VKQueryCache& query_cache_,
-                               VKQueryCache& query_cache_, const VKDevice& device_,
+                               const VKDevice& device_, VKScheduler& scheduler_)
-                               VKScheduler& scheduler_)
     : GenericFenceManager{rasterizer_, gpu_, texture_cache_, buffer_cache_, query_cache_},
       device{device_}, scheduler{scheduler_} {}
 

src/video_core/renderer_vulkan/vk_fence_manager.h

@@ -8,6 +8,7 @@
 
 #include "video_core/fence_manager.h"
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
+#include "video_core/renderer_vulkan/vk_texture_cache.h"
 #include "video_core/renderer_vulkan/wrapper.h"
 
 namespace Core {
@@ -24,7 +25,6 @@ class VKBufferCache;
 class VKDevice;
 class VKQueryCache;
 class VKScheduler;
-class VKTextureCache;
 
 class InnerFence : public VideoCommon::FenceBase {
 public:
@@ -51,12 +51,12 @@ private:
 using Fence = std::shared_ptr<InnerFence>;
 
 using GenericFenceManager =
-    VideoCommon::FenceManager<Fence, VKTextureCache, VKBufferCache, VKQueryCache>;
+    VideoCommon::FenceManager<Fence, TextureCache, VKBufferCache, VKQueryCache>;
 
 class VKFenceManager final : public GenericFenceManager {
 public:
     explicit VKFenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_,
-                            Tegra::MemoryManager& memory_manager_, VKTextureCache& texture_cache_,
+                            Tegra::MemoryManager& memory_manager_, TextureCache& texture_cache_,
                             VKBufferCache& buffer_cache_, VKQueryCache& query_cache_,
                             const VKDevice& device_, VKScheduler& scheduler_);
 

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -15,7 +15,6 @@
 #include "video_core/renderer_vulkan/vk_device.h"
 #include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
 #include "video_core/renderer_vulkan/vk_pipeline_cache.h"
-#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
 #include "video_core/renderer_vulkan/vk_update_descriptor.h"
 #include "video_core/renderer_vulkan/wrapper.h"
@@ -69,23 +68,45 @@ VkViewportSwizzleNV UnpackViewportSwizzle(u16 swizzle) {
     };
 }
 
+VkSampleCountFlagBits ConvertMsaaMode(Tegra::Texture::MsaaMode msaa_mode) {
+    switch (msaa_mode) {
+    case Tegra::Texture::MsaaMode::Msaa1x1:
+        return VK_SAMPLE_COUNT_1_BIT;
+    case Tegra::Texture::MsaaMode::Msaa2x1:
+    case Tegra::Texture::MsaaMode::Msaa2x1_D3D:
+        return VK_SAMPLE_COUNT_2_BIT;
+    case Tegra::Texture::MsaaMode::Msaa2x2:
+    case Tegra::Texture::MsaaMode::Msaa2x2_VC4:
+    case Tegra::Texture::MsaaMode::Msaa2x2_VC12:
+        return VK_SAMPLE_COUNT_4_BIT;
+    case Tegra::Texture::MsaaMode::Msaa4x2:
+    case Tegra::Texture::MsaaMode::Msaa4x2_D3D:
+    case Tegra::Texture::MsaaMode::Msaa4x2_VC8:
+    case Tegra::Texture::MsaaMode::Msaa4x2_VC24:
+        return VK_SAMPLE_COUNT_8_BIT;
+    case Tegra::Texture::MsaaMode::Msaa4x4:
+        return VK_SAMPLE_COUNT_16_BIT;
+    default:
+        UNREACHABLE_MSG("Invalid msaa_mode={}", static_cast<int>(msaa_mode));
+        return VK_SAMPLE_COUNT_1_BIT;
+    }
+}
+
 } // Anonymous namespace
 
 VKGraphicsPipeline::VKGraphicsPipeline(const VKDevice& device_, VKScheduler& scheduler_,
                                        VKDescriptorPool& descriptor_pool_,
                                        VKUpdateDescriptorQueue& update_descriptor_queue_,
-                                       VKRenderPassCache& renderpass_cache_,
+                                       const GraphicsPipelineCacheKey& key,
-                                       const GraphicsPipelineCacheKey& key_,
+                                       vk::Span<VkDescriptorSetLayoutBinding> bindings,
-                                       vk::Span<VkDescriptorSetLayoutBinding> bindings_,
+                                       const SPIRVProgram& program, u32 num_color_buffers)
-                                       const SPIRVProgram& program_)
+    : device{device_}, scheduler{scheduler_}, cache_key{key}, hash{cache_key.Hash()},
-    : device{device_}, scheduler{scheduler_}, cache_key{key_}, hash{cache_key.Hash()},
+      descriptor_set_layout{CreateDescriptorSetLayout(bindings)},
-      descriptor_set_layout{CreateDescriptorSetLayout(bindings_)},
       descriptor_allocator{descriptor_pool_, *descriptor_set_layout},
       update_descriptor_queue{update_descriptor_queue_}, layout{CreatePipelineLayout()},
-      descriptor_template{CreateDescriptorUpdateTemplate(program_)}, modules{CreateShaderModules(
+      descriptor_template{CreateDescriptorUpdateTemplate(program)},
-                                                                         program_)},
+      modules(CreateShaderModules(program)),
-      renderpass{renderpass_cache_.GetRenderPass(cache_key.renderpass_params)},
+      pipeline(CreatePipeline(program, cache_key.renderpass, num_color_buffers)) {}
-      pipeline{CreatePipeline(cache_key.renderpass_params, program_)} {}
 
 VKGraphicsPipeline::~VKGraphicsPipeline() = default;
 
@@ -179,8 +200,9 @@ std::vector<vk::ShaderModule> VKGraphicsPipeline::CreateShaderModules(
     return shader_modules;
 }
 
-vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpass_params,
+vk::Pipeline VKGraphicsPipeline::CreatePipeline(const SPIRVProgram& program,
-                                                const SPIRVProgram& program) const {
+                                                VkRenderPass renderpass,
+                                                u32 num_color_buffers) const {
     const auto& state = cache_key.fixed_state;
     const auto& viewport_swizzles = state.viewport_swizzles;
 
@@ -290,8 +312,7 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa
     };
 
     std::array<VkViewportSwizzleNV, Maxwell::NumViewports> swizzles;
-    std::transform(viewport_swizzles.begin(), viewport_swizzles.end(), swizzles.begin(),
+    std::ranges::transform(viewport_swizzles, swizzles.begin(), UnpackViewportSwizzle);
-                   UnpackViewportSwizzle);
     VkPipelineViewportSwizzleStateCreateInfoNV swizzle_ci{
         .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_SWIZZLE_STATE_CREATE_INFO_NV,
         .pNext = nullptr,
@@ -326,7 +347,7 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa
         .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
         .pNext = nullptr,
         .flags = 0,
-        .rasterizationSamples = VK_SAMPLE_COUNT_1_BIT,
+        .rasterizationSamples = ConvertMsaaMode(state.msaa_mode),
         .sampleShadingEnable = VK_FALSE,
         .minSampleShading = 0.0f,
         .pSampleMask = nullptr,
@@ -352,8 +373,7 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa
     };
 
     std::array<VkPipelineColorBlendAttachmentState, Maxwell::NumRenderTargets> cb_attachments;
-    const auto num_attachments = static_cast<std::size_t>(renderpass_params.num_color_attachments);
+    for (std::size_t index = 0; index < num_color_buffers; ++index) {
-    for (std::size_t index = 0; index < num_attachments; ++index) {
         static constexpr std::array COMPONENT_TABLE{
             VK_COLOR_COMPONENT_R_BIT,
             VK_COLOR_COMPONENT_G_BIT,
@@ -387,7 +407,7 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa
         .flags = 0,
         .logicOpEnable = VK_FALSE,
         .logicOp = VK_LOGIC_OP_COPY,
-        .attachmentCount = static_cast<u32>(num_attachments),
+        .attachmentCount = num_color_buffers,
         .pAttachments = cb_attachments.data(),
         .blendConstants = {},
     };
@@ -447,8 +467,7 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa
             stage_ci.pNext = &subgroup_size_ci;
         }
     }
-
+    return device.GetLogical().CreateGraphicsPipeline(VkGraphicsPipelineCreateInfo{
-    const VkGraphicsPipelineCreateInfo ci{
         .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
         .pNext = nullptr,
         .flags = 0,
@@ -468,8 +487,7 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa
         .subpass = 0,
         .basePipelineHandle = nullptr,
         .basePipelineIndex = 0,
-    };
+    });
-    return device.GetLogical().CreateGraphicsPipeline(ci);
 }
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_graphics_pipeline.h

@@ -8,10 +8,10 @@
 #include <optional>
 #include <vector>
 
+#include "common/common_types.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_vulkan/fixed_pipeline_state.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
-#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
 #include "video_core/renderer_vulkan/vk_shader_decompiler.h"
 #include "video_core/renderer_vulkan/wrapper.h"
 
@@ -20,8 +20,7 @@ namespace Vulkan {
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 
 struct GraphicsPipelineCacheKey {
-    RenderPassParams renderpass_params;
+    VkRenderPass renderpass;
-    u32 padding;
     std::array<GPUVAddr, Maxwell::MaxShaderProgram> shaders;
     FixedPipelineState fixed_state;
 
@@ -34,7 +33,7 @@ struct GraphicsPipelineCacheKey {
     }
 
     std::size_t Size() const noexcept {
-        return sizeof(renderpass_params) + sizeof(padding) + sizeof(shaders) + fixed_state.Size();
+        return sizeof(renderpass) + sizeof(shaders) + fixed_state.Size();
     }
 };
 static_assert(std::has_unique_object_representations_v<GraphicsPipelineCacheKey>);
@@ -43,7 +42,6 @@ static_assert(std::is_trivially_constructible_v<GraphicsPipelineCacheKey>);
 
 class VKDescriptorPool;
 class VKDevice;
-class VKRenderPassCache;
 class VKScheduler;
 class VKUpdateDescriptorQueue;
 
@@ -52,12 +50,11 @@ using SPIRVProgram = std::array<std::optional<SPIRVShader>, Maxwell::MaxShaderSt
 class VKGraphicsPipeline final {
 public:
     explicit VKGraphicsPipeline(const VKDevice& device_, VKScheduler& scheduler_,
-                                VKDescriptorPool& descriptor_pool_,
+                                VKDescriptorPool& descriptor_pool,
                                 VKUpdateDescriptorQueue& update_descriptor_queue_,
-                                VKRenderPassCache& renderpass_cache_,
+                                const GraphicsPipelineCacheKey& key,
-                                const GraphicsPipelineCacheKey& key_,
+                                vk::Span<VkDescriptorSetLayoutBinding> bindings,
-                                vk::Span<VkDescriptorSetLayoutBinding> bindings_,
+                                const SPIRVProgram& program, u32 num_color_buffers);
-                                const SPIRVProgram& program_);
     ~VKGraphicsPipeline();
 
     VkDescriptorSet CommitDescriptorSet();
@@ -70,10 +67,6 @@ public:
         return *layout;
     }
 
-    VkRenderPass GetRenderPass() const {
-        return renderpass;
-    }
-
     GraphicsPipelineCacheKey GetCacheKey() const {
         return cache_key;
     }
@@ -89,8 +82,8 @@ private:
 
     std::vector<vk::ShaderModule> CreateShaderModules(const SPIRVProgram& program) const;
 
-    vk::Pipeline CreatePipeline(const RenderPassParams& renderpass_params,
+    vk::Pipeline CreatePipeline(const SPIRVProgram& program, VkRenderPass renderpass,
-                                const SPIRVProgram& program) const;
+                                u32 num_color_buffers) const;
 
     const VKDevice& device;
     VKScheduler& scheduler;
@@ -104,7 +97,6 @@ private:
     vk::DescriptorUpdateTemplateKHR descriptor_template;
     std::vector<vk::ShaderModule> modules;
 
-    VkRenderPass renderpass;
     vk::Pipeline pipeline;
 };
 

src/video_core/renderer_vulkan/vk_image.cpp

@@ -1,135 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <memory>
-#include <vector>
-
-#include "common/assert.h"
-#include "video_core/renderer_vulkan/vk_device.h"
-#include "video_core/renderer_vulkan/vk_image.h"
-#include "video_core/renderer_vulkan/vk_scheduler.h"
-#include "video_core/renderer_vulkan/wrapper.h"
-
-namespace Vulkan {
-
-VKImage::VKImage(const VKDevice& device_, VKScheduler& scheduler_,
-                 const VkImageCreateInfo& image_ci_, VkImageAspectFlags aspect_mask_)
-    : device{device_}, scheduler{scheduler_}, format{image_ci_.format}, aspect_mask{aspect_mask_},
-      image_num_layers{image_ci_.arrayLayers}, image_num_levels{image_ci_.mipLevels} {
-    UNIMPLEMENTED_IF_MSG(image_ci_.queueFamilyIndexCount != 0,
-                         "Queue family tracking is not implemented");
-
-    image = device_.GetLogical().CreateImage(image_ci_);
-
-    const u32 num_ranges = image_num_layers * image_num_levels;
-    barriers.resize(num_ranges);
-    subrange_states.resize(num_ranges, {{}, image_ci_.initialLayout});
-}
-
-VKImage::~VKImage() = default;
-
-void VKImage::Transition(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels,
-                         VkPipelineStageFlags new_stage_mask, VkAccessFlags new_access,
-                         VkImageLayout new_layout) {
-    if (!HasChanged(base_layer, num_layers, base_level, num_levels, new_access, new_layout)) {
-        return;
-    }
-
-    std::size_t cursor = 0;
-    for (u32 layer_it = 0; layer_it < num_layers; ++layer_it) {
-        for (u32 level_it = 0; level_it < num_levels; ++level_it, ++cursor) {
-            const u32 layer = base_layer + layer_it;
-            const u32 level = base_level + level_it;
-            auto& state = GetSubrangeState(layer, level);
-            auto& barrier = barriers[cursor];
-            barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER;
-            barrier.pNext = nullptr;
-            barrier.srcAccessMask = state.access;
-            barrier.dstAccessMask = new_access;
-            barrier.oldLayout = state.layout;
-            barrier.newLayout = new_layout;
-            barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-            barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-            barrier.image = *image;
-            barrier.subresourceRange.aspectMask = aspect_mask;
-            barrier.subresourceRange.baseMipLevel = level;
-            barrier.subresourceRange.levelCount = 1;
-            barrier.subresourceRange.baseArrayLayer = layer;
-            barrier.subresourceRange.layerCount = 1;
-            state.access = new_access;
-            state.layout = new_layout;
-        }
-    }
-
-    scheduler.RequestOutsideRenderPassOperationContext();
-
-    scheduler.Record([barriers = barriers, cursor](vk::CommandBuffer cmdbuf) {
-        // TODO(Rodrigo): Implement a way to use the latest stage across subresources.
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
-                               VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, {}, {},
-                               vk::Span(barriers.data(), cursor));
-    });
-}
-
-bool VKImage::HasChanged(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels,
-                         VkAccessFlags new_access, VkImageLayout new_layout) noexcept {
-    const bool is_full_range = base_layer == 0 && num_layers == image_num_layers &&
-                               base_level == 0 && num_levels == image_num_levels;
-    if (!is_full_range) {
-        state_diverged = true;
-    }
-
-    if (!state_diverged) {
-        auto& state = GetSubrangeState(0, 0);
-        if (state.access != new_access || state.layout != new_layout) {
-            return true;
-        }
-    }
-
-    for (u32 layer_it = 0; layer_it < num_layers; ++layer_it) {
-        for (u32 level_it = 0; level_it < num_levels; ++level_it) {
-            const u32 layer = base_layer + layer_it;
-            const u32 level = base_level + level_it;
-            auto& state = GetSubrangeState(layer, level);
-            if (state.access != new_access || state.layout != new_layout) {
-                return true;
-            }
-        }
-    }
-    return false;
-}
-
-void VKImage::CreatePresentView() {
-    // Image type has to be 2D to be presented.
-    present_view = device.GetLogical().CreateImageView({
-        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .image = *image,
-        .viewType = VK_IMAGE_VIEW_TYPE_2D,
-        .format = format,
-        .components =
-            {
-                .r = VK_COMPONENT_SWIZZLE_IDENTITY,
-                .g = VK_COMPONENT_SWIZZLE_IDENTITY,
-                .b = VK_COMPONENT_SWIZZLE_IDENTITY,
-                .a = VK_COMPONENT_SWIZZLE_IDENTITY,
-            },
-        .subresourceRange =
-            {
-                .aspectMask = aspect_mask,
-                .baseMipLevel = 0,
-                .levelCount = 1,
-                .baseArrayLayer = 0,
-                .layerCount = 1,
-            },
-    });
-}
-
-VKImage::SubrangeState& VKImage::GetSubrangeState(u32 layer, u32 level) noexcept {
-    return subrange_states[static_cast<std::size_t>(layer * image_num_levels) +
-                           static_cast<std::size_t>(level)];
-}
-
-} // namespace Vulkan

src/video_core/renderer_vulkan/vk_image.h

@@ -1,84 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <memory>
-#include <vector>
-
-#include "common/common_types.h"
-#include "video_core/renderer_vulkan/wrapper.h"
-
-namespace Vulkan {
-
-class VKDevice;
-class VKScheduler;
-
-class VKImage {
-public:
-    explicit VKImage(const VKDevice& device_, VKScheduler& scheduler_,
-                     const VkImageCreateInfo& image_ci_, VkImageAspectFlags aspect_mask_);
-    ~VKImage();
-
-    /// Records in the passed command buffer an image transition and updates the state of the image.
-    void Transition(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels,
-                    VkPipelineStageFlags new_stage_mask, VkAccessFlags new_access,
-                    VkImageLayout new_layout);
-
-    /// Returns a view compatible with presentation, the image has to be 2D.
-    VkImageView GetPresentView() {
-        if (!present_view) {
-            CreatePresentView();
-        }
-        return *present_view;
-    }
-
-    /// Returns the Vulkan image handler.
-    const vk::Image& GetHandle() const {
-        return image;
-    }
-
-    /// Returns the Vulkan format for this image.
-    VkFormat GetFormat() const {
-        return format;
-    }
-
-    /// Returns the Vulkan aspect mask.
-    VkImageAspectFlags GetAspectMask() const {
-        return aspect_mask;
-    }
-
-private:
-    struct SubrangeState final {
-        VkAccessFlags access = 0;                         ///< Current access bits.
-        VkImageLayout layout = VK_IMAGE_LAYOUT_UNDEFINED; ///< Current image layout.
-    };
-
-    bool HasChanged(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels,
-                    VkAccessFlags new_access, VkImageLayout new_layout) noexcept;
-
-    /// Creates a presentation view.
-    void CreatePresentView();
-
-    /// Returns the subrange state for a layer and layer.
-    SubrangeState& GetSubrangeState(u32 layer, u32 level) noexcept;
-
-    const VKDevice& device; ///< Device handler.
-    VKScheduler& scheduler; ///< Device scheduler.
-
-    const VkFormat format;                ///< Vulkan format.
-    const VkImageAspectFlags aspect_mask; ///< Vulkan aspect mask.
-    const u32 image_num_layers;           ///< Number of layers.
-    const u32 image_num_levels;           ///< Number of mipmap levels.
-
-    vk::Image image;            ///< Image handle.
-    vk::ImageView present_view; ///< Image view compatible with presentation.
-
-    std::vector<VkImageMemoryBarrier> barriers; ///< Pool of barriers.
-    std::vector<SubrangeState> subrange_states; ///< Current subrange state.
-
-    bool state_diverged = false; ///< True when subresources mismatch in layout.
-};
-
-} // namespace Vulkan

src/video_core/renderer_vulkan/vk_memory_manager.cpp

@@ -216,7 +216,7 @@ VKMemoryCommitImpl::~VKMemoryCommitImpl() {
 }
 
 MemoryMap VKMemoryCommitImpl::Map(u64 size, u64 offset_) const {
-    return MemoryMap{this, memory.Map(interval.first + offset_, size)};
+    return MemoryMap(this, std::span<u8>(memory.Map(interval.first + offset_, size), size));
 }
 
 void VKMemoryCommitImpl::Unmap() const {

src/video_core/renderer_vulkan/vk_memory_manager.h

@@ -5,6 +5,7 @@
 #pragma once
 
 #include <memory>
+#include <span>
 #include <utility>
 #include <vector>
 #include "common/common_types.h"
@@ -93,8 +94,8 @@ private:
 /// Holds ownership of a memory map.
 class MemoryMap final {
 public:
-    explicit MemoryMap(const VKMemoryCommitImpl* commit_, u8* address_)
+    explicit MemoryMap(const VKMemoryCommitImpl* commit_, std::span<u8> span_)
-        : commit{commit_}, address{address_} {}
+        : commit{commit_}, span{span_} {}
 
     ~MemoryMap() {
         if (commit) {
@@ -108,19 +109,24 @@ public:
         commit = nullptr;
     }
 
+    /// Returns a span to the memory map.
+    [[nodiscard]] std::span<u8> Span() const noexcept {
+        return span;
+    }
+
     /// Returns the address of the memory map.
-    u8* GetAddress() const {
+    [[nodiscard]] u8* Address() const noexcept {
-        return address;
+        return span.data();
     }
 
     /// Returns the address of the memory map;
-    operator u8*() const {
+    [[nodiscard]] operator u8*() const noexcept {
-        return address;
+        return span.data();
     }
 
 private:
     const VKMemoryCommitImpl* commit{}; ///< Mapped memory commit.
-    u8* address{};                      ///< Address to the mapped memory.
+    std::span<u8> span;                 ///< Address to the mapped memory.
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -8,6 +8,7 @@
 #include <vector>
 
 #include "common/bit_cast.h"
+#include "common/cityhash.h"
 #include "common/microprofile.h"
 #include "core/core.h"
 #include "core/memory.h"
@@ -22,7 +23,6 @@
 #include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
 #include "video_core/renderer_vulkan/vk_pipeline_cache.h"
 #include "video_core/renderer_vulkan/vk_rasterizer.h"
-#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
 #include "video_core/renderer_vulkan/vk_update_descriptor.h"
 #include "video_core/renderer_vulkan/wrapper.h"
@@ -52,7 +52,9 @@ constexpr VkDescriptorType STORAGE_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_TEX
 constexpr VkDescriptorType STORAGE_IMAGE = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
 
 constexpr VideoCommon::Shader::CompilerSettings compiler_settings{
-    VideoCommon::Shader::CompileDepth::FullDecompile};
+    .depth = VideoCommon::Shader::CompileDepth::FullDecompile,
+    .disable_else_derivation = true,
+};
 
 constexpr std::size_t GetStageFromProgram(std::size_t program) {
     return program == 0 ? 0 : program - 1;
@@ -149,12 +151,11 @@ VKPipelineCache::VKPipelineCache(RasterizerVulkan& rasterizer_, Tegra::GPU& gpu_
                                  Tegra::Engines::KeplerCompute& kepler_compute_,
                                  Tegra::MemoryManager& gpu_memory_, const VKDevice& device_,
                                  VKScheduler& scheduler_, VKDescriptorPool& descriptor_pool_,
-                                 VKUpdateDescriptorQueue& update_descriptor_queue_,
+                                 VKUpdateDescriptorQueue& update_descriptor_queue_)
-                                 VKRenderPassCache& renderpass_cache_)
+    : VideoCommon::ShaderCache<Shader>{rasterizer_}, gpu{gpu_}, maxwell3d{maxwell3d_},
-    : ShaderCache{rasterizer_}, gpu{gpu_}, maxwell3d{maxwell3d_}, kepler_compute{kepler_compute_},
+      kepler_compute{kepler_compute_}, gpu_memory{gpu_memory_}, device{device_},
-      gpu_memory{gpu_memory_}, device{device_}, scheduler{scheduler_},
+      scheduler{scheduler_}, descriptor_pool{descriptor_pool_}, update_descriptor_queue{
-      descriptor_pool{descriptor_pool_}, update_descriptor_queue{update_descriptor_queue_},
+                                                                    update_descriptor_queue_} {}
-      renderpass_cache{renderpass_cache_} {}
 
 VKPipelineCache::~VKPipelineCache() = default;
 
@@ -199,7 +200,8 @@ std::array<Shader*, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {
 }
 
 VKGraphicsPipeline* VKPipelineCache::GetGraphicsPipeline(
-    const GraphicsPipelineCacheKey& key, VideoCommon::Shader::AsyncShaders& async_shaders) {
+    const GraphicsPipelineCacheKey& key, u32 num_color_buffers,
+    VideoCommon::Shader::AsyncShaders& async_shaders) {
     MICROPROFILE_SCOPE(Vulkan_PipelineCache);
 
     if (last_graphics_pipeline && last_graphics_key == key) {
@@ -215,8 +217,8 @@ VKGraphicsPipeline* VKPipelineCache::GetGraphicsPipeline(
             LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
             const auto [program, bindings] = DecompileShaders(key.fixed_state);
             async_shaders.QueueVulkanShader(this, device, scheduler, descriptor_pool,
-                                            update_descriptor_queue, renderpass_cache, bindings,
+                                            update_descriptor_queue, bindings, program, key,
-                                            program, key);
+                                            num_color_buffers);
         }
         last_graphics_pipeline = pair->second.get();
         return last_graphics_pipeline;
@@ -229,8 +231,8 @@ VKGraphicsPipeline* VKPipelineCache::GetGraphicsPipeline(
         LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
         const auto [program, bindings] = DecompileShaders(key.fixed_state);
         entry = std::make_unique<VKGraphicsPipeline>(device, scheduler, descriptor_pool,
-                                                     update_descriptor_queue, renderpass_cache, key,
+                                                     update_descriptor_queue, key, bindings,
-                                                     bindings, program);
+                                                     program, num_color_buffers);
         gpu.ShaderNotify().MarkShaderComplete();
     }
     last_graphics_pipeline = entry.get();

src/video_core/renderer_vulkan/vk_pipeline_cache.h

@@ -19,7 +19,6 @@
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_vulkan/fixed_pipeline_state.h"
 #include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
-#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
 #include "video_core/renderer_vulkan/vk_shader_decompiler.h"
 #include "video_core/renderer_vulkan/wrapper.h"
 #include "video_core/shader/async_shaders.h"
@@ -119,18 +118,18 @@ private:
 
 class VKPipelineCache final : public VideoCommon::ShaderCache<Shader> {
 public:
-    explicit VKPipelineCache(RasterizerVulkan& rasterizer_, Tegra::GPU& gpu_,
+    explicit VKPipelineCache(RasterizerVulkan& rasterizer, Tegra::GPU& gpu,
-                             Tegra::Engines::Maxwell3D& maxwell3d_,
+                             Tegra::Engines::Maxwell3D& maxwell3d,
-                             Tegra::Engines::KeplerCompute& kepler_compute_,
+                             Tegra::Engines::KeplerCompute& kepler_compute,
-                             Tegra::MemoryManager& gpu_memory_, const VKDevice& device_,
+                             Tegra::MemoryManager& gpu_memory, const VKDevice& device,
-                             VKScheduler& scheduler_, VKDescriptorPool& descriptor_pool_,
+                             VKScheduler& scheduler, VKDescriptorPool& descriptor_pool,
-                             VKUpdateDescriptorQueue& update_descriptor_queue_,
+                             VKUpdateDescriptorQueue& update_descriptor_queue);
-                             VKRenderPassCache& renderpass_cache_);
     ~VKPipelineCache() override;
 
     std::array<Shader*, Maxwell::MaxShaderProgram> GetShaders();
 
     VKGraphicsPipeline* GetGraphicsPipeline(const GraphicsPipelineCacheKey& key,
+                                            u32 num_color_buffers,
                                             VideoCommon::Shader::AsyncShaders& async_shaders);
 
     VKComputePipeline& GetComputePipeline(const ComputePipelineCacheKey& key);
@@ -153,7 +152,6 @@ private:
     VKScheduler& scheduler;
     VKDescriptorPool& descriptor_pool;
     VKUpdateDescriptorQueue& update_descriptor_queue;
-    VKRenderPassCache& renderpass_cache;
 
     std::unique_ptr<Shader> null_shader;
     std::unique_ptr<Shader> null_kernel;

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -11,11 +11,11 @@
 #include <vector>
 
 #include <boost/container/static_vector.hpp>
-#include <boost/functional/hash.hpp>
 
 #include "common/common_types.h"
 #include "video_core/rasterizer_accelerated.h"
 #include "video_core/rasterizer_interface.h"
+#include "video_core/renderer_vulkan/blit_image.h"
 #include "video_core/renderer_vulkan/fixed_pipeline_state.h"
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
 #include "video_core/renderer_vulkan/vk_compute_pass.h"
@@ -24,10 +24,9 @@
 #include "video_core/renderer_vulkan/vk_memory_manager.h"
 #include "video_core/renderer_vulkan/vk_pipeline_cache.h"
 #include "video_core/renderer_vulkan/vk_query_cache.h"
-#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
-#include "video_core/renderer_vulkan/vk_sampler_cache.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
 #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
+#include "video_core/renderer_vulkan/vk_stream_buffer.h"
 #include "video_core/renderer_vulkan/vk_texture_cache.h"
 #include "video_core/renderer_vulkan/vk_update_descriptor.h"
 #include "video_core/renderer_vulkan/wrapper.h"
@@ -49,60 +48,9 @@ namespace Vulkan {
 
 struct VKScreenInfo;
 
-using ImageViewsPack = boost::container::static_vector<VkImageView, Maxwell::NumRenderTargets + 1>;
-
-struct FramebufferCacheKey {
-    VkRenderPass renderpass{};
-    u32 width = 0;
-    u32 height = 0;
-    u32 layers = 0;
-    ImageViewsPack views;
-
-    std::size_t Hash() const noexcept {
-        std::size_t hash = 0;
-        boost::hash_combine(hash, static_cast<VkRenderPass>(renderpass));
-        for (const auto& view : views) {
-            boost::hash_combine(hash, static_cast<VkImageView>(view));
-        }
-        boost::hash_combine(hash, width);
-        boost::hash_combine(hash, height);
-        boost::hash_combine(hash, layers);
-        return hash;
-    }
-
-    bool operator==(const FramebufferCacheKey& rhs) const noexcept {
-        return std::tie(renderpass, views, width, height, layers) ==
-               std::tie(rhs.renderpass, rhs.views, rhs.width, rhs.height, rhs.layers);
-    }
-
-    bool operator!=(const FramebufferCacheKey& rhs) const noexcept {
-        return !operator==(rhs);
-    }
-};
-
-} // namespace Vulkan
-
-namespace std {
-
-template <>
-struct hash<Vulkan::FramebufferCacheKey> {
-    std::size_t operator()(const Vulkan::FramebufferCacheKey& k) const noexcept {
-        return k.Hash();
-    }
-};
-
-} // namespace std
-
-namespace Vulkan {
-
 class StateTracker;
 class BufferBindings;
 
-struct ImageView {
-    View view;
-    VkImageLayout* layout = nullptr;
-};
-
 class RasterizerVulkan final : public VideoCore::RasterizerAccelerated {
 public:
     explicit RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
@@ -123,15 +71,18 @@ public:
     void InvalidateRegion(VAddr addr, u64 size) override;
     void OnCPUWrite(VAddr addr, u64 size) override;
     void SyncGuestHost() override;
+    void UnmapMemory(VAddr addr, u64 size) override;
     void SignalSemaphore(GPUVAddr addr, u32 value) override;
     void SignalSyncPoint(u32 value) override;
     void ReleaseFences() override;
     void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
     void WaitForIdle() override;
+    void FragmentBarrier() override;
+    void TiledCacheBarrier() override;
     void FlushCommands() override;
     void TickFrame() override;
-    bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
+    bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surface& src,
-                               const Tegra::Engines::Fermi2D::Regs::Surface& dst,
+                               const Tegra::Engines::Fermi2D::Surface& dst,
                                const Tegra::Engines::Fermi2D::Config& copy_config) override;
     bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
                            u32 pixel_stride) override;
@@ -145,11 +96,17 @@ public:
     }
 
     /// Maximum supported size that a constbuffer can have in bytes.
-    static constexpr std::size_t MaxConstbufferSize = 0x10000;
+    static constexpr size_t MaxConstbufferSize = 0x10000;
     static_assert(MaxConstbufferSize % (4 * sizeof(float)) == 0,
                   "The maximum size of a constbuffer must be a multiple of the size of GLvec4");
 
 private:
+    static constexpr size_t MAX_TEXTURES = 192;
+    static constexpr size_t MAX_IMAGES = 48;
+    static constexpr size_t MAX_IMAGE_VIEWS = MAX_TEXTURES + MAX_IMAGES;
+
+    static constexpr VkDeviceSize DEFAULT_BUFFER_SIZE = 4 * sizeof(float);
+
     struct DrawParameters {
         void Draw(vk::CommandBuffer cmdbuf) const;
 
@@ -160,23 +117,8 @@ private:
         bool is_indexed = 0;
     };
 
-    using ColorAttachments = std::array<View, Maxwell::NumRenderTargets>;
-    using ZetaAttachment = View;
-
-    using Texceptions = std::bitset<Maxwell::NumRenderTargets + 1>;
-
-    static constexpr std::size_t ZETA_TEXCEPTION_INDEX = 8;
-    static constexpr VkDeviceSize DEFAULT_BUFFER_SIZE = 4 * sizeof(float);
-
     void FlushWork();
 
-    /// @brief Updates the currently bound attachments
-    /// @param is_clear True when the framebuffer is updated as a clear
-    /// @return Bitfield of attachments being used as sampled textures
-    Texceptions UpdateAttachments(bool is_clear);
-
-    std::tuple<VkFramebuffer, VkExtent2D> ConfigureFramebuffers(VkRenderPass renderpass);
-
     /// Setups geometry buffers and state.
     DrawParameters SetupGeometry(FixedPipelineState& fixed_state, BufferBindings& buffer_bindings,
                                  bool is_indexed, bool is_instanced);
@@ -184,17 +126,12 @@ private:
     /// Setup descriptors in the graphics pipeline.
     void SetupShaderDescriptors(const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders);
 
-    void SetupImageTransitions(Texceptions texceptions, const ColorAttachments& color,
-                               const ZetaAttachment& zeta);
-
     void UpdateDynamicStates();
 
     void BeginTransformFeedback();
 
     void EndTransformFeedback();
 
-    bool WalkAttachmentOverlaps(const CachedSurfaceView& attachment);
-
     void SetupVertexArrays(BufferBindings& buffer_bindings);
 
     void SetupIndexBuffer(BufferBindings& buffer_bindings, DrawParameters& params, bool is_indexed);
@@ -240,14 +177,6 @@ private:
 
     void SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAddr address);
 
-    void SetupUniformTexels(const Tegra::Texture::TICEntry& image, const UniformTexelEntry& entry);
-
-    void SetupTexture(const Tegra::Texture::FullTextureInfo& texture, const SamplerEntry& entry);
-
-    void SetupStorageTexel(const Tegra::Texture::TICEntry& tic, const StorageTexelEntry& entry);
-
-    void SetupImage(const Tegra::Texture::TICEntry& tic, const ImageEntry& entry);
-
     void UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs);
     void UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs);
     void UpdateDepthBias(Tegra::Engines::Maxwell3D::Regs& regs);
@@ -264,18 +193,16 @@ private:
     void UpdateStencilOp(Tegra::Engines::Maxwell3D::Regs& regs);
     void UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs& regs);
 
-    std::size_t CalculateGraphicsStreamBufferSize(bool is_indexed) const;
+    size_t CalculateGraphicsStreamBufferSize(bool is_indexed) const;
 
-    std::size_t CalculateComputeStreamBufferSize() const;
+    size_t CalculateComputeStreamBufferSize() const;
 
-    std::size_t CalculateVertexArraysSize() const;
+    size_t CalculateVertexArraysSize() const;
 
-    std::size_t CalculateIndexBufferSize() const;
+    size_t CalculateIndexBufferSize() const;
 
-    std::size_t CalculateConstBufferSize(const ConstBufferEntry& entry,
+    size_t CalculateConstBufferSize(const ConstBufferEntry& entry,
-                                         const Tegra::Engines::ConstBufferInfo& buffer) const;
+                                    const Tegra::Engines::ConstBufferInfo& buffer) const;
-
-    RenderPassParams GetRenderPassParams(Texceptions texceptions) const;
 
     VkBuffer DefaultBuffer();
 
@@ -290,18 +217,19 @@ private:
     StateTracker& state_tracker;
     VKScheduler& scheduler;
 
+    VKStreamBuffer stream_buffer;
     VKStagingBufferPool staging_pool;
     VKDescriptorPool descriptor_pool;
     VKUpdateDescriptorQueue update_descriptor_queue;
-    VKRenderPassCache renderpass_cache;
+    BlitImageHelper blit_image;
     QuadArrayPass quad_array_pass;
     QuadIndexedPass quad_indexed_pass;
     Uint8Pass uint8_pass;
 
-    VKTextureCache texture_cache;
+    TextureCacheRuntime texture_cache_runtime;
+    TextureCache texture_cache;
     VKPipelineCache pipeline_cache;
     VKBufferCache buffer_cache;
-    VKSamplerCache sampler_cache;
     VKQueryCache query_cache;
     VKFenceManager fence_manager;
 
@@ -310,16 +238,11 @@ private:
     vk::Event wfi_event;
     VideoCommon::Shader::AsyncShaders async_shaders;
 
-    ColorAttachments color_attachments;
+    boost::container::static_vector<u32, MAX_IMAGE_VIEWS> image_view_indices;
-    ZetaAttachment zeta_attachment;
+    std::array<VideoCommon::ImageViewId, MAX_IMAGE_VIEWS> image_view_ids;
-
+    boost::container::static_vector<VkSampler, MAX_TEXTURES> sampler_handles;
-    std::vector<ImageView> sampled_views;
-    std::vector<ImageView> image_views;
 
     u32 draw_counter = 0;
-
-    // TODO(Rodrigo): Invalidate on image destruction
-    std::unordered_map<FramebufferCacheKey, vk::Framebuffer> framebuffer_cache;
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_renderpass_cache.cpp

@@ -1,158 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <cstring>
-#include <memory>
-#include <vector>
-
-#include "common/cityhash.h"
-#include "video_core/engines/maxwell_3d.h"
-#include "video_core/renderer_vulkan/maxwell_to_vk.h"
-#include "video_core/renderer_vulkan/vk_device.h"
-#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
-#include "video_core/renderer_vulkan/wrapper.h"
-
-namespace Vulkan {
-
-std::size_t RenderPassParams::Hash() const noexcept {
-    const u64 hash = Common::CityHash64(reinterpret_cast<const char*>(this), sizeof *this);
-    return static_cast<std::size_t>(hash);
-}
-
-bool RenderPassParams::operator==(const RenderPassParams& rhs) const noexcept {
-    return std::memcmp(&rhs, this, sizeof *this) == 0;
-}
-
-VKRenderPassCache::VKRenderPassCache(const VKDevice& device_) : device{device_} {}
-
-VKRenderPassCache::~VKRenderPassCache() = default;
-
-VkRenderPass VKRenderPassCache::GetRenderPass(const RenderPassParams& params) {
-    const auto [pair, is_cache_miss] = cache.try_emplace(params);
-    auto& entry = pair->second;
-    if (is_cache_miss) {
-        entry = CreateRenderPass(params);
-    }
-    return *entry;
-}
-
-vk::RenderPass VKRenderPassCache::CreateRenderPass(const RenderPassParams& params) const {
-    using namespace VideoCore::Surface;
-    const std::size_t num_attachments = static_cast<std::size_t>(params.num_color_attachments);
-
-    std::vector<VkAttachmentDescription> descriptors;
-    descriptors.reserve(num_attachments);
-
-    std::vector<VkAttachmentReference> color_references;
-    color_references.reserve(num_attachments);
-
-    for (std::size_t rt = 0; rt < num_attachments; ++rt) {
-        const auto guest_format = static_cast<Tegra::RenderTargetFormat>(params.color_formats[rt]);
-        const PixelFormat pixel_format = PixelFormatFromRenderTargetFormat(guest_format);
-        const auto format = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, pixel_format);
-        ASSERT_MSG(format.attachable, "Trying to attach a non-attachable format with format={}",
-                   static_cast<int>(pixel_format));
-
-        // TODO(Rodrigo): Add MAY_ALIAS_BIT when it's needed.
-        const VkImageLayout color_layout = ((params.texceptions >> rt) & 1) != 0
-                                               ? VK_IMAGE_LAYOUT_GENERAL
-                                               : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
-        descriptors.push_back({
-            .flags = VK_ATTACHMENT_DESCRIPTION_MAY_ALIAS_BIT,
-            .format = format.format,
-            .samples = VK_SAMPLE_COUNT_1_BIT,
-            .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
-            .storeOp = VK_ATTACHMENT_STORE_OP_STORE,
-            .stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE,
-            .stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE,
-            .initialLayout = color_layout,
-            .finalLayout = color_layout,
-        });
-
-        color_references.push_back({
-            .attachment = static_cast<u32>(rt),
-            .layout = color_layout,
-        });
-    }
-
-    VkAttachmentReference zeta_attachment_ref;
-    const bool has_zeta = params.zeta_format != 0;
-    if (has_zeta) {
-        const auto guest_format = static_cast<Tegra::DepthFormat>(params.zeta_format);
-        const PixelFormat pixel_format = PixelFormatFromDepthFormat(guest_format);
-        const auto format = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, pixel_format);
-        ASSERT_MSG(format.attachable, "Trying to attach a non-attachable format with format={}",
-                   static_cast<int>(pixel_format));
-
-        const VkImageLayout zeta_layout = params.zeta_texception != 0
-                                              ? VK_IMAGE_LAYOUT_GENERAL
-                                              : VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
-        descriptors.push_back({
-            .flags = 0,
-            .format = format.format,
-            .samples = VK_SAMPLE_COUNT_1_BIT,
-            .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
-            .storeOp = VK_ATTACHMENT_STORE_OP_STORE,
-            .stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
-            .stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE,
-            .initialLayout = zeta_layout,
-            .finalLayout = zeta_layout,
-        });
-
-        zeta_attachment_ref = {
-            .attachment = static_cast<u32>(num_attachments),
-            .layout = zeta_layout,
-        };
-    }
-
-    const VkSubpassDescription subpass_description{
-        .flags = 0,
-        .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
-        .inputAttachmentCount = 0,
-        .pInputAttachments = nullptr,
-        .colorAttachmentCount = static_cast<u32>(color_references.size()),
-        .pColorAttachments = color_references.data(),
-        .pResolveAttachments = nullptr,
-        .pDepthStencilAttachment = has_zeta ? &zeta_attachment_ref : nullptr,
-        .preserveAttachmentCount = 0,
-        .pPreserveAttachments = nullptr,
-    };
-
-    VkAccessFlags access = 0;
-    VkPipelineStageFlags stage = 0;
-    if (!color_references.empty()) {
-        access |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
-        stage |= VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
-    }
-
-    if (has_zeta) {
-        access |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT |
-                  VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
-        stage |= VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
-    }
-
-    const VkSubpassDependency subpass_dependency{
-        .srcSubpass = VK_SUBPASS_EXTERNAL,
-        .dstSubpass = 0,
-        .srcStageMask = stage,
-        .dstStageMask = stage,
-        .srcAccessMask = 0,
-        .dstAccessMask = access,
-        .dependencyFlags = 0,
-    };
-
-    return device.GetLogical().CreateRenderPass({
-        .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .attachmentCount = static_cast<u32>(descriptors.size()),
-        .pAttachments = descriptors.data(),
-        .subpassCount = 1,
-        .pSubpasses = &subpass_description,
-        .dependencyCount = 1,
-        .pDependencies = &subpass_dependency,
-    });
-}
-
-} // namespace Vulkan

src/video_core/renderer_vulkan/vk_renderpass_cache.h

@@ -1,70 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <type_traits>
-#include <unordered_map>
-
-#include <boost/container/static_vector.hpp>
-#include <boost/functional/hash.hpp>
-
-#include "video_core/engines/maxwell_3d.h"
-#include "video_core/renderer_vulkan/wrapper.h"
-#include "video_core/surface.h"
-
-namespace Vulkan {
-
-class VKDevice;
-
-struct RenderPassParams {
-    std::array<u8, Tegra::Engines::Maxwell3D::Regs::NumRenderTargets> color_formats;
-    u8 num_color_attachments;
-    u8 texceptions;
-
-    u8 zeta_format;
-    u8 zeta_texception;
-
-    std::size_t Hash() const noexcept;
-
-    bool operator==(const RenderPassParams& rhs) const noexcept;
-
-    bool operator!=(const RenderPassParams& rhs) const noexcept {
-        return !operator==(rhs);
-    }
-};
-static_assert(std::has_unique_object_representations_v<RenderPassParams>);
-static_assert(std::is_trivially_copyable_v<RenderPassParams>);
-static_assert(std::is_trivially_constructible_v<RenderPassParams>);
-
-} // namespace Vulkan
-
-namespace std {
-
-template <>
-struct hash<Vulkan::RenderPassParams> {
-    std::size_t operator()(const Vulkan::RenderPassParams& k) const noexcept {
-        return k.Hash();
-    }
-};
-
-} // namespace std
-
-namespace Vulkan {
-
-class VKRenderPassCache final {
-public:
-    explicit VKRenderPassCache(const VKDevice& device_);
-    ~VKRenderPassCache();
-
-    VkRenderPass GetRenderPass(const RenderPassParams& params);
-
-private:
-    vk::RenderPass CreateRenderPass(const RenderPassParams& params) const;
-
-    const VKDevice& device;
-    std::unordered_map<RenderPassParams, vk::RenderPass> cache;
-};
-
-} // namespace Vulkan

src/video_core/renderer_vulkan/vk_sampler_cache.cpp

@@ -1,83 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <unordered_map>
-
-#include "video_core/renderer_vulkan/maxwell_to_vk.h"
-#include "video_core/renderer_vulkan/vk_sampler_cache.h"
-#include "video_core/renderer_vulkan/wrapper.h"
-#include "video_core/textures/texture.h"
-
-using Tegra::Texture::TextureMipmapFilter;
-
-namespace Vulkan {
-
-namespace {
-
-VkBorderColor ConvertBorderColor(std::array<float, 4> color) {
-    // TODO(Rodrigo): Manage integer border colors
-    if (color == std::array<float, 4>{0, 0, 0, 0}) {
-        return VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
-    } else if (color == std::array<float, 4>{0, 0, 0, 1}) {
-        return VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK;
-    } else if (color == std::array<float, 4>{1, 1, 1, 1}) {
-        return VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE;
-    }
-    if (color[0] + color[1] + color[2] > 1.35f) {
-        // If color elements are brighter than roughly 0.5 average, use white border
-        return VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE;
-    } else if (color[3] > 0.5f) {
-        return VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK;
-    } else {
-        return VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
-    }
-}
-
-} // Anonymous namespace
-
-VKSamplerCache::VKSamplerCache(const VKDevice& device_) : device{device_} {}
-
-VKSamplerCache::~VKSamplerCache() = default;
-
-vk::Sampler VKSamplerCache::CreateSampler(const Tegra::Texture::TSCEntry& tsc) const {
-    const bool arbitrary_borders = device.IsExtCustomBorderColorSupported();
-    const std::array color = tsc.GetBorderColor();
-
-    VkSamplerCustomBorderColorCreateInfoEXT border{
-        .sType = VK_STRUCTURE_TYPE_SAMPLER_CUSTOM_BORDER_COLOR_CREATE_INFO_EXT,
-        .pNext = nullptr,
-        .customBorderColor = {},
-        .format = VK_FORMAT_UNDEFINED,
-    };
-    std::memcpy(&border.customBorderColor, color.data(), sizeof(color));
-
-    return device.GetLogical().CreateSampler({
-        .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
-        .pNext = arbitrary_borders ? &border : nullptr,
-        .flags = 0,
-        .magFilter = MaxwellToVK::Sampler::Filter(tsc.mag_filter),
-        .minFilter = MaxwellToVK::Sampler::Filter(tsc.min_filter),
-        .mipmapMode = MaxwellToVK::Sampler::MipmapMode(tsc.mipmap_filter),
-        .addressModeU = MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_u, tsc.mag_filter),
-        .addressModeV = MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_v, tsc.mag_filter),
-        .addressModeW = MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_p, tsc.mag_filter),
-        .mipLodBias = tsc.GetLodBias(),
-        .anisotropyEnable =
-            static_cast<VkBool32>(tsc.GetMaxAnisotropy() > 1.0f ? VK_TRUE : VK_FALSE),
-        .maxAnisotropy = tsc.GetMaxAnisotropy(),
-        .compareEnable = tsc.depth_compare_enabled,
-        .compareOp = MaxwellToVK::Sampler::DepthCompareFunction(tsc.depth_compare_func),
-        .minLod = tsc.mipmap_filter == TextureMipmapFilter::None ? 0.0f : tsc.GetMinLod(),
-        .maxLod = tsc.mipmap_filter == TextureMipmapFilter::None ? 0.25f : tsc.GetMaxLod(),
-        .borderColor =
-            arbitrary_borders ? VK_BORDER_COLOR_INT_CUSTOM_EXT : ConvertBorderColor(color),
-        .unnormalizedCoordinates = VK_FALSE,
-    });
-}
-
-VkSampler VKSamplerCache::ToSamplerType(const vk::Sampler& sampler) const {
-    return *sampler;
-}
-
-} // namespace Vulkan

src/video_core/renderer_vulkan/vk_sampler_cache.h

@@ -1,29 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include "video_core/renderer_vulkan/wrapper.h"
-#include "video_core/sampler_cache.h"
-#include "video_core/textures/texture.h"
-
-namespace Vulkan {
-
-class VKDevice;
-
-class VKSamplerCache final : public VideoCommon::SamplerCache<VkSampler, vk::Sampler> {
-public:
-    explicit VKSamplerCache(const VKDevice& device_);
-    ~VKSamplerCache();
-
-protected:
-    vk::Sampler CreateSampler(const Tegra::Texture::TSCEntry& tsc) const override;
-
-    VkSampler ToSamplerType(const vk::Sampler& sampler) const override;
-
-private:
-    const VKDevice& device;
-};
-
-} // namespace Vulkan