Texture Cache: Improve documentation

This commit adds a series of HLE methods for handling 3D textures in
general. This helps games that generate 3D textures on every frame and
may reduce loading times for certain games.

.appveyor/UtilityFunctions.ps1

@@ -1,39 +0,0 @@
-# Set-up Visual Studio Command Prompt environment for PowerShell
-pushd "C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\Common7\Tools\"
-cmd /c "VsDevCmd.bat -arch=x64 & set" | foreach {
-    if ($_ -match "=") {
-        $v = $_.split("="); Set-Item -Force -Path "ENV:\$($v[0])" -Value "$($v[1])"
-    }
-}
-popd
-
-function Which ($search_path, $name) {
-    ($search_path).Split(";") | Get-ChildItem -Filter $name | Select -First 1 -Exp FullName
-}
-
-function GetDeps ($search_path, $binary) {
-    ((dumpbin /dependents $binary).Where({ $_ -match "dependencies:"}, "SkipUntil") | Select-String "[^ ]*\.dll").Matches | foreach {
-        Which $search_path $_.Value
-    }
-}
-
-function RecursivelyGetDeps ($search_path, $binary) {
-    $final_deps = @()
-    $deps_to_process = GetDeps $search_path $binary
-    while ($deps_to_process.Count -gt 0) {
-        $current, $deps_to_process = $deps_to_process
-        if ($final_deps -contains $current) { continue }
-
-        # Is this a system dll file?
-        # We use the same algorithm that cmake uses to determine this.
-        if ($current -match "$([regex]::Escape($env:SystemRoot))\\sys") { continue }
-        if ($current -match "$([regex]::Escape($env:WinDir))\\sys") { continue }
-        if ($current -match "\\msvc[^\\]+dll") { continue }
-        if ($current -match "\\api-ms-win-[^\\]+dll") { continue }
-
-        $final_deps += $current
-        $new_deps = GetDeps $search_path $current
-        $deps_to_process += ($new_deps | ?{-not ($final_deps -contains $_)})
-    }
-    return $final_deps
-}

appveyor.yml

@@ -1,178 +0,0 @@
-# shallow clone
-clone_depth: 10
-
-cache:
-  - C:\ProgramData\chocolatey\bin -> appveyor.yml
-  - C:\ProgramData\chocolatey\lib -> appveyor.yml
-
-os: Visual Studio 2017
-
-environment:
-  # Tell msys2 to add mingw64 to the path
-  MSYSTEM: MINGW64
-  # Tell msys2 to inherit the current directory when starting the shell
-  CHERE_INVOKING: 1
-  matrix:
-    - BUILD_TYPE: msvc
-    - BUILD_TYPE: mingw
-
-platform:
-  - x64
-
-configuration:
-  - Release
-
-install:
-  - git submodule update --init --recursive
-  - ps: |
-        if ($env:BUILD_TYPE -eq 'mingw') {
-          $dependencies = "mingw64/mingw-w64-x86_64-cmake",
-                          "mingw64/mingw-w64-x86_64-qt5",
-                          "mingw64/mingw-w64-x86_64-SDL2"
-          # redirect err to null to prevent warnings from becoming errors
-          # workaround to prevent pacman from failing due to cyclical dependencies
-          C:\msys64\usr\bin\bash -lc "pacman --noconfirm -S mingw64/mingw-w64-x86_64-freetype mingw64/mingw-w64-x86_64-fontconfig" 2> $null
-          C:\msys64\usr\bin\bash -lc "pacman --noconfirm -S $dependencies" 2> $null
-        }
-
-before_build:
-  - mkdir %BUILD_TYPE%_build
-  - cd %BUILD_TYPE%_build
-  - ps: |
-        $COMPAT = if ($env:ENABLE_COMPATIBILITY_REPORTING -eq $null) {0} else {$env:ENABLE_COMPATIBILITY_REPORTING}
-        if ($env:BUILD_TYPE -eq 'msvc') {
-          # redirect stderr and change the exit code to prevent powershell from cancelling the build if cmake prints a warning
-          cmd /C 'cmake -G "Visual Studio 15 2017 Win64" -DYUZU_USE_BUNDLED_QT=1 -DYUZU_USE_BUNDLED_SDL2=1 -DYUZU_USE_BUNDLED_UNICORN=1 -DYUZU_USE_QT_WEB_ENGINE=ON -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${COMPAT} -DUSE_DISCORD_PRESENCE=ON .. 2>&1 && exit 0'
-        } else {
-          C:\msys64\usr\bin\bash.exe -lc "cmake -G 'MSYS Makefiles' -DYUZU_BUILD_UNICORN=1 -DCMAKE_BUILD_TYPE=Release -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${COMPAT} -DUSE_DISCORD_PRESENCE=ON .. 2>&1"
-        }
-  - cd ..
-
-build_script:
-  - ps: |
-        if ($env:BUILD_TYPE -eq 'msvc') {
-          # https://www.appveyor.com/docs/build-phase
-          msbuild msvc_build/yuzu.sln /maxcpucount /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll"
-        } else {
-          C:\msys64\usr\bin\bash.exe -lc 'mingw32-make -C mingw_build/ 2>&1'
-        }
-
-after_build:
-  - ps: |
-        $GITDATE = $(git show -s --date=short --format='%ad') -replace "-",""
-        $GITREV = $(git show -s --format='%h')
-
-        # Find out which kind of release we are producing by tag name
-        if ($env:APPVEYOR_REPO_TAG_NAME) {
-          $RELEASE_DIST, $RELEASE_VERSION = $env:APPVEYOR_REPO_TAG_NAME.split('-')
-        } else {
-          # There is no repo tag - make assumptions
-          $RELEASE_DIST = "head"
-        }
-
-        if ($env:BUILD_TYPE -eq 'msvc') {
-          # Where are these spaces coming from? Regardless, let's remove them
-          $MSVC_BUILD_ZIP = "yuzu-windows-msvc-$GITDATE-$GITREV.zip" -replace " ", ""
-          $MSVC_BUILD_PDB = "yuzu-windows-msvc-$GITDATE-$GITREV-debugsymbols.zip" -replace " ", ""
-          $MSVC_SEVENZIP = "yuzu-windows-msvc-$GITDATE-$GITREV.7z" -replace " ", ""
-
-          # set the build names as env vars so the artifacts can upload them
-          $env:BUILD_ZIP = $MSVC_BUILD_ZIP
-          $env:BUILD_SYMBOLS = $MSVC_BUILD_PDB
-          $env:BUILD_UPDATE = $MSVC_SEVENZIP
-
-          $BUILD_DIR = ".\msvc_build\bin\Release"
-
-          # Make a debug symbol upload
-          mkdir pdb
-          Get-ChildItem "$BUILD_DIR\" -Recurse -Filter "*.pdb" | Copy-Item -destination .\pdb
-          7z a -tzip $MSVC_BUILD_PDB .\pdb\*.pdb
-          rm "$BUILD_DIR\*.pdb"
-
-          mkdir $RELEASE_DIST
-          # get rid of extra exes by copying everything over, then deleting all the exes, then copying just the exes we want
-          Copy-Item "$BUILD_DIR\*" -Destination $RELEASE_DIST -Recurse
-          rm "$RELEASE_DIST\*.exe"
-          Get-ChildItem "$BUILD_DIR" -Recurse -Filter "yuzu*.exe" | Copy-Item -destination $RELEASE_DIST
-          Get-ChildItem "$BUILD_DIR" -Recurse -Filter "QtWebEngineProcess*.exe" | Copy-Item -destination $RELEASE_DIST
-          Copy-Item .\license.txt -Destination $RELEASE_DIST
-          Copy-Item .\README.md -Destination $RELEASE_DIST
-          7z a -tzip $MSVC_BUILD_ZIP $RELEASE_DIST\*
-          7z a $MSVC_SEVENZIP $RELEASE_DIST
-        } else {
-          $MINGW_BUILD_ZIP = "yuzu-windows-mingw-$GITDATE-$GITREV.zip" -replace " ", ""
-          $MINGW_SEVENZIP = "yuzu-windows-mingw-$GITDATE-$GITREV.7z" -replace " ", ""
-          # not going to bother adding separate debug symbols for mingw, so just upload a README for it
-          # if someone wants to add them, change mingw to compile with -g and use objdump and strip to separate the symbols from the binary
-          $MINGW_NO_DEBUG_SYMBOLS = "README_No_Debug_Symbols.txt"
-          Set-Content -Path $MINGW_NO_DEBUG_SYMBOLS -Value "This is a workaround for Appveyor since msvc has debug symbols but mingw doesnt" -Force
-
-          # store the build information in env vars so we can use them as artifacts
-          $env:BUILD_ZIP = $MINGW_BUILD_ZIP
-          $env:BUILD_SYMBOLS = $MINGW_NO_DEBUG_SYMBOLS
-          $env:BUILD_UPDATE = $MINGW_SEVENZIP
-
-          $CMAKE_SOURCE_DIR = "$env:APPVEYOR_BUILD_FOLDER"
-          $CMAKE_BINARY_DIR = "$CMAKE_SOURCE_DIR/mingw_build/bin"
-          $RELEASE_DIST = $RELEASE_DIST + "-mingw"
-
-          mkdir $RELEASE_DIST
-          mkdir $RELEASE_DIST/platforms
-          mkdir $RELEASE_DIST/styles
-          mkdir $RELEASE_DIST/imageformats
-
-          # copy the compiled binaries and other release files to the release folder
-          Get-ChildItem "$CMAKE_BINARY_DIR" -Filter "yuzu*.exe" | Copy-Item -destination $RELEASE_DIST
-          Copy-Item -path "$CMAKE_SOURCE_DIR/license.txt" -destination $RELEASE_DIST
-          Copy-Item -path "$CMAKE_SOURCE_DIR/README.md" -destination $RELEASE_DIST
-
-          # copy the qt windows plugin dll to platforms
-          Copy-Item -path "C:/msys64/mingw64/share/qt5/plugins/platforms/qwindows.dll" -force -destination "$RELEASE_DIST/platforms"
-
-          # copy the qt windows vista style dll to platforms
-          Copy-Item -path "C:/msys64/mingw64/share/qt5/plugins/styles/qwindowsvistastyle.dll" -force -destination "$RELEASE_DIST/styles"
-
-          # copy the qt jpeg imageformat dll to platforms
-          Copy-Item -path "C:/msys64/mingw64/share/qt5/plugins/imageformats/qjpeg.dll" -force -destination "$RELEASE_DIST/imageformats"
-
-          # copy all the dll dependencies to the release folder
-          . "./.appveyor/UtilityFunctions.ps1"
-          $DLLSearchPath = "C:\msys64\mingw64\bin;$env:PATH"
-          $MingwDLLs = RecursivelyGetDeps $DLLSearchPath "$RELEASE_DIST\yuzu.exe"
-          $MingwDLLs += RecursivelyGetDeps $DLLSearchPath  "$RELEASE_DIST\yuzu_cmd.exe"
-          $MingwDLLs += RecursivelyGetDeps $DLLSearchPath  "$RELEASE_DIST\imageformats\qjpeg.dll"
-          Write-Host "Detected the following dependencies:"
-          Write-Host $MingwDLLs
-          foreach ($file in $MingwDLLs) {
-            Copy-Item -path "$file" -force -destination "$RELEASE_DIST"
-          }
-
-          7z a -tzip $MINGW_BUILD_ZIP $RELEASE_DIST\*
-          7z a $MINGW_SEVENZIP $RELEASE_DIST
-        }
-
-test_script:
-  - cd %BUILD_TYPE%_build
-  - ps: |
-        if ($env:BUILD_TYPE -eq 'msvc') {
-          ctest -VV -C Release
-        } else {
-          C:\msys64\usr\bin\bash.exe -lc "ctest -VV -C Release"
-        }
-  - cd ..
-
-artifacts:
-  - path: $(BUILD_ZIP)
-    name: build
-    type: zip
-
-deploy:
-  provider: GitHub
-  release: $(appveyor_repo_tag_name)
-  auth_token:
-    secure: QqePPnXbkzmXct5c8hZ2X5AbsthbI6cS1Sr+VBzcD8oUOIjfWJJKXVAQGUbQAbb0
-  artifact: update,build
-  draft: false
-  prerelease: false
-  on:
-    appveyor_repo_tag: true

src/video_core/engines/shader_bytecode.h

@@ -1973,7 +1973,7 @@ private:
             INST("1101-01---------", Id::TLDS, Type::Texture, "TLDS"),
             INST("110010----111---", Id::TLD4, Type::Texture, "TLD4"),
             INST("1101111011111---", Id::TLD4_B, Type::Texture, "TLD4_B"),
-            INST("11011111--00----", Id::TLD4S, Type::Texture, "TLD4S"),
+            INST("11011111-0------", Id::TLD4S, Type::Texture, "TLD4S"),
             INST("110111110110----", Id::TMML_B, Type::Texture, "TMML_B"),
             INST("1101111101011---", Id::TMML, Type::Texture, "TMML"),
             INST("11011110011110--", Id::TXD_B, Type::Texture, "TXD_B"),

src/video_core/renderer_vulkan/shaders/blit.frag

@@ -0,0 +1,24 @@
+// Copyright 2019 yuzu Emulator Project
+// 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;
+
+layout (location = 0) out vec4 color;
+
+layout (binding = 1) uniform sampler2D color_texture;
+
+void main() {
+    color = texture(color_texture, frag_tex_coord);
+}

src/video_core/renderer_vulkan/shaders/blit.vert

@@ -0,0 +1,28 @@
+// Copyright 2019 yuzu Emulator Project
+// 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;
+layout (location = 1) in vec2 vert_tex_coord;
+
+layout (location = 0) out vec2 frag_tex_coord;
+
+layout (set = 0, binding = 0) uniform MatrixBlock {
+    mat4 modelview_matrix;
+};
+
+void main() {
+    gl_Position = modelview_matrix * vec4(vert_position, 0.0, 1.0);
+    frag_tex_coord = vert_tex_coord;
+}

src/video_core/renderer_vulkan/shaders/quad_array.comp

@@ -0,0 +1,37 @@
+// Copyright 2019 yuzu Emulator Project
+// 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;
+
+layout (std430, set = 0, binding = 0) buffer OutputBuffer {
+    uint output_indexes[];
+};
+
+layout (push_constant) uniform PushConstants {
+    uint first;
+};
+
+void main() {
+    uint primitive = gl_GlobalInvocationID.x;
+    if (primitive * 6 >= output_indexes.length()) {
+        return;
+    }
+
+    const uint quad_map[6] = uint[](0, 1, 2, 0, 2, 3);
+    for (uint vertex = 0; vertex < 6; ++vertex) {
+        uint index = first + primitive * 4 + quad_map[vertex];
+        output_indexes[primitive * 6 + vertex] = index;
+    }
+}

src/video_core/renderer_vulkan/shaders/uint8.comp

@@ -0,0 +1,33 @@
+// Copyright 2019 yuzu Emulator Project
+// 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
+
+layout (local_size_x = 1024) in;
+
+layout (std430, set = 0, binding = 0) readonly buffer InputBuffer {
+    uint8_t input_indexes[];
+};
+
+layout (std430, set = 0, binding = 1) writeonly buffer OutputBuffer {
+    uint16_t output_indexes[];
+};
+
+void main() {
+    uint id = gl_GlobalInvocationID.x;
+    if (id < input_indexes.length()) {
+        output_indexes[id] = uint16_t(input_indexes[id]);
+    }
+}

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -3,7 +3,7 @@
 // Refer to the license.txt file included.
 
 #include "common/assert.h"
-#include "common/logging/log.h"
+#include "common/microprofile.h"
 #include "video_core/renderer_vulkan/declarations.h"
 #include "video_core/renderer_vulkan/vk_device.h"
 #include "video_core/renderer_vulkan/vk_resource_manager.h"
@@ -11,46 +11,172 @@
 
 namespace Vulkan {
 
-VKScheduler::VKScheduler(const VKDevice& device, VKResourceManager& resource_manager)
-    : device{device}, resource_manager{resource_manager} {
-    next_fence = &resource_manager.CommitFence();
-    AllocateNewContext();
+MICROPROFILE_DECLARE(Vulkan_WaitForWorker);
+
+void VKScheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf,
+                                           const vk::DispatchLoaderDynamic& dld) {
+    auto command = first;
+    while (command != nullptr) {
+        auto next = command->GetNext();
+        command->Execute(cmdbuf, dld);
+        command->~Command();
+        command = next;
+    }
+
+    command_offset = 0;
+    first = nullptr;
+    last = nullptr;
 }
 
-VKScheduler::~VKScheduler() = default;
+VKScheduler::VKScheduler(const VKDevice& device, VKResourceManager& resource_manager)
+    : device{device}, resource_manager{resource_manager}, next_fence{
+                                                              &resource_manager.CommitFence()} {
+    AcquireNewChunk();
+    AllocateNewContext();
+    worker_thread = std::thread(&VKScheduler::WorkerThread, this);
+}
+
+VKScheduler::~VKScheduler() {
+    quit = true;
+    cv.notify_all();
+    worker_thread.join();
+}
 
 void VKScheduler::Flush(bool release_fence, vk::Semaphore semaphore) {
     SubmitExecution(semaphore);
-    if (release_fence)
+    if (release_fence) {
         current_fence->Release();
+    }
     AllocateNewContext();
 }
 
 void VKScheduler::Finish(bool release_fence, vk::Semaphore semaphore) {
     SubmitExecution(semaphore);
     current_fence->Wait();
-    if (release_fence)
+    if (release_fence) {
         current_fence->Release();
+    }
     AllocateNewContext();
 }
 
+void VKScheduler::WaitWorker() {
+    MICROPROFILE_SCOPE(Vulkan_WaitForWorker);
+    DispatchWork();
+
+    bool finished = false;
+    do {
+        cv.notify_all();
+        std::unique_lock lock{mutex};
+        finished = chunk_queue.Empty();
+    } while (!finished);
+}
+
+void VKScheduler::DispatchWork() {
+    if (chunk->Empty()) {
+        return;
+    }
+    chunk_queue.Push(std::move(chunk));
+    cv.notify_all();
+    AcquireNewChunk();
+}
+
+void VKScheduler::RequestRenderpass(const vk::RenderPassBeginInfo& renderpass_bi) {
+    if (state.renderpass && renderpass_bi == *state.renderpass) {
+        return;
+    }
+    const bool end_renderpass = state.renderpass.has_value();
+    state.renderpass = renderpass_bi;
+    Record([renderpass_bi, end_renderpass](auto cmdbuf, auto& dld) {
+        if (end_renderpass) {
+            cmdbuf.endRenderPass(dld);
+        }
+        cmdbuf.beginRenderPass(renderpass_bi, vk::SubpassContents::eInline, dld);
+    });
+}
+
+void VKScheduler::RequestOutsideRenderPassOperationContext() {
+    EndRenderPass();
+}
+
+void VKScheduler::BindGraphicsPipeline(vk::Pipeline pipeline) {
+    if (state.graphics_pipeline == pipeline) {
+        return;
+    }
+    state.graphics_pipeline = pipeline;
+    Record([pipeline](auto cmdbuf, auto& dld) {
+        cmdbuf.bindPipeline(vk::PipelineBindPoint::eGraphics, pipeline, dld);
+    });
+}
+
+void VKScheduler::WorkerThread() {
+    std::unique_lock lock{mutex};
+    do {
+        cv.wait(lock, [this] { return !chunk_queue.Empty() || quit; });
+        if (quit) {
+            continue;
+        }
+        auto extracted_chunk = std::move(chunk_queue.Front());
+        chunk_queue.Pop();
+        extracted_chunk->ExecuteAll(current_cmdbuf, device.GetDispatchLoader());
+        chunk_reserve.Push(std::move(extracted_chunk));
+    } while (!quit);
+}
+
 void VKScheduler::SubmitExecution(vk::Semaphore semaphore) {
+    EndPendingOperations();
+    InvalidateState();
+    WaitWorker();
+
+    std::unique_lock lock{mutex};
+
+    const auto queue = device.GetGraphicsQueue();
     const auto& dld = device.GetDispatchLoader();
     current_cmdbuf.end(dld);
 
-    const auto queue = device.GetGraphicsQueue();
-    const vk::SubmitInfo submit_info(0, nullptr, nullptr, 1, &current_cmdbuf, semaphore ? 1u : 0u,
+    const vk::SubmitInfo submit_info(0, nullptr, nullptr, 1, &current_cmdbuf, semaphore ? 1U : 0U,
                                      &semaphore);
-    queue.submit({submit_info}, *current_fence, dld);
+    queue.submit({submit_info}, static_cast<vk::Fence>(*current_fence), dld);
 }
 
 void VKScheduler::AllocateNewContext() {
+    std::unique_lock lock{mutex};
     current_fence = next_fence;
-    current_cmdbuf = resource_manager.CommitCommandBuffer(*current_fence);
     next_fence = &resource_manager.CommitFence();
 
-    const auto& dld = device.GetDispatchLoader();
-    current_cmdbuf.begin({vk::CommandBufferUsageFlagBits::eOneTimeSubmit}, dld);
+    current_cmdbuf = resource_manager.CommitCommandBuffer(*current_fence);
+    current_cmdbuf.begin({vk::CommandBufferUsageFlagBits::eOneTimeSubmit},
+                         device.GetDispatchLoader());
+}
+
+void VKScheduler::InvalidateState() {
+    state.graphics_pipeline = nullptr;
+    state.viewports = false;
+    state.scissors = false;
+    state.depth_bias = false;
+    state.blend_constants = false;
+    state.depth_bounds = false;
+    state.stencil_values = false;
+}
+
+void VKScheduler::EndPendingOperations() {
+    EndRenderPass();
+}
+
+void VKScheduler::EndRenderPass() {
+    if (!state.renderpass) {
+        return;
+    }
+    state.renderpass = std::nullopt;
+    Record([](auto cmdbuf, auto& dld) { cmdbuf.endRenderPass(dld); });
+}
+
+void VKScheduler::AcquireNewChunk() {
+    if (chunk_reserve.Empty()) {
+        chunk = std::make_unique<CommandChunk>();
+        return;
+    }
+    chunk = std::move(chunk_reserve.Front());
+    chunk_reserve.Pop();
 }
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_scheduler.h

@@ -4,7 +4,14 @@
 
 #pragma once
 
+#include <condition_variable>
+#include <memory>
+#include <optional>
+#include <stack>
+#include <thread>
+#include <utility>
 #include "common/common_types.h"
+#include "common/threadsafe_queue.h"
 #include "video_core/renderer_vulkan/declarations.h"
 
 namespace Vulkan {
@@ -30,23 +37,6 @@ private:
     VKFence* const& fence;
 };
 
-class VKCommandBufferView {
-public:
-    VKCommandBufferView() = default;
-    VKCommandBufferView(const vk::CommandBuffer& cmdbuf) : cmdbuf{cmdbuf} {}
-
-    const vk::CommandBuffer* operator->() const noexcept {
-        return &cmdbuf;
-    }
-
-    operator vk::CommandBuffer() const noexcept {
-        return cmdbuf;
-    }
-
-private:
-    const vk::CommandBuffer& cmdbuf;
-};
-
 /// The scheduler abstracts command buffer and fence management with an interface that's able to do
 /// OpenGL-like operations on Vulkan command buffers.
 class VKScheduler {
@@ -54,32 +44,190 @@ public:
     explicit VKScheduler(const VKDevice& device, VKResourceManager& resource_manager);
     ~VKScheduler();
 
-    /// Gets a reference to the current fence.
-    VKFenceView GetFence() const {
-        return current_fence;
-    }
-
-    /// Gets a reference to the current command buffer.
-    VKCommandBufferView GetCommandBuffer() const {
-        return current_cmdbuf;
-    }
-
     /// Sends the current execution context to the GPU.
     void Flush(bool release_fence = true, vk::Semaphore semaphore = nullptr);
 
     /// Sends the current execution context to the GPU and waits for it to complete.
     void Finish(bool release_fence = true, vk::Semaphore semaphore = nullptr);
 
+    /// Waits for the worker thread to finish executing everything. After this function returns it's
+    /// safe to touch worker resources.
+    void WaitWorker();
+
+    /// Sends currently recorded work to the worker thread.
+    void DispatchWork();
+
+    /// Requests to begin a renderpass.
+    void RequestRenderpass(const vk::RenderPassBeginInfo& renderpass_bi);
+
+    /// Requests the current executino context to be able to execute operations only allowed outside
+    /// of a renderpass.
+    void RequestOutsideRenderPassOperationContext();
+
+    /// Binds a pipeline to the current execution context.
+    void BindGraphicsPipeline(vk::Pipeline pipeline);
+
+    /// Returns true when viewports have been set in the current command buffer.
+    bool TouchViewports() {
+        return std::exchange(state.viewports, true);
+    }
+
+    /// Returns true when scissors have been set in the current command buffer.
+    bool TouchScissors() {
+        return std::exchange(state.scissors, true);
+    }
+
+    /// Returns true when depth bias have been set in the current command buffer.
+    bool TouchDepthBias() {
+        return std::exchange(state.depth_bias, true);
+    }
+
+    /// Returns true when blend constants have been set in the current command buffer.
+    bool TouchBlendConstants() {
+        return std::exchange(state.blend_constants, true);
+    }
+
+    /// Returns true when depth bounds have been set in the current command buffer.
+    bool TouchDepthBounds() {
+        return std::exchange(state.depth_bounds, true);
+    }
+
+    /// Returns true when stencil values have been set in the current command buffer.
+    bool TouchStencilValues() {
+        return std::exchange(state.stencil_values, true);
+    }
+
+    /// Send work to a separate thread.
+    template <typename T>
+    void Record(T&& command) {
+        if (chunk->Record(command)) {
+            return;
+        }
+        DispatchWork();
+        (void)chunk->Record(command);
+    }
+
+    /// Gets a reference to the current fence.
+    VKFenceView GetFence() const {
+        return current_fence;
+    }
+
 private:
+    class Command {
+    public:
+        virtual ~Command() = default;
+
+        virtual void Execute(vk::CommandBuffer cmdbuf,
+                             const vk::DispatchLoaderDynamic& dld) const = 0;
+
+        Command* GetNext() const {
+            return next;
+        }
+
+        void SetNext(Command* next_) {
+            next = next_;
+        }
+
+    private:
+        Command* next = nullptr;
+    };
+
+    template <typename T>
+    class TypedCommand final : public Command {
+    public:
+        explicit TypedCommand(T&& command) : command{std::move(command)} {}
+        ~TypedCommand() override = default;
+
+        TypedCommand(TypedCommand&&) = delete;
+        TypedCommand& operator=(TypedCommand&&) = delete;
+
+        void Execute(vk::CommandBuffer cmdbuf,
+                     const vk::DispatchLoaderDynamic& dld) const override {
+            command(cmdbuf, dld);
+        }
+
+    private:
+        T command;
+    };
+
+    class CommandChunk final {
+    public:
+        void ExecuteAll(vk::CommandBuffer cmdbuf, const vk::DispatchLoaderDynamic& dld);
+
+        template <typename T>
+        bool Record(T& command) {
+            using FuncType = TypedCommand<T>;
+            static_assert(sizeof(FuncType) < sizeof(data), "Lambda is too large");
+
+            if (command_offset > sizeof(data) - sizeof(FuncType)) {
+                return false;
+            }
+
+            Command* current_last = last;
+
+            last = new (data.data() + command_offset) FuncType(std::move(command));
+
+            if (current_last) {
+                current_last->SetNext(last);
+            } else {
+                first = last;
+            }
+
+            command_offset += sizeof(FuncType);
+            return true;
+        }
+
+        bool Empty() const {
+            return command_offset == 0;
+        }
+
+    private:
+        Command* first = nullptr;
+        Command* last = nullptr;
+
+        std::size_t command_offset = 0;
+        std::array<u8, 0x8000> data{};
+    };
+
+    void WorkerThread();
+
     void SubmitExecution(vk::Semaphore semaphore);
 
     void AllocateNewContext();
 
+    void InvalidateState();
+
+    void EndPendingOperations();
+
+    void EndRenderPass();
+
+    void AcquireNewChunk();
+
     const VKDevice& device;
     VKResourceManager& resource_manager;
     vk::CommandBuffer current_cmdbuf;
     VKFence* current_fence = nullptr;
     VKFence* next_fence = nullptr;
+
+    struct State {
+        std::optional<vk::RenderPassBeginInfo> renderpass;
+        vk::Pipeline graphics_pipeline;
+        bool viewports = false;
+        bool scissors = false;
+        bool depth_bias = false;
+        bool blend_constants = false;
+        bool depth_bounds = false;
+        bool stencil_values = false;
+    } state;
+
+    std::unique_ptr<CommandChunk> chunk;
+    std::thread worker_thread;
+
+    Common::SPSCQueue<std::unique_ptr<CommandChunk>> chunk_queue;
+    Common::SPSCQueue<std::unique_ptr<CommandChunk>> chunk_reserve;
+    std::mutex mutex;
+    std::condition_variable cv;
+    bool quit = false;
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -2552,29 +2552,7 @@ public:
     }
 
     Id operator()(const ExprCondCode& expr) {
-        const Node cc = decomp.ir.GetConditionCode(expr.cc);
-        Id target;
-
-        if (const auto pred = std::get_if<PredicateNode>(&*cc)) {
-            const auto index = pred->GetIndex();
-            switch (index) {
-            case Tegra::Shader::Pred::NeverExecute:
-                target = decomp.v_false;
-                break;
-            case Tegra::Shader::Pred::UnusedIndex:
-                target = decomp.v_true;
-                break;
-            default:
-                target = decomp.predicates.at(index);
-                break;
-            }
-        } else if (const auto flag = std::get_if<InternalFlagNode>(&*cc)) {
-            target = decomp.internal_flags.at(static_cast<u32>(flag->GetFlag()));
-        } else {
-            UNREACHABLE();
-        }
-
-        return decomp.OpLoad(decomp.t_bool, target);
+        return decomp.AsBool(decomp.Visit(decomp.ir.GetConditionCode(expr.cc)));
     }
 
     Id operator()(const ExprVar& expr) {

src/video_core/shader/decode/conversion.cpp

@@ -63,12 +63,11 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
     case OpCode::Id::I2F_R:
     case OpCode::Id::I2F_C:
     case OpCode::Id::I2F_IMM: {
-        UNIMPLEMENTED_IF(instr.conversion.int_src.selector != 0);
         UNIMPLEMENTED_IF(instr.conversion.dst_size == Register::Size::Long);
         UNIMPLEMENTED_IF_MSG(instr.generates_cc,
                              "Condition codes generation in I2F is not implemented");
 
-        Node value = [&]() {
+        Node value = [&] {
             switch (opcode->get().GetId()) {
             case OpCode::Id::I2F_R:
                 return GetRegister(instr.gpr20);
@@ -81,7 +80,19 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
                 return Immediate(0);
             }
         }();
+
         const bool input_signed = instr.conversion.is_input_signed;
+
+        if (instr.conversion.src_size == Register::Size::Byte) {
+            const u32 offset = static_cast<u32>(instr.conversion.int_src.selector) * 8;
+            if (offset > 0) {
+                value = SignedOperation(OperationCode::ILogicalShiftRight, input_signed,
+                                        std::move(value), Immediate(offset));
+            }
+        } else {
+            UNIMPLEMENTED_IF(instr.conversion.int_src.selector != 0);
+        }
+
         value = ConvertIntegerSize(value, instr.conversion.src_size, input_signed);
         value = GetOperandAbsNegInteger(value, instr.conversion.abs_a, false, input_signed);
         value = SignedOperation(OperationCode::FCastInteger, input_signed, PRECISE, value);

src/video_core/shader/decode/memory.cpp

@@ -22,7 +22,23 @@ using Tegra::Shader::Register;
 
 namespace {
 
-u32 GetUniformTypeElementsCount(Tegra::Shader::UniformType uniform_type) {
+u32 GetLdgMemorySize(Tegra::Shader::UniformType uniform_type) {
+    switch (uniform_type) {
+    case Tegra::Shader::UniformType::UnsignedByte:
+    case Tegra::Shader::UniformType::Single:
+        return 1;
+    case Tegra::Shader::UniformType::Double:
+        return 2;
+    case Tegra::Shader::UniformType::Quad:
+    case Tegra::Shader::UniformType::UnsignedQuad:
+        return 4;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented size={}!", static_cast<u32>(uniform_type));
+        return 1;
+    }
+}
+
+u32 GetStgMemorySize(Tegra::Shader::UniformType uniform_type) {
     switch (uniform_type) {
     case Tegra::Shader::UniformType::Single:
         return 1;
@@ -170,7 +186,7 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         const auto [real_address_base, base_address, descriptor] =
             TrackGlobalMemory(bb, instr, false);
 
-        const u32 count = GetUniformTypeElementsCount(type);
+        const u32 count = GetLdgMemorySize(type);
         if (!real_address_base || !base_address) {
             // Tracking failed, load zeroes.
             for (u32 i = 0; i < count; ++i) {
@@ -181,12 +197,22 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
 
         for (u32 i = 0; i < count; ++i) {
             const Node it_offset = Immediate(i * 4);
-            const Node real_address =
-                Operation(OperationCode::UAdd, NO_PRECISE, real_address_base, it_offset);
-            const Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
+            const Node real_address = Operation(OperationCode::UAdd, real_address_base, it_offset);
+            Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
+
+            if (type == Tegra::Shader::UniformType::UnsignedByte) {
+                // To handle unaligned loads get the byte used to dereferenced global memory
+                // and extract that byte from the loaded uint32.
+                Node byte = Operation(OperationCode::UBitwiseAnd, real_address, Immediate(3));
+                byte = Operation(OperationCode::ULogicalShiftLeft, std::move(byte), Immediate(3));
+
+                gmem = Operation(OperationCode::UBitfieldExtract, std::move(gmem), std::move(byte),
+                                 Immediate(8));
+            }
 
             SetTemporary(bb, i, gmem);
         }
+
         for (u32 i = 0; i < count; ++i) {
             SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
         }
@@ -276,7 +302,7 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
             break;
         }
 
-        const u32 count = GetUniformTypeElementsCount(type);
+        const u32 count = GetStgMemorySize(type);
         for (u32 i = 0; i < count; ++i) {
             const Node it_offset = Immediate(i * 4);
             const Node real_address = Operation(OperationCode::UAdd, real_address_base, it_offset);

src/video_core/shader/decode/texture.cpp

@@ -743,13 +743,18 @@ Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is
     // When lod is used always is in gpr20
     const Node lod = lod_enabled ? GetRegister(instr.gpr20) : Immediate(0);
 
-    // Fill empty entries from the guest sampler.
+    // Fill empty entries from the guest sampler
     const std::size_t entry_coord_count = GetCoordCount(sampler.GetType());
     if (type_coord_count != entry_coord_count) {
         LOG_WARNING(HW_GPU, "Bound and built texture types mismatch");
-    }
-    for (std::size_t i = type_coord_count; i < entry_coord_count; ++i) {
-        coords.push_back(GetRegister(Register::ZeroIndex));
+
+        // When the size is higher we insert zeroes
+        for (std::size_t i = type_coord_count; i < entry_coord_count; ++i) {
+            coords.push_back(GetRegister(Register::ZeroIndex));
+        }
+
+        // Then we ensure the size matches the number of entries (dropping unused values)
+        coords.resize(entry_coord_count);
     }
 
     Node4 values;

src/video_core/texture_cache/surface_params.cpp

@@ -392,4 +392,42 @@ std::string SurfaceParams::TargetName() const {
     }
 }
 
+u32 SurfaceParams::GetBlockSize() const {
+    const u32 x = 64U << block_width;
+    const u32 y = 8U << block_height;
+    const u32 z = 1U << block_depth;
+    return x * y * z;
+}
+
+std::pair<u32, u32> SurfaceParams::GetBlockXY() const {
+    const u32 x_pixels = 64U / GetBytesPerPixel();
+    const u32 x = x_pixels << block_width;
+    const u32 y = 8U << block_height;
+    return {x, y};
+}
+
+std::tuple<u32, u32, u32> SurfaceParams::GetBlockOffsetXYZ(u32 offset) const {
+    const auto div_ceil = [](const u32 x, const u32 y) { return ((x + y - 1) / y); };
+    const u32 block_size = GetBlockSize();
+    const u32 block_index = offset / block_size;
+    const u32 gob_offset = offset % block_size;
+    const u32 gob_index = gob_offset / static_cast<u32>(Tegra::Texture::GetGOBSize());
+    const u32 x_gob_pixels = 64U / GetBytesPerPixel();
+    const u32 x_block_pixels = x_gob_pixels << block_width;
+    const u32 y_block_pixels = 8U << block_height;
+    const u32 z_block_pixels = 1U << block_depth;
+    const u32 x_blocks = div_ceil(width, x_block_pixels);
+    const u32 y_blocks = div_ceil(height, y_block_pixels);
+    const u32 z_blocks = div_ceil(depth, z_block_pixels);
+    const u32 base_x = block_index % x_blocks;
+    const u32 base_y = (block_index / x_blocks) % y_blocks;
+    const u32 base_z = (block_index / (x_blocks * y_blocks)) % z_blocks;
+    u32 x = base_x * x_block_pixels;
+    u32 y = base_y * y_block_pixels;
+    u32 z = base_z * z_block_pixels;
+    z += gob_index >> block_height;
+    y += (gob_index * 8U) % y_block_pixels;
+    return {x, y, z};
+}
+
 } // namespace VideoCommon

src/video_core/texture_cache/surface_params.h

@@ -4,6 +4,8 @@
 
 #pragma once
 
+#include <utility>
+
 #include "common/alignment.h"
 #include "common/bit_util.h"
 #include "common/cityhash.h"
@@ -136,6 +138,15 @@ public:
 
     std::size_t GetConvertedMipmapSize(u32 level) const;
 
+    /// Get this texture Tegra Block size in guest memory layout
+    u32 GetBlockSize() const;
+
+    /// Get X, Y coordinates max sizes of a single block.
+    std::pair<u32, u32> GetBlockXY() const;
+
+    /// Get the offset in x, y, z coordinates from a memory offset
+    std::tuple<u32, u32, u32> GetBlockOffsetXYZ(u32 offset) const;
+
     /// Returns the size of a layer in bytes in guest memory.
     std::size_t GetGuestLayerSize() const {
         return GetLayerSize(false, false);
@@ -269,7 +280,8 @@ private:
 
     /// Returns the size of all mipmap levels and aligns as needed.
     std::size_t GetInnerMemorySize(bool as_host_size, bool layer_only, bool uncompressed) const {
-        return GetLayerSize(as_host_size, uncompressed) * (layer_only ? 1U : depth);
+        return GetLayerSize(as_host_size, uncompressed) *
+               (layer_only ? 1U : (is_layered ? depth : 1U));
     }
 
     /// Returns the size of a layer

src/video_core/texture_cache/texture_cache.h

@@ -615,6 +615,86 @@ private:
         return {{new_surface, new_surface->GetMainView()}};
     }
 
+    /**
+     * Takes care of managing 3D textures and its slices. Does HLE methods for reconstructing the 3D
+     * textures within the GPU if possible. Falls back to LLE when it isn't possible to use any of
+     * the HLE methods.
+     *
+     * @param overlaps          The overlapping surfaces registered in the cache.
+     * @param params            The parameters on the new surface.
+     * @param gpu_addr          The starting address of the new surface.
+     * @param cache_addr        The starting address of the new surface on physical memory.
+     * @param preserve_contents Indicates that the new surface should be loaded from memory or
+     *                          left blank.
+     */
+    std::optional<std::pair<TSurface, TView>> Manage3DSurfaces(std::vector<TSurface>& overlaps,
+                                                               const SurfaceParams& params,
+                                                               const GPUVAddr gpu_addr,
+                                                               const CacheAddr cache_addr,
+                                                               bool preserve_contents) {
+        if (params.target == SurfaceTarget::Texture3D) {
+            bool failed = false;
+            if (params.num_levels > 1) {
+                // We can't handle mipmaps in 3D textures yet, better fallback to LLE approach
+                return std::nullopt;
+            }
+            TSurface new_surface = GetUncachedSurface(gpu_addr, params);
+            bool modified = false;
+            for (auto& surface : overlaps) {
+                const SurfaceParams& src_params = surface->GetSurfaceParams();
+                if (src_params.target != SurfaceTarget::Texture2D) {
+                    failed = true;
+                    break;
+                }
+                if (src_params.height != params.height) {
+                    failed = true;
+                    break;
+                }
+                if (src_params.block_depth != params.block_depth ||
+                    src_params.block_height != params.block_height) {
+                    failed = true;
+                    break;
+                }
+                const u32 offset = static_cast<u32>(surface->GetCacheAddr() - cache_addr);
+                const auto [x, y, z] = params.GetBlockOffsetXYZ(offset);
+                modified |= surface->IsModified();
+                const CopyParams copy_params(0, 0, 0, 0, 0, z, 0, 0, params.width, params.height,
+                                             1);
+                ImageCopy(surface, new_surface, copy_params);
+            }
+            if (failed) {
+                return std::nullopt;
+            }
+            for (const auto& surface : overlaps) {
+                Unregister(surface);
+            }
+            new_surface->MarkAsModified(modified, Tick());
+            Register(new_surface);
+            auto view = new_surface->GetMainView();
+            return {{std::move(new_surface), view}};
+        } else {
+            for (const auto& surface : overlaps) {
+                if (!surface->MatchTarget(params.target)) {
+                    if (overlaps.size() == 1 && surface->GetCacheAddr() == cache_addr) {
+                        if (Settings::values.use_accurate_gpu_emulation) {
+                            return std::nullopt;
+                        }
+                        Unregister(surface);
+                        return InitializeSurface(gpu_addr, params, preserve_contents);
+                    }
+                    return std::nullopt;
+                }
+                if (surface->GetCacheAddr() != cache_addr) {
+                    continue;
+                }
+                if (surface->MatchesStructure(params) == MatchStructureResult::FullMatch) {
+                    return {{surface, surface->GetMainView()}};
+                }
+            }
+            return InitializeSurface(gpu_addr, params, preserve_contents);
+        }
+    }
+
     /**
      * Gets the starting address and parameters of a candidate surface and tries
      * to find a matching surface within the cache. This is done in 3 big steps:
@@ -687,6 +767,15 @@ private:
             }
         }
 
+        // Check if it's a 3D texture
+        if (params.block_depth > 0) {
+            auto surface =
+                Manage3DSurfaces(overlaps, params, gpu_addr, cache_addr, preserve_contents);
+            if (surface) {
+                return *surface;
+            }
+        }
+
         // Split cases between 1 overlap or many.
         if (overlaps.size() == 1) {
             TSurface current_surface = overlaps[0];

src/video_core/textures/decoders.h

@@ -12,6 +12,10 @@ namespace Tegra::Texture {
 
 // GOBSize constant. Calculated by 64 bytes in x multiplied by 8 y coords, represents
 // an small rect of (64/bytes_per_pixel)X8.
+inline std::size_t GetGOBSize() {
+    return 512;
+}
+
 inline std::size_t GetGOBSizeShift() {
     return 9;
 }