engines/const_buffer_engine_interface: Store image format type

This information is required to properly implement SULD.B. It might also
be handy for all image operations, since it would allow us to implement
them on devices that require the image format to be specified (on
desktop, this would be AMD on OpenGL and Intel on OpenGL and Vulkan).

CMakeModules/CopyYuzuQt5Deps.cmake

@@ -6,9 +6,9 @@ function(copy_yuzu_Qt5_deps target_dir)
     set(Qt5_STYLES_DIR "${Qt5_DIR}/../../../plugins/styles/")
     set(Qt5_IMAGEFORMATS_DIR "${Qt5_DIR}/../../../plugins/imageformats/")
     set(Qt5_RESOURCES_DIR "${Qt5_DIR}/../../../resources/")
-    set(PLATFORMS ${DLL_DEST}plugins/platforms/)
-    set(STYLES ${DLL_DEST}plugins/styles/)
-    set(IMAGEFORMATS ${DLL_DEST}plugins/imageformats/)
+    set(PLATFORMS ${DLL_DEST}platforms/)
+    set(STYLES ${DLL_DEST}styles/)
+    set(IMAGEFORMATS ${DLL_DEST}imageformats/)
     windows_copy_files(${target_dir} ${Qt5_DLL_DIR} ${DLL_DEST}
         icudt*.dll
         icuin*.dll
@@ -42,15 +42,11 @@ function(copy_yuzu_Qt5_deps target_dir)
             icudtl.dat
         )
     endif ()
+
     windows_copy_files(yuzu ${Qt5_PLATFORMS_DIR} ${PLATFORMS} qwindows$<$<CONFIG:Debug>:d>.*)
     windows_copy_files(yuzu ${Qt5_STYLES_DIR} ${STYLES} qwindowsvistastyle$<$<CONFIG:Debug>:d>.*)
     windows_copy_files(yuzu ${Qt5_IMAGEFORMATS_DIR} ${IMAGEFORMATS}
         qjpeg$<$<CONFIG:Debug>:d>.*
         qgif$<$<CONFIG:Debug>:d>.*
         )
-    # Create an empty qt.conf file. Qt will detect that this file exists, and use the folder that its in as the root folder.
-    # This way it'll look for plugins in the root/plugins/ folder
-    add_custom_command(TARGET yuzu POST_BUILD
-        COMMAND ${CMAKE_COMMAND} -E touch ${DLL_DEST}qt.conf
-    )
 endfunction(copy_yuzu_Qt5_deps)

src/core/CMakeLists.txt

@@ -131,6 +131,8 @@ add_library(core STATIC
     frontend/framebuffer_layout.cpp
     frontend/framebuffer_layout.h
     frontend/input.h
+    frontend/scope_acquire_context.cpp
+    frontend/scope_acquire_context.h
     gdbstub/gdbstub.cpp
     gdbstub/gdbstub.h
     hardware_interrupt_manager.cpp

src/core/core.cpp

@@ -24,6 +24,7 @@
 #include "core/file_sys/sdmc_factory.h"
 #include "core/file_sys/vfs_concat.h"
 #include "core/file_sys/vfs_real.h"
+#include "core/frontend/scope_acquire_context.h"
 #include "core/gdbstub/gdbstub.h"
 #include "core/hardware_interrupt_manager.h"
 #include "core/hle/kernel/client_port.h"
@@ -165,14 +166,15 @@ struct System::Impl {
         service_manager = std::make_shared<Service::SM::ServiceManager>();
 
         Service::Init(service_manager, system);
-        GDBStub::Init();
+        GDBStub::DeferStart();
 
-        interrupt_manager = std::make_unique<Core::Hardware::InterruptManager>(system);
-        gpu_core = VideoCore::CreateGPU(emu_window, system);
-        if (!gpu_core) {
+        renderer = VideoCore::CreateRenderer(emu_window, system);
+        if (!renderer->Init()) {
             return ResultStatus::ErrorVideoCore;
         }
-        gpu_core->Renderer().Rasterizer().SetupDirtyFlags();
+        interrupt_manager = std::make_unique<Core::Hardware::InterruptManager>(system);
+        gpu_core = VideoCore::CreateGPU(system);
+        renderer->Rasterizer().SetupDirtyFlags();
 
         is_powered_on = true;
         exit_lock = false;
@@ -184,6 +186,8 @@ struct System::Impl {
 
     ResultStatus Load(System& system, Frontend::EmuWindow& emu_window,
                       const std::string& filepath) {
+        Core::Frontend::ScopeAcquireContext acquire_context{emu_window};
+
         app_loader = Loader::GetLoader(GetGameFileFromPath(virtual_filesystem, filepath));
         if (!app_loader) {
             LOG_CRITICAL(Core, "Failed to obtain loader for {}!", filepath);
@@ -212,6 +216,10 @@ struct System::Impl {
         AddGlueRegistrationForProcess(*app_loader, *main_process);
         kernel.MakeCurrentProcess(main_process.get());
 
+        // Main process has been loaded and been made current.
+        // Begin GPU and CPU execution.
+        gpu_core->Start();
+
         // Initialize cheat engine
         if (cheat_engine) {
             cheat_engine->Initialize();
@@ -269,6 +277,7 @@ struct System::Impl {
         }
 
         // Shutdown emulation session
+        renderer.reset();
         GDBStub::Shutdown();
         Service::Shutdown();
         service_manager.reset();
@@ -344,6 +353,7 @@ struct System::Impl {
     Service::FileSystem::FileSystemController fs_controller;
     /// AppLoader used to load the current executing application
     std::unique_ptr<Loader::AppLoader> app_loader;
+    std::unique_ptr<VideoCore::RendererBase> renderer;
     std::unique_ptr<Tegra::GPU> gpu_core;
     std::unique_ptr<Hardware::InterruptManager> interrupt_manager;
     Memory::Memory memory;
@@ -526,11 +536,11 @@ const Core::Hardware::InterruptManager& System::InterruptManager() const {
 }
 
 VideoCore::RendererBase& System::Renderer() {
-    return impl->gpu_core->Renderer();
+    return *impl->renderer;
 }
 
 const VideoCore::RendererBase& System::Renderer() const {
-    return impl->gpu_core->Renderer();
+    return *impl->renderer;
 }
 
 Kernel::KernelCore& System::Kernel() {

src/core/frontend/emu_window.h

@@ -13,39 +13,19 @@
 namespace Core::Frontend {
 
 /**
- * Represents a drawing context that supports graphics operations.
+ * Represents a graphics context that can be used for background computation or drawing. If the
+ * graphics backend doesn't require the context, then the implementation of these methods can be
+ * stubs
  */
 class GraphicsContext {
 public:
     virtual ~GraphicsContext();
 
-    /// Inform the driver to swap the front/back buffers and present the current image
-    virtual void SwapBuffers() {}
-
     /// Makes the graphics context current for the caller thread
-    virtual void MakeCurrent() {}
+    virtual void MakeCurrent() = 0;
 
     /// Releases (dunno if this is the "right" word) the context from the caller thread
-    virtual void DoneCurrent() {}
-
-    class Scoped {
-    public:
-        explicit Scoped(GraphicsContext& context_) : context(context_) {
-            context.MakeCurrent();
-        }
-        ~Scoped() {
-            context.DoneCurrent();
-        }
-
-    private:
-        GraphicsContext& context;
-    };
-
-    /// Calls MakeCurrent on the context and calls DoneCurrent when the scope for the returned value
-    /// ends
-    Scoped Acquire() {
-        return Scoped{*this};
-    }
+    virtual void DoneCurrent() = 0;
 };
 
 /**
@@ -66,7 +46,7 @@ public:
  * - DO NOT TREAT THIS CLASS AS A GUI TOOLKIT ABSTRACTION LAYER. That's not what it is. Please
  *   re-read the upper points again and think about it if you don't see this.
  */
-class EmuWindow {
+class EmuWindow : public GraphicsContext {
 public:
     /// Data structure to store emuwindow configuration
     struct WindowConfig {
@@ -80,9 +60,17 @@ public:
     virtual void PollEvents() = 0;
 
     /**
-     * Returns a GraphicsContext that the frontend provides to be used for rendering.
+     * Returns a GraphicsContext that the frontend provides that is shared with the emu window. This
+     * context can be used from other threads for background graphics computation. If the frontend
+     * is using a graphics backend that doesn't need anything specific to run on a different thread,
+     * then it can use a stubbed implemenation for GraphicsContext.
+     *
+     * If the return value is null, then the core should assume that the frontend cannot provide a
+     * Shared Context
      */
-    virtual std::unique_ptr<GraphicsContext> CreateSharedContext() const = 0;
+    virtual std::unique_ptr<GraphicsContext> CreateSharedContext() const {
+        return nullptr;
+    }
 
     /// Returns if window is shown (not minimized)
     virtual bool IsShown() const = 0;

src/core/frontend/scope_acquire_context.cpp

@@ -0,0 +1,18 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "core/frontend/emu_window.h"
+#include "core/frontend/scope_acquire_context.h"
+
+namespace Core::Frontend {
+
+ScopeAcquireContext::ScopeAcquireContext(Core::Frontend::GraphicsContext& context)
+    : context{context} {
+    context.MakeCurrent();
+}
+ScopeAcquireContext::~ScopeAcquireContext() {
+    context.DoneCurrent();
+}
+
+} // namespace Core::Frontend

src/core/frontend/scope_acquire_context.h

@@ -0,0 +1,23 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/common_types.h"
+
+namespace Core::Frontend {
+
+class GraphicsContext;
+
+/// Helper class to acquire/release window context within a given scope
+class ScopeAcquireContext : NonCopyable {
+public:
+    explicit ScopeAcquireContext(Core::Frontend::GraphicsContext& context);
+    ~ScopeAcquireContext();
+
+private:
+    Core::Frontend::GraphicsContext& context;
+};
+
+} // namespace Core::Frontend

src/core/gdbstub/gdbstub.cpp

@@ -141,6 +141,7 @@ constexpr char target_xml[] =
 )";
 
 int gdbserver_socket = -1;
+bool defer_start = false;
 
 u8 command_buffer[GDB_BUFFER_SIZE];
 u32 command_length;
@@ -1166,6 +1167,9 @@ static void RemoveBreakpoint() {
 
 void HandlePacket() {
     if (!IsConnected()) {
+        if (defer_start) {
+            ToggleServer(true);
+        }
         return;
     }
 
@@ -1256,6 +1260,10 @@ void ToggleServer(bool status) {
     }
 }
 
+void DeferStart() {
+    defer_start = true;
+}
+
 static void Init(u16 port) {
     if (!server_enabled) {
         // Set the halt loop to false in case the user enabled the gdbstub mid-execution.
@@ -1341,6 +1349,7 @@ void Shutdown() {
     if (!server_enabled) {
         return;
     }
+    defer_start = false;
 
     LOG_INFO(Debug_GDBStub, "Stopping GDB ...");
     if (gdbserver_socket != -1) {

src/core/gdbstub/gdbstub.h

@@ -43,6 +43,13 @@ void ToggleServer(bool status);
 /// Start the gdbstub server.
 void Init();
 
+/**
+ * Defer initialization of the gdbstub to the first packet processing functions.
+ * This avoids a case where the gdbstub thread is frozen after initialization
+ * and fails to respond in time to packets.
+ */
+void DeferStart();
+
 /// Stop gdbstub server.
 void Shutdown();
 

src/video_core/engines/const_buffer_engine_interface.h

@@ -18,10 +18,14 @@ struct SamplerDescriptor {
     union {
         u32 raw = 0;
         BitField<0, 2, Tegra::Shader::TextureType> texture_type;
-        BitField<2, 3, Tegra::Texture::ComponentType> component_type;
+        BitField<2, 3, Tegra::Texture::ComponentType> r_type;
         BitField<5, 1, u32> is_array;
         BitField<6, 1, u32> is_buffer;
         BitField<7, 1, u32> is_shadow;
+        BitField<8, 3, Tegra::Texture::ComponentType> g_type;
+        BitField<11, 3, Tegra::Texture::ComponentType> b_type;
+        BitField<14, 3, Tegra::Texture::ComponentType> a_type;
+        BitField<17, 7, Tegra::Texture::TextureFormat> format;
     };
 
     bool operator==(const SamplerDescriptor& rhs) const noexcept {
@@ -36,9 +40,11 @@ struct SamplerDescriptor {
         using Tegra::Shader::TextureType;
         SamplerDescriptor result;
 
-        // This is going to be used to determine the shading language type.
-        // Because of that we don't care about all component types on color textures.
-        result.component_type.Assign(tic.r_type.Value());
+        result.format.Assign(tic.format.Value());
+        result.r_type.Assign(tic.r_type.Value());
+        result.g_type.Assign(tic.g_type.Value());
+        result.b_type.Assign(tic.b_type.Value());
+        result.a_type.Assign(tic.a_type.Value());
 
         switch (tic.texture_type.Value()) {
         case Tegra::Texture::TextureType::Texture1D:

src/video_core/engines/shader_bytecode.h

@@ -82,6 +82,10 @@ union Attribute {
         Position = 7,
         Attribute_0 = 8,
         Attribute_31 = 39,
+        FrontColor = 40,
+        FrontSecondaryColor = 41,
+        BackColor = 42,
+        BackSecondaryColor = 43,
         ClipDistances0123 = 44,
         ClipDistances4567 = 45,
         PointCoord = 46,
@@ -89,6 +93,8 @@ union Attribute {
         // shader, and a tuple of (TessCoord.x, TessCoord.y, TessCoord.z, ~) when inside a Tess Eval
         // shader.
         TessCoordInstanceIDVertexID = 47,
+        TexCoord_0 = 48,
+        TexCoord_7 = 55,
         // This attribute contains a tuple of (Unk, Unk, Unk, gl_FrontFacing) when inside a fragment
         // shader. It is unknown what the other values contain.
         FrontFacing = 63,

src/video_core/gpu.cpp

@@ -7,7 +7,6 @@
 #include "core/core.h"
 #include "core/core_timing.h"
 #include "core/core_timing_util.h"
-#include "core/frontend/emu_window.h"
 #include "core/memory.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/engines/kepler_compute.h"
@@ -17,15 +16,14 @@
 #include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
 #include "video_core/renderer_base.h"
-#include "video_core/video_core.h"
 
 namespace Tegra {
 
 MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
 
-GPU::GPU(Core::System& system, std::unique_ptr<VideoCore::RendererBase>&& renderer_, bool is_async)
-    : system{system}, renderer{std::move(renderer_)}, is_async{is_async} {
-    auto& rasterizer{renderer->Rasterizer()};
+GPU::GPU(Core::System& system, VideoCore::RendererBase& renderer, bool is_async)
+    : system{system}, renderer{renderer}, is_async{is_async} {
+    auto& rasterizer{renderer.Rasterizer()};
     memory_manager = std::make_unique<Tegra::MemoryManager>(system, rasterizer);
     dma_pusher = std::make_unique<Tegra::DmaPusher>(*this);
     maxwell_3d = std::make_unique<Engines::Maxwell3D>(system, rasterizer, *memory_manager);
@@ -139,7 +137,7 @@ u64 GPU::GetTicks() const {
 }
 
 void GPU::FlushCommands() {
-    renderer->Rasterizer().FlushCommands();
+    renderer.Rasterizer().FlushCommands();
 }
 
 // Note that, traditionally, methods are treated as 4-byte addressable locations, and hence

src/video_core/gpu.h

@@ -25,11 +25,8 @@ inline u8* FromCacheAddr(CacheAddr cache_addr) {
 }
 
 namespace Core {
-namespace Frontend {
-class EmuWindow;
-}
 class System;
-} // namespace Core
+}
 
 namespace VideoCore {
 class RendererBase;
@@ -132,8 +129,7 @@ class MemoryManager;
 
 class GPU {
 public:
-    explicit GPU(Core::System& system, std::unique_ptr<VideoCore::RendererBase>&& renderer,
-                 bool is_async);
+    explicit GPU(Core::System& system, VideoCore::RendererBase& renderer, bool is_async);
 
     virtual ~GPU();
 
@@ -178,14 +174,6 @@ public:
     /// Returns a reference to the GPU DMA pusher.
     Tegra::DmaPusher& DmaPusher();
 
-    VideoCore::RendererBase& Renderer() {
-        return *renderer;
-    }
-
-    const VideoCore::RendererBase& Renderer() const {
-        return *renderer;
-    }
-
     // Waits for the GPU to finish working
     virtual void WaitIdle() const = 0;
 
@@ -299,7 +287,7 @@ private:
 protected:
     std::unique_ptr<Tegra::DmaPusher> dma_pusher;
     Core::System& system;
-    std::unique_ptr<VideoCore::RendererBase> renderer;
+    VideoCore::RendererBase& renderer;
 
 private:
     std::unique_ptr<Tegra::MemoryManager> memory_manager;

src/video_core/gpu_asynch.cpp

@@ -10,16 +10,13 @@
 
 namespace VideoCommon {
 
-GPUAsynch::GPUAsynch(Core::System& system, std::unique_ptr<VideoCore::RendererBase>&& renderer_,
-                     std::unique_ptr<Core::Frontend::GraphicsContext>&& context)
-    : GPU(system, std::move(renderer_), true), gpu_thread{system}, gpu_context(std::move(context)),
-      cpu_context(renderer->GetRenderWindow().CreateSharedContext()) {}
+GPUAsynch::GPUAsynch(Core::System& system, VideoCore::RendererBase& renderer)
+    : GPU(system, renderer, true), gpu_thread{system} {}
 
 GPUAsynch::~GPUAsynch() = default;
 
 void GPUAsynch::Start() {
-    cpu_context->MakeCurrent();
-    gpu_thread.StartThread(*renderer, *gpu_context, *dma_pusher);
+    gpu_thread.StartThread(renderer, *dma_pusher);
 }
 
 void GPUAsynch::PushGPUEntries(Tegra::CommandList&& entries) {

src/video_core/gpu_asynch.h

@@ -7,10 +7,6 @@
 #include "video_core/gpu.h"
 #include "video_core/gpu_thread.h"
 
-namespace Core::Frontend {
-class GraphicsContext;
-}
-
 namespace VideoCore {
 class RendererBase;
 } // namespace VideoCore
@@ -20,8 +16,7 @@ namespace VideoCommon {
 /// Implementation of GPU interface that runs the GPU asynchronously
 class GPUAsynch final : public Tegra::GPU {
 public:
-    explicit GPUAsynch(Core::System& system, std::unique_ptr<VideoCore::RendererBase>&& renderer,
-                       std::unique_ptr<Core::Frontend::GraphicsContext>&& context);
+    explicit GPUAsynch(Core::System& system, VideoCore::RendererBase& renderer);
     ~GPUAsynch() override;
 
     void Start() override;
@@ -37,8 +32,6 @@ protected:
 
 private:
     GPUThread::ThreadManager gpu_thread;
-    std::unique_ptr<Core::Frontend::GraphicsContext> cpu_context;
-    std::unique_ptr<Core::Frontend::GraphicsContext> gpu_context;
 };
 
 } // namespace VideoCommon

src/video_core/gpu_synch.cpp

@@ -7,15 +7,12 @@
 
 namespace VideoCommon {
 
-GPUSynch::GPUSynch(Core::System& system, std::unique_ptr<VideoCore::RendererBase>&& renderer,
-                   std::unique_ptr<Core::Frontend::GraphicsContext>&& context)
-    : GPU(system, std::move(renderer), false), context{std::move(context)} {}
+GPUSynch::GPUSynch(Core::System& system, VideoCore::RendererBase& renderer)
+    : GPU(system, renderer, false) {}
 
 GPUSynch::~GPUSynch() = default;
 
-void GPUSynch::Start() {
-    context->MakeCurrent();
-}
+void GPUSynch::Start() {}
 
 void GPUSynch::PushGPUEntries(Tegra::CommandList&& entries) {
     dma_pusher->Push(std::move(entries));
@@ -23,19 +20,19 @@ void GPUSynch::PushGPUEntries(Tegra::CommandList&& entries) {
 }
 
 void GPUSynch::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
-    renderer->SwapBuffers(framebuffer);
+    renderer.SwapBuffers(framebuffer);
 }
 
 void GPUSynch::FlushRegion(CacheAddr addr, u64 size) {
-    renderer->Rasterizer().FlushRegion(addr, size);
+    renderer.Rasterizer().FlushRegion(addr, size);
 }
 
 void GPUSynch::InvalidateRegion(CacheAddr addr, u64 size) {
-    renderer->Rasterizer().InvalidateRegion(addr, size);
+    renderer.Rasterizer().InvalidateRegion(addr, size);
 }
 
 void GPUSynch::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
-    renderer->Rasterizer().FlushAndInvalidateRegion(addr, size);
+    renderer.Rasterizer().FlushAndInvalidateRegion(addr, size);
 }
 
 } // namespace VideoCommon

src/video_core/gpu_synch.h

@@ -6,10 +6,6 @@
 
 #include "video_core/gpu.h"
 
-namespace Core::Frontend {
-class GraphicsContext;
-}
-
 namespace VideoCore {
 class RendererBase;
 } // namespace VideoCore
@@ -19,8 +15,7 @@ namespace VideoCommon {
 /// Implementation of GPU interface that runs the GPU synchronously
 class GPUSynch final : public Tegra::GPU {
 public:
-    explicit GPUSynch(Core::System& system, std::unique_ptr<VideoCore::RendererBase>&& renderer,
-                      std::unique_ptr<Core::Frontend::GraphicsContext>&& context);
+    explicit GPUSynch(Core::System& system, VideoCore::RendererBase& renderer);
     ~GPUSynch() override;
 
     void Start() override;
@@ -34,9 +29,6 @@ public:
 protected:
     void TriggerCpuInterrupt([[maybe_unused]] u32 syncpoint_id,
                              [[maybe_unused]] u32 value) const override {}
-
-private:
-    std::unique_ptr<Core::Frontend::GraphicsContext> context;
 };
 
 } // namespace VideoCommon

src/video_core/gpu_thread.cpp

@@ -5,7 +5,7 @@
 #include "common/assert.h"
 #include "common/microprofile.h"
 #include "core/core.h"
-#include "core/frontend/emu_window.h"
+#include "core/frontend/scope_acquire_context.h"
 #include "video_core/dma_pusher.h"
 #include "video_core/gpu.h"
 #include "video_core/gpu_thread.h"
@@ -14,8 +14,8 @@
 namespace VideoCommon::GPUThread {
 
 /// Runs the GPU thread
-static void RunThread(VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context,
-                      Tegra::DmaPusher& dma_pusher, SynchState& state) {
+static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_pusher,
+                      SynchState& state) {
     MicroProfileOnThreadCreate("GpuThread");
 
     // Wait for first GPU command before acquiring the window context
@@ -27,7 +27,7 @@ static void RunThread(VideoCore::RendererBase& renderer, Core::Frontend::Graphic
         return;
     }
 
-    auto current_context = context.Acquire();
+    Core::Frontend::ScopeAcquireContext acquire_context{renderer.GetRenderWindow()};
 
     CommandDataContainer next;
     while (state.is_running) {
@@ -62,11 +62,8 @@ ThreadManager::~ThreadManager() {
     thread.join();
 }
 
-void ThreadManager::StartThread(VideoCore::RendererBase& renderer,
-                                Core::Frontend::GraphicsContext& context,
-                                Tegra::DmaPusher& dma_pusher) {
-    thread = std::thread{RunThread, std::ref(renderer), std::ref(context), std::ref(dma_pusher),
-                         std::ref(state)};
+void ThreadManager::StartThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_pusher) {
+    thread = std::thread{RunThread, std::ref(renderer), std::ref(dma_pusher), std::ref(state)};
 }
 
 void ThreadManager::SubmitList(Tegra::CommandList&& entries) {

src/video_core/gpu_thread.h

@@ -10,6 +10,7 @@
 #include <optional>
 #include <thread>
 #include <variant>
+
 #include "common/threadsafe_queue.h"
 #include "video_core/gpu.h"
 
@@ -19,9 +20,6 @@ class DmaPusher;
 } // namespace Tegra
 
 namespace Core {
-namespace Frontend {
-class GraphicsContext;
-}
 class System;
 } // namespace Core
 
@@ -101,8 +99,7 @@ public:
     ~ThreadManager();
 
     /// Creates and starts the GPU thread.
-    void StartThread(VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context,
-                     Tegra::DmaPusher& dma_pusher);
+    void StartThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_pusher);
 
     /// Push GPU command entries to be processed
     void SubmitList(Tegra::CommandList&& entries);

src/video_core/renderer_base.h

@@ -46,8 +46,7 @@ public:
 
     /// Draws the latest frame to the window waiting timeout_ms for a frame to arrive (Renderer
     /// specific implementation)
-    /// Returns true if a frame was drawn
-    virtual bool TryPresent(int timeout_ms) = 0;
+    virtual void TryPresent(int timeout_ms) = 0;
 
     // Getter/setter functions:
     // ------------------------

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -327,7 +327,8 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
 
     const auto worker = [&](Core::Frontend::GraphicsContext* context, std::size_t begin,
                             std::size_t end) {
-        const auto scope = context->Acquire();
+        context->MakeCurrent();
+        SCOPE_EXIT({ return context->DoneCurrent(); });
 
         for (std::size_t i = begin; i < end; ++i) {
             if (stop_loading) {

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -366,10 +366,19 @@ constexpr bool IsGenericAttribute(Attribute::Index index) {
     return index >= Attribute::Index::Attribute_0 && index <= Attribute::Index::Attribute_31;
 }
 
+constexpr bool IsLegacyTexCoord(Attribute::Index index) {
+    return static_cast<int>(index) >= static_cast<int>(Attribute::Index::TexCoord_0) &&
+           static_cast<int>(index) <= static_cast<int>(Attribute::Index::TexCoord_7);
+}
+
 constexpr Attribute::Index ToGenericAttribute(u64 value) {
     return static_cast<Attribute::Index>(value + static_cast<u64>(Attribute::Index::Attribute_0));
 }
 
+constexpr int GetLegacyTexCoordIndex(Attribute::Index index) {
+    return static_cast<int>(index) - static_cast<int>(Attribute::Index::TexCoord_0);
+}
+
 u32 GetGenericAttributeIndex(Attribute::Index index) {
     ASSERT(IsGenericAttribute(index));
     return static_cast<u32>(index) - static_cast<u32>(Attribute::Index::Attribute_0);
@@ -498,7 +507,7 @@ private:
         if (!identifier.empty()) {
             code.AddLine("// {}", identifier);
         }
-        code.AddLine("#version 440 core");
+        code.AddLine("#version 440 {}", ir.UsesLegacyVaryings() ? "compatibility" : "core");
         code.AddLine("#extension GL_ARB_separate_shader_objects : enable");
         if (device.HasShaderBallot()) {
             code.AddLine("#extension GL_ARB_shader_ballot : require");
@@ -561,6 +570,16 @@ private:
         if (stage != ShaderType::Fragment) {
             return;
         }
+        if (ir.UsesLegacyVaryings()) {
+            code.AddLine("in gl_PerFragment {{");
+            ++code.scope;
+            code.AddLine("vec4 gl_TexCoord[8];");
+            code.AddLine("vec4 gl_Color;");
+            code.AddLine("vec4 gl_SecondaryColor;");
+            --code.scope;
+            code.AddLine("}};");
+        }
+
         for (u32 rt = 0; rt < Maxwell::NumRenderTargets; ++rt) {
             code.AddLine("layout (location = {}) out vec4 frag_color{};", rt, rt);
         }
@@ -617,12 +636,12 @@ private:
             code.AddLine("float gl_PointSize;");
         }
 
-        if (ir.UsesInstanceId()) {
-            code.AddLine("int gl_InstanceID;");
-        }
-
-        if (ir.UsesVertexId()) {
-            code.AddLine("int gl_VertexID;");
+        if (ir.UsesLegacyVaryings()) {
+            code.AddLine("vec4 gl_TexCoord[8];");
+            code.AddLine("vec4 gl_FrontColor;");
+            code.AddLine("vec4 gl_FrontSecondaryColor;");
+            code.AddLine("vec4 gl_BackColor;");
+            code.AddLine("vec4 gl_BackSecondaryColor;");
         }
 
         --code.scope;
@@ -1128,6 +1147,10 @@ private:
             default:
                 UNREACHABLE();
             }
+        case Attribute::Index::FrontColor:
+            return {"gl_Color"s + GetSwizzle(element), Type::Float};
+        case Attribute::Index::FrontSecondaryColor:
+            return {"gl_SecondaryColor"s + GetSwizzle(element), Type::Float};
         case Attribute::Index::PointCoord:
             switch (element) {
             case 0:
@@ -1168,6 +1191,12 @@ private:
                 return {GeometryPass(GetGenericInputAttribute(attribute)) + GetSwizzle(element),
                         Type::Float};
             }
+            if (IsLegacyTexCoord(attribute)) {
+                UNIMPLEMENTED_IF(stage == ShaderType::Geometry);
+                return {fmt::format("gl_TexCoord[{}]{}", GetLegacyTexCoordIndex(attribute),
+                                    GetSwizzle(element)),
+                        Type::Float};
+            }
             break;
         }
         UNIMPLEMENTED_MSG("Unhandled input attribute: {}", static_cast<u32>(attribute));
@@ -1206,11 +1235,12 @@ private:
     }
 
     std::optional<Expression> GetOutputAttribute(const AbufNode* abuf) {
+        const u32 element = abuf->GetElement();
         switch (const auto attribute = abuf->GetIndex()) {
         case Attribute::Index::Position:
-            return {{"gl_Position"s + GetSwizzle(abuf->GetElement()), Type::Float}};
+            return {{"gl_Position"s + GetSwizzle(element), Type::Float}};
         case Attribute::Index::LayerViewportPointSize:
-            switch (abuf->GetElement()) {
+            switch (element) {
             case 0:
                 UNIMPLEMENTED();
                 return {};
@@ -1228,13 +1258,26 @@ private:
                 return {{"gl_PointSize", Type::Float}};
             }
             return {};
+        case Attribute::Index::FrontColor:
+            return {{"gl_FrontColor"s + GetSwizzle(element), Type::Float}};
+        case Attribute::Index::FrontSecondaryColor:
+            return {{"gl_FrontSecondaryColor"s + GetSwizzle(element), Type::Float}};
+        case Attribute::Index::BackColor:
+            return {{"gl_BackColor"s + GetSwizzle(element), Type::Float}};
+        case Attribute::Index::BackSecondaryColor:
+            return {{"gl_BackSecondaryColor"s + GetSwizzle(element), Type::Float}};
         case Attribute::Index::ClipDistances0123:
-            return {{fmt::format("gl_ClipDistance[{}]", abuf->GetElement()), Type::Float}};
+            return {{fmt::format("gl_ClipDistance[{}]", element), Type::Float}};
         case Attribute::Index::ClipDistances4567:
-            return {{fmt::format("gl_ClipDistance[{}]", abuf->GetElement() + 4), Type::Float}};
+            return {{fmt::format("gl_ClipDistance[{}]", element + 4), Type::Float}};
         default:
             if (IsGenericAttribute(attribute)) {
-                return {{GetGenericOutputAttribute(attribute, abuf->GetElement()), Type::Float}};
+                return {{GetGenericOutputAttribute(attribute, element), Type::Float}};
+            }
+            if (IsLegacyTexCoord(attribute)) {
+                return {{fmt::format("gl_TexCoord[{}]{}", GetLegacyTexCoordIndex(attribute),
+                                     GetSwizzle(element)),
+                         Type::Float}};
             }
             UNIMPLEMENTED_MSG("Unhandled output attribute: {}", static_cast<u32>(attribute));
             return {};

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -30,6 +30,8 @@ namespace OpenGL {
 
 namespace {
 
+// If the size of this is too small, it ends up creating a soft cap on FPS as the renderer will have
+// to wait on available presentation frames.
 constexpr std::size_t SWAP_CHAIN_SIZE = 3;
 
 struct Frame {
@@ -212,7 +214,7 @@ public:
     std::deque<Frame*> present_queue;
     Frame* previous_frame{};
 
-    FrameMailbox() {
+    FrameMailbox() : has_debug_tool{HasDebugTool()} {
         for (auto& frame : swap_chain) {
             free_queue.push(&frame);
         }
@@ -283,9 +285,13 @@ public:
         std::unique_lock lock{swap_chain_lock};
         present_queue.push_front(frame);
         present_cv.notify_one();
+
+        DebugNotifyNextFrame();
     }
 
     Frame* TryGetPresentFrame(int timeout_ms) {
+        DebugWaitForNextFrame();
+
         std::unique_lock lock{swap_chain_lock};
         // wait for new entries in the present_queue
         present_cv.wait_for(lock, std::chrono::milliseconds(timeout_ms),
@@ -311,12 +317,38 @@ public:
         previous_frame = frame;
         return frame;
     }
+
+private:
+    std::mutex debug_synch_mutex;
+    std::condition_variable debug_synch_condition;
+    std::atomic_int frame_for_debug{};
+    const bool has_debug_tool; // When true, using a GPU debugger, so keep frames in lock-step
+
+    /// Signal that a new frame is available (called from GPU thread)
+    void DebugNotifyNextFrame() {
+        if (!has_debug_tool) {
+            return;
+        }
+        frame_for_debug++;
+        std::lock_guard lock{debug_synch_mutex};
+        debug_synch_condition.notify_one();
+    }
+
+    /// Wait for a new frame to be available (called from presentation thread)
+    void DebugWaitForNextFrame() {
+        if (!has_debug_tool) {
+            return;
+        }
+        const int last_frame = frame_for_debug;
+        std::unique_lock lock{debug_synch_mutex};
+        debug_synch_condition.wait(lock,
+                                   [this, last_frame] { return frame_for_debug > last_frame; });
+    }
 };
 
-RendererOpenGL::RendererOpenGL(Core::Frontend::EmuWindow& emu_window, Core::System& system,
-                               Core::Frontend::GraphicsContext& context)
+RendererOpenGL::RendererOpenGL(Core::Frontend::EmuWindow& emu_window, Core::System& system)
     : VideoCore::RendererBase{emu_window}, emu_window{emu_window}, system{system},
-      frame_mailbox{}, context{context}, has_debug_tool{HasDebugTool()} {}
+      frame_mailbox{std::make_unique<FrameMailbox>()} {}
 
 RendererOpenGL::~RendererOpenGL() = default;
 
@@ -324,6 +356,8 @@ MICROPROFILE_DEFINE(OpenGL_RenderFrame, "OpenGL", "Render Frame", MP_RGB(128, 12
 MICROPROFILE_DEFINE(OpenGL_WaitPresent, "OpenGL", "Wait For Present", MP_RGB(128, 128, 128));
 
 void RendererOpenGL::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
+    render_window.PollEvents();
+
     if (!framebuffer) {
         return;
     }
@@ -379,13 +413,6 @@ void RendererOpenGL::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
         m_current_frame++;
         rasterizer->TickFrame();
     }
-
-    render_window.PollEvents();
-    if (has_debug_tool) {
-        glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
-        Present(0);
-        context.SwapBuffers();
-    }
 }
 
 void RendererOpenGL::PrepareRendertarget(const Tegra::FramebufferConfig* framebuffer) {
@@ -453,8 +480,6 @@ void RendererOpenGL::LoadColorToActiveGLTexture(u8 color_r, u8 color_g, u8 color
 }
 
 void RendererOpenGL::InitOpenGLObjects() {
-    frame_mailbox = std::make_unique<FrameMailbox>();
-
     glClearColor(Settings::values.bg_red, Settings::values.bg_green, Settings::values.bg_blue,
                  0.0f);
 
@@ -667,21 +692,12 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) {
     glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
 }
 
-bool RendererOpenGL::TryPresent(int timeout_ms) {
-    if (has_debug_tool) {
-        LOG_DEBUG(Render_OpenGL,
-                  "Skipping presentation because we are presenting on the main context");
-        return false;
-    }
-    return Present(timeout_ms);
-}
-
-bool RendererOpenGL::Present(int timeout_ms) {
+void RendererOpenGL::TryPresent(int timeout_ms) {
     const auto& layout = render_window.GetFramebufferLayout();
     auto frame = frame_mailbox->TryGetPresentFrame(timeout_ms);
     if (!frame) {
         LOG_DEBUG(Render_OpenGL, "TryGetPresentFrame returned no frame to present");
-        return false;
+        return;
     }
 
     // Clearing before a full overwrite of a fbo can signal to drivers that they can avoid a
@@ -709,7 +725,6 @@ bool RendererOpenGL::Present(int timeout_ms) {
     glFlush();
 
     glBindFramebuffer(GL_READ_FRAMEBUFFER, 0);
-    return true;
 }
 
 void RendererOpenGL::RenderScreenshot() {

src/video_core/renderer_opengl/renderer_opengl.h

@@ -55,14 +55,13 @@ class FrameMailbox;
 
 class RendererOpenGL final : public VideoCore::RendererBase {
 public:
-    explicit RendererOpenGL(Core::Frontend::EmuWindow& emu_window, Core::System& system,
-                            Core::Frontend::GraphicsContext& context);
+    explicit RendererOpenGL(Core::Frontend::EmuWindow& emu_window, Core::System& system);
     ~RendererOpenGL() override;
 
     bool Init() override;
     void ShutDown() override;
     void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override;
-    bool TryPresent(int timeout_ms) override;
+    void TryPresent(int timeout_ms) override;
 
 private:
     /// Initializes the OpenGL state and creates persistent objects.
@@ -90,11 +89,8 @@ private:
 
     void PrepareRendertarget(const Tegra::FramebufferConfig* framebuffer);
 
-    bool Present(int timeout_ms);
-
     Core::Frontend::EmuWindow& emu_window;
     Core::System& system;
-    Core::Frontend::GraphicsContext& context;
 
     StateTracker state_tracker{system};
 
@@ -119,8 +115,6 @@ private:
 
     /// Frame presentation mailbox
     std::unique_ptr<FrameMailbox> frame_mailbox;
-
-    bool has_debug_tool = false;
 };
 
 } // namespace OpenGL

src/video_core/renderer_vulkan/renderer_vulkan.cpp

@@ -141,9 +141,8 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
     render_window.PollEvents();
 }
 
-bool RendererVulkan::TryPresent(int /*timeout_ms*/) {
+void RendererVulkan::TryPresent(int /*timeout_ms*/) {
     // TODO (bunnei): ImplementMe
-    return true;
 }
 
 bool RendererVulkan::Init() {

src/video_core/renderer_vulkan/renderer_vulkan.h

@@ -42,7 +42,7 @@ public:
     bool Init() override;
     void ShutDown() override;
     void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override;
-    bool TryPresent(int timeout_ms) override;
+    void TryPresent(int timeout_ms) override;
 
 private:
     std::optional<vk::DebugUtilsMessengerEXT> CreateDebugCallback(

src/video_core/shader/decode/xmad.cpp

@@ -31,7 +31,7 @@ u32 ShaderIR::DecodeXmad(NodeBlock& bb, u32 pc) {
     const bool is_signed_b = instr.xmad.sign_b == 1;
     const bool is_signed_c = is_signed_a;
 
-    auto [is_merge, is_psl, is_high_b, mode, op_b,
+    auto [is_merge, is_psl, is_high_b, mode, op_b_binding,
           op_c] = [&]() -> std::tuple<bool, bool, bool, Tegra::Shader::XmadMode, Node, Node> {
         switch (opcode->get().GetId()) {
         case OpCode::Id::XMAD_CR:
@@ -67,9 +67,10 @@ u32 ShaderIR::DecodeXmad(NodeBlock& bb, u32 pc) {
     op_a = SignedOperation(OperationCode::IBitfieldExtract, is_signed_a, std::move(op_a),
                            instr.xmad.high_a ? Immediate(16) : Immediate(0), Immediate(16));
 
-    const Node original_b = op_b;
-    op_b = SignedOperation(OperationCode::IBitfieldExtract, is_signed_b, std::move(op_b),
-                           is_high_b ? Immediate(16) : Immediate(0), Immediate(16));
+    const Node original_b = op_b_binding;
+    const Node op_b =
+        SignedOperation(OperationCode::IBitfieldExtract, is_signed_b, std::move(op_b_binding),
+                        is_high_b ? Immediate(16) : Immediate(0), Immediate(16));
 
     // we already check sign_a and sign_b is difference or not before so just use one in here.
     Node product = SignedOperation(OperationCode::IMul, is_signed_a, op_a, op_b);

src/video_core/shader/shader_ir.cpp

@@ -96,6 +96,7 @@ Node ShaderIR::GetPredicate(bool immediate) {
 }
 
 Node ShaderIR::GetInputAttribute(Attribute::Index index, u64 element, Node buffer) {
+    MarkAttributeUsage(index, element);
     used_input_attributes.emplace(index);
     return MakeNode<AbufNode>(index, static_cast<u32>(element), std::move(buffer));
 }
@@ -106,42 +107,8 @@ Node ShaderIR::GetPhysicalInputAttribute(Tegra::Shader::Register physical_addres
 }
 
 Node ShaderIR::GetOutputAttribute(Attribute::Index index, u64 element, Node buffer) {
-    if (index == Attribute::Index::LayerViewportPointSize) {
-        switch (element) {
-        case 0:
-            UNIMPLEMENTED();
-            break;
-        case 1:
-            uses_layer = true;
-            break;
-        case 2:
-            uses_viewport_index = true;
-            break;
-        case 3:
-            uses_point_size = true;
-            break;
-        }
-    }
-    if (index == Attribute::Index::TessCoordInstanceIDVertexID) {
-        switch (element) {
-        case 2:
-            uses_instance_id = true;
-            break;
-        case 3:
-            uses_vertex_id = true;
-            break;
-        default:
-            break;
-        }
-    }
-    if (index == Attribute::Index::ClipDistances0123 ||
-        index == Attribute::Index::ClipDistances4567) {
-        const auto clip_index =
-            static_cast<u32>((index == Attribute::Index::ClipDistances4567 ? 1 : 0) + element);
-        used_clip_distances.at(clip_index) = true;
-    }
+    MarkAttributeUsage(index, element);
     used_output_attributes.insert(index);
-
     return MakeNode<AbufNode>(index, static_cast<u32>(element), std::move(buffer));
 }
 
@@ -452,6 +419,54 @@ Node ShaderIR::BitfieldInsert(Node base, Node insert, u32 offset, u32 bits) {
                      Immediate(bits));
 }
 
+void ShaderIR::MarkAttributeUsage(Attribute::Index index, u64 element) {
+    switch (index) {
+    case Attribute::Index::LayerViewportPointSize:
+        switch (element) {
+        case 0:
+            UNIMPLEMENTED();
+            break;
+        case 1:
+            uses_layer = true;
+            break;
+        case 2:
+            uses_viewport_index = true;
+            break;
+        case 3:
+            uses_point_size = true;
+            break;
+        }
+        break;
+    case Attribute::Index::TessCoordInstanceIDVertexID:
+        switch (element) {
+        case 2:
+            uses_instance_id = true;
+            break;
+        case 3:
+            uses_vertex_id = true;
+            break;
+        }
+        break;
+    case Attribute::Index::ClipDistances0123:
+    case Attribute::Index::ClipDistances4567: {
+        const u64 clip_index = (index == Attribute::Index::ClipDistances4567 ? 4 : 0) + element;
+        used_clip_distances.at(clip_index) = true;
+        break;
+    }
+    case Attribute::Index::FrontColor:
+    case Attribute::Index::FrontSecondaryColor:
+    case Attribute::Index::BackColor:
+    case Attribute::Index::BackSecondaryColor:
+        uses_legacy_varyings = true;
+        break;
+    default:
+        if (index >= Attribute::Index::TexCoord_0 && index <= Attribute::Index::TexCoord_7) {
+            uses_legacy_varyings = true;
+        }
+        break;
+    }
+}
+
 std::size_t ShaderIR::DeclareAmend(Node new_amend) {
     const std::size_t id = amend_code.size();
     amend_code.push_back(new_amend);

src/video_core/shader/shader_ir.h

@@ -137,6 +137,10 @@ public:
         return uses_vertex_id;
     }
 
+    bool UsesLegacyVaryings() const {
+        return uses_legacy_varyings;
+    }
+
     bool UsesWarps() const {
         return uses_warps;
     }
@@ -343,6 +347,9 @@ private:
     /// Inserts a sequence of bits from a node
     Node BitfieldInsert(Node base, Node insert, u32 offset, u32 bits);
 
+    /// Marks the usage of a input or output attribute.
+    void MarkAttributeUsage(Tegra::Shader::Attribute::Index index, u64 element);
+
     void WriteTexInstructionFloat(NodeBlock& bb, Tegra::Shader::Instruction instr,
                                   const Node4& components);
 
@@ -443,6 +450,7 @@ private:
     bool uses_physical_attributes{}; // Shader uses AL2P or physical attribute read/writes
     bool uses_instance_id{};
     bool uses_vertex_id{};
+    bool uses_legacy_varyings{};
     bool uses_warps{};
     bool uses_indexed_samplers{};
 

src/video_core/video_core.cpp

@@ -15,13 +15,13 @@
 #endif
 #include "video_core/video_core.h"
 
-namespace {
-std::unique_ptr<VideoCore::RendererBase> CreateRenderer(Core::Frontend::EmuWindow& emu_window,
-                                                        Core::System& system,
-                                                        Core::Frontend::GraphicsContext& context) {
+namespace VideoCore {
+
+std::unique_ptr<RendererBase> CreateRenderer(Core::Frontend::EmuWindow& emu_window,
+                                             Core::System& system) {
     switch (Settings::values.renderer_backend) {
     case Settings::RendererBackend::OpenGL:
-        return std::make_unique<OpenGL::RendererOpenGL>(emu_window, system, context);
+        return std::make_unique<OpenGL::RendererOpenGL>(emu_window, system);
 #ifdef HAS_VULKAN
     case Settings::RendererBackend::Vulkan:
         return std::make_unique<Vulkan::RendererVulkan>(emu_window, system);
@@ -30,23 +30,13 @@ std::unique_ptr<VideoCore::RendererBase> CreateRenderer(Core::Frontend::EmuWindo
         return nullptr;
     }
 }
-} // Anonymous namespace
-
-namespace VideoCore {
-
-std::unique_ptr<Tegra::GPU> CreateGPU(Core::Frontend::EmuWindow& emu_window, Core::System& system) {
-    auto context = emu_window.CreateSharedContext();
-    const auto scope = context->Acquire();
-    auto renderer = CreateRenderer(emu_window, system, *context);
-    if (!renderer->Init()) {
-        return nullptr;
-    }
 
+std::unique_ptr<Tegra::GPU> CreateGPU(Core::System& system) {
     if (Settings::values.use_asynchronous_gpu_emulation) {
-        return std::make_unique<VideoCommon::GPUAsynch>(system, std::move(renderer),
-                                                        std::move(context));
+        return std::make_unique<VideoCommon::GPUAsynch>(system, system.Renderer());
     }
-    return std::make_unique<VideoCommon::GPUSynch>(system, std::move(renderer), std::move(context));
+
+    return std::make_unique<VideoCommon::GPUSynch>(system, system.Renderer());
 }
 
 u16 GetResolutionScaleFactor(const RendererBase& renderer) {

src/video_core/video_core.h

@@ -22,8 +22,17 @@ namespace VideoCore {
 
 class RendererBase;
 
+/**
+ * Creates a renderer instance.
+ *
+ * @note The returned renderer instance is simply allocated. Its Init()
+ *       function still needs to be called to fully complete its setup.
+ */
+std::unique_ptr<RendererBase> CreateRenderer(Core::Frontend::EmuWindow& emu_window,
+                                             Core::System& system);
+
 /// Creates an emulated GPU instance using the given system context.
-std::unique_ptr<Tegra::GPU> CreateGPU(Core::Frontend::EmuWindow& emu_window, Core::System& system);
+std::unique_ptr<Tegra::GPU> CreateGPU(Core::System& system);
 
 u16 GetResolutionScaleFactor(const RendererBase& renderer);
 

src/yuzu/bootmanager.cpp

@@ -10,6 +10,9 @@
 #include <QMessageBox>
 #include <QOffscreenSurface>
 #include <QOpenGLContext>
+#include <QOpenGLFunctions>
+#include <QOpenGLFunctions_4_3_Core>
+#include <QOpenGLWindow>
 #include <QPainter>
 #include <QScreen>
 #include <QStringList>
@@ -26,6 +29,7 @@
 #include "common/scope_exit.h"
 #include "core/core.h"
 #include "core/frontend/framebuffer_layout.h"
+#include "core/frontend/scope_acquire_context.h"
 #include "core/settings.h"
 #include "input_common/keyboard.h"
 #include "input_common/main.h"
@@ -35,16 +39,26 @@
 #include "yuzu/bootmanager.h"
 #include "yuzu/main.h"
 
-EmuThread::EmuThread() = default;
+EmuThread::EmuThread(GRenderWindow& window)
+    : shared_context{window.CreateSharedContext()},
+      context{(Settings::values.use_asynchronous_gpu_emulation && shared_context) ? *shared_context
+                                                                                  : window} {}
 
 EmuThread::~EmuThread() = default;
 
+static GMainWindow* GetMainWindow() {
+    for (QWidget* w : qApp->topLevelWidgets()) {
+        if (GMainWindow* main = qobject_cast<GMainWindow*>(w)) {
+            return main;
+        }
+    }
+    return nullptr;
+}
+
 void EmuThread::run() {
     MicroProfileOnThreadCreate("EmuThread");
 
-    // Main process has been loaded. Make the context current to this thread and begin GPU and CPU
-    // execution.
-    Core::System::GetInstance().GPU().Start();
+    Core::Frontend::ScopeAcquireContext acquire_context{context};
 
     emit LoadProgress(VideoCore::LoadCallbackStage::Prepare, 0, 0);
 
@@ -97,156 +111,162 @@ void EmuThread::run() {
 #endif
 }
 
-class OpenGLSharedContext : public Core::Frontend::GraphicsContext {
+class GGLContext : public Core::Frontend::GraphicsContext {
 public:
-    /// Create the original context that should be shared from
-    explicit OpenGLSharedContext(QSurface* surface) : surface(surface) {
-        QSurfaceFormat format;
-        format.setVersion(4, 3);
-        format.setProfile(QSurfaceFormat::CompatibilityProfile);
-        format.setOption(QSurfaceFormat::FormatOption::DeprecatedFunctions);
-        // TODO: expose a setting for buffer value (ie default/single/double/triple)
-        format.setSwapBehavior(QSurfaceFormat::DefaultSwapBehavior);
-        format.setSwapInterval(0);
-
-        context = std::make_unique<QOpenGLContext>();
-        context->setFormat(format);
-        if (!context->create()) {
-            LOG_ERROR(Frontend, "Unable to create main openGL context");
-        }
-    }
-
-    /// Create the shared contexts for rendering and presentation
-    explicit OpenGLSharedContext(QOpenGLContext* share_context, QSurface* main_surface = nullptr) {
+    explicit GGLContext(QOpenGLContext* shared_context)
+        : context(new QOpenGLContext(shared_context->parent())),
+          surface(new QOffscreenSurface(nullptr)) {
 
         // disable vsync for any shared contexts
-        auto format = share_context->format();
-        format.setSwapInterval(main_surface ? Settings::values.use_vsync : 0);
+        auto format = shared_context->format();
+        format.setSwapInterval(0);
 
-        context = std::make_unique<QOpenGLContext>();
-        context->setShareContext(share_context);
+        context->setShareContext(shared_context);
         context->setFormat(format);
-        if (!context->create()) {
-            LOG_ERROR(Frontend, "Unable to create shared openGL context");
-        }
-
-        if (!main_surface) {
-            offscreen_surface = std::make_unique<QOffscreenSurface>(nullptr);
-            offscreen_surface->setFormat(format);
-            offscreen_surface->create();
-            surface = offscreen_surface.get();
-        } else {
-            surface = main_surface;
-        }
-    }
-
-    ~OpenGLSharedContext() {
-        DoneCurrent();
-    }
-
-    void SwapBuffers() override {
-        context->swapBuffers(surface);
+        context->create();
+        surface->setParent(shared_context->parent());
+        surface->setFormat(format);
+        surface->create();
     }
 
     void MakeCurrent() override {
-        if (is_current) {
-            return;
-        }
-        is_current = context->makeCurrent(surface);
+        context->makeCurrent(surface);
     }
 
     void DoneCurrent() override {
-        if (!is_current) {
-            return;
-        }
         context->doneCurrent();
-        is_current = false;
-    }
-
-    QOpenGLContext* GetShareContext() {
-        return context.get();
-    }
-
-    const QOpenGLContext* GetShareContext() const {
-        return context.get();
     }
 
 private:
-    // Avoid using Qt parent system here since we might move the QObjects to new threads
-    // As a note, this means we should avoid using slots/signals with the objects too
-    std::unique_ptr<QOpenGLContext> context;
-    std::unique_ptr<QOffscreenSurface> offscreen_surface{};
-    QSurface* surface;
-    bool is_current = false;
+    QOpenGLContext* context;
+    QOffscreenSurface* surface;
 };
 
-class DummyContext : public Core::Frontend::GraphicsContext {};
-
-class RenderWidget : public QWidget {
+class ChildRenderWindow : public QWindow {
 public:
-    explicit RenderWidget(GRenderWindow* parent) : QWidget(parent), render_window(parent) {
-        setAttribute(Qt::WA_NativeWindow);
-        setAttribute(Qt::WA_PaintOnScreen);
+    ChildRenderWindow(QWindow* parent, QWidget* event_handler)
+        : QWindow{parent}, event_handler{event_handler} {}
+
+    virtual ~ChildRenderWindow() = default;
+
+    virtual void Present() = 0;
+
+protected:
+    bool event(QEvent* event) override {
+        switch (event->type()) {
+        case QEvent::UpdateRequest:
+            Present();
+            return true;
+        case QEvent::MouseButtonPress:
+        case QEvent::MouseButtonRelease:
+        case QEvent::MouseButtonDblClick:
+        case QEvent::MouseMove:
+        case QEvent::KeyPress:
+        case QEvent::KeyRelease:
+        case QEvent::FocusIn:
+        case QEvent::FocusOut:
+        case QEvent::FocusAboutToChange:
+        case QEvent::Enter:
+        case QEvent::Leave:
+        case QEvent::Wheel:
+        case QEvent::TabletMove:
+        case QEvent::TabletPress:
+        case QEvent::TabletRelease:
+        case QEvent::TabletEnterProximity:
+        case QEvent::TabletLeaveProximity:
+        case QEvent::TouchBegin:
+        case QEvent::TouchUpdate:
+        case QEvent::TouchEnd:
+        case QEvent::InputMethodQuery:
+        case QEvent::TouchCancel:
+            return QCoreApplication::sendEvent(event_handler, event);
+        case QEvent::Drop:
+            GetMainWindow()->DropAction(static_cast<QDropEvent*>(event));
+            return true;
+        case QEvent::DragResponse:
+        case QEvent::DragEnter:
+        case QEvent::DragLeave:
+        case QEvent::DragMove:
+            GetMainWindow()->AcceptDropEvent(static_cast<QDropEvent*>(event));
+            return true;
+        default:
+            return QWindow::event(event);
+        }
     }
 
-    virtual ~RenderWidget() = default;
-
-    /// Called on the UI thread when this Widget is ready to draw
-    /// Dervied classes can override this to draw the latest frame.
-    virtual void Present() {}
-
-    void paintEvent(QPaintEvent* event) override {
-        Present();
-        update();
-    }
-
-    QPaintEngine* paintEngine() const override {
-        return nullptr;
+    void exposeEvent(QExposeEvent* event) override {
+        QWindow::requestUpdate();
+        QWindow::exposeEvent(event);
     }
 
 private:
-    GRenderWindow* render_window;
+    QWidget* event_handler{};
 };
 
-class OpenGLRenderWidget : public RenderWidget {
+class OpenGLWindow final : public ChildRenderWindow {
 public:
-    explicit OpenGLRenderWidget(GRenderWindow* parent) : RenderWidget(parent) {
-        windowHandle()->setSurfaceType(QWindow::OpenGLSurface);
+    OpenGLWindow(QWindow* parent, QWidget* event_handler, QOpenGLContext* shared_context)
+        : ChildRenderWindow{parent, event_handler},
+          context(new QOpenGLContext(shared_context->parent())) {
+
+        // disable vsync for any shared contexts
+        auto format = shared_context->format();
+        format.setSwapInterval(Settings::values.use_vsync ? 1 : 0);
+        this->setFormat(format);
+
+        context->setShareContext(shared_context);
+        context->setScreen(this->screen());
+        context->setFormat(format);
+        context->create();
+
+        setSurfaceType(QWindow::OpenGLSurface);
+
+        // TODO: One of these flags might be interesting: WA_OpaquePaintEvent, WA_NoBackground,
+        // WA_DontShowOnScreen, WA_DeleteOnClose
     }
 
-    void SetContext(std::unique_ptr<Core::Frontend::GraphicsContext>&& context_) {
-        context = std::move(context_);
+    ~OpenGLWindow() override {
+        context->doneCurrent();
     }
 
     void Present() override {
-        if (!isVisible()) {
+        if (!isExposed()) {
             return;
         }
 
-        context->MakeCurrent();
-        glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
-        if (Core::System::GetInstance().Renderer().TryPresent(100)) {
-            context->SwapBuffers();
-            glFinish();
-        }
+        context->makeCurrent(this);
+        Core::System::GetInstance().Renderer().TryPresent(100);
+        context->swapBuffers(this);
+        auto f = context->versionFunctions<QOpenGLFunctions_4_3_Core>();
+        f->glFinish();
+        QWindow::requestUpdate();
     }
 
 private:
-    std::unique_ptr<Core::Frontend::GraphicsContext> context{};
+    QOpenGLContext* context{};
 };
 
 #ifdef HAS_VULKAN
-class VulkanRenderWidget : public RenderWidget {
+class VulkanWindow final : public ChildRenderWindow {
 public:
-    explicit VulkanRenderWidget(GRenderWindow* parent, QVulkanInstance* instance)
-        : RenderWidget(parent) {
-        windowHandle()->setSurfaceType(QWindow::VulkanSurface);
-        windowHandle()->setVulkanInstance(instance);
+    VulkanWindow(QWindow* parent, QWidget* event_handler, QVulkanInstance* instance)
+        : ChildRenderWindow{parent, event_handler} {
+        setSurfaceType(QSurface::SurfaceType::VulkanSurface);
+        setVulkanInstance(instance);
     }
+
+    ~VulkanWindow() override = default;
+
+    void Present() override {
+        // TODO(bunnei): ImplementMe
+    }
+
+private:
+    QWidget* event_handler{};
 };
 #endif
 
-GRenderWindow::GRenderWindow(GMainWindow* parent_, EmuThread* emu_thread)
+GRenderWindow::GRenderWindow(QWidget* parent_, EmuThread* emu_thread)
     : QWidget(parent_), emu_thread(emu_thread) {
     setWindowTitle(QStringLiteral("yuzu %1 | %2-%3")
                        .arg(QString::fromUtf8(Common::g_build_name),
@@ -258,13 +278,26 @@ GRenderWindow::GRenderWindow(GMainWindow* parent_, EmuThread* emu_thread)
     setLayout(layout);
     InputCommon::Init();
 
-    connect(this, &GRenderWindow::FirstFrameDisplayed, parent_, &GMainWindow::OnLoadComplete);
+    GMainWindow* parent = GetMainWindow();
+    connect(this, &GRenderWindow::FirstFrameDisplayed, parent, &GMainWindow::OnLoadComplete);
 }
 
 GRenderWindow::~GRenderWindow() {
     InputCommon::Shutdown();
 }
 
+void GRenderWindow::MakeCurrent() {
+    if (core_context) {
+        core_context->MakeCurrent();
+    }
+}
+
+void GRenderWindow::DoneCurrent() {
+    if (core_context) {
+        core_context->DoneCurrent();
+    }
+}
+
 void GRenderWindow::PollEvents() {
     if (!first_frame) {
         first_frame = true;
@@ -276,6 +309,21 @@ bool GRenderWindow::IsShown() const {
     return !isMinimized();
 }
 
+void GRenderWindow::RetrieveVulkanHandlers(void* get_instance_proc_addr, void* instance,
+                                           void* surface) const {
+#ifdef HAS_VULKAN
+    const auto instance_proc_addr = vk_instance->getInstanceProcAddr("vkGetInstanceProcAddr");
+    const VkInstance instance_copy = vk_instance->vkInstance();
+    const VkSurfaceKHR surface_copy = vk_instance->surfaceForWindow(child_window);
+
+    std::memcpy(get_instance_proc_addr, &instance_proc_addr, sizeof(instance_proc_addr));
+    std::memcpy(instance, &instance_copy, sizeof(instance_copy));
+    std::memcpy(surface, &surface_copy, sizeof(surface_copy));
+#else
+    UNREACHABLE_MSG("Executing Vulkan code without compiling Vulkan");
+#endif
+}
+
 // On Qt 5.0+, this correctly gets the size of the framebuffer (pixels).
 //
 // Older versions get the window size (density independent pixels),
@@ -319,7 +367,7 @@ qreal GRenderWindow::windowPixelRatio() const {
     return devicePixelRatio();
 }
 
-std::pair<u32, u32> GRenderWindow::ScaleTouch(const QPointF& pos) const {
+std::pair<u32, u32> GRenderWindow::ScaleTouch(const QPointF pos) const {
     const qreal pixel_ratio = windowPixelRatio();
     return {static_cast<u32>(std::max(std::round(pos.x() * pixel_ratio), qreal{0.0})),
             static_cast<u32>(std::max(std::round(pos.y() * pixel_ratio), qreal{0.0}))};
@@ -339,10 +387,8 @@ void GRenderWindow::keyReleaseEvent(QKeyEvent* event) {
 }
 
 void GRenderWindow::mousePressEvent(QMouseEvent* event) {
-    // touch input is handled in TouchBeginEvent
-    if (event->source() == Qt::MouseEventSynthesizedBySystem) {
-        return;
-    }
+    if (event->source() == Qt::MouseEventSynthesizedBySystem)
+        return; // touch input is handled in TouchBeginEvent
 
     auto pos = event->pos();
     if (event->button() == Qt::LeftButton) {
@@ -354,10 +400,8 @@ void GRenderWindow::mousePressEvent(QMouseEvent* event) {
 }
 
 void GRenderWindow::mouseMoveEvent(QMouseEvent* event) {
-    // touch input is handled in TouchUpdateEvent
-    if (event->source() == Qt::MouseEventSynthesizedBySystem) {
-        return;
-    }
+    if (event->source() == Qt::MouseEventSynthesizedBySystem)
+        return; // touch input is handled in TouchUpdateEvent
 
     auto pos = event->pos();
     const auto [x, y] = ScaleTouch(pos);
@@ -366,16 +410,13 @@ void GRenderWindow::mouseMoveEvent(QMouseEvent* event) {
 }
 
 void GRenderWindow::mouseReleaseEvent(QMouseEvent* event) {
-    // touch input is handled in TouchEndEvent
-    if (event->source() == Qt::MouseEventSynthesizedBySystem) {
-        return;
-    }
+    if (event->source() == Qt::MouseEventSynthesizedBySystem)
+        return; // touch input is handled in TouchEndEvent
 
-    if (event->button() == Qt::LeftButton) {
+    if (event->button() == Qt::LeftButton)
         this->TouchReleased();
-    } else if (event->button() == Qt::RightButton) {
+    else if (event->button() == Qt::RightButton)
         InputCommon::GetMotionEmu()->EndTilt();
-    }
 }
 
 void GRenderWindow::TouchBeginEvent(const QTouchEvent* event) {
@@ -433,13 +474,9 @@ void GRenderWindow::resizeEvent(QResizeEvent* event) {
 
 std::unique_ptr<Core::Frontend::GraphicsContext> GRenderWindow::CreateSharedContext() const {
     if (Settings::values.renderer_backend == Settings::RendererBackend::OpenGL) {
-        auto c = static_cast<OpenGLSharedContext*>(main_context.get());
-        // Bind the shared contexts to the main surface in case the backend wants to take over
-        // presentation
-        return std::make_unique<OpenGLSharedContext>(c->GetShareContext(),
-                                                     child_widget->windowHandle());
+        return std::make_unique<GGLContext>(QOpenGLContext::globalShareContext());
     }
-    return std::make_unique<DummyContext>();
+    return {};
 }
 
 bool GRenderWindow::InitRenderTarget() {
@@ -460,11 +497,14 @@ bool GRenderWindow::InitRenderTarget() {
         break;
     }
 
-    child_widget->resize(Layout::ScreenUndocked::Width, Layout::ScreenUndocked::Height);
-    layout()->addWidget(child_widget);
     // Reset minimum required size to avoid resizing issues on the main window after restarting.
     setMinimumSize(1, 1);
 
+    // Show causes the window to actually be created and the gl context as well, but we don't want
+    // the widget to be shown yet, so immediately hide it.
+    show();
+    hide();
+
     resize(Layout::ScreenUndocked::Width, Layout::ScreenUndocked::Height);
 
     OnMinimalClientAreaChangeRequest(GetActiveConfig().min_client_area_size);
@@ -483,10 +523,9 @@ bool GRenderWindow::InitRenderTarget() {
 void GRenderWindow::ReleaseRenderTarget() {
     if (child_widget) {
         layout()->removeWidget(child_widget);
-        child_widget->deleteLater();
+        delete child_widget;
         child_widget = nullptr;
     }
-    main_context.reset();
 }
 
 void GRenderWindow::CaptureScreenshot(u32 res_scale, const QString& screenshot_path) {
@@ -518,13 +557,24 @@ void GRenderWindow::OnMinimalClientAreaChangeRequest(std::pair<u32, u32> minimal
 bool GRenderWindow::InitializeOpenGL() {
     // TODO: One of these flags might be interesting: WA_OpaquePaintEvent, WA_NoBackground,
     // WA_DontShowOnScreen, WA_DeleteOnClose
-    auto child = new OpenGLRenderWidget(this);
-    child_widget = child;
-    child_widget->windowHandle()->create();
-    auto context = std::make_shared<OpenGLSharedContext>(child->windowHandle());
-    main_context = context;
-    child->SetContext(
-        std::make_unique<OpenGLSharedContext>(context->GetShareContext(), child->windowHandle()));
+    QSurfaceFormat fmt;
+    fmt.setVersion(4, 3);
+    fmt.setProfile(QSurfaceFormat::CompatibilityProfile);
+    fmt.setOption(QSurfaceFormat::FormatOption::DeprecatedFunctions);
+    // TODO: expose a setting for buffer value (ie default/single/double/triple)
+    fmt.setSwapBehavior(QSurfaceFormat::DefaultSwapBehavior);
+    fmt.setSwapInterval(0);
+    QSurfaceFormat::setDefaultFormat(fmt);
+
+    GMainWindow* parent = GetMainWindow();
+    QWindow* parent_win_handle = parent ? parent->windowHandle() : nullptr;
+    child_window = new OpenGLWindow(parent_win_handle, this, QOpenGLContext::globalShareContext());
+    child_window->create();
+    child_widget = createWindowContainer(child_window, this);
+    child_widget->resize(Layout::ScreenUndocked::Width, Layout::ScreenUndocked::Height);
+    layout()->addWidget(child_widget);
+
+    core_context = CreateSharedContext();
 
     return true;
 }
@@ -554,10 +604,13 @@ bool GRenderWindow::InitializeVulkan() {
         return false;
     }
 
-    auto child = new VulkanRenderWidget(this, vk_instance.get());
-    child_widget = child;
-    child_widget->windowHandle()->create();
-    main_context = std::make_unique<DummyContext>();
+    GMainWindow* parent = GetMainWindow();
+    QWindow* parent_win_handle = parent ? parent->windowHandle() : nullptr;
+    child_window = new VulkanWindow(parent_win_handle, this, vk_instance.get());
+    child_window->create();
+    child_widget = createWindowContainer(child_window, this);
+    child_widget->resize(Layout::ScreenUndocked::Width, Layout::ScreenUndocked::Height);
+    layout()->addWidget(child_widget);
 
     return true;
 #else
@@ -567,24 +620,8 @@ bool GRenderWindow::InitializeVulkan() {
 #endif
 }
 
-void GRenderWindow::RetrieveVulkanHandlers(void* get_instance_proc_addr, void* instance,
-                                           void* surface) const {
-#ifdef HAS_VULKAN
-    const auto instance_proc_addr = vk_instance->getInstanceProcAddr("vkGetInstanceProcAddr");
-    const VkInstance instance_copy = vk_instance->vkInstance();
-    const VkSurfaceKHR surface_copy = vk_instance->surfaceForWindow(child_widget->windowHandle());
-
-    std::memcpy(get_instance_proc_addr, &instance_proc_addr, sizeof(instance_proc_addr));
-    std::memcpy(instance, &instance_copy, sizeof(instance_copy));
-    std::memcpy(surface, &surface_copy, sizeof(surface_copy));
-#else
-    UNREACHABLE_MSG("Executing Vulkan code without compiling Vulkan");
-#endif
-}
-
 bool GRenderWindow::LoadOpenGL() {
-    auto context = CreateSharedContext();
-    auto scope = context->Acquire();
+    Core::Frontend::ScopeAcquireContext acquire_context{*this};
     if (!gladLoadGL()) {
         QMessageBox::critical(this, tr("Error while initializing OpenGL 4.3!"),
                               tr("Your GPU may not support OpenGL 4.3, or you do not have the "

src/yuzu/bootmanager.h

@@ -18,10 +18,12 @@
 #include "core/frontend/emu_window.h"
 
 class GRenderWindow;
-class GMainWindow;
 class QKeyEvent;
+class QScreen;
 class QTouchEvent;
 class QStringList;
+class QSurface;
+class QOpenGLContext;
 #ifdef HAS_VULKAN
 class QVulkanInstance;
 #endif
@@ -34,7 +36,7 @@ class EmuThread final : public QThread {
     Q_OBJECT
 
 public:
-    explicit EmuThread();
+    explicit EmuThread(GRenderWindow& window);
     ~EmuThread() override;
 
     /**
@@ -88,6 +90,12 @@ private:
     std::mutex running_mutex;
     std::condition_variable running_cv;
 
+    /// Only used in asynchronous GPU mode
+    std::unique_ptr<Core::Frontend::GraphicsContext> shared_context;
+
+    /// This is shared_context in asynchronous GPU mode, core_context in synchronous GPU mode
+    Core::Frontend::GraphicsContext& context;
+
 signals:
     /**
      * Emitted when the CPU has halted execution
@@ -116,10 +124,12 @@ class GRenderWindow : public QWidget, public Core::Frontend::EmuWindow {
     Q_OBJECT
 
 public:
-    GRenderWindow(GMainWindow* parent, EmuThread* emu_thread);
+    GRenderWindow(QWidget* parent, EmuThread* emu_thread);
     ~GRenderWindow() override;
 
     // EmuWindow implementation.
+    void MakeCurrent() override;
+    void DoneCurrent() override;
     void PollEvents() override;
     bool IsShown() const override;
     void RetrieveVulkanHandlers(void* get_instance_proc_addr, void* instance,
@@ -155,8 +165,6 @@ public:
 
     void CaptureScreenshot(u32 res_scale, const QString& screenshot_path);
 
-    std::pair<u32, u32> ScaleTouch(const QPointF& pos) const;
-
 public slots:
     void OnEmulationStarting(EmuThread* emu_thread);
     void OnEmulationStopping();
@@ -168,6 +176,7 @@ signals:
     void FirstFrameDisplayed();
 
 private:
+    std::pair<u32, u32> ScaleTouch(QPointF pos) const;
     void TouchBeginEvent(const QTouchEvent* event);
     void TouchUpdateEvent(const QTouchEvent* event);
     void TouchEndEvent();
@@ -181,10 +190,7 @@ private:
 
     EmuThread* emu_thread;
 
-    // Main context that will be shared with all other contexts that are requested.
-    // If this is used in a shared context setting, then this should not be used directly, but
-    // should instead be shared from
-    std::shared_ptr<Core::Frontend::GraphicsContext> main_context;
+    std::unique_ptr<GraphicsContext> core_context;
 
 #ifdef HAS_VULKAN
     std::unique_ptr<QVulkanInstance> vk_instance;
@@ -195,6 +201,12 @@ private:
 
     QByteArray geometry;
 
+    /// Native window handle that backs this presentation widget
+    QWindow* child_window = nullptr;
+
+    /// In order to embed the window into GRenderWindow, you need to use createWindowContainer to
+    /// put the child_window into a widget then add it to the layout. This child_widget can be
+    /// parented to GRenderWindow and use Qt's lifetime system
     QWidget* child_widget = nullptr;
 
     bool first_frame = false;

src/yuzu/main.cpp

@@ -984,7 +984,7 @@ void GMainWindow::BootGame(const QString& filename) {
         return;
 
     // Create and start the emulation thread
-    emu_thread = std::make_unique<EmuThread>();
+    emu_thread = std::make_unique<EmuThread>(*render_window);
     emit EmulationStarting(emu_thread.get());
     emu_thread->start();
 
@@ -2375,6 +2375,7 @@ int main(int argc, char* argv[]) {
 
     // Enables the core to make the qt created contexts current on std::threads
     QCoreApplication::setAttribute(Qt::AA_DontCheckOpenGLContextThreadAffinity);
+    QCoreApplication::setAttribute(Qt::AA_ShareOpenGLContexts);
     QApplication app(argc, argv);
 
     // Qt changes the locale and causes issues in float conversion using std::to_string() when

src/yuzu_cmd/emu_window/emu_window_sdl2_gl.cpp

@@ -37,24 +37,16 @@ public:
     }
 
     void MakeCurrent() override {
-        if (is_current) {
-            return;
-        }
-        is_current = SDL_GL_MakeCurrent(window, context) == 0;
+        SDL_GL_MakeCurrent(window, context);
     }
 
     void DoneCurrent() override {
-        if (!is_current) {
-            return;
-        }
         SDL_GL_MakeCurrent(window, nullptr);
-        is_current = false;
     }
 
 private:
     SDL_Window* window;
     SDL_GLContext context;
-    bool is_current = false;
 };
 
 bool EmuWindow_SDL2_GL::SupportsRequiredGLExtensions() {
@@ -156,6 +148,14 @@ EmuWindow_SDL2_GL::~EmuWindow_SDL2_GL() {
     SDL_GL_DeleteContext(window_context);
 }
 
+void EmuWindow_SDL2_GL::MakeCurrent() {
+    core_context->MakeCurrent();
+}
+
+void EmuWindow_SDL2_GL::DoneCurrent() {
+    core_context->DoneCurrent();
+}
+
 void EmuWindow_SDL2_GL::RetrieveVulkanHandlers(void* get_instance_proc_addr, void* instance,
                                                void* surface) const {
     // Should not have been called from OpenGL

src/yuzu_cmd/emu_window/emu_window_sdl2_gl.h

@@ -13,6 +13,8 @@ public:
     explicit EmuWindow_SDL2_GL(Core::System& system, bool fullscreen);
     ~EmuWindow_SDL2_GL();
 
+    void MakeCurrent() override;
+    void DoneCurrent() override;
     void Present() override;
 
     /// Ignored in OpenGL

src/yuzu_cmd/emu_window/emu_window_sdl2_vk.cpp

@@ -111,6 +111,14 @@ EmuWindow_SDL2_VK::~EmuWindow_SDL2_VK() {
     vkDestroyInstance(vk_instance, nullptr);
 }
 
+void EmuWindow_SDL2_VK::MakeCurrent() {
+    // Unused on Vulkan
+}
+
+void EmuWindow_SDL2_VK::DoneCurrent() {
+    // Unused on Vulkan
+}
+
 void EmuWindow_SDL2_VK::RetrieveVulkanHandlers(void* get_instance_proc_addr, void* instance,
                                                void* surface) const {
     const auto instance_proc_addr = vkGetInstanceProcAddr;

src/yuzu_cmd/emu_window/emu_window_sdl2_vk.h

@@ -13,6 +13,8 @@ public:
     explicit EmuWindow_SDL2_VK(Core::System& system, bool fullscreen);
     ~EmuWindow_SDL2_VK();
 
+    void MakeCurrent() override;
+    void DoneCurrent() override;
     void Present() override;
     void RetrieveVulkanHandlers(void* get_instance_proc_addr, void* instance,
                                 void* surface) const override;

src/yuzu_cmd/yuzu.cpp

@@ -230,11 +230,18 @@ int main(int argc, char** argv) {
 
     system.TelemetrySession().AddField(Telemetry::FieldType::App, "Frontend", "SDL");
 
-    // Core is loaded, start the GPU (makes the GPU contexts current to this thread)
-    system.GPU().Start();
-
     system.Renderer().Rasterizer().LoadDiskResources();
 
+    // Acquire render context for duration of the thread if this is the rendering thread
+    if (!Settings::values.use_asynchronous_gpu_emulation) {
+        emu_window->MakeCurrent();
+    }
+    SCOPE_EXIT({
+        if (!Settings::values.use_asynchronous_gpu_emulation) {
+            emu_window->DoneCurrent();
+        }
+    });
+
     std::thread render_thread([&emu_window] { emu_window->Present(); });
     while (emu_window->IsOpen()) {
         system.RunLoop();

src/yuzu_tester/emu_window/emu_window_sdl2_hide.cpp

@@ -102,6 +102,8 @@ EmuWindow_SDL2_Hide::EmuWindow_SDL2_Hide() {
     LOG_INFO(Frontend, "yuzu-tester Version: {} | {}-{}", Common::g_build_fullname,
              Common::g_scm_branch, Common::g_scm_desc);
     Settings::LogSettings();
+
+    DoneCurrent();
 }
 
 EmuWindow_SDL2_Hide::~EmuWindow_SDL2_Hide() {
@@ -112,6 +114,14 @@ EmuWindow_SDL2_Hide::~EmuWindow_SDL2_Hide() {
 
 void EmuWindow_SDL2_Hide::PollEvents() {}
 
+void EmuWindow_SDL2_Hide::MakeCurrent() {
+    SDL_GL_MakeCurrent(render_window, gl_context);
+}
+
+void EmuWindow_SDL2_Hide::DoneCurrent() {
+    SDL_GL_MakeCurrent(render_window, nullptr);
+}
+
 bool EmuWindow_SDL2_Hide::IsShown() const {
     return false;
 }
@@ -119,35 +129,3 @@ bool EmuWindow_SDL2_Hide::IsShown() const {
 void EmuWindow_SDL2_Hide::RetrieveVulkanHandlers(void*, void*, void*) const {
     UNREACHABLE();
 }
-
-class SDLGLContext : public Core::Frontend::GraphicsContext {
-public:
-    explicit SDLGLContext() {
-        // create a hidden window to make the shared context against
-        window = SDL_CreateWindow(NULL, SDL_WINDOWPOS_UNDEFINED, SDL_WINDOWPOS_UNDEFINED, 0, 0,
-                                  SDL_WINDOW_HIDDEN | SDL_WINDOW_OPENGL);
-        context = SDL_GL_CreateContext(window);
-    }
-
-    ~SDLGLContext() {
-        DoneCurrent();
-        SDL_GL_DeleteContext(context);
-        SDL_DestroyWindow(window);
-    }
-
-    void MakeCurrent() override {
-        SDL_GL_MakeCurrent(window, context);
-    }
-
-    void DoneCurrent() override {
-        SDL_GL_MakeCurrent(window, nullptr);
-    }
-
-private:
-    SDL_Window* window;
-    SDL_GLContext context;
-};
-
-std::unique_ptr<Core::Frontend::GraphicsContext> EmuWindow_SDL2_Hide::CreateSharedContext() const {
-    return std::make_unique<SDLGLContext>();
-}

src/yuzu_tester/emu_window/emu_window_sdl2_hide.h

@@ -16,6 +16,12 @@ public:
     /// Polls window events
     void PollEvents() override;
 
+    /// Makes the graphics context current for the caller thread
+    void MakeCurrent() override;
+
+    /// Releases the GL context from the caller thread
+    void DoneCurrent() override;
+
     /// Whether the screen is being shown or not.
     bool IsShown() const override;
 
@@ -23,7 +29,8 @@ public:
     void RetrieveVulkanHandlers(void* get_instance_proc_addr, void* instance,
                                 void* surface) const override;
 
-    std::unique_ptr<Core::Frontend::GraphicsContext> CreateSharedContext() const override;
+    /// Whether the window is still open, and a close request hasn't yet been sent
+    bool IsOpen() const;
 
 private:
     /// Whether the GPU and driver supports the OpenGL extension required

src/yuzu_tester/yuzu.cpp

@@ -164,6 +164,11 @@ int main(int argc, char** argv) {
 
     std::unique_ptr<EmuWindow_SDL2_Hide> emu_window{std::make_unique<EmuWindow_SDL2_Hide>()};
 
+    if (!Settings::values.use_multi_core) {
+        // Single core mode must acquire OpenGL context for entire emulation session
+        emu_window->MakeCurrent();
+    }
+
     bool finished = false;
     int return_value = 0;
     const auto callback = [&finished,
@@ -252,7 +257,6 @@ int main(int argc, char** argv) {
 
     system.TelemetrySession().AddField(Telemetry::FieldType::App, "Frontend", "SDLHideTester");
 
-    system.GPU().Start();
     system.Renderer().Rasterizer().LoadDiskResources();
 
     while (!finished) {