video_core: Refactor GPU interface (#7272)

* video_core: Refactor GPU interface

* citra_qt: Better debug widget lifetime

CMakeModules/GenerateSCMRev.cmake

@@ -26,16 +26,14 @@ set(HASH_FILES
     "${VIDEO_CORE}/shader/generator/spv_fs_shader_gen.h"
     "${VIDEO_CORE}/shader/shader.cpp"
     "${VIDEO_CORE}/shader/shader.h"
-    "${VIDEO_CORE}/pica.cpp"
+    "${VIDEO_CORE}/pica/regs_framebuffer.h"
-    "${VIDEO_CORE}/pica.h"
+    "${VIDEO_CORE}/pica/regs_lighting.h"
-    "${VIDEO_CORE}/regs_framebuffer.h"
+    "${VIDEO_CORE}/pica/regs_pipeline.h"
-    "${VIDEO_CORE}/regs_lighting.h"
+    "${VIDEO_CORE}/pica/regs_rasterizer.h"
-    "${VIDEO_CORE}/regs_pipeline.h"
+    "${VIDEO_CORE}/pica/regs_shader.h"
-    "${VIDEO_CORE}/regs_rasterizer.h"
+    "${VIDEO_CORE}/pica/regs_texturing.h"
-    "${VIDEO_CORE}/regs_shader.h"
+    "${VIDEO_CORE}/pica/regs_internal.cpp"
-    "${VIDEO_CORE}/regs_texturing.h"
+    "${VIDEO_CORE}/pica/regs_internal.h"
-    "${VIDEO_CORE}/regs.cpp"
-    "${VIDEO_CORE}/regs.h"
 )
 set(COMBINED "")
 foreach (F IN LISTS HASH_FILES)

src/android/app/src/main/jni/emu_window/emu_window.cpp

@@ -15,7 +15,6 @@
 #include "jni/input_manager.h"
 #include "network/network.h"
 #include "video_core/renderer_base.h"
-#include "video_core/video_core.h"
 
 static bool IsPortraitMode() {
     return JNI_FALSE != IDCache::GetEnvForThread()->CallStaticBooleanMethod(

src/android/app/src/main/jni/emu_window/emu_window.h

@@ -7,6 +7,10 @@
 #include <vector>
 #include "core/frontend/emu_window.h"
 
+namespace Core {
+class System;
+}
+
 class EmuWindow_Android : public Frontend::EmuWindow {
 public:
     EmuWindow_Android(ANativeWindow* surface);

src/android/app/src/main/jni/emu_window/emu_window_gl.cpp

@@ -12,10 +12,11 @@
 
 #include "common/logging/log.h"
 #include "common/settings.h"
+#include "core/core.h"
 #include "input_common/main.h"
 #include "jni/emu_window/emu_window_gl.h"
+#include "video_core/gpu.h"
 #include "video_core/renderer_base.h"
-#include "video_core/video_core.h"
 
 static constexpr std::array<EGLint, 15> egl_attribs{EGL_SURFACE_TYPE,
                                                     EGL_WINDOW_BIT,
@@ -71,8 +72,8 @@ private:
     EGLContext egl_context{};
 };
 
-EmuWindow_Android_OpenGL::EmuWindow_Android_OpenGL(ANativeWindow* surface)
+EmuWindow_Android_OpenGL::EmuWindow_Android_OpenGL(Core::System& system_, ANativeWindow* surface)
-    : EmuWindow_Android{surface} {
+    : EmuWindow_Android{surface}, system{system_} {
     if (egl_display = eglGetDisplay(EGL_DEFAULT_DISPLAY); egl_display == EGL_NO_DISPLAY) {
         LOG_CRITICAL(Frontend, "eglGetDisplay() failed");
         return;
@@ -199,6 +200,9 @@ void EmuWindow_Android_OpenGL::StopPresenting() {
 }
 
 void EmuWindow_Android_OpenGL::TryPresenting() {
+    if (!system.IsPoweredOn()) {
+        return;
+    }
     if (presenting_state == PresentingState::Initial) [[unlikely]] {
         eglMakeCurrent(egl_display, egl_surface, egl_surface, egl_context);
         glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
@@ -208,8 +212,6 @@ void EmuWindow_Android_OpenGL::TryPresenting() {
         return;
     }
     eglSwapInterval(egl_display, Settings::values.use_vsync_new ? 1 : 0);
-    if (VideoCore::g_renderer) {
+    system.GPU().Renderer().TryPresent(0);
-        VideoCore::g_renderer->TryPresent(0);
+    eglSwapBuffers(egl_display, egl_surface);
-        eglSwapBuffers(egl_display, egl_surface);
-    }
 }

src/android/app/src/main/jni/emu_window/emu_window_gl.h

@@ -11,11 +11,15 @@
 
 #include "jni/emu_window/emu_window.h"
 
+namespace Core {
+class System;
+}
+
 struct ANativeWindow;
 
 class EmuWindow_Android_OpenGL : public EmuWindow_Android {
 public:
-    EmuWindow_Android_OpenGL(ANativeWindow* surface);
+    EmuWindow_Android_OpenGL(Core::System& system, ANativeWindow* surface);
     ~EmuWindow_Android_OpenGL() override = default;
 
     void TryPresenting() override;
@@ -30,6 +34,7 @@ private:
     void DestroyContext() override;
 
 private:
+    Core::System& system;
     EGLConfig egl_config;
     EGLSurface egl_surface{};
     EGLContext egl_context{};

src/android/app/src/main/jni/emu_window/emu_window_vk.cpp

@@ -7,7 +7,6 @@
 #include "common/logging/log.h"
 #include "common/settings.h"
 #include "jni/emu_window/emu_window_vk.h"
-#include "video_core/video_core.h"
 
 class GraphicsContext_Android final : public Frontend::GraphicsContext {
 public:

src/android/app/src/main/jni/native.cpp

@@ -51,8 +51,9 @@
 #include "jni/id_cache.h"
 #include "jni/input_manager.h"
 #include "jni/ndk_motion.h"
+#include "video_core/debug_utils/debug_utils.h"
+#include "video_core/gpu.h"
 #include "video_core/renderer_base.h"
-#include "video_core/video_core.h"
 
 #if CITRA_ARCH(arm64)
 #include <adrenotools/driver.h>
@@ -126,7 +127,7 @@ static bool CheckMicPermission() {
 
 static Core::System::ResultStatus RunCitra(const std::string& filepath) {
     // Citra core only supports a single running instance
-    std::lock_guard<std::mutex> lock(running_mutex);
+    std::scoped_lock lock(running_mutex);
 
     LOG_INFO(Frontend, "Citra starting...");
 
@@ -137,10 +138,12 @@ static Core::System::ResultStatus RunCitra(const std::string& filepath) {
         return Core::System::ResultStatus::ErrorLoader;
     }
 
+    Core::System& system{Core::System::GetInstance()};
+
     const auto graphics_api = Settings::values.graphics_api.GetValue();
     switch (graphics_api) {
     case Settings::GraphicsAPI::OpenGL:
-        window = std::make_unique<EmuWindow_Android_OpenGL>(s_surf);
+        window = std::make_unique<EmuWindow_Android_OpenGL>(system, s_surf);
         break;
     case Settings::GraphicsAPI::Vulkan:
         window = std::make_unique<EmuWindow_Android_Vulkan>(s_surf, vulkan_library);
@@ -150,8 +153,6 @@ static Core::System::ResultStatus RunCitra(const std::string& filepath) {
         window = std::make_unique<EmuWindow_Android_Vulkan>(s_surf, vulkan_library);
     }
 
-    Core::System& system{Core::System::GetInstance()};
-
     // Forces a config reload on game boot, if the user changed settings in the UI
     Config{};
     // Replace with game-specific settings
@@ -179,6 +180,7 @@ static Core::System::ResultStatus RunCitra(const std::string& filepath) {
     // Register microphone permission check
     system.RegisterMicPermissionCheck(&CheckMicPermission);
 
+    Pica::g_debug_context = Pica::DebugContext::Construct();
     InputManager::Init();
 
     window->MakeCurrent();
@@ -196,7 +198,7 @@ static Core::System::ResultStatus RunCitra(const std::string& filepath) {
     LoadDiskCacheProgress(VideoCore::LoadCallbackStage::Prepare, 0, 0);
 
     std::unique_ptr<Frontend::GraphicsContext> cpu_context;
-    system.Renderer().Rasterizer()->LoadDiskResources(stop_run, &LoadDiskCacheProgress);
+    system.GPU().Renderer().Rasterizer()->LoadDiskResources(stop_run, &LoadDiskCacheProgress);
 
     LoadDiskCacheProgress(VideoCore::LoadCallbackStage::Complete, 0, 0);
 
@@ -275,8 +277,10 @@ void Java_org_citra_citra_1emu_NativeLibrary_surfaceChanged(JNIEnv* env,
     if (window) {
         window->OnSurfaceChanged(s_surf);
     }
-    if (VideoCore::g_renderer) {
+
-        VideoCore::g_renderer->NotifySurfaceChanged();
+    auto& system = Core::System::GetInstance();
+    if (system.IsPoweredOn()) {
+        system.GPU().Renderer().NotifySurfaceChanged();
     }
 
     LOG_INFO(Frontend, "Surface changed");
@@ -311,8 +315,9 @@ void Java_org_citra_citra_1emu_NativeLibrary_notifyOrientationChange([[maybe_unu
                                                                      jint layout_option,
                                                                      jint rotation) {
     Settings::values.layout_option = static_cast<Settings::LayoutOption>(layout_option);
-    if (VideoCore::g_renderer) {
+    auto& system = Core::System::GetInstance();
-        VideoCore::g_renderer->UpdateCurrentFramebufferLayout(!(rotation % 2));
+    if (system.IsPoweredOn()) {
+        system.GPU().Renderer().UpdateCurrentFramebufferLayout(!(rotation % 2));
     }
     InputManager::screen_rotation = rotation;
     Camera::NDK::g_rotation = rotation;
@@ -322,8 +327,9 @@ void Java_org_citra_citra_1emu_NativeLibrary_swapScreens([[maybe_unused]] JNIEnv
                                                          [[maybe_unused]] jobject obj,
                                                          jboolean swap_screens, jint rotation) {
     Settings::values.swap_screen = swap_screens;
-    if (VideoCore::g_renderer) {
+    auto& system = Core::System::GetInstance();
-        VideoCore::g_renderer->UpdateCurrentFramebufferLayout(!(rotation % 2));
+    if (system.IsPoweredOn()) {
+        system.GPU().Renderer().UpdateCurrentFramebufferLayout(!(rotation % 2));
     }
     InputManager::screen_rotation = rotation;
     Camera::NDK::g_rotation = rotation;

src/citra/citra.cpp

@@ -36,6 +36,7 @@
 #include "core/telemetry_session.h"
 #include "input_common/main.h"
 #include "network/network.h"
+#include "video_core/gpu.h"
 #include "video_core/renderer_base.h"
 
 #ifdef __unix__
@@ -438,9 +439,10 @@ int main(int argc, char** argv) {
         movie.StartRecording(movie_record, movie_record_author);
     }
     if (!dump_video.empty() && DynamicLibrary::FFmpeg::LoadFFmpeg()) {
+        auto& renderer = system.GPU().Renderer();
         const auto layout{
-            Layout::FrameLayoutFromResolutionScale(system.Renderer().GetResolutionScaleFactor())};
+            Layout::FrameLayoutFromResolutionScale(renderer.GetResolutionScaleFactor())};
-        auto dumper = std::make_shared<VideoDumper::FFmpegBackend>();
+        auto dumper = std::make_shared<VideoDumper::FFmpegBackend>(renderer);
         if (dumper->StartDumping(dump_video, layout)) {
             system.RegisterVideoDumper(dumper);
         }
@@ -458,7 +460,7 @@ int main(int argc, char** argv) {
     });
 
     std::atomic_bool stop_run;
-    system.Renderer().Rasterizer()->LoadDiskResources(
+    system.GPU().Renderer().Rasterizer()->LoadDiskResources(
         stop_run, [](VideoCore::LoadCallbackStage stage, std::size_t value, std::size_t total) {
             LOG_DEBUG(Frontend, "Loading stage {} progress {} {}", static_cast<u32>(stage), value,
                       total);

src/citra/emu_window/emu_window_sdl2_gl.cpp

@@ -11,8 +11,9 @@
 #include "citra/emu_window/emu_window_sdl2_gl.h"
 #include "common/scm_rev.h"
 #include "common/settings.h"
+#include "core/core.h"
+#include "video_core/gpu.h"
 #include "video_core/renderer_base.h"
-#include "video_core/video_core.h"
 
 class SDLGLContext : public Frontend::GraphicsContext {
 public:
@@ -159,7 +160,7 @@ void EmuWindow_SDL2_GL::Present() {
     SDL_GL_MakeCurrent(render_window, window_context);
     SDL_GL_SetSwapInterval(1);
     while (IsOpen()) {
-        VideoCore::g_renderer->TryPresent(100, is_secondary);
+        system.GPU().Renderer().TryPresent(100, is_secondary);
         SDL_GL_SwapWindow(render_window);
     }
     SDL_GL_MakeCurrent(render_window, nullptr);

src/citra/emu_window/emu_window_sdl2_sw.cpp

@@ -13,6 +13,7 @@
 #include "common/settings.h"
 #include "core/core.h"
 #include "core/frontend/emu_window.h"
+#include "video_core/gpu.h"
 #include "video_core/renderer_software/renderer_software.h"
 
 class DummyContext : public Frontend::GraphicsContext {};
@@ -94,7 +95,7 @@ void EmuWindow_SDL2_SW::Present() {
 }
 
 SDL_Surface* EmuWindow_SDL2_SW::LoadFramebuffer(VideoCore::ScreenId screen_id) {
-    const auto& renderer = static_cast<SwRenderer::RendererSoftware&>(system.Renderer());
+    const auto& renderer = static_cast<SwRenderer::RendererSoftware&>(system.GPU().Renderer());
     const auto& info = renderer.Screen(screen_id);
     const int width = static_cast<int>(info.width);
     const int height = static_cast<int>(info.height);

src/citra_qt/bootmanager.cpp

@@ -22,9 +22,9 @@
 #include "input_common/main.h"
 #include "input_common/motion_emu.h"
 #include "video_core/custom_textures/custom_tex_manager.h"
+#include "video_core/gpu.h"
 #include "video_core/renderer_base.h"
 #include "video_core/renderer_software/renderer_software.h"
-#include "video_core/video_core.h"
 
 #ifdef HAS_OPENGL
 #include <glad/glad.h>
@@ -73,7 +73,7 @@ void EmuThread::run() {
 
     emit LoadProgress(VideoCore::LoadCallbackStage::Prepare, 0, 0);
 
-    system.Renderer().Rasterizer()->LoadDiskResources(
+    system.GPU().Renderer().Rasterizer()->LoadDiskResources(
         stop_run, [this](VideoCore::LoadCallbackStage stage, std::size_t value, std::size_t total) {
             emit LoadProgress(stage, value, total);
         });
@@ -284,9 +284,7 @@ public:
         }
         context->MakeCurrent();
         glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
-        if (VideoCore::g_renderer) {
+        system.GPU().Renderer().TryPresent(100, is_secondary);
-            VideoCore::g_renderer->TryPresent(100, is_secondary);
-        }
         context->SwapBuffers();
         glFinish();
     }
@@ -367,7 +365,7 @@ struct SoftwareRenderWidget : public RenderWidget {
     }
 
     QImage LoadFramebuffer(VideoCore::ScreenId screen_id) {
-        const auto& renderer = static_cast<SwRenderer::RendererSoftware&>(system.Renderer());
+        const auto& renderer = static_cast<SwRenderer::RendererSoftware&>(system.GPU().Renderer());
         const auto& info = renderer.Screen(screen_id);
         const int width = static_cast<int>(info.width);
         const int height = static_cast<int>(info.height);
@@ -678,13 +676,14 @@ void GRenderWindow::ReleaseRenderTarget() {
 }
 
 void GRenderWindow::CaptureScreenshot(u32 res_scale, const QString& screenshot_path) {
+    auto& renderer = system.GPU().Renderer();
     if (res_scale == 0) {
-        res_scale = system.Renderer().GetResolutionScaleFactor();
+        res_scale = renderer.GetResolutionScaleFactor();
     }
 
     const auto layout{Layout::FrameLayoutFromResolutionScale(res_scale, is_secondary)};
     screenshot_image = QImage(QSize(layout.width, layout.height), QImage::Format_RGB32);
-    system.Renderer().RequestScreenshot(
+    renderer.RequestScreenshot(
         screenshot_image.bits(),
         [this, screenshot_path](bool invert_y) {
             const std::string std_screenshot_path = screenshot_path.toStdString();

src/citra_qt/configuration/configure_graphics.cpp

@@ -20,7 +20,6 @@ ConfigureGraphics::ConfigureGraphics(std::span<const QString> physical_devices,
         ui->physical_device_combo->addItem(name);
     }
 
-    ui->toggle_vsync_new->setEnabled(!is_powered_on);
     ui->graphics_api_combo->setEnabled(!is_powered_on);
     ui->physical_device_combo->setEnabled(!is_powered_on);
     ui->toggle_async_shaders->setEnabled(!is_powered_on);

src/citra_qt/debugger/graphics/graphics.cpp

@@ -5,8 +5,8 @@
 #include <QListView>
 #include "citra_qt/debugger/graphics/graphics.h"
 #include "citra_qt/util/util.h"
-
+#include "core/core.h"
-extern GraphicsDebugger g_debugger;
+#include "video_core/gpu.h"
 
 GPUCommandStreamItemModel::GPUCommandStreamItemModel(QObject* parent)
     : QAbstractListModel(parent), command_count(0) {
@@ -19,19 +19,19 @@ int GPUCommandStreamItemModel::rowCount([[maybe_unused]] const QModelIndex& pare
 }
 
 QVariant GPUCommandStreamItemModel::data(const QModelIndex& index, int role) const {
-    if (!index.isValid())
+    if (!index.isValid() || !GetDebugger())
         return QVariant();
 
     int command_index = index.row();
     const Service::GSP::Command& command = GetDebugger()->ReadGXCommandHistory(command_index);
     if (role == Qt::DisplayRole) {
         std::map<Service::GSP::CommandId, const char*> command_names = {
-            {Service::GSP::CommandId::REQUEST_DMA, "REQUEST_DMA"},
+            {Service::GSP::CommandId::RequestDma, "REQUEST_DMA"},
-            {Service::GSP::CommandId::SUBMIT_GPU_CMDLIST, "SUBMIT_GPU_CMDLIST"},
+            {Service::GSP::CommandId::SubmitCmdList, "SUBMIT_GPU_CMDLIST"},
-            {Service::GSP::CommandId::SET_MEMORY_FILL, "SET_MEMORY_FILL"},
+            {Service::GSP::CommandId::MemoryFill, "SET_MEMORY_FILL"},
-            {Service::GSP::CommandId::SET_DISPLAY_TRANSFER, "SET_DISPLAY_TRANSFER"},
+            {Service::GSP::CommandId::DisplayTransfer, "SET_DISPLAY_TRANSFER"},
-            {Service::GSP::CommandId::SET_TEXTURE_COPY, "SET_TEXTURE_COPY"},
+            {Service::GSP::CommandId::TextureCopy, "SET_TEXTURE_COPY"},
-            {Service::GSP::CommandId::CACHE_FLUSH, "CACHE_FLUSH"},
+            {Service::GSP::CommandId::CacheFlush, "CACHE_FLUSH"},
         };
         const u32* command_data = reinterpret_cast<const u32*>(&command);
         QString str = QStringLiteral("%1 %2 %3 %4 %5 %6 %7 %8 %9")
@@ -63,8 +63,8 @@ void GPUCommandStreamItemModel::OnGXCommandFinishedInternal(int total_command_co
     emit dataChanged(index(prev_command_count, 0), index(total_command_count - 1, 0));
 }
 
-GPUCommandStreamWidget::GPUCommandStreamWidget(QWidget* parent)
+GPUCommandStreamWidget::GPUCommandStreamWidget(Core::System& system_, QWidget* parent)
-    : QDockWidget(tr("Graphics Debugger"), parent), model(this) {
+    : QDockWidget(tr("Graphics Debugger"), parent), system{system_}, model(this) {
     setObjectName(QStringLiteral("GraphicsDebugger"));
 
     auto* command_list = new QListView;
@@ -74,12 +74,26 @@ GPUCommandStreamWidget::GPUCommandStreamWidget(QWidget* parent)
     setWidget(command_list);
 }
 
+void GPUCommandStreamWidget::Register() {
+    auto& debugger = system.GPU().Debugger();
+    debugger.RegisterObserver(&model);
+}
+
+void GPUCommandStreamWidget::Unregister() {
+    auto& debugger = system.GPU().Debugger();
+    debugger.UnregisterObserver(&model);
+}
+
 void GPUCommandStreamWidget::showEvent(QShowEvent* event) {
-    g_debugger.RegisterObserver(&model);
+    if (system.IsPoweredOn()) {
+        Register();
+    }
     QDockWidget::showEvent(event);
 }
 
 void GPUCommandStreamWidget::hideEvent(QHideEvent* event) {
-    g_debugger.UnregisterObserver(&model);
+    if (system.IsPoweredOn()) {
+        Unregister();
+    }
     QDockWidget::hideEvent(event);
 }

src/citra_qt/debugger/graphics/graphics.h

@@ -8,8 +8,12 @@
 #include <QDockWidget>
 #include "video_core/gpu_debugger.h"
 
+namespace Core {
+class System;
+}
+
 class GPUCommandStreamItemModel : public QAbstractListModel,
-                                  public GraphicsDebugger::DebuggerObserver {
+                                  public VideoCore::GraphicsDebugger::DebuggerObserver {
     Q_OBJECT
 
 public:
@@ -35,12 +39,16 @@ class GPUCommandStreamWidget : public QDockWidget {
     Q_OBJECT
 
 public:
-    GPUCommandStreamWidget(QWidget* parent = nullptr);
+    GPUCommandStreamWidget(Core::System& system, QWidget* parent = nullptr);
+
+    void Register();
+    void Unregister();
 
 protected:
     void showEvent(QShowEvent* event) override;
     void hideEvent(QHideEvent* event) override;
 
 private:
+    Core::System& system;
     GPUCommandStreamItemModel model;
 };

src/citra_qt/debugger/graphics/graphics_breakpoint_observer.cpp

@@ -18,7 +18,8 @@ BreakPointObserverDock::BreakPointObserverDock(std::shared_ptr<Pica::DebugContex
             &BreakPointObserverDock::OnBreakPointHit, Qt::BlockingQueuedConnection);
 }
 
-void BreakPointObserverDock::OnPicaBreakPointHit(Pica::DebugContext::Event event, void* data) {
+void BreakPointObserverDock::OnPicaBreakPointHit(Pica::DebugContext::Event event,
+                                                 const void* data) {
     emit BreakPointHit(event, data);
 }
 

src/citra_qt/debugger/graphics/graphics_breakpoint_observer.h

@@ -20,14 +20,14 @@ public:
     BreakPointObserverDock(std::shared_ptr<Pica::DebugContext> debug_context, const QString& title,
                            QWidget* parent = nullptr);
 
-    void OnPicaBreakPointHit(Pica::DebugContext::Event event, void* data) override;
+    void OnPicaBreakPointHit(Pica::DebugContext::Event event, const void* data) override;
     void OnPicaResume() override;
 
 signals:
     void Resumed();
-    void BreakPointHit(Pica::DebugContext::Event event, void* data);
+    void BreakPointHit(Pica::DebugContext::Event event, const void* data);
 
 private:
-    virtual void OnBreakPointHit(Pica::DebugContext::Event event, void* data) = 0;
+    virtual void OnBreakPointHit(Pica::DebugContext::Event event, const void* data) = 0;
     virtual void OnResumed() = 0;
 };

src/citra_qt/debugger/graphics/graphics_breakpoints.cpp

@@ -191,12 +191,12 @@ GraphicsBreakPointsWidget::GraphicsBreakPointsWidget(
     setWidget(main_widget);
 }
 
-void GraphicsBreakPointsWidget::OnPicaBreakPointHit(Event event, void* data) {
+void GraphicsBreakPointsWidget::OnPicaBreakPointHit(Event event, const void* data) {
     // Process in GUI thread
     emit BreakPointHit(event, data);
 }
 
-void GraphicsBreakPointsWidget::OnBreakPointHit(Pica::DebugContext::Event event, void* data) {
+void GraphicsBreakPointsWidget::OnBreakPointHit(Pica::DebugContext::Event event, const void* data) {
     status_text->setText(tr("Emulation halted at breakpoint"));
     resume_button->setEnabled(true);
 }

src/citra_qt/debugger/graphics/graphics_breakpoints.h

@@ -23,16 +23,16 @@ public:
     explicit GraphicsBreakPointsWidget(std::shared_ptr<Pica::DebugContext> debug_context,
                                        QWidget* parent = nullptr);
 
-    void OnPicaBreakPointHit(Pica::DebugContext::Event event, void* data) override;
+    void OnPicaBreakPointHit(Pica::DebugContext::Event event, const void* data) override;
     void OnPicaResume() override;
 
 signals:
     void Resumed();
-    void BreakPointHit(Pica::DebugContext::Event event, void* data);
+    void BreakPointHit(Pica::DebugContext::Event event, const void* data);
     void BreakPointsChanged(const QModelIndex& topLeft, const QModelIndex& bottomRight);
 
 private:
-    void OnBreakPointHit(Pica::DebugContext::Event event, void* data);
+    void OnBreakPointHit(Pica::DebugContext::Event event, const void* data);
     void OnItemDoubleClicked(const QModelIndex&);
     void OnResumeRequested();
     void OnResumed();

src/citra_qt/debugger/graphics/graphics_cmdlists.cpp

@@ -19,8 +19,8 @@
 #include "core/core.h"
 #include "core/memory.h"
 #include "video_core/debug_utils/debug_utils.h"
-#include "video_core/pica_state.h"
+#include "video_core/gpu.h"
-#include "video_core/regs.h"
+#include "video_core/pica/pica_core.h"
 #include "video_core/texture/texture_decode.h"
 
 namespace {
@@ -73,7 +73,7 @@ QVariant GPUCommandListModel::data(const QModelIndex& index, int role) const {
     if (role == Qt::DisplayRole) {
         switch (index.column()) {
         case 0:
-            return QString::fromLatin1(Pica::Regs::GetRegisterName(write.cmd_id));
+            return QString::fromLatin1(Pica::RegsInternal::GetRegisterName(write.cmd_id));
         case 1:
             return QStringLiteral("%1").arg(write.cmd_id, 3, 16, QLatin1Char('0'));
         case 2:
@@ -119,8 +119,7 @@ void GPUCommandListModel::OnPicaTraceFinished(const Pica::DebugUtils::PicaTrace&
 }
 
 #define COMMAND_IN_RANGE(cmd_id, reg_name)                                                         \
-    (cmd_id >= PICA_REG_INDEX(reg_name) &&                                                         \
+    (cmd_id >= PICA_REG_INDEX(reg_name) && cmd_id <= PICA_REG_INDEX(reg_name))
-     cmd_id < PICA_REG_INDEX(reg_name) + sizeof(decltype(Pica::g_state.regs.reg_name)) / 4)
 
 void GPUCommandListWidget::OnCommandDoubleClicked(const QModelIndex& index) {
     const unsigned int command_id =
@@ -147,13 +146,13 @@ void GPUCommandListWidget::OnCommandDoubleClicked(const QModelIndex& index) {
 void GPUCommandListWidget::SetCommandInfo(const QModelIndex& index) {
     QWidget* new_info_widget = nullptr;
 
-    const unsigned int command_id =
+    const u32 command_id =
         list_widget->model()->data(index, GPUCommandListModel::CommandIdRole).toUInt();
     if (COMMAND_IN_RANGE(command_id, texturing.texture0) ||
         COMMAND_IN_RANGE(command_id, texturing.texture1) ||
         COMMAND_IN_RANGE(command_id, texturing.texture2)) {
 
-        unsigned texture_index;
+        u32 texture_index;
         if (COMMAND_IN_RANGE(command_id, texturing.texture0)) {
             texture_index = 0;
         } else if (COMMAND_IN_RANGE(command_id, texturing.texture1)) {
@@ -162,7 +161,8 @@ void GPUCommandListWidget::SetCommandInfo(const QModelIndex& index) {
             texture_index = 2;
         }
 
-        const auto texture = Pica::g_state.regs.texturing.GetTextures()[texture_index];
+        auto& pica = system.GPU().PicaCore();
+        const auto texture = pica.regs.internal.texturing.GetTextures()[texture_index];
         const auto config = texture.config;
         const auto format = texture.format;
 

src/citra_qt/debugger/graphics/graphics_surface.cpp

@@ -15,10 +15,10 @@
 #include "citra_qt/debugger/graphics/graphics_surface.h"
 #include "citra_qt/util/spinbox.h"
 #include "common/color.h"
+#include "core/core.h"
 #include "core/memory.h"
-#include "video_core/pica_state.h"
+#include "video_core/gpu.h"
-#include "video_core/regs_framebuffer.h"
+#include "video_core/pica/pica_core.h"
-#include "video_core/regs_texturing.h"
 #include "video_core/texture/texture_decode.h"
 #include "video_core/utils.h"
 
@@ -49,10 +49,10 @@ void SurfacePicture::mouseMoveEvent(QMouseEvent* event) {
     mousePressEvent(event);
 }
 
-GraphicsSurfaceWidget::GraphicsSurfaceWidget(Memory::MemorySystem& memory_,
+GraphicsSurfaceWidget::GraphicsSurfaceWidget(Core::System& system_,
                                              std::shared_ptr<Pica::DebugContext> debug_context,
                                              QWidget* parent)
-    : BreakPointObserverDock(debug_context, tr("Pica Surface Viewer"), parent), memory{memory_},
+    : BreakPointObserverDock(debug_context, tr("Pica Surface Viewer"), parent), system{system_},
       surface_source(Source::ColorBuffer) {
     setObjectName(QStringLiteral("PicaSurface"));
 
@@ -214,7 +214,7 @@ GraphicsSurfaceWidget::GraphicsSurfaceWidget(Memory::MemorySystem& memory_,
     }
 }
 
-void GraphicsSurfaceWidget::OnBreakPointHit(Pica::DebugContext::Event event, void* data) {
+void GraphicsSurfaceWidget::OnBreakPointHit(Pica::DebugContext::Event event, const void* data) {
     emit Update();
     widget()->setEnabled(true);
 }
@@ -289,7 +289,7 @@ void GraphicsSurfaceWidget::Pick(int x, int y) {
         return;
     }
 
-    const u8* buffer = memory.GetPhysicalPointer(surface_address);
+    const u8* buffer = system.Memory().GetPhysicalPointer(surface_address);
     if (!buffer) {
         surface_info_label->setText(tr("(unable to access pixel data)"));
         surface_info_label->setAlignment(Qt::AlignCenter);
@@ -410,13 +410,13 @@ void GraphicsSurfaceWidget::Pick(int x, int y) {
 void GraphicsSurfaceWidget::OnUpdate() {
     QPixmap pixmap;
 
+    const auto& regs = system.GPU().PicaCore().regs.internal;
     switch (surface_source) {
     case Source::ColorBuffer: {
         // TODO: Store a reference to the registers in the debug context instead of accessing them
         // directly...
 
-        const auto& framebuffer = Pica::g_state.regs.framebuffer.framebuffer;
+        const auto& framebuffer = regs.framebuffer.framebuffer;
-
         surface_address = framebuffer.GetColorBufferPhysicalAddress();
         surface_width = framebuffer.GetWidth();
         surface_height = framebuffer.GetHeight();
@@ -451,8 +451,7 @@ void GraphicsSurfaceWidget::OnUpdate() {
     }
 
     case Source::DepthBuffer: {
-        const auto& framebuffer = Pica::g_state.regs.framebuffer.framebuffer;
+        const auto& framebuffer = regs.framebuffer.framebuffer;
-
         surface_address = framebuffer.GetDepthBufferPhysicalAddress();
         surface_width = framebuffer.GetWidth();
         surface_height = framebuffer.GetHeight();
@@ -479,8 +478,7 @@ void GraphicsSurfaceWidget::OnUpdate() {
     }
 
     case Source::StencilBuffer: {
-        const auto& framebuffer = Pica::g_state.regs.framebuffer.framebuffer;
+        const auto& framebuffer = regs.framebuffer.framebuffer;
-
         surface_address = framebuffer.GetDepthBufferPhysicalAddress();
         surface_width = framebuffer.GetWidth();
         surface_height = framebuffer.GetHeight();
@@ -513,7 +511,7 @@ void GraphicsSurfaceWidget::OnUpdate() {
             break;
         }
 
-        const auto texture = Pica::g_state.regs.texturing.GetTextures()[texture_index];
+        const auto texture = regs.texturing.GetTextures()[texture_index];
         auto info = Pica::Texture::TextureInfo::FromPicaRegister(texture.config, texture.format);
 
         surface_address = info.physical_address;
@@ -545,7 +543,7 @@ void GraphicsSurfaceWidget::OnUpdate() {
     // TODO: Implement a good way to visualize alpha components!
 
     QImage decoded_image(surface_width, surface_height, QImage::Format_ARGB32);
-    const u8* buffer = memory.GetPhysicalPointer(surface_address);
+    const u8* buffer = system.Memory().GetPhysicalPointer(surface_address);
 
     if (!buffer) {
         surface_picture_label->hide();
@@ -681,7 +679,7 @@ void GraphicsSurfaceWidget::SaveSurface() {
                                  tr("Failed to save surface data to file '%1'").arg(filename));
         }
     } else if (selected_filter == bin_filter) {
-        const u8* const buffer = memory.GetPhysicalPointer(surface_address);
+        const u8* const buffer = system.Memory().GetPhysicalPointer(surface_address);
         ASSERT_MSG(buffer, "Memory not accessible");
 
         QFile file{filename};

src/citra_qt/debugger/graphics/graphics_surface.h

@@ -14,8 +14,8 @@ class CSpinBox;
 
 class GraphicsSurfaceWidget;
 
-namespace Memory {
+namespace Core {
-class MemorySystem;
+class System;
 }
 
 class SurfacePicture : public QLabel {
@@ -76,7 +76,7 @@ class GraphicsSurfaceWidget : public BreakPointObserverDock {
     static unsigned int NibblesPerPixel(Format format);
 
 public:
-    explicit GraphicsSurfaceWidget(Memory::MemorySystem& memory,
+    explicit GraphicsSurfaceWidget(Core::System& system,
                                    std::shared_ptr<Pica::DebugContext> debug_context,
                                    QWidget* parent = nullptr);
     void Pick(int x, int y);
@@ -95,12 +95,12 @@ signals:
     void Update();
 
 private:
-    void OnBreakPointHit(Pica::DebugContext::Event event, void* data) override;
+    void OnBreakPointHit(Pica::DebugContext::Event event, const void* data) override;
     void OnResumed() override;
 
     void SaveSurface();
 
-    Memory::MemorySystem& memory;
+    Core::System& system;
     QComboBox* surface_source_list;
     CSpinBox* surface_address_control;
     QSpinBox* surface_width_control;

src/citra_qt/debugger/graphics/graphics_tracing.cpp

@@ -14,14 +14,15 @@
 #include <nihstro/float24.h>
 #include "citra_qt/debugger/graphics/graphics_tracing.h"
 #include "common/common_types.h"
-#include "core/hw/gpu.h"
+#include "core/core.h"
-#include "core/hw/lcd.h"
 #include "core/tracer/recorder.h"
-#include "video_core/pica_state.h"
+#include "video_core/gpu.h"
+#include "video_core/pica/pica_core.h"
 
-GraphicsTracingWidget::GraphicsTracingWidget(std::shared_ptr<Pica::DebugContext> debug_context,
+GraphicsTracingWidget::GraphicsTracingWidget(Core::System& system_,
+                                             std::shared_ptr<Pica::DebugContext> debug_context,
                                              QWidget* parent)
-    : BreakPointObserverDock(debug_context, tr("CiTrace Recorder"), parent) {
+    : BreakPointObserverDock(debug_context, tr("CiTrace Recorder"), parent), system{system_} {
 
     setObjectName(QStringLiteral("CiTracing"));
 
@@ -61,45 +62,46 @@ void GraphicsTracingWidget::StartRecording() {
     if (!context)
         return;
 
-    auto shader_binary = Pica::g_state.vs.program_code;
+    auto& pica = system.GPU().PicaCore();
-    auto swizzle_data = Pica::g_state.vs.swizzle_data;
+    auto shader_binary = pica.vs_setup.program_code;
+    auto swizzle_data = pica.vs_setup.swizzle_data;
 
     // Encode floating point numbers to 24-bit values
     // TODO: Drop this explicit conversion once we store float24 values bit-correctly internally.
     std::array<u32, 4 * 16> default_attributes;
-    for (unsigned i = 0; i < 16; ++i) {
+    for (u32 i = 0; i < 16; ++i) {
-        for (unsigned comp = 0; comp < 3; ++comp) {
+        for (u32 comp = 0; comp < 3; ++comp) {
-            default_attributes[4 * i + comp] = nihstro::to_float24(
+            default_attributes[4 * i + comp] =
-                Pica::g_state.input_default_attributes.attr[i][comp].ToFloat32());
+                nihstro::to_float24(pica.input_default_attributes[i][comp].ToFloat32());
         }
     }
 
     std::array<u32, 4 * 96> vs_float_uniforms;
-    for (unsigned i = 0; i < 96; ++i)
+    for (u32 i = 0; i < 96; ++i) {
-        for (unsigned comp = 0; comp < 3; ++comp)
+        for (u32 comp = 0; comp < 3; ++comp) {
             vs_float_uniforms[4 * i + comp] =
-                nihstro::to_float24(Pica::g_state.vs.uniforms.f[i][comp].ToFloat32());
+                nihstro::to_float24(pica.vs_setup.uniforms.f[i][comp].ToFloat32());
+        }
+    }
 
     CiTrace::Recorder::InitialState state;
-    std::copy_n((u32*)&GPU::g_regs, sizeof(GPU::g_regs) / sizeof(u32),
-                std::back_inserter(state.gpu_registers));
-    std::copy_n((u32*)&LCD::g_regs, sizeof(LCD::g_regs) / sizeof(u32),
-                std::back_inserter(state.lcd_registers));
-    std::copy_n((u32*)&Pica::g_state.regs, sizeof(Pica::g_state.regs) / sizeof(u32),
-                std::back_inserter(state.pica_registers));
-    std::copy(default_attributes.begin(), default_attributes.end(),
-              std::back_inserter(state.default_attributes));
-    std::copy(shader_binary.begin(), shader_binary.end(),
-              std::back_inserter(state.vs_program_binary));
-    std::copy(swizzle_data.begin(), swizzle_data.end(), std::back_inserter(state.vs_swizzle_data));
-    std::copy(vs_float_uniforms.begin(), vs_float_uniforms.end(),
-              std::back_inserter(state.vs_float_uniforms));
-    // boost::copy(TODO: Not implemented, std::back_inserter(state.gs_program_binary));
-    // boost::copy(TODO: Not implemented, std::back_inserter(state.gs_swizzle_data));
-    // boost::copy(TODO: Not implemented, std::back_inserter(state.gs_float_uniforms));
 
-    auto recorder = new CiTrace::Recorder(state);
+    const auto copy = [&](std::vector<u32>& dest, auto& data) {
-    context->recorder = std::shared_ptr<CiTrace::Recorder>(recorder);
+        dest.resize(sizeof(data));
+        std::memcpy(dest.data(), std::addressof(data), sizeof(data));
+    };
+
+    copy(state.pica_registers, pica.regs);
+    copy(state.lcd_registers, pica.regs_lcd);
+    copy(state.default_attributes, default_attributes);
+    copy(state.vs_program_binary, shader_binary);
+    copy(state.vs_swizzle_data, swizzle_data);
+    copy(state.vs_float_uniforms, vs_float_uniforms);
+    // copy(TODO: Not implemented, std::back_inserter(state.gs_program_binary));
+    // copy(TODO: Not implemented, std::back_inserter(state.gs_swizzle_data));
+    // copy(TODO: Not implemented, std::back_inserter(state.gs_float_uniforms));
+
+    context->recorder = std::make_shared<CiTrace::Recorder>(state);
 
     emit SetStartTracingButtonEnabled(false);
     emit SetStopTracingButtonEnabled(true);
@@ -139,7 +141,7 @@ void GraphicsTracingWidget::AbortRecording() {
     emit SetStartTracingButtonEnabled(true);
 }
 
-void GraphicsTracingWidget::OnBreakPointHit(Pica::DebugContext::Event event, void* data) {
+void GraphicsTracingWidget::OnBreakPointHit(Pica::DebugContext::Event event, const void* data) {
     widget()->setEnabled(true);
 }
 

src/citra_qt/debugger/graphics/graphics_tracing.h

@@ -6,13 +6,18 @@
 
 #include "citra_qt/debugger/graphics/graphics_breakpoint_observer.h"
 
+namespace Core {
+class System;
+}
+
 class EmuThread;
 
 class GraphicsTracingWidget : public BreakPointObserverDock {
     Q_OBJECT
 
 public:
-    explicit GraphicsTracingWidget(std::shared_ptr<Pica::DebugContext> debug_context,
+    explicit GraphicsTracingWidget(Core::System& system,
+                                   std::shared_ptr<Pica::DebugContext> debug_context,
                                    QWidget* parent = nullptr);
 
     void OnEmulationStarting(EmuThread* emu_thread);
@@ -23,11 +28,14 @@ private slots:
     void StopRecording();
     void AbortRecording();
 
-    void OnBreakPointHit(Pica::DebugContext::Event event, void* data) override;
+    void OnBreakPointHit(Pica::DebugContext::Event event, const void* data) override;
     void OnResumed() override;
 
 signals:
     void SetStartTracingButtonEnabled(bool enable);
     void SetStopTracingButtonEnabled(bool enable);
     void SetAbortTracingButtonEnabled(bool enable);
+
+private:
+    Core::System& system;
 };

src/citra_qt/debugger/graphics/graphics_vertex_shader.cpp

@@ -16,9 +16,9 @@
 #include <QTreeView>
 #include "citra_qt/debugger/graphics/graphics_vertex_shader.h"
 #include "citra_qt/util/util.h"
-#include "video_core/pica_state.h"
+#include "core/core.h"
-#include "video_core/shader/debug_data.h"
+#include "video_core/gpu.h"
-#include "video_core/shader/shader.h"
+#include "video_core/pica/pica_core.h"
 #include "video_core/shader/shader_interpreter.h"
 
 using nihstro::Instruction;
@@ -352,16 +352,14 @@ void GraphicsVertexShaderWidget::DumpShader() {
         return;
     }
 
-    auto& setup = Pica::g_state.vs;
+    auto& pica = system.GPU().PicaCore();
-    auto& config = Pica::g_state.regs.vs;
+    Pica::DebugUtils::DumpShader(filename.toStdString(), pica.regs.internal.vs, pica.vs_setup,
-
+                                 pica.regs.internal.rasterizer.vs_output_attributes);
-    Pica::DebugUtils::DumpShader(filename.toStdString(), config, setup,
-                                 Pica::g_state.regs.rasterizer.vs_output_attributes);
 }
 
 GraphicsVertexShaderWidget::GraphicsVertexShaderWidget(
-    std::shared_ptr<Pica::DebugContext> debug_context, QWidget* parent)
+    Core::System& system_, std::shared_ptr<Pica::DebugContext> debug_context, QWidget* parent)
-    : BreakPointObserverDock(debug_context, tr("Pica Vertex Shader"), parent) {
+    : BreakPointObserverDock(debug_context, tr("Pica Vertex Shader"), parent), system{system_} {
     setObjectName(QStringLiteral("PicaVertexShader"));
 
     // Clear input vertex data so that it contains valid float values in case a debug shader
@@ -472,7 +470,8 @@ GraphicsVertexShaderWidget::GraphicsVertexShaderWidget(
     widget()->setEnabled(false);
 }
 
-void GraphicsVertexShaderWidget::OnBreakPointHit(Pica::DebugContext::Event event, void* data) {
+void GraphicsVertexShaderWidget::OnBreakPointHit(Pica::DebugContext::Event event,
+                                                 const void* data) {
     if (event == Pica::DebugContext::Event::VertexShaderInvocation) {
         Reload(true, data);
     } else {
@@ -482,7 +481,7 @@ void GraphicsVertexShaderWidget::OnBreakPointHit(Pica::DebugContext::Event event
     widget()->setEnabled(true);
 }
 
-void GraphicsVertexShaderWidget::Reload(bool replace_vertex_data, void* vertex_data) {
+void GraphicsVertexShaderWidget::Reload(bool replace_vertex_data, const void* vertex_data) {
     model->beginResetModel();
 
     if (replace_vertex_data) {
@@ -491,7 +490,7 @@ void GraphicsVertexShaderWidget::Reload(bool replace_vertex_data, void* vertex_d
             for (unsigned attr = 0; attr < 16; ++attr) {
                 for (unsigned comp = 0; comp < 4; ++comp) {
                     input_data[4 * attr + comp]->setText(
-                        QStringLiteral("%1").arg(input_vertex.attr[attr][comp].ToFloat32()));
+                        QStringLiteral("%1").arg(input_vertex[attr][comp].ToFloat32()));
                 }
             }
             breakpoint_warning->hide();
@@ -508,28 +507,27 @@ void GraphicsVertexShaderWidget::Reload(bool replace_vertex_data, void* vertex_d
     // Reload shader code
     info.Clear();
 
-    auto& shader_setup = Pica::g_state.vs;
+    auto& pica = system.GPU().PicaCore();
-    auto& shader_config = Pica::g_state.regs.vs;
+    for (auto instr : pica.vs_setup.program_code)
-    for (auto instr : shader_setup.program_code)
         info.code.push_back({instr});
-    int num_attributes = shader_config.max_input_attribute_index + 1;
+    int num_attributes = pica.regs.internal.vs.max_input_attribute_index + 1;
 
-    for (auto pattern : shader_setup.swizzle_data) {
+    for (auto pattern : pica.vs_setup.swizzle_data) {
         const nihstro::SwizzleInfo swizzle_info = {.pattern = nihstro::SwizzlePattern{pattern}};
         info.swizzle_info.push_back(swizzle_info);
     }
 
-    u32 entry_point = Pica::g_state.regs.vs.main_offset;
+    u32 entry_point = pica.regs.internal.vs.main_offset;
     info.labels.insert({entry_point, "main"});
 
     // Generate debug information
     Pica::Shader::InterpreterEngine shader_engine;
-    shader_engine.SetupBatch(shader_setup, entry_point);
+    shader_engine.SetupBatch(pica.vs_setup, entry_point);
-    debug_data = shader_engine.ProduceDebugInfo(shader_setup, input_vertex, shader_config);
+    debug_data = shader_engine.ProduceDebugInfo(pica.vs_setup, input_vertex, pica.regs.internal.vs);
 
     // Reload widget state
     for (int attr = 0; attr < num_attributes; ++attr) {
-        unsigned source_attr = shader_config.GetRegisterForAttribute(attr);
+        unsigned source_attr = pica.regs.internal.vs.GetRegisterForAttribute(attr);
         input_data_mapping[attr]->setText(QStringLiteral("-> v%1").arg(source_attr));
         input_data_container[attr]->setVisible(true);
     }
@@ -551,7 +549,7 @@ void GraphicsVertexShaderWidget::OnResumed() {
 
 void GraphicsVertexShaderWidget::OnInputAttributeChanged(int index) {
     const f32 value = input_data[index]->text().toFloat();
-    input_vertex.attr[index / 4][index % 4] = Pica::f24::FromFloat32(value);
+    input_vertex[index / 4][index % 4] = Pica::f24::FromFloat32(value);
     // Re-execute shader with updated value
     Reload();
 }

src/citra_qt/debugger/graphics/graphics_vertex_shader.h

@@ -8,8 +8,12 @@
 #include <QTreeView>
 #include <nihstro/parser_shbin.h>
 #include "citra_qt/debugger/graphics/graphics_breakpoint_observer.h"
+#include "video_core/pica/output_vertex.h"
 #include "video_core/shader/debug_data.h"
-#include "video_core/shader/shader.h"
+
+namespace Core {
+class System;
+}
 
 class QLabel;
 class QSpinBox;
@@ -40,11 +44,12 @@ class GraphicsVertexShaderWidget : public BreakPointObserverDock {
     using Event = Pica::DebugContext::Event;
 
 public:
-    GraphicsVertexShaderWidget(std::shared_ptr<Pica::DebugContext> debug_context,
+    GraphicsVertexShaderWidget(Core::System& system,
+                               std::shared_ptr<Pica::DebugContext> debug_context,
                                QWidget* parent = nullptr);
 
 private slots:
-    void OnBreakPointHit(Pica::DebugContext::Event event, void* data) override;
+    void OnBreakPointHit(Pica::DebugContext::Event event, const void* data) override;
     void OnResumed() override;
 
     void OnInputAttributeChanged(int index);
@@ -60,9 +65,10 @@ private slots:
      * specify that no valid vertex data can be retrieved currently. Only used if
      * replace_vertex_data is true.
      */
-    void Reload(bool replace_vertex_data = false, void* vertex_data = nullptr);
+    void Reload(bool replace_vertex_data = false, const void* vertex_data = nullptr);
 
 private:
+    Core::System& system;
     QLabel* instruction_description;
     QTreeView* binary_list;
     GraphicsVertexShaderModel* model;
@@ -83,7 +89,7 @@ private:
 
     nihstro::ShaderInfo info;
     Pica::Shader::DebugData<true> debug_data;
-    Pica::Shader::AttributeBuffer input_vertex;
+    Pica::AttributeBuffer input_vertex;
 
     friend class GraphicsVertexShaderModel;
 };

src/citra_qt/main.cpp

@@ -74,7 +74,6 @@
 #include "common/logging/backend.h"
 #include "common/logging/log.h"
 #include "common/memory_detect.h"
-#include "common/microprofile.h"
 #include "common/scm_rev.h"
 #include "common/scope_exit.h"
 #if CITRA_ARCH(x86_64)
@@ -96,8 +95,8 @@
 #include "input_common/main.h"
 #include "network/network_settings.h"
 #include "ui_main.h"
+#include "video_core/gpu.h"
 #include "video_core/renderer_base.h"
-#include "video_core/video_core.h"
 
 #ifdef __APPLE__
 #include "common/apple_authorization.h"
@@ -458,7 +457,7 @@ void GMainWindow::InitializeDebugWidgets() {
     connect(this, &GMainWindow::EmulationStopping, registersWidget,
             &RegistersWidget::OnEmulationStopping);
 
-    graphicsWidget = new GPUCommandStreamWidget(this);
+    graphicsWidget = new GPUCommandStreamWidget(system, this);
     addDockWidget(Qt::RightDockWidgetArea, graphicsWidget);
     graphicsWidget->hide();
     debug_menu->addAction(graphicsWidget->toggleViewAction());
@@ -473,12 +472,13 @@ void GMainWindow::InitializeDebugWidgets() {
     graphicsBreakpointsWidget->hide();
     debug_menu->addAction(graphicsBreakpointsWidget->toggleViewAction());
 
-    graphicsVertexShaderWidget = new GraphicsVertexShaderWidget(Pica::g_debug_context, this);
+    graphicsVertexShaderWidget =
+        new GraphicsVertexShaderWidget(system, Pica::g_debug_context, this);
     addDockWidget(Qt::RightDockWidgetArea, graphicsVertexShaderWidget);
     graphicsVertexShaderWidget->hide();
     debug_menu->addAction(graphicsVertexShaderWidget->toggleViewAction());
 
-    graphicsTracingWidget = new GraphicsTracingWidget(Pica::g_debug_context, this);
+    graphicsTracingWidget = new GraphicsTracingWidget(system, Pica::g_debug_context, this);
     addDockWidget(Qt::RightDockWidgetArea, graphicsTracingWidget);
     graphicsTracingWidget->hide();
     debug_menu->addAction(graphicsTracingWidget->toggleViewAction());
@@ -1237,6 +1237,11 @@ void GMainWindow::BootGame(const QString& filename) {
         video_dumping_path.clear();
     }
 
+    // Register debug widgets
+    if (graphicsWidget->isVisible()) {
+        graphicsWidget->Register();
+    }
+
     // Create and start the emulation thread
     emu_thread = std::make_unique<EmuThread>(system, *render_window);
     emit EmulationStarting(emu_thread.get());
@@ -1315,6 +1320,11 @@ void GMainWindow::ShutdownGame() {
     // TODO(bunnei): This function is not thread safe, but it's being used as if it were
     Pica::g_debug_context->ClearBreakpoints();
 
+    // Unregister debug widgets
+    if (graphicsWidget->isVisible()) {
+        graphicsWidget->Unregister();
+    }
+
     // Frame advancing must be cancelled in order to release the emu thread from waiting
     system.frame_limiter.SetFrameAdvancing(false);
 
@@ -2214,7 +2224,7 @@ void GMainWindow::OnToggleFilterBar() {
 
 void GMainWindow::OnCreateGraphicsSurfaceViewer() {
     auto graphicsSurfaceViewerWidget =
-        new GraphicsSurfaceWidget(system.Memory(), Pica::g_debug_context, this);
+        new GraphicsSurfaceWidget(system, Pica::g_debug_context, this);
     addDockWidget(Qt::RightDockWidgetArea, graphicsSurfaceViewerWidget);
     // TODO: Maybe graphicsSurfaceViewerWidget->setFloating(true);
     graphicsSurfaceViewerWidget->show();
@@ -2434,10 +2444,10 @@ void GMainWindow::OnStartVideoDumping() {
 }
 
 void GMainWindow::StartVideoDumping(const QString& path) {
-    Layout::FramebufferLayout layout{
+    auto& renderer = system.GPU().Renderer();
-        Layout::FrameLayoutFromResolutionScale(VideoCore::g_renderer->GetResolutionScaleFactor())};
+    const auto layout{Layout::FrameLayoutFromResolutionScale(renderer.GetResolutionScaleFactor())};
 
-    auto dumper = std::make_shared<VideoDumper::FFmpegBackend>();
+    auto dumper = std::make_shared<VideoDumper::FFmpegBackend>(renderer);
     if (dumper->StartDumping(path.toStdString(), layout)) {
         system.RegisterVideoDumper(dumper);
     } else {

src/common/CMakeLists.txt

@@ -37,16 +37,14 @@ add_custom_command(OUTPUT scm_rev.cpp
       "${VIDEO_CORE}/shader/generator/spv_fs_shader_gen.h"
       "${VIDEO_CORE}/shader/shader.cpp"
       "${VIDEO_CORE}/shader/shader.h"
-      "${VIDEO_CORE}/pica.cpp"
+      "${VIDEO_CORE}/pica/regs_framebuffer.h"
-      "${VIDEO_CORE}/pica.h"
+      "${VIDEO_CORE}/pica/regs_lighting.h"
-      "${VIDEO_CORE}/regs_framebuffer.h"
+      "${VIDEO_CORE}/pica/regs_pipeline.h"
-      "${VIDEO_CORE}/regs_lighting.h"
+      "${VIDEO_CORE}/pica/regs_rasterizer.h"
-      "${VIDEO_CORE}/regs_pipeline.h"
+      "${VIDEO_CORE}/pica/regs_shader.h"
-      "${VIDEO_CORE}/regs_rasterizer.h"
+      "${VIDEO_CORE}/pica/regs_texturing.h"
-      "${VIDEO_CORE}/regs_shader.h"
+      "${VIDEO_CORE}/pica/regs_internal.cpp"
-      "${VIDEO_CORE}/regs_texturing.h"
+      "${VIDEO_CORE}/pica/regs_internal.h"
-      "${VIDEO_CORE}/regs.cpp"
-      "${VIDEO_CORE}/regs.h"
       # and also check that the scm_rev files haven't changed
       "${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.cpp.in"
       "${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.h"

src/common/memory_ref.h

@@ -110,8 +110,10 @@ public:
         return std::span{cptr, std::min(size, csize)};
     }
 
-    std::span<const u8> GetReadBytes(std::size_t size) const {
+    template <typename T>
-        return std::span{cptr, std::min(size, csize)};
+    std::span<const T> GetReadBytes(std::size_t size) const {
+        const auto* cptr_t = reinterpret_cast<T*>(cptr);
+        return std::span{cptr_t, std::min(size, csize) / sizeof(T)};
     }
 
     std::size_t GetSize() const {

src/core/CMakeLists.txt

@@ -299,8 +299,10 @@ add_library(citra_core STATIC
     hle/service/fs/fs_user.h
     hle/service/gsp/gsp.cpp
     hle/service/gsp/gsp.h
+    hle/service/gsp/gsp_command.h
     hle/service/gsp/gsp_gpu.cpp
     hle/service/gsp/gsp_gpu.h
+    hle/service/gsp/gsp_interrupt.h
     hle/service/gsp/gsp_lcd.cpp
     hle/service/gsp/gsp_lcd.h
     hle/service/hid/hid.cpp
@@ -433,12 +435,6 @@ add_library(citra_core STATIC
     hw/aes/ccm.h
     hw/aes/key.cpp
     hw/aes/key.h
-    hw/gpu.cpp
-    hw/gpu.h
-    hw/hw.cpp
-    hw/hw.h
-    hw/lcd.cpp
-    hw/lcd.h
     hw/rsa/rsa.cpp
     hw/rsa/rsa.h
     hw/y2r.cpp

src/core/cheats/cheats.cpp

@@ -11,7 +11,6 @@
 #include "core/core.h"
 #include "core/core_timing.h"
 #include "core/hle/kernel/process.h"
-#include "core/hw/gpu.h"
 
 namespace Cheats {
 

src/core/core.cpp

@@ -35,14 +35,13 @@
 #include "core/hle/service/cam/cam.h"
 #include "core/hle/service/fs/archive.h"
 #include "core/hle/service/gsp/gsp.h"
+#include "core/hle/service/gsp/gsp_gpu.h"
 #include "core/hle/service/ir/ir_rst.h"
 #include "core/hle/service/mic/mic_u.h"
 #include "core/hle/service/plgldr/plgldr.h"
 #include "core/hle/service/service.h"
 #include "core/hle/service/sm/sm.h"
-#include "core/hw/gpu.h"
+#include "core/hw/aes/key.h"
-#include "core/hw/hw.h"
-#include "core/hw/lcd.h"
 #include "core/loader/loader.h"
 #include "core/movie.h"
 #ifdef ENABLE_SCRIPTING
@@ -51,8 +50,8 @@
 #include "core/telemetry_session.h"
 #include "network/network.h"
 #include "video_core/custom_textures/custom_tex_manager.h"
+#include "video_core/gpu.h"
 #include "video_core/renderer_base.h"
-#include "video_core/video_core.h"
 
 namespace Core {
 
@@ -235,7 +234,6 @@ System::ResultStatus System::RunLoop(bool tight_loop) {
         GDBStub::SetCpuStepFlag(false);
     }
 
-    HW::Update();
     Reschedule();
 
     return status;
@@ -433,7 +431,7 @@ System::ResultStatus System::Init(Frontend::EmuWindow& emu_window,
     service_manager = std::make_unique<Service::SM::ServiceManager>(*this);
     archive_manager = std::make_unique<Service::FS::ArchiveManager>(*this);
 
-    HW::Init(*memory);
+    HW::AES::InitKeys();
     Service::Init(*this);
     GDBStub::DeferStart();
 
@@ -443,7 +441,10 @@ System::ResultStatus System::Init(Frontend::EmuWindow& emu_window,
 
     custom_tex_manager = std::make_unique<VideoCore::CustomTexManager>(*this);
 
-    VideoCore::Init(emu_window, secondary_window, *this);
+    auto gsp = service_manager->GetService<Service::GSP::GSP_GPU>("gsp::Gpu");
+    gpu = std::make_unique<VideoCore::GPU>(*this, emu_window, secondary_window);
+    gpu->SetInterruptHandler(
+        [gsp](Service::GSP::InterruptId interrupt_id) { gsp->SignalInterrupt(interrupt_id); });
 
     LOG_DEBUG(Core, "Initialized OK");
 
@@ -452,8 +453,8 @@ System::ResultStatus System::Init(Frontend::EmuWindow& emu_window,
     return ResultStatus::Success;
 }
 
-VideoCore::RendererBase& System::Renderer() {
+VideoCore::GPU& System::GPU() {
-    return *VideoCore::g_renderer;
+    return *gpu;
 }
 
 Service::SM::ServiceManager& System::ServiceManager() {
@@ -555,8 +556,7 @@ void System::Shutdown(bool is_deserializing) {
     // Shutdown emulation session
     is_powered_on = false;
 
-    VideoCore::Shutdown();
+    gpu.reset();
-    HW::Shutdown();
     if (!is_deserializing) {
         GDBStub::Shutdown();
         perf_stats.reset();
@@ -626,18 +626,9 @@ void System::ApplySettings() {
     GDBStub::SetServerPort(Settings::values.gdbstub_port.GetValue());
     GDBStub::ToggleServer(Settings::values.use_gdbstub.GetValue());
 
-    VideoCore::g_shader_jit_enabled = Settings::values.use_shader_jit.GetValue();
+    if (gpu) {
-    VideoCore::g_hw_shader_enabled = Settings::values.use_hw_shader.GetValue();
+        gpu->Renderer().UpdateCurrentFramebufferLayout();
-    VideoCore::g_hw_shader_accurate_mul = Settings::values.shaders_accurate_mul.GetValue();
+        auto& settings = gpu->Renderer().Settings();
-
-#ifndef ANDROID
-    if (VideoCore::g_renderer) {
-        VideoCore::g_renderer->UpdateCurrentFramebufferLayout();
-    }
-#endif
-
-    if (VideoCore::g_renderer) {
-        auto& settings = VideoCore::g_renderer->Settings();
         settings.bg_color_update_requested = true;
         settings.shader_update_requested = true;
     }
@@ -699,17 +690,15 @@ void System::serialize(Archive& ar, const unsigned int file_version) {
             *m_emu_window, m_secondary_window, *memory_mode.first, *n3ds_hw_caps.first, num_cores);
     }
 
-    // flush on save, don't flush on load
+    // Flush on save, don't flush on load
-    bool should_flush = !Archive::is_loading::value;
+    const bool should_flush = !Archive::is_loading::value;
-    Memory::RasterizerClearAll(should_flush);
+    gpu->ClearAll(should_flush);
     ar&* timing.get();
     for (u32 i = 0; i < num_cores; i++) {
         ar&* cpu_cores[i].get();
     }
     ar&* service_manager.get();
     ar&* archive_manager.get();
-    ar& GPU::g_regs;
-    ar& LCD::g_regs;
 
     // NOTE: DSP doesn't like being destroyed and recreated. So instead we do an inline
     // serialization; this means that the DSP Settings need to match for loading to work.
@@ -722,16 +711,21 @@ void System::serialize(Archive& ar, const unsigned int file_version) {
 
     ar&* memory.get();
     ar&* kernel.get();
-    VideoCore::serialize(ar, file_version);
+    ar&* gpu.get();
     ar& movie;
 
     // This needs to be set from somewhere - might as well be here!
     if (Archive::is_loading::value) {
         timing->UnlockEventQueue();
-        Service::GSP::SetGlobalModule(*this);
         memory->SetDSP(*dsp_core);
         cheat_engine->Connect();
-        VideoCore::g_renderer->Sync();
+        gpu->Sync();
+
+        // Re-register gpu callback, because gsp service changed after service_manager got
+        // serialized
+        auto gsp = service_manager->GetService<Service::GSP::GSP_GPU>("gsp::Gpu");
+        gpu->SetInterruptHandler(
+            [gsp](Service::GSP::InterruptId interrupt_id) { gsp->SignalInterrupt(interrupt_id); });
     }
 }
 

src/core/core.h

@@ -58,9 +58,13 @@ class Backend;
 
 namespace VideoCore {
 class CustomTexManager;
-class RendererBase;
+class GPU;
 } // namespace VideoCore
 
+namespace Pica {
+class DebugContext;
+}
+
 namespace Loader {
 class AppLoader;
 }
@@ -217,7 +221,7 @@ public:
         return *dsp_core;
     }
 
-    [[nodiscard]] VideoCore::RendererBase& Renderer();
+    [[nodiscard]] VideoCore::GPU& GPU();
 
     /**
      * Gets a reference to the service manager.
@@ -384,6 +388,8 @@ private:
     /// Telemetry session for this emulation session
     std::unique_ptr<Core::TelemetrySession> telemetry_session;
 
+    std::unique_ptr<VideoCore::GPU> gpu;
+
     /// Service manager
     std::unique_ptr<Service::SM::ServiceManager> service_manager;
 

src/core/core_timing.h

@@ -37,6 +37,10 @@
 constexpr u64 BASE_CLOCK_RATE_ARM11 = 268111856;
 constexpr u64 MAX_VALUE_TO_MULTIPLY = std::numeric_limits<s64>::max() / BASE_CLOCK_RATE_ARM11;
 
+/// Refresh rate defined by ratio of ARM11 frequency to ARM11 ticks per frame
+/// (268,111,856) / (4,481,136) = 59.83122493939037Hz
+constexpr double SCREEN_REFRESH_RATE = BASE_CLOCK_RATE_ARM11 / static_cast<double>(4481136ull);
+
 constexpr s64 msToCycles(int ms) {
     // since ms is int there is no way to overflow
     return BASE_CLOCK_RATE_ARM11 * static_cast<s64>(ms) / 1000;

src/core/dumping/ffmpeg_backend.cpp

@@ -11,10 +11,10 @@
 #include "common/scope_exit.h"
 #include "common/settings.h"
 #include "common/string_util.h"
+#include "core/core_timing.h"
 #include "core/dumping/ffmpeg_backend.h"
-#include "core/hw/gpu.h"
+#include "video_core/gpu.h"
 #include "video_core/renderer_base.h"
-#include "video_core/video_core.h"
 
 using namespace DynamicLibrary;
 
@@ -381,7 +381,7 @@ bool FFmpegVideoStream::InitFilters() {
     }
 
     // Configure buffer source
-    static constexpr AVRational src_time_base{static_cast<int>(GPU::frame_ticks),
+    static constexpr AVRational src_time_base{static_cast<int>(VideoCore::FRAME_TICKS),
                                               static_cast<int>(BASE_CLOCK_RATE_ARM11)};
     const std::string in_args =
         fmt::format("video_size={}x{}:pix_fmt={}:time_base={}/{}:pixel_aspect=1", layout.width,
@@ -732,7 +732,7 @@ void FFmpegMuxer::WriteTrailer() {
     FFmpeg::av_write_trailer(format_context.get());
 }
 
-FFmpegBackend::FFmpegBackend() = default;
+FFmpegBackend::FFmpegBackend(VideoCore::RendererBase& renderer_) : renderer{renderer_} {}
 
 FFmpegBackend::~FFmpegBackend() {
     ASSERT_MSG(!IsDumping(), "Dumping must be stopped first");
@@ -796,7 +796,7 @@ bool FFmpegBackend::StartDumping(const std::string& path, const Layout::Framebuf
         }
     });
 
-    VideoCore::g_renderer->PrepareVideoDumping();
+    renderer.PrepareVideoDumping();
     is_dumping = true;
 
     return true;
@@ -829,7 +829,7 @@ void FFmpegBackend::AddAudioSample(const std::array<s16, 2>& sample) {
 
 void FFmpegBackend::StopDumping() {
     is_dumping = false;
-    VideoCore::g_renderer->CleanupVideoDumping();
+    renderer.CleanupVideoDumping();
 
     // Flush the video processing queue
     AddVideoFrame(VideoFrame());

src/core/dumping/ffmpeg_backend.h

@@ -18,6 +18,10 @@
 #include "common/threadsafe_queue.h"
 #include "core/dumping/backend.h"
 
+namespace VideoCore {
+class RendererBase;
+}
+
 namespace VideoDumper {
 
 using VariableAudioFrame = std::vector<s16>;
@@ -181,7 +185,7 @@ private:
  */
 class FFmpegBackend : public Backend {
 public:
-    FFmpegBackend();
+    FFmpegBackend(VideoCore::RendererBase& renderer);
     ~FFmpegBackend() override;
     bool StartDumping(const std::string& path, const Layout::FramebufferLayout& layout) override;
     void AddVideoFrame(VideoFrame frame) override;
@@ -194,6 +198,7 @@ public:
 private:
     void EndDumping();
 
+    VideoCore::RendererBase& renderer;
     std::atomic_bool is_dumping = false; ///< Whether the backend is currently dumping
 
     FFmpegMuxer ffmpeg{};

src/core/hle/service/gsp/gsp.cpp

@@ -2,33 +2,17 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <vector>
 #include "core/core.h"
-#include "core/hle/kernel/event.h"
-#include "core/hle/kernel/shared_memory.h"
 #include "core/hle/service/gsp/gsp.h"
+#include "core/hle/service/gsp/gsp_gpu.h"
+#include "core/hle/service/gsp/gsp_lcd.h"
 
 namespace Service::GSP {
 
-static std::weak_ptr<GSP_GPU> gsp_gpu;
-
-void SignalInterrupt(InterruptId interrupt_id) {
-    auto gpu = gsp_gpu.lock();
-    ASSERT(gpu != nullptr);
-    return gpu->SignalInterrupt(interrupt_id);
-}
-
 void InstallInterfaces(Core::System& system) {
     auto& service_manager = system.ServiceManager();
-    auto gpu = std::make_shared<GSP_GPU>(system);
+    std::make_shared<GSP_GPU>(system)->InstallAsService(service_manager);
-    gpu->InstallAsService(service_manager);
-    gsp_gpu = gpu;
-
     std::make_shared<GSP_LCD>()->InstallAsService(service_manager);
 }
 
-void SetGlobalModule(Core::System& system) {
-    gsp_gpu = system.ServiceManager().GetService<GSP_GPU>("gsp::Gpu");
-}
-
 } // namespace Service::GSP

src/core/hle/service/gsp/gsp.h

@@ -4,25 +4,12 @@
 
 #pragma once
 
-#include <cstddef>
-#include <string>
-#include "common/common_types.h"
-#include "core/hle/result.h"
-#include "core/hle/service/gsp/gsp_gpu.h"
-#include "core/hle/service/gsp/gsp_lcd.h"
-
 namespace Core {
 class System;
 }
 
 namespace Service::GSP {
-/**
- * Signals that the specified interrupt type has occurred to userland code
- * @param interrupt_id ID of interrupt that is being signalled
- */
-void SignalInterrupt(InterruptId interrupt_id);
 
 void InstallInterfaces(Core::System& system);
 
-void SetGlobalModule(Core::System& system);
 } // namespace Service::GSP

src/core/hle/service/gsp/gsp_command.h

@@ -0,0 +1,110 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/bit_field.h"
+
+namespace Service::GSP {
+
+/// GSP command ID
+enum class CommandId : u32 {
+    RequestDma = 0x00,
+    SubmitCmdList = 0x01,
+    MemoryFill = 0x02,
+    DisplayTransfer = 0x03,
+    TextureCopy = 0x04,
+    CacheFlush = 0x05,
+};
+
+struct DmaCommand {
+    u32 source_address;
+    u32 dest_address;
+    u32 size;
+};
+
+struct SubmitCmdListCommand {
+    u32 address;
+    u32 size;
+    u32 flags;
+    u32 unused[3];
+    u32 do_flush;
+};
+
+struct MemoryFillCommand {
+    u32 start1;
+    u32 value1;
+    u32 end1;
+
+    u32 start2;
+    u32 value2;
+    u32 end2;
+
+    u16 control1;
+    u16 control2;
+};
+
+struct DisplayTransferCommand {
+    u32 in_buffer_address;
+    u32 out_buffer_address;
+    u32 in_buffer_size;
+    u32 out_buffer_size;
+    u32 flags;
+};
+
+struct TextureCopyCommand {
+    u32 in_buffer_address;
+    u32 out_buffer_address;
+    u32 size;
+    u32 in_width_gap;
+    u32 out_width_gap;
+    u32 flags;
+};
+
+struct CacheFlushCommand {
+    struct {
+        u32 address;
+        u32 size;
+    } regions[3];
+};
+
+/// GSP command
+struct Command {
+    BitField<0, 8, CommandId> id;
+    union {
+        DmaCommand dma_request;
+        SubmitCmdListCommand submit_gpu_cmdlist;
+        MemoryFillCommand memory_fill;
+        DisplayTransferCommand display_transfer;
+        TextureCopyCommand texture_copy;
+        CacheFlushCommand cache_flush;
+        std::array<u8, 0x1C> raw_data;
+    };
+};
+static_assert(sizeof(Command) == 0x20, "Command struct has incorrect size");
+
+/// GSP shared memory GX command buffer header
+struct CommandBuffer {
+    union {
+        u32 hex;
+
+        // Current command index. This index is updated by GSP module after loading the command
+        // data, right before the command is processed. When this index is updated by GSP module,
+        // the total commands field is decreased by one as well.
+        BitField<0, 8, u32> index;
+
+        // Total commands to process, must not be value 0 when GSP module handles commands. This
+        // must be <=15 when writing a command to shared memory. This is incremented by the
+        // application when writing a command to shared memory, after increasing this value
+        // TriggerCmdReqQueue is only used if this field is value 1.
+        BitField<8, 8, u32> number_commands;
+    };
+
+    u32 unk[7];
+
+    Command commands[0xF];
+};
+static_assert(sizeof(CommandBuffer) == 0x200, "CommandBuffer struct has incorrect size");
+
+} // namespace Service::GSP

src/core/hle/service/gsp/gsp_gpu.cpp

@@ -9,30 +9,21 @@
 #include <boost/serialization/shared_ptr.hpp>
 #include "common/archives.h"
 #include "common/bit_field.h"
-#include "common/microprofile.h"
-#include "common/swap.h"
 #include "core/core.h"
-#include "core/file_sys/plugin_3gx.h"
-#include "core/hle/ipc.h"
 #include "core/hle/ipc_helpers.h"
 #include "core/hle/kernel/shared_memory.h"
 #include "core/hle/kernel/shared_page.h"
 #include "core/hle/result.h"
 #include "core/hle/service/gsp/gsp_gpu.h"
-#include "core/hw/gpu.h"
-#include "core/hw/hw.h"
-#include "core/hw/lcd.h"
 #include "core/memory.h"
-#include "video_core/debug_utils/debug_utils.h"
+#include "video_core/gpu.h"
 #include "video_core/gpu_debugger.h"
+#include "video_core/pica/regs_lcd.h"
 
 SERIALIZE_EXPORT_IMPL(Service::GSP::SessionData)
 SERIALIZE_EXPORT_IMPL(Service::GSP::GSP_GPU)
 SERVICE_CONSTRUCT_IMPL(Service::GSP::GSP_GPU)
 
-// Main graphics debugger object - TODO: Here is probably not the best place for this
-GraphicsDebugger g_debugger;
-
 namespace Service::GSP {
 
 // Beginning address of HW regs
@@ -59,60 +50,32 @@ constexpr ResultCode ERR_REGS_INVALID_SIZE(ErrorDescription::InvalidSize, ErrorM
                                            ErrorSummary::InvalidArgument,
                                            ErrorLevel::Usage); // 0xE0E02BEC
 
-static PAddr VirtualToPhysicalAddress(VAddr addr) {
-    if (addr == 0) {
-        return 0;
-    }
-
-    // Note: the region end check is inclusive because the game can pass in an address that
-    // represents an open right boundary
-    if (addr >= Memory::VRAM_VADDR && addr <= Memory::VRAM_VADDR_END) {
-        return addr - Memory::VRAM_VADDR + Memory::VRAM_PADDR;
-    }
-    if (addr >= Memory::LINEAR_HEAP_VADDR && addr <= Memory::LINEAR_HEAP_VADDR_END) {
-        return addr - Memory::LINEAR_HEAP_VADDR + Memory::FCRAM_PADDR;
-    }
-    if (addr >= Memory::NEW_LINEAR_HEAP_VADDR && addr <= Memory::NEW_LINEAR_HEAP_VADDR_END) {
-        return addr - Memory::NEW_LINEAR_HEAP_VADDR + Memory::FCRAM_PADDR;
-    }
-    if (addr >= Memory::PLUGIN_3GX_FB_VADDR && addr <= Memory::PLUGIN_3GX_FB_VADDR_END) {
-        return addr - Memory::PLUGIN_3GX_FB_VADDR + Service::PLGLDR::PLG_LDR::GetPluginFBAddr();
-    }
-
-    LOG_ERROR(HW_Memory, "Unknown virtual address @ 0x{:08X}", addr);
-    // To help with debugging, set bit on address so that it's obviously invalid.
-    // TODO: find the correct way to handle this error
-    return addr | 0x80000000;
-}
-
 u32 GSP_GPU::GetUnusedThreadId() const {
     for (u32 id = 0; id < MaxGSPThreads; ++id) {
-        if (!used_thread_ids[id])
+        if (!used_thread_ids[id]) {
             return id;
+        }
     }
 
     UNREACHABLE_MSG("All GSP threads are in use");
     return 0;
 }
 
-/// Gets a pointer to a thread command buffer in GSP shared memory
+CommandBuffer* GSP_GPU::GetCommandBuffer(u32 thread_id) {
-static inline u8* GetCommandBuffer(std::shared_ptr<Kernel::SharedMemory> shared_memory,
+    auto* ptr = shared_memory->GetPointer(0x800 + (thread_id * sizeof(CommandBuffer)));
-                                   u32 thread_id) {
+    return reinterpret_cast<CommandBuffer*>(ptr);
-    return shared_memory->GetPointer(0x800 + (thread_id * sizeof(CommandBuffer)));
 }
 
 FrameBufferUpdate* GSP_GPU::GetFrameBufferInfo(u32 thread_id, u32 screen_index) {
     DEBUG_ASSERT_MSG(screen_index < 2, "Invalid screen index");
 
     // For each thread there are two FrameBufferUpdate fields
-    u32 offset = 0x200 + (2 * thread_id + screen_index) * sizeof(FrameBufferUpdate);
+    const u32 offset = 0x200 + (2 * thread_id + screen_index) * sizeof(FrameBufferUpdate);
     u8* ptr = shared_memory->GetPointer(offset);
     return reinterpret_cast<FrameBufferUpdate*>(ptr);
 }
 
-/// Gets a pointer to the interrupt relay queue for a given thread index
+InterruptRelayQueue* GSP_GPU::GetInterruptRelayQueue(u32 thread_id) {
-static inline InterruptRelayQueue* GetInterruptRelayQueue(
-    std::shared_ptr<Kernel::SharedMemory> shared_memory, u32 thread_id) {
     u8* ptr = shared_memory->GetPointer(sizeof(InterruptRelayQueue) * thread_id);
     return reinterpret_cast<InterruptRelayQueue*>(ptr);
 }
@@ -125,19 +88,6 @@ void GSP_GPU::ClientDisconnected(std::shared_ptr<Kernel::ServerSession> server_s
     SessionRequestHandler::ClientDisconnected(server_session);
 }
 
-/**
- * Writes a single GSP GPU hardware registers with a single u32 value
- * (For internal use.)
- *
- * @param base_address The address of the register in question
- * @param data Data to be written
- */
-static void WriteSingleHWReg(u32 base_address, u32 data) {
-    DEBUG_ASSERT_MSG((base_address & 3) == 0 && base_address < 0x420000,
-                     "Write address out of range or misaligned");
-    HW::Write<u32>(base_address + REGS_BEGIN, data);
-}
-
 /**
  * Writes sequential GSP GPU hardware registers using an array of source data
  *
@@ -146,7 +96,8 @@ static void WriteSingleHWReg(u32 base_address, u32 data) {
  * @param data A vector containing the source data
  * @return RESULT_SUCCESS if the parameters are valid, error code otherwise
  */
-static ResultCode WriteHWRegs(u32 base_address, u32 size_in_bytes, std::span<const u8> data) {
+static ResultCode WriteHWRegs(u32 base_address, u32 size_in_bytes, std::span<const u8> data,
+                              VideoCore::GPU& gpu) {
     // This magic number is verified to be done by the gsp module
     const u32 max_size_in_bytes = 0x80;
 
@@ -155,28 +106,30 @@ static ResultCode WriteHWRegs(u32 base_address, u32 size_in_bytes, std::span<con
                   "Write address was out of range or misaligned! (address=0x{:08x}, size=0x{:08x})",
                   base_address, size_in_bytes);
         return ERR_REGS_OUTOFRANGE_OR_MISALIGNED;
-    } else if (size_in_bytes <= max_size_in_bytes) {
+    }
-        if (size_in_bytes & 3) {
-            LOG_ERROR(Service_GSP, "Misaligned size 0x{:08x}", size_in_bytes);
-            return ERR_REGS_MISALIGNED;
-        } else {
-            std::size_t offset = 0;
-            while (size_in_bytes > 0) {
-                u32 value;
-                std::memcpy(&value, &data[offset], sizeof(u32));
-                WriteSingleHWReg(base_address, value);
 
-                size_in_bytes -= 4;
+    if (size_in_bytes > max_size_in_bytes) {
-                offset += 4;
-                base_address += 4;
-            }
-            return RESULT_SUCCESS;
-        }
-
-    } else {
         LOG_ERROR(Service_GSP, "Out of range size 0x{:08x}", size_in_bytes);
         return ERR_REGS_INVALID_SIZE;
     }
+
+    if (size_in_bytes & 3) {
+        LOG_ERROR(Service_GSP, "Misaligned size 0x{:08x}", size_in_bytes);
+        return ERR_REGS_MISALIGNED;
+    }
+
+    std::size_t offset = 0;
+    while (size_in_bytes > 0) {
+        u32 value;
+        std::memcpy(&value, &data[offset], sizeof(u32));
+        gpu.WriteReg(REGS_BEGIN + base_address, value);
+
+        size_in_bytes -= 4;
+        offset += 4;
+        base_address += 4;
+    }
+
+    return RESULT_SUCCESS;
 }
 
 /**
@@ -190,7 +143,7 @@ static ResultCode WriteHWRegs(u32 base_address, u32 size_in_bytes, std::span<con
  * @return RESULT_SUCCESS if the parameters are valid, error code otherwise
  */
 static ResultCode WriteHWRegsWithMask(u32 base_address, u32 size_in_bytes, std::span<const u8> data,
-                                      std::span<const u8> masks) {
+                                      std::span<const u8> masks, VideoCore::GPU& gpu) {
     // This magic number is verified to be done by the gsp module
     const u32 max_size_in_bytes = 0x80;
 
@@ -199,60 +152,58 @@ static ResultCode WriteHWRegsWithMask(u32 base_address, u32 size_in_bytes, std::
                   "Write address was out of range or misaligned! (address=0x{:08x}, size=0x{:08x})",
                   base_address, size_in_bytes);
         return ERR_REGS_OUTOFRANGE_OR_MISALIGNED;
-    } else if (size_in_bytes <= max_size_in_bytes) {
+    }
-        if (size_in_bytes & 3) {
-            LOG_ERROR(Service_GSP, "Misaligned size 0x{:08x}", size_in_bytes);
-            return ERR_REGS_MISALIGNED;
-        } else {
-            std::size_t offset = 0;
-            while (size_in_bytes > 0) {
-                const u32 reg_address = base_address + REGS_BEGIN;
 
-                u32 reg_value;
+    if (size_in_bytes > max_size_in_bytes) {
-                HW::Read<u32>(reg_value, reg_address);
-
-                u32 value, mask;
-                std::memcpy(&value, &data[offset], sizeof(u32));
-                std::memcpy(&mask, &masks[offset], sizeof(u32));
-
-                // Update the current value of the register only for set mask bits
-                reg_value = (reg_value & ~mask) | (value & mask);
-
-                WriteSingleHWReg(base_address, reg_value);
-
-                size_in_bytes -= 4;
-                offset += 4;
-                base_address += 4;
-            }
-            return RESULT_SUCCESS;
-        }
-
-    } else {
         LOG_ERROR(Service_GSP, "Out of range size 0x{:08x}", size_in_bytes);
         return ERR_REGS_INVALID_SIZE;
     }
+
+    if (size_in_bytes & 3) {
+        LOG_ERROR(Service_GSP, "Misaligned size 0x{:08x}", size_in_bytes);
+        return ERR_REGS_MISALIGNED;
+    }
+
+    std::size_t offset = 0;
+    while (size_in_bytes > 0) {
+        const u32 reg_address = base_address + REGS_BEGIN;
+        u32 reg_value = gpu.ReadReg(reg_address);
+
+        u32 value, mask;
+        std::memcpy(&value, &data[offset], sizeof(u32));
+        std::memcpy(&mask, &masks[offset], sizeof(u32));
+
+        // Update the current value of the register only for set mask bits
+        reg_value = (reg_value & ~mask) | (value & mask);
+        gpu.WriteReg(reg_address, reg_value);
+
+        size_in_bytes -= 4;
+        offset += 4;
+        base_address += 4;
+    }
+
+    return RESULT_SUCCESS;
 }
 
 void GSP_GPU::WriteHWRegs(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp(ctx);
-    u32 reg_addr = rp.Pop<u32>();
+    const u32 reg_addr = rp.Pop<u32>();
-    u32 size = rp.Pop<u32>();
+    const u32 size = rp.Pop<u32>();
-    std::vector<u8> src_data = rp.PopStaticBuffer();
+    const auto src_data = rp.PopStaticBuffer();
 
     IPC::RequestBuilder rb = rp.MakeBuilder(1, 0);
-    rb.Push(GSP::WriteHWRegs(reg_addr, size, src_data));
+    rb.Push(GSP::WriteHWRegs(reg_addr, size, src_data, system.GPU()));
 }
 
 void GSP_GPU::WriteHWRegsWithMask(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp(ctx);
-    u32 reg_addr = rp.Pop<u32>();
+    const u32 reg_addr = rp.Pop<u32>();
-    u32 size = rp.Pop<u32>();
+    const u32 size = rp.Pop<u32>();
-
+    const auto src_data = rp.PopStaticBuffer();
-    std::vector<u8> src_data = rp.PopStaticBuffer();
+    const auto mask_data = rp.PopStaticBuffer();
-    std::vector<u8> mask_data = rp.PopStaticBuffer();
 
     IPC::RequestBuilder rb = rp.MakeBuilder(1, 0);
-    rb.Push(GSP::WriteHWRegsWithMask(reg_addr, size, src_data, mask_data));
+    rb.Push(GSP::WriteHWRegsWithMask(reg_addr, size, src_data, mask_data, system.GPU()));
 }
 
 void GSP_GPU::ReadHWRegs(Kernel::HLERequestContext& ctx) {
@@ -270,7 +221,7 @@ void GSP_GPU::ReadHWRegs(Kernel::HLERequestContext& ctx) {
         return;
     }
 
-    // size should be word-aligned
+    // Size should be word-aligned
     if ((size % 4) != 0) {
         IPC::RequestBuilder rb = rp.MakeBuilder(1, 0);
         rb.Push(ERR_REGS_MISALIGNED);
@@ -279,8 +230,9 @@ void GSP_GPU::ReadHWRegs(Kernel::HLERequestContext& ctx) {
     }
 
     std::vector<u8> buffer(size);
-    for (u32 offset = 0; offset < size; ++offset) {
+    for (u32 word = 0; word < size / sizeof(u32); ++word) {
-        HW::Read<u8>(buffer[offset], REGS_BEGIN + reg_addr + offset);
+        const u32 data = system.GPU().ReadReg(REGS_BEGIN + reg_addr + word * sizeof(u32));
+        std::memcpy(buffer.data() + word * sizeof(u32), &data, sizeof(u32));
     }
 
     IPC::RequestBuilder rb = rp.MakeBuilder(1, 2);
@@ -288,53 +240,15 @@ void GSP_GPU::ReadHWRegs(Kernel::HLERequestContext& ctx) {
     rb.PushStaticBuffer(std::move(buffer), 0);
 }
 
-ResultCode SetBufferSwap(u32 screen_id, const FrameBufferInfo& info) {
-    u32 base_address = 0x400000;
-    PAddr phys_address_left = VirtualToPhysicalAddress(info.address_left);
-    PAddr phys_address_right = VirtualToPhysicalAddress(info.address_right);
-    if (info.active_fb == 0) {
-        WriteSingleHWReg(base_address + 4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(
-                                                screen_id, address_left1)),
-                         phys_address_left);
-        WriteSingleHWReg(base_address + 4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(
-                                                screen_id, address_right1)),
-                         phys_address_right);
-    } else {
-        WriteSingleHWReg(base_address + 4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(
-                                                screen_id, address_left2)),
-                         phys_address_left);
-        WriteSingleHWReg(base_address + 4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(
-                                                screen_id, address_right2)),
-                         phys_address_right);
-    }
-    WriteSingleHWReg(base_address +
-                         4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(screen_id, stride)),
-                     info.stride);
-    WriteSingleHWReg(base_address +
-                         4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(screen_id, color_format)),
-                     info.format);
-    WriteSingleHWReg(base_address +
-                         4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(screen_id, active_fb)),
-                     info.shown_fb);
-
-    if (Pica::g_debug_context)
-        Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::BufferSwapped, nullptr);
-
-    if (screen_id == 0) {
-        MicroProfileFlip();
-        Core::System::GetInstance().perf_stats->EndGameFrame();
-    }
-
-    return RESULT_SUCCESS;
-}
-
 void GSP_GPU::SetBufferSwap(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp(ctx);
     u32 screen_id = rp.Pop<u32>();
     auto fb_info = rp.PopRaw<FrameBufferInfo>();
 
+    system.GPU().SetBufferSwap(screen_id, fb_info);
+
     IPC::RequestBuilder rb = rp.MakeBuilder(1, 0);
-    rb.Push(GSP::SetBufferSwap(screen_id, fb_info));
+    rb.Push(RESULT_SUCCESS);
 }
 
 void GSP_GPU::FlushDataCache(Kernel::HLERequestContext& ctx) {
@@ -382,10 +296,9 @@ void GSP_GPU::RegisterInterruptRelayQueue(Kernel::HLERequestContext& ctx) {
     u32 flags = rp.Pop<u32>();
 
     auto interrupt_event = rp.PopObject<Kernel::Event>();
-    // TODO(mailwl): return right error code instead assert
+    ASSERT_MSG(interrupt_event, "handle is not valid!");
-    ASSERT_MSG((interrupt_event != nullptr), "handle is not valid!");
 
-    interrupt_event->SetName("GSP_GSP_GPU::interrupt_event");
+    interrupt_event->SetName("GSP_GPU::interrupt_event");
 
     SessionData* session_data = GetSessionData(ctx.Session());
     session_data->interrupt_event = std::move(interrupt_event);
@@ -422,15 +335,17 @@ void GSP_GPU::UnregisterInterruptRelayQueue(Kernel::HLERequestContext& ctx) {
 
 void GSP_GPU::SignalInterruptForThread(InterruptId interrupt_id, u32 thread_id) {
     SessionData* session_data = FindRegisteredThreadData(thread_id);
-    if (session_data == nullptr)
+    if (!session_data) {
         return;
+    }
 
     auto interrupt_event = session_data->interrupt_event;
     if (interrupt_event == nullptr) {
         LOG_WARNING(Service_GSP, "cannot synchronize until GSP event has been created!");
         return;
     }
-    InterruptRelayQueue* interrupt_relay_queue = GetInterruptRelayQueue(shared_memory, thread_id);
+
+    auto* interrupt_relay_queue = GetInterruptRelayQueue(thread_id);
     u8 next = interrupt_relay_queue->index;
     next += interrupt_relay_queue->number_interrupts;
     next = next % 0x34; // 0x34 is the number of interrupt slots
@@ -441,29 +356,20 @@ void GSP_GPU::SignalInterruptForThread(InterruptId interrupt_id, u32 thread_id)
     interrupt_relay_queue->error_code = 0x0; // No error
 
     // Update framebuffer information if requested
-    // TODO(yuriks): Confirm where this code should be called. It is definitely updated without
+    const s32 screen_id = (interrupt_id == InterruptId::PDC0)   ? 0
-    //               executing any GSP commands, only waiting on the event.
+                          : (interrupt_id == InterruptId::PDC1) ? 1
-    // TODO(Subv): The real GSP module triggers PDC0 after updating both the top and bottom
+                                                                : -1;
-    // screen, it is currently unknown what PDC1 does.
-    int screen_id = (interrupt_id == InterruptId::PDC0)   ? 0
-                    : (interrupt_id == InterruptId::PDC1) ? 1
-                                                          : -1;
     if (screen_id != -1) {
-        FrameBufferUpdate* info = GetFrameBufferInfo(thread_id, screen_id);
+        auto* info = GetFrameBufferInfo(thread_id, screen_id);
         if (info->is_dirty) {
-            GSP::SetBufferSwap(screen_id, info->framebuffer_info[info->index]);
+            system.GPU().SetBufferSwap(screen_id, info->framebuffer_info[info->index]);
             info->is_dirty.Assign(false);
         }
     }
+
     interrupt_event->Signal();
 }
 
-/**
- * Signals that the specified interrupt type has occurred to userland code
- * @param interrupt_id ID of interrupt that is being signalled
- * @todo This should probably take a thread_id parameter and only signal this thread?
- * @todo This probably does not belong in the GSP module, instead move to video_core
- */
 void GSP_GPU::SignalInterrupt(InterruptId interrupt_id) {
     if (nullptr == shared_memory) {
         LOG_WARNING(Service_GSP, "cannot synchronize until GSP shared memory has been created!");
@@ -488,154 +394,13 @@ void GSP_GPU::SignalInterrupt(InterruptId interrupt_id) {
     SignalInterruptForThread(interrupt_id, active_thread_id);
 }
 
-MICROPROFILE_DEFINE(GPU_GSP_DMA, "GPU", "GSP DMA", MP_RGB(100, 0, 255));
-
-/// Executes the next GSP command
-static void ExecuteCommand(const Command& command, u32 thread_id) {
-    // Utility function to convert register ID to address
-    static auto WriteGPURegister = [](u32 id, u32 data) {
-        GPU::Write<u32>(0x1EF00000 + 4 * id, data);
-    };
-
-    switch (command.id) {
-
-    // GX request DMA - typically used for copying memory from GSP heap to VRAM
-    case CommandId::REQUEST_DMA: {
-        MICROPROFILE_SCOPE(GPU_GSP_DMA);
-        Memory::MemorySystem& memory = Core::System::GetInstance().Memory();
-
-        // TODO: Consider attempting rasterizer-accelerated surface blit if that usage is ever
-        // possible/likely
-        Memory::RasterizerFlushVirtualRegion(command.dma_request.source_address,
-                                             command.dma_request.size, Memory::FlushMode::Flush);
-        Memory::RasterizerFlushVirtualRegion(command.dma_request.dest_address,
-                                             command.dma_request.size,
-                                             Memory::FlushMode::Invalidate);
-
-        // TODO(Subv): These memory accesses should not go through the application's memory mapping.
-        // They should go through the GSP module's memory mapping.
-        memory.CopyBlock(*Core::System::GetInstance().Kernel().GetCurrentProcess(),
-                         command.dma_request.dest_address, command.dma_request.source_address,
-                         command.dma_request.size);
-        SignalInterrupt(InterruptId::DMA);
-        break;
-    }
-    // TODO: This will need some rework in the future. (why?)
-    case CommandId::SUBMIT_GPU_CMDLIST: {
-        auto& params = command.submit_gpu_cmdlist;
-
-        if (params.do_flush) {
-            // This flag flushes the command list (params.address, params.size) from the cache.
-            // Command lists are not processed by the hardware renderer, so we don't need to
-            // actually flush them in Citra.
-        }
-
-        WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.address)),
-                         VirtualToPhysicalAddress(params.address) >> 3);
-        WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.size)),
-                         params.size);
-
-        // TODO: Not sure if we are supposed to always write this .. seems to trigger processing
-        // though
-        WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.trigger)), 1);
-
-        // TODO(yuriks): Figure out the meaning of the `flags` field.
-
-        break;
-    }
-
-    // It's assumed that the two "blocks" behave equivalently.
-    // Presumably this is done simply to allow two memory fills to run in parallel.
-    case CommandId::SET_MEMORY_FILL: {
-        auto& params = command.memory_fill;
-
-        if (params.start1 != 0) {
-            WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].address_start)),
-                             VirtualToPhysicalAddress(params.start1) >> 3);
-            WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].address_end)),
-                             VirtualToPhysicalAddress(params.end1) >> 3);
-            WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].value_32bit)),
-                             params.value1);
-            WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].control)),
-                             params.control1);
-        }
-
-        if (params.start2 != 0) {
-            WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].address_start)),
-                             VirtualToPhysicalAddress(params.start2) >> 3);
-            WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].address_end)),
-                             VirtualToPhysicalAddress(params.end2) >> 3);
-            WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].value_32bit)),
-                             params.value2);
-            WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].control)),
-                             params.control2);
-        }
-        break;
-    }
-
-    case CommandId::SET_DISPLAY_TRANSFER: {
-        auto& params = command.display_transfer;
-        WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(display_transfer_config.input_address)),
-                         VirtualToPhysicalAddress(params.in_buffer_address) >> 3);
-        WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(display_transfer_config.output_address)),
-                         VirtualToPhysicalAddress(params.out_buffer_address) >> 3);
-        WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(display_transfer_config.input_size)),
-                         params.in_buffer_size);
-        WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(display_transfer_config.output_size)),
-                         params.out_buffer_size);
-        WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(display_transfer_config.flags)),
-                         params.flags);
-        WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(display_transfer_config.trigger)), 1);
-        break;
-    }
-
-    case CommandId::SET_TEXTURE_COPY: {
-        auto& params = command.texture_copy;
-        WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.input_address),
-                         VirtualToPhysicalAddress(params.in_buffer_address) >> 3);
-        WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.output_address),
-                         VirtualToPhysicalAddress(params.out_buffer_address) >> 3);
-        WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.texture_copy.size),
-                         params.size);
-        WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.texture_copy.input_size),
-                         params.in_width_gap);
-        WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.texture_copy.output_size),
-                         params.out_width_gap);
-        WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.flags), params.flags);
-
-        // NOTE: Actual GSP ORs 1 with current register instead of overwriting. Doesn't seem to
-        // matter.
-        WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.trigger), 1);
-        break;
-    }
-
-    case CommandId::CACHE_FLUSH: {
-        // NOTE: Rasterizer flushing handled elsewhere in CPU read/write and other GPU handlers
-        // Use command.cache_flush.regions to implement this handler
-        break;
-    }
-
-    default:
-        LOG_ERROR(Service_GSP, "unknown command 0x{:08X}", (int)command.id.Value());
-    }
-
-    if (Pica::g_debug_context)
-        Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::GSPCommandProcessed,
-                                       (void*)&command);
-}
-
 void GSP_GPU::SetLcdForceBlack(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp(ctx);
+    const bool enable_black = rp.Pop<bool>();
 
-    bool enable_black = rp.Pop<bool>();
+    Pica::ColorFill data{};
-    LCD::Regs::ColorFill data = {0};
-
-    // Since data is already zeroed, there is no need to explicitly set
-    // the color to black (all zero).
     data.is_enabled.Assign(enable_black);
-
+    system.GPU().SetColorFill(data);
-    LCD::Write(HW::VADDR_LCD + 4 * LCD_REG_INDEX(color_fill_top), data.raw);    // Top LCD
-    LCD::Write(HW::VADDR_LCD + 4 * LCD_REG_INDEX(color_fill_bottom), data.raw); // Bottom LCD
 
     IPC::RequestBuilder rb = rp.MakeBuilder(1, 0);
     rb.Push(RESULT_SUCCESS);
@@ -644,20 +409,17 @@ void GSP_GPU::SetLcdForceBlack(Kernel::HLERequestContext& ctx) {
 void GSP_GPU::TriggerCmdReqQueue(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp(ctx);
 
-    // Iterate through each thread's command queue...
+    // Iterate through each command.
-    for (unsigned thread_id = 0; thread_id < 0x4; ++thread_id) {
+    auto* command_buffer = GetCommandBuffer(active_thread_id);
-        CommandBuffer* command_buffer = (CommandBuffer*)GetCommandBuffer(shared_memory, thread_id);
+    auto& gpu = system.GPU();
+    for (u32 i = 0; i < command_buffer->number_commands; i++) {
+        gpu.Debugger().GXCommandProcessed(command_buffer->commands[i]);
 
-        // Iterate through each command...
+        // Decode and execute command
-        for (unsigned i = 0; i < command_buffer->number_commands; ++i) {
+        gpu.Execute(command_buffer->commands[i]);
-            g_debugger.GXCommandProcessed((u8*)&command_buffer->commands[i]);
 
-            // Decode and execute command
+        // Indicates that command has completed
-            ExecuteCommand(command_buffer->commands[i], thread_id);
+        command_buffer->number_commands.Assign(command_buffer->number_commands - 1);
-
-            // Indicates that command has completed
-            command_buffer->number_commands.Assign(command_buffer->number_commands - 1);
-        }
     }
 
     IPC::RequestBuilder rb = rp.MakeBuilder(1, 0);

src/core/hle/service/gsp/gsp_gpu.h

@@ -13,7 +13,8 @@
 #include "common/common_types.h"
 #include "core/hle/kernel/event.h"
 #include "core/hle/kernel/hle_ipc.h"
-#include "core/hle/result.h"
+#include "core/hle/service/gsp/gsp_command.h"
+#include "core/hle/service/gsp/gsp_interrupt.h"
 #include "core/hle/service/service.h"
 
 namespace Core {
@@ -28,53 +29,6 @@ class SharedMemory;
 
 namespace Service::GSP {
 
-/// GSP interrupt ID
-enum class InterruptId : u8 {
-    PSC0 = 0x00,
-    PSC1 = 0x01,
-    PDC0 = 0x02, // Seems called every vertical screen line
-    PDC1 = 0x03, // Seems called every frame
-    PPF = 0x04,
-    P3D = 0x05,
-    DMA = 0x06,
-};
-
-/// GSP command ID
-enum class CommandId : u32 {
-    REQUEST_DMA = 0x00,
-    /// Submits a commandlist for execution by the GPU.
-    SUBMIT_GPU_CMDLIST = 0x01,
-
-    // Fills a given memory range with a particular value
-    SET_MEMORY_FILL = 0x02,
-
-    // Copies an image and optionally performs color-conversion or scaling.
-    // This is highly similar to the GameCube's EFB copy feature
-    SET_DISPLAY_TRANSFER = 0x03,
-
-    // Conceptionally similar to SET_DISPLAY_TRANSFER and presumable uses the same hardware path
-    SET_TEXTURE_COPY = 0x04,
-    /// Flushes up to 3 cache regions in a single command.
-    CACHE_FLUSH = 0x05,
-};
-
-/// GSP thread interrupt relay queue
-struct InterruptRelayQueue {
-    // Index of last interrupt in the queue
-    u8 index;
-    // Number of interrupts remaining to be processed by the userland code
-    u8 number_interrupts;
-    // Error code - zero on success, otherwise an error has occurred
-    u8 error_code;
-    u8 padding1;
-
-    u32 missed_PDC0;
-    u32 missed_PDC1;
-
-    InterruptId slot[0x34]; ///< Interrupt ID slots
-};
-static_assert(sizeof(InterruptRelayQueue) == 0x40, "InterruptRelayQueue struct has incorrect size");
-
 struct FrameBufferInfo {
     u32 active_fb; // 0 = first, 1 = second
     u32 address_left;
@@ -96,95 +50,9 @@ struct FrameBufferUpdate {
     u32 pad2;
 };
 static_assert(sizeof(FrameBufferUpdate) == 0x40, "Struct has incorrect size");
-// TODO: Not sure if this padding is correct.
-// Chances are the second block is stored at offset 0x24 rather than 0x20.
 static_assert(offsetof(FrameBufferUpdate, framebuffer_info[1]) == 0x20,
               "FrameBufferInfo element has incorrect alignment");
 
-/// GSP command
-struct Command {
-    BitField<0, 8, CommandId> id;
-
-    union {
-        struct {
-            u32 source_address;
-            u32 dest_address;
-            u32 size;
-        } dma_request;
-
-        struct {
-            u32 address;
-            u32 size;
-            u32 flags;
-            u32 unused[3];
-            u32 do_flush;
-        } submit_gpu_cmdlist;
-
-        struct {
-            u32 start1;
-            u32 value1;
-            u32 end1;
-
-            u32 start2;
-            u32 value2;
-            u32 end2;
-
-            u16 control1;
-            u16 control2;
-        } memory_fill;
-
-        struct {
-            u32 in_buffer_address;
-            u32 out_buffer_address;
-            u32 in_buffer_size;
-            u32 out_buffer_size;
-            u32 flags;
-        } display_transfer;
-
-        struct {
-            u32 in_buffer_address;
-            u32 out_buffer_address;
-            u32 size;
-            u32 in_width_gap;
-            u32 out_width_gap;
-            u32 flags;
-        } texture_copy;
-
-        struct {
-            struct {
-                u32 address;
-                u32 size;
-            } regions[3];
-        } cache_flush;
-
-        u8 raw_data[0x1C];
-    };
-};
-static_assert(sizeof(Command) == 0x20, "Command struct has incorrect size");
-
-/// GSP shared memory GX command buffer header
-struct CommandBuffer {
-    union {
-        u32 hex;
-
-        // Current command index. This index is updated by GSP module after loading the command
-        // data, right before the command is processed. When this index is updated by GSP module,
-        // the total commands field is decreased by one as well.
-        BitField<0, 8, u32> index;
-
-        // Total commands to process, must not be value 0 when GSP module handles commands. This
-        // must be <=15 when writing a command to shared memory. This is incremented by the
-        // application when writing a command to shared memory, after increasing this value
-        // TriggerCmdReqQueue is only used if this field is value 1.
-        BitField<8, 8, u32> number_commands;
-    };
-
-    u32 unk[7];
-
-    Command commands[0xF];
-};
-static_assert(sizeof(CommandBuffer) == 0x200, "CommandBuffer struct has incorrect size");
-
 constexpr u32 FRAMEBUFFER_WIDTH = 240;
 constexpr u32 FRAMEBUFFER_WIDTH_POW2 = 256;
 constexpr u32 TOP_FRAMEBUFFER_HEIGHT = 400;
@@ -242,6 +110,12 @@ public:
      */
     FrameBufferUpdate* GetFrameBufferInfo(u32 thread_id, u32 screen_index);
 
+    /// Gets a pointer to a thread command buffer in GSP shared memory
+    CommandBuffer* GetCommandBuffer(u32 thread_id);
+
+    /// Gets a pointer to the interrupt relay queue for a given thread index
+    InterruptRelayQueue* GetInterruptRelayQueue(u32 thread_id);
+
     /**
      * Retreives the ID of the thread with GPU rights.
      */
@@ -513,7 +387,7 @@ private:
     static constexpr u32 MaxGSPThreads = 4;
 
     /// Thread ids currently in use by the sessions connected to the GSPGPU service.
-    std::array<bool, MaxGSPThreads> used_thread_ids = {false, false, false, false};
+    std::array<bool, MaxGSPThreads> used_thread_ids{};
 
     friend class SessionData;
 
@@ -522,8 +396,6 @@ private:
     friend class boost::serialization::access;
 };
 
-ResultCode SetBufferSwap(u32 screen_id, const FrameBufferInfo& info);
-
 } // namespace Service::GSP
 
 BOOST_CLASS_EXPORT_KEY(Service::GSP::SessionData)

src/core/hle/service/gsp/gsp_interrupt.h

@@ -0,0 +1,42 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <functional>
+#include "common/common_types.h"
+
+namespace Service::GSP {
+
+/// GSP interrupt ID
+enum class InterruptId : u8 {
+    PSC0 = 0x00,
+    PSC1 = 0x01,
+    PDC0 = 0x02,
+    PDC1 = 0x03,
+    PPF = 0x04,
+    P3D = 0x05,
+    DMA = 0x06,
+};
+
+/// GSP thread interrupt relay queue
+struct InterruptRelayQueue {
+    // Index of last interrupt in the queue
+    u8 index;
+    // Number of interrupts remaining to be processed by the userland code
+    u8 number_interrupts;
+    // Error code - zero on success, otherwise an error has occurred
+    u8 error_code;
+    u8 padding1;
+
+    u32 missed_PDC0;
+    u32 missed_PDC1;
+
+    InterruptId slot[0x34]; ///< Interrupt ID slots
+};
+static_assert(sizeof(InterruptRelayQueue) == 0x40, "InterruptRelayQueue struct has incorrect size");
+
+using InterruptHandler = std::function<void(InterruptId)>;
+
+} // namespace Service::GSP

src/core/hle/service/hid/hid.cpp

@@ -22,7 +22,6 @@
 #include "core/hle/service/hid/hid_user.h"
 #include "core/hle/service/service.h"
 #include "core/movie.h"
-#include "video_core/video_core.h"
 
 SERVICE_CONSTRUCT_IMPL(Service::HID::Module)
 SERIALIZE_EXPORT_IMPL(Service::HID::Module)

src/core/hw/gpu.cpp

@@ -1,572 +0,0 @@
-// Copyright 2014 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <cstring>
-#include <numeric>
-#include <type_traits>
-#include "common/alignment.h"
-#include "common/color.h"
-#include "common/common_types.h"
-#include "common/logging/log.h"
-#include "common/microprofile.h"
-#include "common/vector_math.h"
-#include "core/core.h"
-#include "core/core_timing.h"
-#include "core/hle/service/gsp/gsp.h"
-#include "core/hw/gpu.h"
-#include "core/hw/hw.h"
-#include "core/memory.h"
-#include "core/tracer/recorder.h"
-#include "video_core/command_processor.h"
-#include "video_core/debug_utils/debug_utils.h"
-#include "video_core/rasterizer_interface.h"
-#include "video_core/renderer_base.h"
-#include "video_core/utils.h"
-#include "video_core/video_core.h"
-
-namespace GPU {
-
-Regs g_regs;
-Memory::MemorySystem* g_memory;
-
-/// Event id for CoreTiming
-static Core::TimingEventType* vblank_event;
-
-template <typename T>
-inline void Read(T& var, const u32 raw_addr) {
-    u32 addr = raw_addr - HW::VADDR_GPU;
-    u32 index = addr / 4;
-
-    // Reads other than u32 are untested, so I'd rather have them abort than silently fail
-    if (index >= Regs::NumIds() || !std::is_same<T, u32>::value) {
-        LOG_ERROR(HW_GPU, "unknown Read{} @ {:#010X}", sizeof(var) * 8, addr);
-        return;
-    }
-
-    var = g_regs[addr / 4];
-}
-
-static Common::Vec4<u8> DecodePixel(Regs::PixelFormat input_format, const u8* src_pixel) {
-    switch (input_format) {
-    case Regs::PixelFormat::RGBA8:
-        return Common::Color::DecodeRGBA8(src_pixel);
-
-    case Regs::PixelFormat::RGB8:
-        return Common::Color::DecodeRGB8(src_pixel);
-
-    case Regs::PixelFormat::RGB565:
-        return Common::Color::DecodeRGB565(src_pixel);
-
-    case Regs::PixelFormat::RGB5A1:
-        return Common::Color::DecodeRGB5A1(src_pixel);
-
-    case Regs::PixelFormat::RGBA4:
-        return Common::Color::DecodeRGBA4(src_pixel);
-
-    default:
-        LOG_ERROR(HW_GPU, "Unknown source framebuffer format {:x}", input_format);
-        return {0, 0, 0, 0};
-    }
-}
-
-MICROPROFILE_DEFINE(GPU_DisplayTransfer, "GPU", "DisplayTransfer", MP_RGB(100, 100, 255));
-MICROPROFILE_DEFINE(GPU_CmdlistProcessing, "GPU", "Cmdlist Processing", MP_RGB(100, 255, 100));
-
-static void MemoryFill(const Regs::MemoryFillConfig& config) {
-    const PAddr start_addr = config.GetStartAddress();
-    const PAddr end_addr = config.GetEndAddress();
-
-    // TODO: do hwtest with these cases
-    if (!g_memory->IsValidPhysicalAddress(start_addr)) {
-        LOG_CRITICAL(HW_GPU, "invalid start address {:#010X}", start_addr);
-        return;
-    }
-
-    if (!g_memory->IsValidPhysicalAddress(end_addr)) {
-        LOG_CRITICAL(HW_GPU, "invalid end address {:#010X}", end_addr);
-        return;
-    }
-
-    if (end_addr <= start_addr) {
-        LOG_CRITICAL(HW_GPU, "invalid memory range from {:#010X} to {:#010X}", start_addr,
-                     end_addr);
-        return;
-    }
-
-    u8* start = g_memory->GetPhysicalPointer(start_addr);
-    u8* end = g_memory->GetPhysicalPointer(end_addr);
-
-    if (VideoCore::g_renderer->Rasterizer()->AccelerateFill(config))
-        return;
-
-    Memory::RasterizerInvalidateRegion(config.GetStartAddress(),
-                                       config.GetEndAddress() - config.GetStartAddress());
-
-    if (config.fill_24bit) {
-        // fill with 24-bit values
-        for (u8* ptr = start; ptr < end; ptr += 3) {
-            ptr[0] = config.value_24bit_r;
-            ptr[1] = config.value_24bit_g;
-            ptr[2] = config.value_24bit_b;
-        }
-    } else if (config.fill_32bit) {
-        // fill with 32-bit values
-        if (end > start) {
-            u32 value = config.value_32bit;
-            std::size_t len = (end - start) / sizeof(u32);
-            for (std::size_t i = 0; i < len; ++i)
-                std::memcpy(&start[i * sizeof(u32)], &value, sizeof(u32));
-        }
-    } else {
-        // fill with 16-bit values
-        u16 value_16bit = config.value_16bit.Value();
-        for (u8* ptr = start; ptr < end; ptr += sizeof(u16))
-            std::memcpy(ptr, &value_16bit, sizeof(u16));
-    }
-}
-
-static void DisplayTransfer(const Regs::DisplayTransferConfig& config) {
-    const PAddr src_addr = config.GetPhysicalInputAddress();
-    PAddr dst_addr = config.GetPhysicalOutputAddress();
-
-    // TODO: do hwtest with these cases
-    if (!g_memory->IsValidPhysicalAddress(src_addr)) {
-        LOG_CRITICAL(HW_GPU, "invalid input address {:#010X}", src_addr);
-        return;
-    }
-
-    if (!g_memory->IsValidPhysicalAddress(dst_addr)) {
-        LOG_CRITICAL(HW_GPU, "invalid output address {:#010X}", dst_addr);
-        return;
-    }
-
-    if (config.input_width == 0) {
-        LOG_CRITICAL(HW_GPU, "zero input width");
-        return;
-    }
-
-    if (config.input_height == 0) {
-        LOG_CRITICAL(HW_GPU, "zero input height");
-        return;
-    }
-
-    if (config.output_width == 0) {
-        LOG_CRITICAL(HW_GPU, "zero output width");
-        return;
-    }
-
-    if (config.output_height == 0) {
-        LOG_CRITICAL(HW_GPU, "zero output height");
-        return;
-    }
-
-    if (VideoCore::g_renderer->Rasterizer()->AccelerateDisplayTransfer(config))
-        return;
-
-    // Using flip_vertically alongside crop_input_lines produces skewed output on hardware.
-    // We have to emulate this because some games rely on this behaviour to render correctly.
-    if (config.flip_vertically && config.crop_input_lines &&
-        config.input_width > config.output_width) {
-        dst_addr += (config.input_width - config.output_width) * (config.output_height - 1) *
-                    GPU::Regs::BytesPerPixel(config.output_format);
-    }
-
-    u8* src_pointer = g_memory->GetPhysicalPointer(src_addr);
-    u8* dst_pointer = g_memory->GetPhysicalPointer(dst_addr);
-
-    if (config.scaling > config.ScaleXY) {
-        LOG_CRITICAL(HW_GPU, "Unimplemented display transfer scaling mode {}",
-                     config.scaling.Value());
-        UNIMPLEMENTED();
-        return;
-    }
-
-    if (config.input_linear && config.scaling != config.NoScale) {
-        LOG_CRITICAL(HW_GPU, "Scaling is only implemented on tiled input");
-        UNIMPLEMENTED();
-        return;
-    }
-
-    int horizontal_scale = config.scaling != config.NoScale ? 1 : 0;
-    int vertical_scale = config.scaling == config.ScaleXY ? 1 : 0;
-
-    u32 output_width = config.output_width >> horizontal_scale;
-    u32 output_height = config.output_height >> vertical_scale;
-
-    u32 input_size =
-        config.input_width * config.input_height * GPU::Regs::BytesPerPixel(config.input_format);
-    u32 output_size = output_width * output_height * GPU::Regs::BytesPerPixel(config.output_format);
-
-    Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(), input_size);
-    Memory::RasterizerInvalidateRegion(config.GetPhysicalOutputAddress(), output_size);
-
-    for (u32 y = 0; y < output_height; ++y) {
-        for (u32 x = 0; x < output_width; ++x) {
-            Common::Vec4<u8> src_color;
-
-            // Calculate the [x,y] position of the input image
-            // based on the current output position and the scale
-            u32 input_x = x << horizontal_scale;
-            u32 input_y = y << vertical_scale;
-
-            u32 output_y;
-            if (config.flip_vertically) {
-                // Flip the y value of the output data,
-                // we do this after calculating the [x,y] position of the input image
-                // to account for the scaling options.
-                output_y = output_height - y - 1;
-            } else {
-                output_y = y;
-            }
-
-            u32 dst_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.output_format);
-            u32 src_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.input_format);
-            u32 src_offset;
-            u32 dst_offset;
-
-            if (config.input_linear) {
-                if (!config.dont_swizzle) {
-                    // Interpret the input as linear and the output as tiled
-                    u32 coarse_y = output_y & ~7;
-                    u32 stride = output_width * dst_bytes_per_pixel;
-
-                    src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
-                    dst_offset = VideoCore::GetMortonOffset(x, output_y, dst_bytes_per_pixel) +
-                                 coarse_y * stride;
-                } else {
-                    // Both input and output are linear
-                    src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
-                    dst_offset = (x + output_y * output_width) * dst_bytes_per_pixel;
-                }
-            } else {
-                if (!config.dont_swizzle) {
-                    // Interpret the input as tiled and the output as linear
-                    u32 coarse_y = input_y & ~7;
-                    u32 stride = config.input_width * src_bytes_per_pixel;
-
-                    src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) +
-                                 coarse_y * stride;
-                    dst_offset = (x + output_y * output_width) * dst_bytes_per_pixel;
-                } else {
-                    // Both input and output are tiled
-                    u32 out_coarse_y = output_y & ~7;
-                    u32 out_stride = output_width * dst_bytes_per_pixel;
-
-                    u32 in_coarse_y = input_y & ~7;
-                    u32 in_stride = config.input_width * src_bytes_per_pixel;
-
-                    src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) +
-                                 in_coarse_y * in_stride;
-                    dst_offset = VideoCore::GetMortonOffset(x, output_y, dst_bytes_per_pixel) +
-                                 out_coarse_y * out_stride;
-                }
-            }
-
-            const u8* src_pixel = src_pointer + src_offset;
-            src_color = DecodePixel(config.input_format, src_pixel);
-            if (config.scaling == config.ScaleX) {
-                Common::Vec4<u8> pixel =
-                    DecodePixel(config.input_format, src_pixel + src_bytes_per_pixel);
-                src_color = ((src_color + pixel) / 2).Cast<u8>();
-            } else if (config.scaling == config.ScaleXY) {
-                Common::Vec4<u8> pixel1 =
-                    DecodePixel(config.input_format, src_pixel + 1 * src_bytes_per_pixel);
-                Common::Vec4<u8> pixel2 =
-                    DecodePixel(config.input_format, src_pixel + 2 * src_bytes_per_pixel);
-                Common::Vec4<u8> pixel3 =
-                    DecodePixel(config.input_format, src_pixel + 3 * src_bytes_per_pixel);
-                src_color = (((src_color + pixel1) + (pixel2 + pixel3)) / 4).Cast<u8>();
-            }
-
-            u8* dst_pixel = dst_pointer + dst_offset;
-            switch (config.output_format) {
-            case Regs::PixelFormat::RGBA8:
-                Common::Color::EncodeRGBA8(src_color, dst_pixel);
-                break;
-
-            case Regs::PixelFormat::RGB8:
-                Common::Color::EncodeRGB8(src_color, dst_pixel);
-                break;
-
-            case Regs::PixelFormat::RGB565:
-                Common::Color::EncodeRGB565(src_color, dst_pixel);
-                break;
-
-            case Regs::PixelFormat::RGB5A1:
-                Common::Color::EncodeRGB5A1(src_color, dst_pixel);
-                break;
-
-            case Regs::PixelFormat::RGBA4:
-                Common::Color::EncodeRGBA4(src_color, dst_pixel);
-                break;
-
-            default:
-                LOG_ERROR(HW_GPU, "Unknown destination framebuffer format {:x}",
-                          static_cast<u32>(config.output_format.Value()));
-                break;
-            }
-        }
-    }
-}
-
-static void TextureCopy(const Regs::DisplayTransferConfig& config) {
-    const PAddr src_addr = config.GetPhysicalInputAddress();
-    const PAddr dst_addr = config.GetPhysicalOutputAddress();
-
-    // TODO: do hwtest with invalid addresses
-    if (!g_memory->IsValidPhysicalAddress(src_addr)) {
-        LOG_CRITICAL(HW_GPU, "invalid input address {:#010X}", src_addr);
-        return;
-    }
-
-    if (!g_memory->IsValidPhysicalAddress(dst_addr)) {
-        LOG_CRITICAL(HW_GPU, "invalid output address {:#010X}", dst_addr);
-        return;
-    }
-
-    if (VideoCore::g_renderer->Rasterizer()->AccelerateTextureCopy(config))
-        return;
-
-    u8* src_pointer = g_memory->GetPhysicalPointer(src_addr);
-    u8* dst_pointer = g_memory->GetPhysicalPointer(dst_addr);
-
-    u32 remaining_size = Common::AlignDown(config.texture_copy.size, 16);
-
-    if (remaining_size == 0) {
-        LOG_CRITICAL(HW_GPU, "zero size. Real hardware freezes on this.");
-        return;
-    }
-
-    u32 input_gap = config.texture_copy.input_gap * 16;
-    u32 output_gap = config.texture_copy.output_gap * 16;
-
-    // Zero gap means contiguous input/output even if width = 0. To avoid infinite loop below, width
-    // is assigned with the total size if gap = 0.
-    u32 input_width = input_gap == 0 ? remaining_size : config.texture_copy.input_width * 16;
-    u32 output_width = output_gap == 0 ? remaining_size : config.texture_copy.output_width * 16;
-
-    if (input_width == 0) {
-        LOG_CRITICAL(HW_GPU, "zero input width. Real hardware freezes on this.");
-        return;
-    }
-
-    if (output_width == 0) {
-        LOG_CRITICAL(HW_GPU, "zero output width. Real hardware freezes on this.");
-        return;
-    }
-
-    std::size_t contiguous_input_size =
-        config.texture_copy.size / input_width * (input_width + input_gap);
-    Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(),
-                                  static_cast<u32>(contiguous_input_size));
-
-    std::size_t contiguous_output_size =
-        config.texture_copy.size / output_width * (output_width + output_gap);
-    // Only need to flush output if it has a gap
-    const auto FlushInvalidate_fn = (output_gap != 0) ? Memory::RasterizerFlushAndInvalidateRegion
-                                                      : Memory::RasterizerInvalidateRegion;
-    FlushInvalidate_fn(config.GetPhysicalOutputAddress(), static_cast<u32>(contiguous_output_size));
-
-    u32 remaining_input = input_width;
-    u32 remaining_output = output_width;
-    while (remaining_size > 0) {
-        u32 copy_size = std::min({remaining_input, remaining_output, remaining_size});
-
-        std::memcpy(dst_pointer, src_pointer, copy_size);
-        src_pointer += copy_size;
-        dst_pointer += copy_size;
-
-        remaining_input -= copy_size;
-        remaining_output -= copy_size;
-        remaining_size -= copy_size;
-
-        if (remaining_input == 0) {
-            remaining_input = input_width;
-            src_pointer += input_gap;
-        }
-        if (remaining_output == 0) {
-            remaining_output = output_width;
-            dst_pointer += output_gap;
-        }
-    }
-}
-
-template <typename T>
-inline void Write(u32 addr, const T data) {
-    addr -= HW::VADDR_GPU;
-    u32 index = addr / 4;
-
-    // Writes other than u32 are untested, so I'd rather have them abort than silently fail
-    if (index >= Regs::NumIds() || !std::is_same<T, u32>::value) {
-        LOG_ERROR(HW_GPU, "unknown Write{} {:#010X} @ {:#010X}", sizeof(data) * 8, (u32)data, addr);
-        return;
-    }
-
-    g_regs[index] = static_cast<u32>(data);
-
-    switch (index) {
-
-    // Memory fills are triggered once the fill value is written.
-    case GPU_REG_INDEX(memory_fill_config[0].trigger):
-    case GPU_REG_INDEX(memory_fill_config[1].trigger): {
-        const bool is_second_filler = (index != GPU_REG_INDEX(memory_fill_config[0].trigger));
-        auto& config = g_regs.memory_fill_config[is_second_filler];
-
-        if (config.trigger) {
-            MemoryFill(config);
-            LOG_TRACE(HW_GPU, "MemoryFill from {:#010X} to {:#010X}", config.GetStartAddress(),
-                      config.GetEndAddress());
-
-            // It seems that it won't signal interrupt if "address_start" is zero.
-            // TODO: hwtest this
-            if (config.GetStartAddress() != 0) {
-                if (!is_second_filler) {
-                    Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PSC0);
-                } else {
-                    Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PSC1);
-                }
-            }
-
-            // Reset "trigger" flag and set the "finish" flag
-            // NOTE: This was confirmed to happen on hardware even if "address_start" is zero.
-            config.trigger.Assign(0);
-            config.finished.Assign(1);
-        }
-        break;
-    }
-
-    case GPU_REG_INDEX(display_transfer_config.trigger): {
-        MICROPROFILE_SCOPE(GPU_DisplayTransfer);
-
-        const auto& config = g_regs.display_transfer_config;
-        if (config.trigger & 1) {
-
-            if (Pica::g_debug_context)
-                Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::IncomingDisplayTransfer,
-                                               nullptr);
-
-            if (config.is_texture_copy) {
-                TextureCopy(config);
-                LOG_TRACE(HW_GPU,
-                          "TextureCopy: {:#X} bytes from {:#010X}({}+{})-> "
-                          "{:#010X}({}+{}), flags {:#010X}",
-                          config.texture_copy.size, config.GetPhysicalInputAddress(),
-                          config.texture_copy.input_width * 16, config.texture_copy.input_gap * 16,
-                          config.GetPhysicalOutputAddress(), config.texture_copy.output_width * 16,
-                          config.texture_copy.output_gap * 16, config.flags);
-            } else {
-                DisplayTransfer(config);
-                LOG_TRACE(HW_GPU,
-                          "DisplayTransfer: {:#010X}({}x{})-> "
-                          "{:#010X}({}x{}), dst format {:x}, flags {:#010X}",
-                          config.GetPhysicalInputAddress(), config.input_width.Value(),
-                          config.input_height.Value(), config.GetPhysicalOutputAddress(),
-                          config.output_width.Value(), config.output_height.Value(),
-                          static_cast<u32>(config.output_format.Value()), config.flags);
-            }
-
-            g_regs.display_transfer_config.trigger = 0;
-            Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PPF);
-        }
-        break;
-    }
-
-    // Seems like writing to this register triggers processing
-    case GPU_REG_INDEX(command_processor_config.trigger): {
-        const auto& config = g_regs.command_processor_config;
-        if (config.trigger & 1) {
-            MICROPROFILE_SCOPE(GPU_CmdlistProcessing);
-
-            Pica::CommandProcessor::ProcessCommandList(config.GetPhysicalAddress(), config.size);
-
-            g_regs.command_processor_config.trigger = 0;
-        }
-        break;
-    }
-
-    default:
-        break;
-    }
-
-    // Notify tracer about the register write
-    // This is happening *after* handling the write to make sure we properly catch all memory reads.
-    if (Pica::g_debug_context && Pica::g_debug_context->recorder) {
-        // addr + GPU VBase - IO VBase + IO PBase
-        Pica::g_debug_context->recorder->RegisterWritten<T>(
-            addr + 0x1EF00000 - 0x1EC00000 + 0x10100000, data);
-    }
-}
-
-// Explicitly instantiate template functions because we aren't defining this in the header:
-
-template void Read<u64>(u64& var, const u32 addr);
-template void Read<u32>(u32& var, const u32 addr);
-template void Read<u16>(u16& var, const u32 addr);
-template void Read<u8>(u8& var, const u32 addr);
-
-template void Write<u64>(u32 addr, const u64 data);
-template void Write<u32>(u32 addr, const u32 data);
-template void Write<u16>(u32 addr, const u16 data);
-template void Write<u8>(u32 addr, const u8 data);
-
-/// Update hardware
-static void VBlankCallback(std::uintptr_t user_data, s64 cycles_late) {
-    VideoCore::g_renderer->SwapBuffers();
-
-    // Signal to GSP that GPU interrupt has occurred
-    // TODO(yuriks): hwtest to determine if PDC0 is for the Top screen and PDC1 for the Sub
-    // screen, or if both use the same interrupts and these two instead determine the
-    // beginning and end of the VBlank period. If needed, split the interrupt firing into
-    // two different intervals.
-    Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PDC0);
-    Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PDC1);
-
-    // Reschedule recurrent event
-    Core::System::GetInstance().CoreTiming().ScheduleEvent(frame_ticks - cycles_late, vblank_event);
-}
-
-/// Initialize hardware
-void Init(Memory::MemorySystem& memory) {
-    g_memory = &memory;
-    std::memset(&g_regs, 0, sizeof(g_regs));
-
-    auto& framebuffer_top = g_regs.framebuffer_config[0];
-    auto& framebuffer_sub = g_regs.framebuffer_config[1];
-
-    // Setup default framebuffer addresses (located in VRAM)
-    // .. or at least these are the ones used by system applets.
-    // There's probably a smarter way to come up with addresses
-    // like this which does not require hardcoding.
-    framebuffer_top.address_left1 = 0x181E6000;
-    framebuffer_top.address_left2 = 0x1822C800;
-    framebuffer_top.address_right1 = 0x18273000;
-    framebuffer_top.address_right2 = 0x182B9800;
-    framebuffer_sub.address_left1 = 0x1848F000;
-    framebuffer_sub.address_left2 = 0x184C7800;
-
-    framebuffer_top.width.Assign(240);
-    framebuffer_top.height.Assign(400);
-    framebuffer_top.stride = 3 * 240;
-    framebuffer_top.color_format.Assign(Regs::PixelFormat::RGB8);
-    framebuffer_top.active_fb = 0;
-
-    framebuffer_sub.width.Assign(240);
-    framebuffer_sub.height.Assign(320);
-    framebuffer_sub.stride = 3 * 240;
-    framebuffer_sub.color_format.Assign(Regs::PixelFormat::RGB8);
-    framebuffer_sub.active_fb = 0;
-
-    Core::Timing& timing = Core::System::GetInstance().CoreTiming();
-    vblank_event = timing.RegisterEvent("GPU::VBlankCallback", VBlankCallback);
-    timing.ScheduleEvent(frame_ticks, vblank_event);
-
-    LOG_DEBUG(HW_GPU, "initialized OK");
-}
-
-/// Shutdown hardware
-void Shutdown() {
-    LOG_DEBUG(HW_GPU, "shutdown OK");
-}
-
-} // namespace GPU

src/core/hw/gpu.h

@@ -1,344 +0,0 @@
-// Copyright 2014 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <cstddef>
-#include <type_traits>
-#include <boost/serialization/access.hpp>
-#include <boost/serialization/binary_object.hpp>
-#include "common/assert.h"
-#include "common/bit_field.h"
-#include "common/common_funcs.h"
-#include "common/common_types.h"
-#include "core/core_timing.h"
-
-namespace Memory {
-class MemorySystem;
-}
-
-namespace GPU {
-
-// Measured on hardware to be 2240568 timer cycles or 4481136 ARM11 cycles
-constexpr u64 frame_ticks = 4481136ull;
-
-// Refresh rate defined by ratio of ARM11 frequency to ARM11 ticks per frame
-// (268,111,856) / (4,481,136) = 59.83122493939037Hz
-constexpr double SCREEN_REFRESH_RATE = BASE_CLOCK_RATE_ARM11 / static_cast<double>(frame_ticks);
-
-// Returns index corresponding to the Regs member labeled by field_name
-#define GPU_REG_INDEX(field_name) (offsetof(GPU::Regs, field_name) / sizeof(u32))
-
-// Returns index corresponding to the Regs::FramebufferConfig labeled by field_name
-// screen_id is a subscript for Regs::framebuffer_config
-#define GPU_FRAMEBUFFER_REG_INDEX(screen_id, field_name)                                           \
-    ((offsetof(GPU::Regs, framebuffer_config) +                                                    \
-      sizeof(GPU::Regs::FramebufferConfig) * (screen_id) +                                         \
-      offsetof(GPU::Regs::FramebufferConfig, field_name)) /                                        \
-     sizeof(u32))
-
-// MMIO region 0x1EFxxxxx
-struct Regs {
-
-// helper macro to make sure the defined structures are of the expected size.
-#define ASSERT_MEMBER_SIZE(name, size_in_bytes)                                                    \
-    static_assert(sizeof(name) == size_in_bytes,                                                   \
-                  "Structure size and register block length don't match")
-
-    // Components are laid out in reverse byte order, most significant bits first.
-    enum class PixelFormat : u32 {
-        RGBA8 = 0,
-        RGB8 = 1,
-        RGB565 = 2,
-        RGB5A1 = 3,
-        RGBA4 = 4,
-    };
-
-    /**
-     * Returns the number of bytes per pixel.
-     */
-    static int BytesPerPixel(PixelFormat format) {
-        switch (format) {
-        case PixelFormat::RGBA8:
-            return 4;
-        case PixelFormat::RGB8:
-            return 3;
-        case PixelFormat::RGB565:
-        case PixelFormat::RGB5A1:
-        case PixelFormat::RGBA4:
-            return 2;
-        default:
-            UNREACHABLE();
-        }
-
-        return 0;
-    }
-
-    INSERT_PADDING_WORDS(0x4);
-
-    struct MemoryFillConfig {
-        u32 address_start;
-        u32 address_end;
-
-        union {
-            u32 value_32bit;
-
-            BitField<0, 16, u32> value_16bit;
-
-            // TODO: Verify component order
-            BitField<0, 8, u32> value_24bit_r;
-            BitField<8, 8, u32> value_24bit_g;
-            BitField<16, 8, u32> value_24bit_b;
-        };
-
-        union {
-            u32 control;
-
-            // Setting this field to 1 triggers the memory fill.
-            // This field also acts as a status flag, and gets reset to 0 upon completion.
-            BitField<0, 1, u32> trigger;
-
-            // Set to 1 upon completion.
-            BitField<1, 1, u32> finished;
-
-            // If both of these bits are unset, then it will fill the memory with a 16 bit value
-            // 1: fill with 24-bit wide values
-            BitField<8, 1, u32> fill_24bit;
-            // 1: fill with 32-bit wide values
-            BitField<9, 1, u32> fill_32bit;
-        };
-
-        inline u32 GetStartAddress() const {
-            return DecodeAddressRegister(address_start);
-        }
-
-        inline u32 GetEndAddress() const {
-            return DecodeAddressRegister(address_end);
-        }
-
-        inline std::string DebugName() const {
-            return fmt::format("from {:#X} to {:#X} with {}-bit value {:#X}", GetStartAddress(),
-                               GetEndAddress(), fill_32bit ? "32" : (fill_24bit ? "24" : "16"),
-                               value_32bit);
-        }
-    } memory_fill_config[2];
-    ASSERT_MEMBER_SIZE(memory_fill_config[0], 0x10);
-
-    INSERT_PADDING_WORDS(0x10b);
-
-    struct FramebufferConfig {
-        union {
-            u32 size;
-
-            BitField<0, 16, u32> width;
-            BitField<16, 16, u32> height;
-        };
-
-        INSERT_PADDING_WORDS(0x2);
-
-        u32 address_left1;
-        u32 address_left2;
-
-        union {
-            u32 format;
-
-            BitField<0, 3, PixelFormat> color_format;
-        };
-
-        INSERT_PADDING_WORDS(0x1);
-
-        union {
-            u32 active_fb;
-
-            // 0: Use parameters ending with "1"
-            // 1: Use parameters ending with "2"
-            BitField<0, 1, u32> second_fb_active;
-        };
-
-        INSERT_PADDING_WORDS(0x5);
-
-        // Distance between two pixel rows, in bytes
-        u32 stride;
-
-        u32 address_right1;
-        u32 address_right2;
-
-        INSERT_PADDING_WORDS(0x30);
-    } framebuffer_config[2];
-    ASSERT_MEMBER_SIZE(framebuffer_config[0], 0x100);
-
-    INSERT_PADDING_WORDS(0x169);
-
-    struct DisplayTransferConfig {
-        u32 input_address;
-        u32 output_address;
-
-        inline u32 GetPhysicalInputAddress() const {
-            return DecodeAddressRegister(input_address);
-        }
-
-        inline u32 GetPhysicalOutputAddress() const {
-            return DecodeAddressRegister(output_address);
-        }
-
-        inline std::string DebugName() const noexcept {
-            return fmt::format("from {:#x} to {:#x} with {} scaling and stride {}, width {}",
-                               GetPhysicalInputAddress(), GetPhysicalOutputAddress(),
-                               scaling == NoScale ? "no" : (scaling == ScaleX ? "X" : "XY"),
-                               input_width.Value(), output_width.Value());
-        }
-
-        union {
-            u32 output_size;
-
-            BitField<0, 16, u32> output_width;
-            BitField<16, 16, u32> output_height;
-        };
-
-        union {
-            u32 input_size;
-
-            BitField<0, 16, u32> input_width;
-            BitField<16, 16, u32> input_height;
-        };
-
-        enum ScalingMode : u32 {
-            NoScale = 0, // Doesn't scale the image
-            ScaleX = 1,  // Downscales the image in half in the X axis and applies a box filter
-            ScaleXY =
-                2, // Downscales the image in half in both the X and Y axes and applies a box filter
-        };
-
-        union {
-            u32 flags;
-
-            BitField<0, 1, u32> flip_vertically; // flips input data vertically
-            BitField<1, 1, u32> input_linear;    // Converts from linear to tiled format
-            BitField<2, 1, u32> crop_input_lines;
-            BitField<3, 1, u32> is_texture_copy; // Copies the data without performing any
-                                                 // processing and respecting texture copy fields
-            BitField<5, 1, u32> dont_swizzle;
-            BitField<8, 3, PixelFormat> input_format;
-            BitField<12, 3, PixelFormat> output_format;
-            /// Uses some kind of 32x32 block swizzling mode, instead of the usual 8x8 one.
-            BitField<16, 1, u32> block_32;        // TODO(yuriks): unimplemented
-            BitField<24, 2, ScalingMode> scaling; // Determines the scaling mode of the transfer
-        };
-
-        INSERT_PADDING_WORDS(0x1);
-
-        // it seems that writing to this field triggers the display transfer
-        u32 trigger;
-
-        INSERT_PADDING_WORDS(0x1);
-
-        struct {
-            u32 size; // The lower 4 bits are ignored
-
-            union {
-                u32 input_size;
-
-                BitField<0, 16, u32> input_width;
-                BitField<16, 16, u32> input_gap;
-            };
-
-            union {
-                u32 output_size;
-
-                BitField<0, 16, u32> output_width;
-                BitField<16, 16, u32> output_gap;
-            };
-        } texture_copy;
-    } display_transfer_config;
-    ASSERT_MEMBER_SIZE(display_transfer_config, 0x2c);
-
-    INSERT_PADDING_WORDS(0x32D);
-
-    struct {
-        // command list size (in bytes)
-        u32 size;
-
-        INSERT_PADDING_WORDS(0x1);
-
-        // command list address
-        u32 address;
-
-        INSERT_PADDING_WORDS(0x1);
-
-        // it seems that writing to this field triggers command list processing
-        u32 trigger;
-
-        inline u32 GetPhysicalAddress() const {
-            return DecodeAddressRegister(address);
-        }
-    } command_processor_config;
-    ASSERT_MEMBER_SIZE(command_processor_config, 0x14);
-
-    INSERT_PADDING_WORDS(0x9c3);
-
-    static constexpr std::size_t NumIds() {
-        return sizeof(Regs) / sizeof(u32);
-    }
-
-    const u32& operator[](int index) const {
-        const u32* content = reinterpret_cast<const u32*>(this);
-        return content[index];
-    }
-
-    u32& operator[](int index) {
-        u32* content = reinterpret_cast<u32*>(this);
-        return content[index];
-    }
-
-#undef ASSERT_MEMBER_SIZE
-
-private:
-    /*
-     * Most physical addresses which GPU registers refer to are 8-byte aligned.
-     * This function should be used to get the address from a raw register value.
-     */
-    static inline u32 DecodeAddressRegister(u32 register_value) {
-        return register_value * 8;
-    }
-
-    template <class Archive>
-    void serialize(Archive& ar, const unsigned int) {
-        ar& boost::serialization::make_binary_object(this, sizeof(Regs));
-    }
-    friend class boost::serialization::access;
-};
-static_assert(std::is_standard_layout<Regs>::value, "Structure does not use standard layout");
-
-#define ASSERT_REG_POSITION(field_name, position)                                                  \
-    static_assert(offsetof(Regs, field_name) == position * 4,                                      \
-                  "Field " #field_name " has invalid position")
-
-ASSERT_REG_POSITION(memory_fill_config[0], 0x00004);
-ASSERT_REG_POSITION(memory_fill_config[1], 0x00008);
-ASSERT_REG_POSITION(framebuffer_config[0], 0x00117);
-ASSERT_REG_POSITION(framebuffer_config[1], 0x00157);
-ASSERT_REG_POSITION(display_transfer_config, 0x00300);
-ASSERT_REG_POSITION(command_processor_config, 0x00638);
-
-#undef ASSERT_REG_POSITION
-
-// The total number of registers is chosen arbitrarily, but let's make sure it's not some odd value
-// anyway.
-static_assert(sizeof(Regs) == 0x1000 * sizeof(u32), "Invalid total size of register set");
-
-extern Regs g_regs;
-
-template <typename T>
-void Read(T& var, const u32 addr);
-
-template <typename T>
-void Write(u32 addr, const T data);
-
-/// Initialize hardware
-void Init(Memory::MemorySystem& memory);
-
-/// Shutdown hardware
-void Shutdown();
-
-} // namespace GPU

src/core/hw/hw.cpp

@@ -1,102 +0,0 @@
-// Copyright 2014 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include "common/common_types.h"
-#include "common/logging/log.h"
-#include "core/hw/aes/key.h"
-#include "core/hw/gpu.h"
-#include "core/hw/hw.h"
-#include "core/hw/lcd.h"
-
-namespace HW {
-
-template <typename T>
-inline void Read(T& var, const u32 addr) {
-    switch (addr & 0xFFFFF000) {
-    case VADDR_GPU:
-    case VADDR_GPU + 0x1000:
-    case VADDR_GPU + 0x2000:
-    case VADDR_GPU + 0x3000:
-    case VADDR_GPU + 0x4000:
-    case VADDR_GPU + 0x5000:
-    case VADDR_GPU + 0x6000:
-    case VADDR_GPU + 0x7000:
-    case VADDR_GPU + 0x8000:
-    case VADDR_GPU + 0x9000:
-    case VADDR_GPU + 0xA000:
-    case VADDR_GPU + 0xB000:
-    case VADDR_GPU + 0xC000:
-    case VADDR_GPU + 0xD000:
-    case VADDR_GPU + 0xE000:
-    case VADDR_GPU + 0xF000:
-        GPU::Read(var, addr);
-        break;
-    case VADDR_LCD:
-        LCD::Read(var, addr);
-        break;
-    default:
-        LOG_ERROR(HW_Memory, "unknown Read{} @ {:#010X}", sizeof(var) * 8, addr);
-    }
-}
-
-template <typename T>
-inline void Write(u32 addr, const T data) {
-    switch (addr & 0xFFFFF000) {
-    case VADDR_GPU:
-    case VADDR_GPU + 0x1000:
-    case VADDR_GPU + 0x2000:
-    case VADDR_GPU + 0x3000:
-    case VADDR_GPU + 0x4000:
-    case VADDR_GPU + 0x5000:
-    case VADDR_GPU + 0x6000:
-    case VADDR_GPU + 0x7000:
-    case VADDR_GPU + 0x8000:
-    case VADDR_GPU + 0x9000:
-    case VADDR_GPU + 0xA000:
-    case VADDR_GPU + 0xB000:
-    case VADDR_GPU + 0xC000:
-    case VADDR_GPU + 0xD000:
-    case VADDR_GPU + 0xE000:
-    case VADDR_GPU + 0xF000:
-        GPU::Write(addr, data);
-        break;
-    case VADDR_LCD:
-        LCD::Write(addr, data);
-        break;
-    default:
-        LOG_ERROR(HW_Memory, "unknown Write{} {:#010X} @ {:#010X}", sizeof(data) * 8, (u32)data,
-                  addr);
-    }
-}
-
-// Explicitly instantiate template functions because we aren't defining this in the header:
-
-template void Read<u64>(u64& var, const u32 addr);
-template void Read<u32>(u32& var, const u32 addr);
-template void Read<u16>(u16& var, const u32 addr);
-template void Read<u8>(u8& var, const u32 addr);
-
-template void Write<u64>(u32 addr, const u64 data);
-template void Write<u32>(u32 addr, const u32 data);
-template void Write<u16>(u32 addr, const u16 data);
-template void Write<u8>(u32 addr, const u8 data);
-
-/// Update hardware
-void Update() {}
-
-/// Initialize hardware
-void Init(Memory::MemorySystem& memory) {
-    AES::InitKeys();
-    GPU::Init(memory);
-    LCD::Init();
-    LOG_DEBUG(HW, "initialized OK");
-}
-
-/// Shutdown hardware
-void Shutdown() {
-    GPU::Shutdown();
-    LCD::Shutdown();
-    LOG_DEBUG(HW, "shutdown OK");
-}
-} // namespace HW

src/core/hw/hw.h

@@ -1,54 +0,0 @@
-// Copyright 2014 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include "common/common_types.h"
-
-namespace Memory {
-class MemorySystem;
-}
-
-namespace HW {
-
-/// Beginnings of IO register regions, in the user VA space.
-enum : u32 {
-    VADDR_HASH = 0x1EC01000,
-    VADDR_CSND = 0x1EC03000,
-    VADDR_DSP = 0x1EC40000,
-    VADDR_PDN = 0x1EC41000,
-    VADDR_CODEC = 0x1EC41000,
-    VADDR_SPI = 0x1EC42000,
-    VADDR_SPI_2 = 0x1EC43000, // Only used under TWL_FIRM?
-    VADDR_I2C = 0x1EC44000,
-    VADDR_CODEC_2 = 0x1EC45000,
-    VADDR_HID = 0x1EC46000,
-    VADDR_GPIO = 0x1EC47000,
-    VADDR_I2C_2 = 0x1EC48000,
-    VADDR_SPI_3 = 0x1EC60000,
-    VADDR_I2C_3 = 0x1EC61000,
-    VADDR_MIC = 0x1EC62000,
-    VADDR_PXI = 0x1EC63000,
-    VADDR_LCD = 0x1ED02000,
-    VADDR_DSP_2 = 0x1ED03000,
-    VADDR_HASH_2 = 0x1EE01000,
-    VADDR_GPU = 0x1EF00000,
-};
-
-template <typename T>
-void Read(T& var, const u32 addr);
-
-template <typename T>
-void Write(u32 addr, const T data);
-
-/// Update hardware
-void Update();
-
-/// Initialize hardware
-void Init(Memory::MemorySystem& memory);
-
-/// Shutdown hardware
-void Shutdown();
-
-} // namespace HW

src/core/hw/lcd.cpp

@@ -1,76 +0,0 @@
-// Copyright 2015 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <cstring>
-#include "common/common_types.h"
-#include "common/logging/log.h"
-#include "core/hw/hw.h"
-#include "core/hw/lcd.h"
-#include "core/tracer/recorder.h"
-#include "video_core/debug_utils/debug_utils.h"
-
-namespace LCD {
-
-Regs g_regs;
-
-template <typename T>
-inline void Read(T& var, const u32 raw_addr) {
-    u32 addr = raw_addr - HW::VADDR_LCD;
-    u32 index = addr / 4;
-
-    // Reads other than u32 are untested, so I'd rather have them abort than silently fail
-    if (index >= 0x400 || !std::is_same<T, u32>::value) {
-        LOG_ERROR(HW_LCD, "unknown Read{} @ {:#010X}", sizeof(var) * 8, addr);
-        return;
-    }
-
-    var = g_regs[index];
-}
-
-template <typename T>
-inline void Write(u32 addr, const T data) {
-    addr -= HW::VADDR_LCD;
-    u32 index = addr / 4;
-
-    // Writes other than u32 are untested, so I'd rather have them abort than silently fail
-    if (index >= 0x400 || !std::is_same<T, u32>::value) {
-        LOG_ERROR(HW_LCD, "unknown Write{} {:#010X} @ {:#010X}", sizeof(data) * 8, (u32)data, addr);
-        return;
-    }
-
-    g_regs[index] = static_cast<u32>(data);
-
-    // Notify tracer about the register write
-    // This is happening *after* handling the write to make sure we properly catch all memory reads.
-    if (Pica::g_debug_context && Pica::g_debug_context->recorder) {
-        // addr + GPU VBase - IO VBase + IO PBase
-        Pica::g_debug_context->recorder->RegisterWritten<T>(
-            addr + HW::VADDR_LCD - 0x1EC00000 + 0x10100000, data);
-    }
-}
-
-// Explicitly instantiate template functions because we aren't defining this in the header:
-
-template void Read<u64>(u64& var, const u32 addr);
-template void Read<u32>(u32& var, const u32 addr);
-template void Read<u16>(u16& var, const u32 addr);
-template void Read<u8>(u8& var, const u32 addr);
-
-template void Write<u64>(u32 addr, const u64 data);
-template void Write<u32>(u32 addr, const u32 data);
-template void Write<u16>(u32 addr, const u16 data);
-template void Write<u8>(u32 addr, const u8 data);
-
-/// Initialize hardware
-void Init() {
-    std::memset(&g_regs, 0, sizeof(g_regs));
-    LOG_DEBUG(HW_LCD, "initialized OK");
-}
-
-/// Shutdown hardware
-void Shutdown() {
-    LOG_DEBUG(HW_LCD, "shutdown OK");
-}
-
-} // namespace LCD

src/core/hw/y2r.cpp

@@ -9,7 +9,7 @@
 #include "common/assert.h"
 #include "common/color.h"
 #include "common/common_types.h"
-#include "common/microprofileui.h"
+#include "common/microprofile.h"
 #include "common/vector_math.h"
 #include "core/core.h"
 #include "core/hle/service/cam/y2r_u.h"

src/core/memory.cpp

@@ -19,10 +19,9 @@
 #include "core/global.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/service/plgldr/plgldr.h"
-#include "core/hw/hw.h"
 #include "core/memory.h"
+#include "video_core/gpu.h"
 #include "video_core/renderer_base.h"
-#include "video_core/video_core.h"
 
 SERIALIZE_EXPORT_IMPL(Memory::MemorySystem::BackingMemImpl<Memory::Region::FCRAM>)
 SERIALIZE_EXPORT_IMPL(Memory::MemorySystem::BackingMemImpl<Memory::Region::VRAM>)
@@ -346,13 +345,52 @@ std::shared_ptr<PageTable> MemorySystem::GetCurrentPageTable() const {
     return impl->current_page_table;
 }
 
+void RasterizerFlushVirtualRegion(VAddr start, u32 size, FlushMode mode) {
+    const VAddr end = start + size;
+
+    auto CheckRegion = [&](VAddr region_start, VAddr region_end, PAddr paddr_region_start) {
+        if (start >= region_end || end <= region_start) {
+            // No overlap with region
+            return;
+        }
+
+        auto& renderer = Core::System::GetInstance().GPU().Renderer();
+        VAddr overlap_start = std::max(start, region_start);
+        VAddr overlap_end = std::min(end, region_end);
+        PAddr physical_start = paddr_region_start + (overlap_start - region_start);
+        u32 overlap_size = overlap_end - overlap_start;
+
+        auto* rasterizer = renderer.Rasterizer();
+        switch (mode) {
+        case FlushMode::Flush:
+            rasterizer->FlushRegion(physical_start, overlap_size);
+            break;
+        case FlushMode::Invalidate:
+            rasterizer->InvalidateRegion(physical_start, overlap_size);
+            break;
+        case FlushMode::FlushAndInvalidate:
+            rasterizer->FlushAndInvalidateRegion(physical_start, overlap_size);
+            break;
+        }
+    };
+
+    CheckRegion(LINEAR_HEAP_VADDR, LINEAR_HEAP_VADDR_END, FCRAM_PADDR);
+    CheckRegion(NEW_LINEAR_HEAP_VADDR, NEW_LINEAR_HEAP_VADDR_END, FCRAM_PADDR);
+    CheckRegion(VRAM_VADDR, VRAM_VADDR_END, VRAM_PADDR);
+    if (Service::PLGLDR::PLG_LDR::GetPluginFBAddr())
+        CheckRegion(PLUGIN_3GX_FB_VADDR, PLUGIN_3GX_FB_VADDR_END,
+                    Service::PLGLDR::PLG_LDR::GetPluginFBAddr());
+}
+
 void MemorySystem::MapPages(PageTable& page_table, u32 base, u32 size, MemoryRef memory,
                             PageType type) {
     LOG_DEBUG(HW_Memory, "Mapping {} onto {:08X}-{:08X}", (void*)memory.GetPtr(),
               base * CITRA_PAGE_SIZE, (base + size) * CITRA_PAGE_SIZE);
 
-    RasterizerFlushVirtualRegion(base << CITRA_PAGE_BITS, size * CITRA_PAGE_SIZE,
+    if (impl->system.IsPoweredOn()) {
-                                 FlushMode::FlushAndInvalidate);
+        RasterizerFlushVirtualRegion(base << CITRA_PAGE_BITS, size * CITRA_PAGE_SIZE,
+                                     FlushMode::FlushAndInvalidate);
+    }
 
     u32 end = base + size;
     while (base != end) {
@@ -421,9 +459,8 @@ T MemorySystem::Read(const VAddr vaddr) {
             return value;
         } else if ((paddr & 0xF0000000) == 0x10000000 &&
                    paddr >= Memory::IO_AREA_PADDR) { // Check MMIO region
-            T ret;
+            return impl->system.GPU().ReadReg(static_cast<VAddr>(paddr) - Memory::IO_AREA_PADDR +
-            HW::Read<T>(ret, static_cast<VAddr>(paddr) - Memory::IO_AREA_PADDR + 0x1EC00000);
+                                              0x1EC00000);
-            return ret;
         }
     }
 
@@ -468,7 +505,10 @@ void MemorySystem::Write(const VAddr vaddr, const T data) {
             return;
         } else if ((paddr & 0xF0000000) == 0x10000000 &&
                    paddr >= Memory::IO_AREA_PADDR) { // Check MMIO region
-            HW::Write<T>(static_cast<VAddr>(paddr) - Memory::IO_AREA_PADDR + 0x1EC00000, data);
+            ASSERT(sizeof(data) == sizeof(u32));
+            impl->system.GPU().WriteReg(static_cast<VAddr>(paddr) - Memory::IO_AREA_PADDR +
+                                            0x1EC00000,
+                                        static_cast<u32>(data));
             return;
         }
     }
@@ -713,84 +753,6 @@ void MemorySystem::RasterizerMarkRegionCached(PAddr start, u32 size, bool cached
     }
 }
 
-void RasterizerFlushRegion(PAddr start, u32 size) {
-    if (VideoCore::g_renderer == nullptr) {
-        return;
-    }
-
-    VideoCore::g_renderer->Rasterizer()->FlushRegion(start, size);
-}
-
-void RasterizerInvalidateRegion(PAddr start, u32 size) {
-    if (VideoCore::g_renderer == nullptr) {
-        return;
-    }
-
-    VideoCore::g_renderer->Rasterizer()->InvalidateRegion(start, size);
-}
-
-void RasterizerFlushAndInvalidateRegion(PAddr start, u32 size) {
-    // Since pages are unmapped on shutdown after video core is shutdown, the renderer may be
-    // null here
-    if (VideoCore::g_renderer == nullptr) {
-        return;
-    }
-
-    VideoCore::g_renderer->Rasterizer()->FlushAndInvalidateRegion(start, size);
-}
-
-void RasterizerClearAll(bool flush) {
-    // Since pages are unmapped on shutdown after video core is shutdown, the renderer may be
-    // null here
-    if (VideoCore::g_renderer == nullptr) {
-        return;
-    }
-
-    VideoCore::g_renderer->Rasterizer()->ClearAll(flush);
-}
-
-void RasterizerFlushVirtualRegion(VAddr start, u32 size, FlushMode mode) {
-    // Since pages are unmapped on shutdown after video core is shutdown, the renderer may be
-    // null here
-    if (VideoCore::g_renderer == nullptr) {
-        return;
-    }
-
-    VAddr end = start + size;
-
-    auto CheckRegion = [&](VAddr region_start, VAddr region_end, PAddr paddr_region_start) {
-        if (start >= region_end || end <= region_start) {
-            // No overlap with region
-            return;
-        }
-
-        VAddr overlap_start = std::max(start, region_start);
-        VAddr overlap_end = std::min(end, region_end);
-        PAddr physical_start = paddr_region_start + (overlap_start - region_start);
-        u32 overlap_size = overlap_end - overlap_start;
-
-        auto* rasterizer = VideoCore::g_renderer->Rasterizer();
-        switch (mode) {
-        case FlushMode::Flush:
-            rasterizer->FlushRegion(physical_start, overlap_size);
-            break;
-        case FlushMode::Invalidate:
-            rasterizer->InvalidateRegion(physical_start, overlap_size);
-            break;
-        case FlushMode::FlushAndInvalidate:
-            rasterizer->FlushAndInvalidateRegion(physical_start, overlap_size);
-            break;
-        }
-    };
-
-    CheckRegion(LINEAR_HEAP_VADDR, LINEAR_HEAP_VADDR_END, FCRAM_PADDR);
-    CheckRegion(NEW_LINEAR_HEAP_VADDR, NEW_LINEAR_HEAP_VADDR_END, FCRAM_PADDR);
-    CheckRegion(VRAM_VADDR, VRAM_VADDR_END, VRAM_PADDR);
-    if (Service::PLGLDR::PLG_LDR::GetPluginFBAddr())
-        CheckRegion(PLUGIN_3GX_FB_VADDR, PLUGIN_3GX_FB_VADDR_END,
-                    Service::PLGLDR::PLG_LDR::GetPluginFBAddr());
-}
-
 u8 MemorySystem::Read8(const VAddr addr) {
     return Read<u8>(addr);
 }

src/core/memory.h

@@ -226,21 +226,6 @@ enum : VAddr {
     PLUGIN_3GX_FB_VADDR_END = PLUGIN_3GX_FB_VADDR + PLUGIN_3GX_FB_SIZE
 };
 
-/**
- * Flushes any externally cached rasterizer resources touching the given region.
- */
-void RasterizerFlushRegion(PAddr start, u32 size);
-
-/**
- * Invalidates any externally cached rasterizer resources touching the given region.
- */
-void RasterizerInvalidateRegion(PAddr start, u32 size);
-
-/**
- * Flushes and invalidates any externally cached rasterizer resources touching the given region.
- */
-void RasterizerFlushAndInvalidateRegion(PAddr start, u32 size);
-
 enum class FlushMode {
     /// Write back modified surfaces to RAM
     Flush,
@@ -250,16 +235,6 @@ enum class FlushMode {
     FlushAndInvalidate,
 };
 
-/**
- * Flushes and invalidates all memory in the rasterizer cache and removes any leftover state
- * If flush is true, the rasterizer should flush any cached resources to RAM before clearing
- */
-void RasterizerClearAll(bool flush);
-
-/**
- * Flushes and invalidates any externally cached rasterizer resources touching the given virtual
- * address region.
- */
 void RasterizerFlushVirtualRegion(VAddr start, u32 size, FlushMode mode);
 
 class MemorySystem {

src/core/movie.cpp

@@ -21,7 +21,6 @@
 #include "core/hle/service/hid/hid.h"
 #include "core/hle/service/ir/extra_hid.h"
 #include "core/hle/service/ir/ir_rst.h"
-#include "core/hw/gpu.h"
 #include "core/loader/loader.h"
 #include "core/movie.h"
 
@@ -218,10 +217,10 @@ Movie::PlayMode Movie::GetPlayMode() const {
 }
 
 u64 Movie::GetCurrentInputIndex() const {
-    return static_cast<u64>(std::nearbyint(current_input / 234.0 * GPU::SCREEN_REFRESH_RATE));
+    return static_cast<u64>(std::nearbyint(current_input / 234.0 * SCREEN_REFRESH_RATE));
 }
 u64 Movie::GetTotalInputCount() const {
-    return static_cast<u64>(std::nearbyint(total_input / 234.0 * GPU::SCREEN_REFRESH_RATE));
+    return static_cast<u64>(std::nearbyint(total_input / 234.0 * SCREEN_REFRESH_RATE));
 }
 
 void Movie::CheckInputEnd() {

src/core/perf_stats.cpp

@@ -13,8 +13,9 @@
 #include <fmt/format.h>
 #include "common/file_util.h"
 #include "common/settings.h"
-#include "core/hw/gpu.h"
+#include "core/core_timing.h"
 #include "core/perf_stats.h"
+#include "video_core/gpu.h"
 
 using namespace std::chrono_literals;
 using DoubleSecs = std::chrono::duration<double, std::chrono::seconds::period>;
@@ -120,7 +121,7 @@ PerfStats::Results PerfStats::GetLastStats() {
 double PerfStats::GetLastFrameTimeScale() const {
     std::scoped_lock lock{object_mutex};
 
-    constexpr double FRAME_LENGTH = 1.0 / GPU::SCREEN_REFRESH_RATE;
+    constexpr double FRAME_LENGTH = 1.0 / SCREEN_REFRESH_RATE;
     return duration_cast<DoubleSecs>(previous_frame_length).count() / FRAME_LENGTH;
 }
 

src/core/tracer/citrace.h

@@ -75,16 +75,7 @@ struct CTMemoryLoad {
 
 struct CTRegisterWrite {
     u32 physical_address;
-
+    u32 value;
-    enum : u32 {
-        SIZE_8 = 0xD1,
-        SIZE_16 = 0xD2,
-        SIZE_32 = 0xD3,
-        SIZE_64 = 0xD4,
-    } size;
-
-    // TODO: Make it clearer which bits of this member are used for sizes other than 32 bits
-    u64 value;
 };
 
 struct CTStreamElement {

src/core/tracer/recorder.cpp

@@ -22,9 +22,8 @@ void Recorder::Finish(const std::string& filename) {
     // Calculate file offsets
     auto& initial = header.initial_state_offsets;
 
-    initial.gpu_registers_size = static_cast<u32>(initial_state.gpu_registers.size());
-    initial.lcd_registers_size = static_cast<u32>(initial_state.lcd_registers.size());
     initial.pica_registers_size = static_cast<u32>(initial_state.pica_registers.size());
+    initial.lcd_registers_size = static_cast<u32>(initial_state.lcd_registers.size());
     initial.default_attributes_size = static_cast<u32>(initial_state.default_attributes.size());
     initial.vs_program_binary_size = static_cast<u32>(initial_state.vs_program_binary.size());
     initial.vs_swizzle_data_size = static_cast<u32>(initial_state.vs_swizzle_data.size());
@@ -81,22 +80,17 @@ void Recorder::Finish(const std::string& filename) {
             throw "Failed to write header";
 
         // Write initial state
-        written =
-            file.WriteArray(initial_state.gpu_registers.data(), initial_state.gpu_registers.size());
-        if (written != initial_state.gpu_registers.size() || file.Tell() != initial.lcd_registers)
-            throw "Failed to write GPU registers";
-
-        written =
-            file.WriteArray(initial_state.lcd_registers.data(), initial_state.lcd_registers.size());
-        if (written != initial_state.lcd_registers.size() || file.Tell() != initial.pica_registers)
-            throw "Failed to write LCD registers";
-
         written = file.WriteArray(initial_state.pica_registers.data(),
                                   initial_state.pica_registers.size());
         if (written != initial_state.pica_registers.size() ||
             file.Tell() != initial.default_attributes)
             throw "Failed to write Pica registers";
 
+        written =
+            file.WriteArray(initial_state.lcd_registers.data(), initial_state.lcd_registers.size());
+        if (written != initial_state.lcd_registers.size() || file.Tell() != initial.pica_registers)
+            throw "Failed to write LCD registers";
+
         written = file.WriteArray(initial_state.default_attributes.data(),
                                   initial_state.default_attributes.size());
         if (written != initial_state.default_attributes.size() ||
@@ -187,21 +181,12 @@ void Recorder::MemoryAccessed(const u8* data, u32 size, u32 physical_address) {
     stream.push_back(element);
 }
 
-template <typename T>
+void Recorder::RegisterWritten(u32 physical_address, u32 value) {
-void Recorder::RegisterWritten(u32 physical_address, T value) {
     StreamElement element = {{RegisterWrite}};
-    element.data.register_write.size = (sizeof(T) == 1)   ? CTRegisterWrite::SIZE_8
-                                       : (sizeof(T) == 2) ? CTRegisterWrite::SIZE_16
-                                       : (sizeof(T) == 4) ? CTRegisterWrite::SIZE_32
-                                                          : CTRegisterWrite::SIZE_64;
     element.data.register_write.physical_address = physical_address;
     element.data.register_write.value = value;
 
     stream.push_back(element);
 }
 
-template void Recorder::RegisterWritten(u32, u8);
-template void Recorder::RegisterWritten(u32, u16);
-template void Recorder::RegisterWritten(u32, u32);
-template void Recorder::RegisterWritten(u32, u64);
 } // namespace CiTrace

src/core/tracer/recorder.h

@@ -4,7 +4,6 @@
 
 #pragma once
 
-#include <span>
 #include <string>
 #include <unordered_map>
 #include <vector>
@@ -17,7 +16,6 @@ namespace CiTrace {
 class Recorder {
 public:
     struct InitialState {
-        std::vector<u32> gpu_registers;
         std::vector<u32> lcd_registers;
         std::vector<u32> pica_registers;
         std::vector<u32> default_attributes;
@@ -52,8 +50,7 @@ public:
      * Record a register write.
      * @note Use this whenever a GPU-related MMIO register has been written to.
      */
-    template <typename T>
+    void RegisterWritten(u32 physical_address, u32 value);
-    void RegisterWritten(u32 physical_address, T value);
 
 private:
     // Initial state of recording start

src/tests/video_core/shader/shader_jit_compiler.cpp

@@ -13,6 +13,8 @@
 #include <catch2/catch_test_macros.hpp>
 #include <fmt/format.h>
 #include <nihstro/inline_assembly.h>
+#include "video_core/pica/shader_setup.h"
+#include "video_core/pica/shader_unit.h"
 #include "video_core/shader/shader_interpreter.h"
 #if CITRA_ARCH(x86_64)
 #include "video_core/shader/shader_jit_x64_compiler.h"
@@ -54,11 +56,11 @@ struct StringMaker<Common::Vec4f> {
 };
 } // namespace Catch
 
-static std::unique_ptr<Pica::Shader::ShaderSetup> CompileShaderSetup(
+static std::unique_ptr<Pica::ShaderSetup> CompileShaderSetup(
     std::initializer_list<nihstro::InlineAsm> code) {
     const auto shbin = nihstro::InlineAsm::CompileToRawBinary(code);
 
-    auto shader = std::make_unique<Pica::Shader::ShaderSetup>();
+    auto shader = std::make_unique<Pica::ShaderSetup>();
 
     std::transform(shbin.program.begin(), shbin.program.end(), shader->program_code.begin(),
                    [](const auto& x) { return x.hex; });
@@ -75,18 +77,16 @@ public:
         shader_jit.Compile(&shader_setup->program_code, &shader_setup->swizzle_data);
     }
 
-    explicit ShaderTest(std::unique_ptr<Pica::Shader::ShaderSetup> input_shader_setup)
+    explicit ShaderTest(std::unique_ptr<Pica::ShaderSetup> input_shader_setup)
         : shader_setup(std::move(input_shader_setup)) {
         shader_jit.Compile(&shader_setup->program_code, &shader_setup->swizzle_data);
     }
 
     Common::Vec4f Run(std::span<const Common::Vec4f> inputs) {
-        Pica::Shader::UnitState shader_unit;
+        Pica::ShaderUnit shader_unit;
         RunJit(shader_unit, inputs);
-        return {shader_unit.registers.output[0].x.ToFloat32(),
+        return {shader_unit.output[0].x.ToFloat32(), shader_unit.output[0].y.ToFloat32(),
-                shader_unit.registers.output[0].y.ToFloat32(),
+                shader_unit.output[0].z.ToFloat32(), shader_unit.output[0].w.ToFloat32()};
-                shader_unit.registers.output[0].z.ToFloat32(),
-                shader_unit.registers.output[0].w.ToFloat32()};
     }
 
     Common::Vec4f Run(std::initializer_list<float> inputs) {
@@ -105,39 +105,36 @@ public:
         return Run(std::vector<Common::Vec4f>{inputs});
     }
 
-    void RunJit(Pica::Shader::UnitState& shader_unit, std::span<const Common::Vec4f> inputs) {
+    void RunJit(Pica::ShaderUnit& shader_unit, std::span<const Common::Vec4f> inputs) {
         for (std::size_t i = 0; i < inputs.size(); ++i) {
             const Common::Vec4f& input = inputs[i];
-            shader_unit.registers.input[i].x = Pica::f24::FromFloat32(input.x);
+            shader_unit.input[i].x = Pica::f24::FromFloat32(input.x);
-            shader_unit.registers.input[i].y = Pica::f24::FromFloat32(input.y);
+            shader_unit.input[i].y = Pica::f24::FromFloat32(input.y);
-            shader_unit.registers.input[i].z = Pica::f24::FromFloat32(input.z);
+            shader_unit.input[i].z = Pica::f24::FromFloat32(input.z);
-            shader_unit.registers.input[i].w = Pica::f24::FromFloat32(input.w);
+            shader_unit.input[i].w = Pica::f24::FromFloat32(input.w);
         }
-        shader_unit.registers.temporary.fill(
+        shader_unit.temporary.fill(Common::Vec4<Pica::f24>::AssignToAll(Pica::f24::Zero()));
-            Common::Vec4<Pica::f24>::AssignToAll(Pica::f24::Zero()));
         shader_jit.Run(*shader_setup, shader_unit, 0);
     }
 
-    void RunJit(Pica::Shader::UnitState& shader_unit, float input) {
+    void RunJit(Pica::ShaderUnit& shader_unit, float input) {
         const Common::Vec4f input_vec(input, 0, 0, 0);
         RunJit(shader_unit, {&input_vec, 1});
     }
 
-    void RunInterpreter(Pica::Shader::UnitState& shader_unit,
+    void RunInterpreter(Pica::ShaderUnit& shader_unit, std::span<const Common::Vec4f> inputs) {
-                        std::span<const Common::Vec4f> inputs) {
         for (std::size_t i = 0; i < inputs.size(); ++i) {
             const Common::Vec4f& input = inputs[i];
-            shader_unit.registers.input[i].x = Pica::f24::FromFloat32(input.x);
+            shader_unit.input[i].x = Pica::f24::FromFloat32(input.x);
-            shader_unit.registers.input[i].y = Pica::f24::FromFloat32(input.y);
+            shader_unit.input[i].y = Pica::f24::FromFloat32(input.y);
-            shader_unit.registers.input[i].z = Pica::f24::FromFloat32(input.z);
+            shader_unit.input[i].z = Pica::f24::FromFloat32(input.z);
-            shader_unit.registers.input[i].w = Pica::f24::FromFloat32(input.w);
+            shader_unit.input[i].w = Pica::f24::FromFloat32(input.w);
         }
-        shader_unit.registers.temporary.fill(
+        shader_unit.temporary.fill(Common::Vec4<Pica::f24>::AssignToAll(Pica::f24::Zero()));
-            Common::Vec4<Pica::f24>::AssignToAll(Pica::f24::Zero()));
         shader_interpreter.Run(*shader_setup, shader_unit);
     }
 
-    void RunInterpreter(Pica::Shader::UnitState& shader_unit, float input) {
+    void RunInterpreter(Pica::ShaderUnit& shader_unit, float input) {
         const Common::Vec4f input_vec(input, 0, 0, 0);
         RunInterpreter(shader_unit, {&input_vec, 1});
     }
@@ -145,7 +142,7 @@ public:
 public:
     JitShader shader_jit;
     ShaderInterpreter shader_interpreter;
-    std::unique_ptr<Pica::Shader::ShaderSetup> shader_setup;
+    std::unique_ptr<Pica::ShaderSetup> shader_setup;
 };
 
 TEST_CASE("ADD", "[video_core][shader][shader_jit]") {
@@ -642,11 +639,11 @@ TEST_CASE("Nested Loop", "[video_core][shader][shader_jit]") {
                                     input) +
                                    input;
 
-        Pica::Shader::UnitState shader_unit_jit;
+        Pica::ShaderUnit shader_unit_jit;
         shader_test.RunJit(shader_unit_jit, input);
 
         REQUIRE(shader_unit_jit.address_registers[2] == expected_aL);
-        REQUIRE(shader_unit_jit.registers.output[0].x.ToFloat32() == Catch::Approx(expected_out));
+        REQUIRE(shader_unit_jit.output[0].x.ToFloat32() == Catch::Approx(expected_out));
     }
     {
         shader_test.shader_setup->uniforms.i[0] = {9, 0, 2, 0};
@@ -659,11 +656,11 @@ TEST_CASE("Nested Loop", "[video_core][shader][shader_jit]") {
                                      (shader_test.shader_setup->uniforms.i[1][0] + 1)) *
                                     input) +
                                    input;
-        Pica::Shader::UnitState shader_unit_jit;
+        Pica::ShaderUnit shader_unit_jit;
         shader_test.RunJit(shader_unit_jit, input);
 
         REQUIRE(shader_unit_jit.address_registers[2] == expected_aL);
-        REQUIRE(shader_unit_jit.registers.output[0].x.ToFloat32() == Catch::Approx(expected_out));
+        REQUIRE(shader_unit_jit.output[0].x.ToFloat32() == Catch::Approx(expected_out));
     }
 }
 

src/video_core/CMakeLists.txt

@@ -1,8 +1,6 @@
 add_subdirectory(host_shaders)
 
 add_library(video_core STATIC
-    command_processor.cpp
-    command_processor.h
     custom_textures/custom_format.cpp
     custom_textures/custom_format.h
     custom_textures/custom_tex_manager.cpp
@@ -11,29 +9,41 @@ add_library(video_core STATIC
     custom_textures/material.h
     debug_utils/debug_utils.cpp
     debug_utils/debug_utils.h
-    geometry_pipeline.cpp
+    gpu.cpp
-    geometry_pipeline.h
+    gpu.h
     gpu_debugger.h
-    pica.cpp
-    pica.h
-    pica_state.h
     pica_types.h
     precompiled_headers.h
-    primitive_assembly.cpp
-    primitive_assembly.h
     rasterizer_accelerated.cpp
     rasterizer_accelerated.h
     rasterizer_interface.h
-    regs.cpp
-    regs.h
-    regs_framebuffer.h
-    regs_lighting.h
-    regs_pipeline.h
-    regs_rasterizer.h
-    regs_shader.h
-    regs_texturing.h
     renderer_base.cpp
     renderer_base.h
+    pica/geometry_pipeline.cpp
+    pica/geometry_pipeline.h
+    pica/pica_core.cpp
+    pica/pica_core.h
+    pica/output_vertex.cpp
+    pica/output_vertex.h
+    pica/primitive_assembly.cpp
+    pica/primitive_assembly.h
+    pica/regs_external.h
+    pica/regs_framebuffer.h
+    pica/regs_internal.cpp
+    pica/regs_internal.h
+    pica/regs_lcd.h
+    pica/regs_lighting.h
+    pica/regs_pipeline.h
+    pica/regs_rasterizer.h
+    pica/regs_shader.h
+    pica/regs_texturing.h
+    pica/shader_setup.cpp
+    pica/shader_setup.h
+    pica/shader_unit.cpp
+    pica/shader_unit.h
+    pica/packed_attribute.h
+    pica/vertex_loader.cpp
+    pica/vertex_loader.h
     rasterizer_cache/framebuffer_base.h
     rasterizer_cache/pixel_format.cpp
     rasterizer_cache/pixel_format.h
@@ -84,6 +94,8 @@ add_library(video_core STATIC
     renderer_opengl/renderer_opengl.h
     renderer_software/renderer_software.cpp
     renderer_software/renderer_software.h
+    renderer_software/sw_blitter.cpp
+    renderer_software/sw_blitter.h
     renderer_software/sw_clipper.cpp
     renderer_software/sw_clipper.h
     renderer_software/sw_framebuffer.cpp
@@ -167,8 +179,6 @@ add_library(video_core STATIC
     texture/texture_decode.cpp
     texture/texture_decode.h
     utils.h
-    vertex_loader.cpp
-    vertex_loader.h
     video_core.cpp
     video_core.h
 )

src/video_core/command_processor.cpp

@@ -1,677 +0,0 @@
-// Copyright 2014 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <array>
-#include <cstddef>
-#include <cstring>
-#include <memory>
-#include <utility>
-#include "common/assert.h"
-#include "common/logging/log.h"
-#include "common/microprofile.h"
-#include "common/vector_math.h"
-#include "core/hle/service/gsp/gsp.h"
-#include "core/hw/gpu.h"
-#include "core/memory.h"
-#include "core/tracer/recorder.h"
-#include "video_core/command_processor.h"
-#include "video_core/debug_utils/debug_utils.h"
-#include "video_core/pica_state.h"
-#include "video_core/pica_types.h"
-#include "video_core/primitive_assembly.h"
-#include "video_core/rasterizer_interface.h"
-#include "video_core/regs.h"
-#include "video_core/regs_pipeline.h"
-#include "video_core/regs_texturing.h"
-#include "video_core/renderer_base.h"
-#include "video_core/shader/shader.h"
-#include "video_core/vertex_loader.h"
-#include "video_core/video_core.h"
-
-namespace Pica::CommandProcessor {
-
-// Expand a 4-bit mask to 4-byte mask, e.g. 0b0101 -> 0x00FF00FF
-constexpr std::array<u32, 16> expand_bits_to_bytes{
-    0x00000000, 0x000000ff, 0x0000ff00, 0x0000ffff, 0x00ff0000, 0x00ff00ff, 0x00ffff00, 0x00ffffff,
-    0xff000000, 0xff0000ff, 0xff00ff00, 0xff00ffff, 0xffff0000, 0xffff00ff, 0xffffff00, 0xffffffff,
-};
-
-MICROPROFILE_DEFINE(GPU_Drawing, "GPU", "Drawing", MP_RGB(50, 50, 240));
-
-static const char* GetShaderSetupTypeName(Shader::ShaderSetup& setup) {
-    if (&setup == &g_state.vs) {
-        return "vertex shader";
-    }
-    if (&setup == &g_state.gs) {
-        return "geometry shader";
-    }
-    return "unknown shader";
-}
-
-static void WriteUniformBoolReg(Shader::ShaderSetup& setup, u32 value) {
-    for (unsigned i = 0; i < setup.uniforms.b.size(); ++i)
-        setup.uniforms.b[i] = (value & (1 << i)) != 0;
-}
-
-static void WriteUniformIntReg(Shader::ShaderSetup& setup, unsigned index,
-                               const Common::Vec4<u8>& values) {
-    ASSERT(index < setup.uniforms.i.size());
-    setup.uniforms.i[index] = values;
-    LOG_TRACE(HW_GPU, "Set {} integer uniform {} to {:02x} {:02x} {:02x} {:02x}",
-              GetShaderSetupTypeName(setup), index, values.x, values.y, values.z, values.w);
-}
-
-static void WriteUniformFloatReg(ShaderRegs& config, Shader::ShaderSetup& setup,
-                                 int& float_regs_counter, std::array<u32, 4>& uniform_write_buffer,
-                                 u32 value) {
-    auto& uniform_setup = config.uniform_setup;
-
-    // TODO: Does actual hardware indeed keep an intermediate buffer or does
-    //       it directly write the values?
-    uniform_write_buffer[float_regs_counter++] = value;
-
-    // Uniforms are written in a packed format such that four float24 values are encoded in
-    // three 32-bit numbers. We write to internal memory once a full such vector is
-    // written.
-    if ((float_regs_counter >= 4 && uniform_setup.IsFloat32()) ||
-        (float_regs_counter >= 3 && !uniform_setup.IsFloat32())) {
-        float_regs_counter = 0;
-
-        if (uniform_setup.index >= setup.uniforms.f.size()) {
-            LOG_ERROR(HW_GPU, "Invalid {} float uniform index {}", GetShaderSetupTypeName(setup),
-                      (int)uniform_setup.index);
-        } else {
-            auto& uniform = setup.uniforms.f[uniform_setup.index];
-
-            // NOTE: The destination component order indeed is "backwards"
-            if (uniform_setup.IsFloat32()) {
-                for (auto i : {0, 1, 2, 3}) {
-                    float buffer_value;
-                    std::memcpy(&buffer_value, &uniform_write_buffer[i], sizeof(float));
-                    uniform[3 - i] = f24::FromFloat32(buffer_value);
-                }
-            } else {
-                // TODO: Untested
-                uniform.w = f24::FromRaw(uniform_write_buffer[0] >> 8);
-                uniform.z = f24::FromRaw(((uniform_write_buffer[0] & 0xFF) << 16) |
-                                         ((uniform_write_buffer[1] >> 16) & 0xFFFF));
-                uniform.y = f24::FromRaw(((uniform_write_buffer[1] & 0xFFFF) << 8) |
-                                         ((uniform_write_buffer[2] >> 24) & 0xFF));
-                uniform.x = f24::FromRaw(uniform_write_buffer[2] & 0xFFFFFF);
-            }
-
-            LOG_TRACE(HW_GPU, "Set {} float uniform {:x} to ({} {} {} {})",
-                      GetShaderSetupTypeName(setup), (int)uniform_setup.index,
-                      uniform.x.ToFloat32(), uniform.y.ToFloat32(), uniform.z.ToFloat32(),
-                      uniform.w.ToFloat32());
-
-            // TODO: Verify that this actually modifies the register!
-            uniform_setup.index.Assign(uniform_setup.index + 1);
-        }
-    }
-}
-
-static void WritePicaReg(u32 id, u32 value, u32 mask) {
-    auto& regs = g_state.regs;
-
-    if (id >= Regs::NUM_REGS) {
-        LOG_ERROR(
-            HW_GPU,
-            "Commandlist tried to write to invalid register 0x{:03X} (value: {:08X}, mask: {:X})",
-            id, value, mask);
-        return;
-    }
-
-    // TODO: Figure out how register masking acts on e.g. vs.uniform_setup.set_value
-    u32 old_value = regs.reg_array[id];
-
-    const u32 write_mask = expand_bits_to_bytes[mask];
-
-    regs.reg_array[id] = (old_value & ~write_mask) | (value & write_mask);
-
-    // Double check for is_pica_tracing to avoid call overhead
-    if (DebugUtils::IsPicaTracing()) {
-        DebugUtils::OnPicaRegWrite({(u16)id, (u16)mask, regs.reg_array[id]});
-    }
-
-    if (g_debug_context)
-        g_debug_context->OnEvent(DebugContext::Event::PicaCommandLoaded,
-                                 reinterpret_cast<void*>(&id));
-
-    switch (id) {
-    // Trigger IRQ
-    case PICA_REG_INDEX(trigger_irq):
-        Service::GSP::SignalInterrupt(Service::GSP::InterruptId::P3D);
-        break;
-
-    case PICA_REG_INDEX(pipeline.triangle_topology):
-        g_state.primitive_assembler.Reconfigure(regs.pipeline.triangle_topology);
-        break;
-
-    case PICA_REG_INDEX(pipeline.restart_primitive):
-        g_state.primitive_assembler.Reset();
-        break;
-
-    case PICA_REG_INDEX(pipeline.vs_default_attributes_setup.index):
-        g_state.immediate.current_attribute = 0;
-        g_state.immediate.reset_geometry_pipeline = true;
-        g_state.default_attr_counter = 0;
-        break;
-
-    // Load default vertex input attributes
-    case PICA_REG_INDEX(pipeline.vs_default_attributes_setup.set_value[0]):
-    case PICA_REG_INDEX(pipeline.vs_default_attributes_setup.set_value[1]):
-    case PICA_REG_INDEX(pipeline.vs_default_attributes_setup.set_value[2]): {
-        // TODO: Does actual hardware indeed keep an intermediate buffer or does
-        //       it directly write the values?
-        g_state.default_attr_write_buffer[g_state.default_attr_counter++] = value;
-
-        // Default attributes are written in a packed format such that four float24 values are
-        // encoded in
-        // three 32-bit numbers. We write to internal memory once a full such vector is
-        // written.
-        if (g_state.default_attr_counter >= 3) {
-            g_state.default_attr_counter = 0;
-
-            auto& setup = regs.pipeline.vs_default_attributes_setup;
-
-            if (setup.index >= 16) {
-                LOG_ERROR(HW_GPU, "Invalid VS default attribute index {}", (int)setup.index);
-                break;
-            }
-
-            Common::Vec4<f24> attribute;
-
-            // NOTE: The destination component order indeed is "backwards"
-            attribute.w = f24::FromRaw(g_state.default_attr_write_buffer[0] >> 8);
-            attribute.z = f24::FromRaw(((g_state.default_attr_write_buffer[0] & 0xFF) << 16) |
-                                       ((g_state.default_attr_write_buffer[1] >> 16) & 0xFFFF));
-            attribute.y = f24::FromRaw(((g_state.default_attr_write_buffer[1] & 0xFFFF) << 8) |
-                                       ((g_state.default_attr_write_buffer[2] >> 24) & 0xFF));
-            attribute.x = f24::FromRaw(g_state.default_attr_write_buffer[2] & 0xFFFFFF);
-
-            LOG_TRACE(HW_GPU, "Set default VS attribute {:x} to ({} {} {} {})", (int)setup.index,
-                      attribute.x.ToFloat32(), attribute.y.ToFloat32(), attribute.z.ToFloat32(),
-                      attribute.w.ToFloat32());
-
-            // TODO: Verify that this actually modifies the register!
-            if (setup.index < 15) {
-                g_state.input_default_attributes.attr[setup.index] = attribute;
-                setup.index++;
-            } else {
-                // Put each attribute into an immediate input buffer.  When all specified immediate
-                // attributes are present, the Vertex Shader is invoked and everything is sent to
-                // the primitive assembler.
-
-                auto& immediate_input = g_state.immediate.input_vertex;
-                auto& immediate_attribute_id = g_state.immediate.current_attribute;
-
-                immediate_input.attr[immediate_attribute_id] = attribute;
-
-                if (immediate_attribute_id < regs.pipeline.max_input_attrib_index) {
-                    immediate_attribute_id += 1;
-                } else {
-                    MICROPROFILE_SCOPE(GPU_Drawing);
-                    immediate_attribute_id = 0;
-
-                    Shader::OutputVertex::ValidateSemantics(regs.rasterizer);
-
-                    auto* shader_engine = Shader::GetEngine();
-                    shader_engine->SetupBatch(g_state.vs, regs.vs.main_offset);
-
-                    // Send to vertex shader
-                    if (g_debug_context)
-                        g_debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation,
-                                                 static_cast<void*>(&immediate_input));
-                    Shader::UnitState shader_unit;
-                    Shader::AttributeBuffer output{};
-
-                    shader_unit.LoadInput(regs.vs, immediate_input);
-                    shader_engine->Run(g_state.vs, shader_unit);
-                    shader_unit.WriteOutput(regs.vs, output);
-
-                    // Send to geometry pipeline
-                    if (g_state.immediate.reset_geometry_pipeline) {
-                        g_state.geometry_pipeline.Reconfigure();
-                        g_state.immediate.reset_geometry_pipeline = false;
-                    }
-                    ASSERT(!g_state.geometry_pipeline.NeedIndexInput());
-                    g_state.geometry_pipeline.Setup(shader_engine);
-                    g_state.geometry_pipeline.SubmitVertex(output);
-
-                    // TODO: If drawing after every immediate mode triangle kills performance,
-                    // change it to flush triangles whenever a drawing config register changes
-                    // See: https://github.com/citra-emu/citra/pull/2866#issuecomment-327011550
-                    VideoCore::g_renderer->Rasterizer()->DrawTriangles();
-                    if (g_debug_context) {
-                        g_debug_context->OnEvent(DebugContext::Event::FinishedPrimitiveBatch,
-                                                 nullptr);
-                    }
-                }
-            }
-        }
-        break;
-    }
-
-    case PICA_REG_INDEX(pipeline.gpu_mode):
-        // This register likely just enables vertex processing and doesn't need any special handling
-        break;
-
-    case PICA_REG_INDEX(pipeline.command_buffer.trigger[0]):
-    case PICA_REG_INDEX(pipeline.command_buffer.trigger[1]): {
-        unsigned index =
-            static_cast<unsigned>(id - PICA_REG_INDEX(pipeline.command_buffer.trigger[0]));
-        u32* head_ptr = (u32*)VideoCore::g_memory->GetPhysicalPointer(
-            regs.pipeline.command_buffer.GetPhysicalAddress(index));
-        g_state.cmd_list.head_ptr = g_state.cmd_list.current_ptr = head_ptr;
-        g_state.cmd_list.length = regs.pipeline.command_buffer.GetSize(index) / sizeof(u32);
-        break;
-    }
-
-    // It seems like these trigger vertex rendering
-    case PICA_REG_INDEX(pipeline.trigger_draw):
-    case PICA_REG_INDEX(pipeline.trigger_draw_indexed): {
-        MICROPROFILE_SCOPE(GPU_Drawing);
-
-#if PICA_LOG_TEV
-        DebugUtils::DumpTevStageConfig(regs.GetTevStages());
-#endif
-        if (g_debug_context)
-            g_debug_context->OnEvent(DebugContext::Event::IncomingPrimitiveBatch, nullptr);
-
-        PrimitiveAssembler<Shader::OutputVertex>& primitive_assembler = g_state.primitive_assembler;
-
-        bool accelerate_draw = VideoCore::g_hw_shader_enabled && primitive_assembler.IsEmpty();
-
-        if (regs.pipeline.use_gs == PipelineRegs::UseGS::No) {
-            auto topology = primitive_assembler.GetTopology();
-            if (topology == PipelineRegs::TriangleTopology::Shader ||
-                topology == PipelineRegs::TriangleTopology::List) {
-                accelerate_draw = accelerate_draw && (regs.pipeline.num_vertices % 3) == 0;
-            }
-            // TODO (wwylele): for Strip/Fan topology, if the primitive assember is not restarted
-            // after this draw call, the buffered vertex from this draw should "leak" to the next
-            // draw, in which case we should buffer the vertex into the software primitive assember,
-            // or disable accelerate draw completely. However, there is not game found yet that does
-            // this, so this is left unimplemented for now. Revisit this when an issue is found in
-            // games.
-        } else {
-            accelerate_draw = false;
-        }
-
-        bool is_indexed = (id == PICA_REG_INDEX(pipeline.trigger_draw_indexed));
-
-        if (accelerate_draw &&
-            VideoCore::g_renderer->Rasterizer()->AccelerateDrawBatch(is_indexed)) {
-            if (g_debug_context) {
-                g_debug_context->OnEvent(DebugContext::Event::FinishedPrimitiveBatch, nullptr);
-            }
-            break;
-        }
-
-        // Processes information about internal vertex attributes to figure out how a vertex is
-        // loaded.
-        // Later, these can be compiled and cached.
-        const u32 base_address = regs.pipeline.vertex_attributes.GetPhysicalBaseAddress();
-        VertexLoader loader(regs.pipeline);
-        Shader::OutputVertex::ValidateSemantics(regs.rasterizer);
-
-        // Load vertices
-        const auto& index_info = regs.pipeline.index_array;
-        const u8* index_address_8 =
-            VideoCore::g_memory->GetPhysicalPointer(base_address + index_info.offset);
-        const u16* index_address_16 = reinterpret_cast<const u16*>(index_address_8);
-        bool index_u16 = index_info.format != 0;
-
-        if (g_debug_context && g_debug_context->recorder) {
-            for (int i = 0; i < 3; ++i) {
-                const auto texture = regs.texturing.GetTextures()[i];
-                if (!texture.enabled)
-                    continue;
-
-                u8* texture_data =
-                    VideoCore::g_memory->GetPhysicalPointer(texture.config.GetPhysicalAddress());
-                g_debug_context->recorder->MemoryAccessed(
-                    texture_data,
-                    Pica::TexturingRegs::NibblesPerPixel(texture.format) * texture.config.width /
-                        2 * texture.config.height,
-                    texture.config.GetPhysicalAddress());
-            }
-        }
-
-        DebugUtils::MemoryAccessTracker memory_accesses;
-
-        // Simple circular-replacement vertex cache
-        // The size has been tuned for optimal balance between hit-rate and the cost of lookup
-        const std::size_t VERTEX_CACHE_SIZE = 32;
-        std::array<bool, VERTEX_CACHE_SIZE> vertex_cache_valid{};
-        std::array<u16, VERTEX_CACHE_SIZE> vertex_cache_ids;
-        std::array<Shader::AttributeBuffer, VERTEX_CACHE_SIZE> vertex_cache;
-        Shader::AttributeBuffer vs_output;
-
-        unsigned int vertex_cache_pos = 0;
-
-        auto* shader_engine = Shader::GetEngine();
-        Shader::UnitState shader_unit;
-
-        shader_engine->SetupBatch(g_state.vs, regs.vs.main_offset);
-
-        g_state.geometry_pipeline.Reconfigure();
-        g_state.geometry_pipeline.Setup(shader_engine);
-        if (g_state.geometry_pipeline.NeedIndexInput())
-            ASSERT(is_indexed);
-
-        for (unsigned int index = 0; index < regs.pipeline.num_vertices; ++index) {
-            // Indexed rendering doesn't use the start offset
-            unsigned int vertex =
-                is_indexed ? (index_u16 ? index_address_16[index] : index_address_8[index])
-                           : (index + regs.pipeline.vertex_offset);
-
-            bool vertex_cache_hit = false;
-
-            if (is_indexed) {
-                if (g_state.geometry_pipeline.NeedIndexInput()) {
-                    g_state.geometry_pipeline.SubmitIndex(vertex);
-                    continue;
-                }
-
-                if (g_debug_context && Pica::g_debug_context->recorder) {
-                    int size = index_u16 ? 2 : 1;
-                    memory_accesses.AddAccess(base_address + index_info.offset + size * index,
-                                              size);
-                }
-
-                for (unsigned int i = 0; i < VERTEX_CACHE_SIZE; ++i) {
-                    if (vertex_cache_valid[i] && vertex == vertex_cache_ids[i]) {
-                        vs_output = vertex_cache[i];
-                        vertex_cache_hit = true;
-                        break;
-                    }
-                }
-            }
-
-            if (!vertex_cache_hit) {
-                // Initialize data for the current vertex
-                Shader::AttributeBuffer input;
-                loader.LoadVertex(base_address, index, vertex, input, memory_accesses);
-
-                // Send to vertex shader
-                if (g_debug_context)
-                    g_debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation,
-                                             (void*)&input);
-                shader_unit.LoadInput(regs.vs, input);
-                shader_engine->Run(g_state.vs, shader_unit);
-                shader_unit.WriteOutput(regs.vs, vs_output);
-
-                if (is_indexed) {
-                    vertex_cache[vertex_cache_pos] = vs_output;
-                    vertex_cache_valid[vertex_cache_pos] = true;
-                    vertex_cache_ids[vertex_cache_pos] = vertex;
-                    vertex_cache_pos = (vertex_cache_pos + 1) % VERTEX_CACHE_SIZE;
-                }
-            }
-
-            // Send to geometry pipeline
-            g_state.geometry_pipeline.SubmitVertex(vs_output);
-        }
-
-        for (auto& range : memory_accesses.ranges) {
-            g_debug_context->recorder->MemoryAccessed(
-                VideoCore::g_memory->GetPhysicalPointer(range.first), range.second, range.first);
-        }
-
-        VideoCore::g_renderer->Rasterizer()->DrawTriangles();
-        if (g_debug_context) {
-            g_debug_context->OnEvent(DebugContext::Event::FinishedPrimitiveBatch, nullptr);
-        }
-
-        break;
-    }
-
-    case PICA_REG_INDEX(gs.bool_uniforms):
-        WriteUniformBoolReg(g_state.gs, g_state.regs.gs.bool_uniforms.Value());
-        break;
-
-    case PICA_REG_INDEX(gs.int_uniforms[0]):
-    case PICA_REG_INDEX(gs.int_uniforms[1]):
-    case PICA_REG_INDEX(gs.int_uniforms[2]):
-    case PICA_REG_INDEX(gs.int_uniforms[3]): {
-        unsigned index = (id - PICA_REG_INDEX(gs.int_uniforms[0]));
-        auto values = regs.gs.int_uniforms[index];
-        WriteUniformIntReg(g_state.gs, index,
-                           Common::Vec4<u8>(values.x, values.y, values.z, values.w));
-        break;
-    }
-
-    case PICA_REG_INDEX(gs.uniform_setup.set_value[0]):
-    case PICA_REG_INDEX(gs.uniform_setup.set_value[1]):
-    case PICA_REG_INDEX(gs.uniform_setup.set_value[2]):
-    case PICA_REG_INDEX(gs.uniform_setup.set_value[3]):
-    case PICA_REG_INDEX(gs.uniform_setup.set_value[4]):
-    case PICA_REG_INDEX(gs.uniform_setup.set_value[5]):
-    case PICA_REG_INDEX(gs.uniform_setup.set_value[6]):
-    case PICA_REG_INDEX(gs.uniform_setup.set_value[7]): {
-        WriteUniformFloatReg(g_state.regs.gs, g_state.gs, g_state.gs_float_regs_counter,
-                             g_state.gs_uniform_write_buffer, value);
-        break;
-    }
-
-    case PICA_REG_INDEX(gs.program.set_word[0]):
-    case PICA_REG_INDEX(gs.program.set_word[1]):
-    case PICA_REG_INDEX(gs.program.set_word[2]):
-    case PICA_REG_INDEX(gs.program.set_word[3]):
-    case PICA_REG_INDEX(gs.program.set_word[4]):
-    case PICA_REG_INDEX(gs.program.set_word[5]):
-    case PICA_REG_INDEX(gs.program.set_word[6]):
-    case PICA_REG_INDEX(gs.program.set_word[7]): {
-        u32& offset = g_state.regs.gs.program.offset;
-        if (offset >= 4096) {
-            LOG_ERROR(HW_GPU, "Invalid GS program offset {}", offset);
-        } else {
-            g_state.gs.program_code[offset] = value;
-            g_state.gs.MarkProgramCodeDirty();
-            offset++;
-        }
-        break;
-    }
-
-    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[0]):
-    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[1]):
-    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[2]):
-    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[3]):
-    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[4]):
-    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[5]):
-    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[6]):
-    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[7]): {
-        u32& offset = g_state.regs.gs.swizzle_patterns.offset;
-        if (offset >= g_state.gs.swizzle_data.size()) {
-            LOG_ERROR(HW_GPU, "Invalid GS swizzle pattern offset {}", offset);
-        } else {
-            g_state.gs.swizzle_data[offset] = value;
-            g_state.gs.MarkSwizzleDataDirty();
-            offset++;
-        }
-        break;
-    }
-
-    case PICA_REG_INDEX(vs.bool_uniforms):
-        // TODO (wwylele): does regs.pipeline.gs_unit_exclusive_configuration affect this?
-        WriteUniformBoolReg(g_state.vs, g_state.regs.vs.bool_uniforms.Value());
-        break;
-
-    case PICA_REG_INDEX(vs.int_uniforms[0]):
-    case PICA_REG_INDEX(vs.int_uniforms[1]):
-    case PICA_REG_INDEX(vs.int_uniforms[2]):
-    case PICA_REG_INDEX(vs.int_uniforms[3]): {
-        // TODO (wwylele): does regs.pipeline.gs_unit_exclusive_configuration affect this?
-        unsigned index = (id - PICA_REG_INDEX(vs.int_uniforms[0]));
-        auto values = regs.vs.int_uniforms[index];
-        WriteUniformIntReg(g_state.vs, index,
-                           Common::Vec4<u8>(values.x, values.y, values.z, values.w));
-        break;
-    }
-
-    case PICA_REG_INDEX(vs.uniform_setup.set_value[0]):
-    case PICA_REG_INDEX(vs.uniform_setup.set_value[1]):
-    case PICA_REG_INDEX(vs.uniform_setup.set_value[2]):
-    case PICA_REG_INDEX(vs.uniform_setup.set_value[3]):
-    case PICA_REG_INDEX(vs.uniform_setup.set_value[4]):
-    case PICA_REG_INDEX(vs.uniform_setup.set_value[5]):
-    case PICA_REG_INDEX(vs.uniform_setup.set_value[6]):
-    case PICA_REG_INDEX(vs.uniform_setup.set_value[7]): {
-        // TODO (wwylele): does regs.pipeline.gs_unit_exclusive_configuration affect this?
-        WriteUniformFloatReg(g_state.regs.vs, g_state.vs, g_state.vs_float_regs_counter,
-                             g_state.vs_uniform_write_buffer, value);
-        break;
-    }
-
-    case PICA_REG_INDEX(vs.program.set_word[0]):
-    case PICA_REG_INDEX(vs.program.set_word[1]):
-    case PICA_REG_INDEX(vs.program.set_word[2]):
-    case PICA_REG_INDEX(vs.program.set_word[3]):
-    case PICA_REG_INDEX(vs.program.set_word[4]):
-    case PICA_REG_INDEX(vs.program.set_word[5]):
-    case PICA_REG_INDEX(vs.program.set_word[6]):
-    case PICA_REG_INDEX(vs.program.set_word[7]): {
-        u32& offset = g_state.regs.vs.program.offset;
-        if (offset >= 512) {
-            LOG_ERROR(HW_GPU, "Invalid VS program offset {}", offset);
-        } else {
-            g_state.vs.program_code[offset] = value;
-            g_state.vs.MarkProgramCodeDirty();
-            if (!g_state.regs.pipeline.gs_unit_exclusive_configuration) {
-                g_state.gs.program_code[offset] = value;
-                g_state.gs.MarkProgramCodeDirty();
-            }
-            offset++;
-        }
-        break;
-    }
-
-    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[0]):
-    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[1]):
-    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[2]):
-    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[3]):
-    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[4]):
-    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[5]):
-    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[6]):
-    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[7]): {
-        u32& offset = g_state.regs.vs.swizzle_patterns.offset;
-        if (offset >= g_state.vs.swizzle_data.size()) {
-            LOG_ERROR(HW_GPU, "Invalid VS swizzle pattern offset {}", offset);
-        } else {
-            g_state.vs.swizzle_data[offset] = value;
-            g_state.vs.MarkSwizzleDataDirty();
-            if (!g_state.regs.pipeline.gs_unit_exclusive_configuration) {
-                g_state.gs.swizzle_data[offset] = value;
-                g_state.gs.MarkSwizzleDataDirty();
-            }
-            offset++;
-        }
-        break;
-    }
-
-    case PICA_REG_INDEX(lighting.lut_data[0]):
-    case PICA_REG_INDEX(lighting.lut_data[1]):
-    case PICA_REG_INDEX(lighting.lut_data[2]):
-    case PICA_REG_INDEX(lighting.lut_data[3]):
-    case PICA_REG_INDEX(lighting.lut_data[4]):
-    case PICA_REG_INDEX(lighting.lut_data[5]):
-    case PICA_REG_INDEX(lighting.lut_data[6]):
-    case PICA_REG_INDEX(lighting.lut_data[7]): {
-        auto& lut_config = regs.lighting.lut_config;
-
-        ASSERT_MSG(lut_config.index < 256, "lut_config.index exceeded maximum value of 255!");
-
-        g_state.lighting.luts[lut_config.type][lut_config.index].raw = value;
-        lut_config.index.Assign(lut_config.index + 1);
-        break;
-    }
-
-    case PICA_REG_INDEX(texturing.fog_lut_data[0]):
-    case PICA_REG_INDEX(texturing.fog_lut_data[1]):
-    case PICA_REG_INDEX(texturing.fog_lut_data[2]):
-    case PICA_REG_INDEX(texturing.fog_lut_data[3]):
-    case PICA_REG_INDEX(texturing.fog_lut_data[4]):
-    case PICA_REG_INDEX(texturing.fog_lut_data[5]):
-    case PICA_REG_INDEX(texturing.fog_lut_data[6]):
-    case PICA_REG_INDEX(texturing.fog_lut_data[7]): {
-        g_state.fog.lut[regs.texturing.fog_lut_offset % 128].raw = value;
-        regs.texturing.fog_lut_offset.Assign(regs.texturing.fog_lut_offset + 1);
-        break;
-    }
-
-    case PICA_REG_INDEX(texturing.proctex_lut_data[0]):
-    case PICA_REG_INDEX(texturing.proctex_lut_data[1]):
-    case PICA_REG_INDEX(texturing.proctex_lut_data[2]):
-    case PICA_REG_INDEX(texturing.proctex_lut_data[3]):
-    case PICA_REG_INDEX(texturing.proctex_lut_data[4]):
-    case PICA_REG_INDEX(texturing.proctex_lut_data[5]):
-    case PICA_REG_INDEX(texturing.proctex_lut_data[6]):
-    case PICA_REG_INDEX(texturing.proctex_lut_data[7]): {
-        auto& index = regs.texturing.proctex_lut_config.index;
-        auto& pt = g_state.proctex;
-
-        switch (regs.texturing.proctex_lut_config.ref_table.Value()) {
-        case TexturingRegs::ProcTexLutTable::Noise:
-            pt.noise_table[index % pt.noise_table.size()].raw = value;
-            break;
-        case TexturingRegs::ProcTexLutTable::ColorMap:
-            pt.color_map_table[index % pt.color_map_table.size()].raw = value;
-            break;
-        case TexturingRegs::ProcTexLutTable::AlphaMap:
-            pt.alpha_map_table[index % pt.alpha_map_table.size()].raw = value;
-            break;
-        case TexturingRegs::ProcTexLutTable::Color:
-            pt.color_table[index % pt.color_table.size()].raw = value;
-            break;
-        case TexturingRegs::ProcTexLutTable::ColorDiff:
-            pt.color_diff_table[index % pt.color_diff_table.size()].raw = value;
-            break;
-        }
-        index.Assign(index + 1);
-        break;
-    }
-    default:
-        break;
-    }
-
-    VideoCore::g_renderer->Rasterizer()->NotifyPicaRegisterChanged(id);
-
-    if (g_debug_context)
-        g_debug_context->OnEvent(DebugContext::Event::PicaCommandProcessed,
-                                 reinterpret_cast<void*>(&id));
-}
-
-void ProcessCommandList(PAddr list, u32 size) {
-
-    u32* buffer = (u32*)VideoCore::g_memory->GetPhysicalPointer(list);
-
-    if (Pica::g_debug_context && Pica::g_debug_context->recorder) {
-        Pica::g_debug_context->recorder->MemoryAccessed((u8*)buffer, size, list);
-    }
-
-    g_state.cmd_list.addr = list;
-    g_state.cmd_list.head_ptr = g_state.cmd_list.current_ptr = buffer;
-    g_state.cmd_list.length = size / sizeof(u32);
-
-    while (g_state.cmd_list.current_ptr < g_state.cmd_list.head_ptr + g_state.cmd_list.length) {
-
-        // Align read pointer to 8 bytes
-        if ((g_state.cmd_list.head_ptr - g_state.cmd_list.current_ptr) % 2 != 0)
-            ++g_state.cmd_list.current_ptr;
-
-        u32 value = *g_state.cmd_list.current_ptr++;
-        const CommandHeader header = {*g_state.cmd_list.current_ptr++};
-
-        WritePicaReg(header.cmd_id, value, header.parameter_mask);
-
-        for (unsigned i = 0; i < header.extra_data_length; ++i) {
-            u32 cmd = header.cmd_id + (header.group_commands ? i + 1 : 0);
-            WritePicaReg(cmd, *g_state.cmd_list.current_ptr++, header.parameter_mask);
-        }
-    }
-}
-
-} // namespace Pica::CommandProcessor

src/video_core/command_processor.h

@@ -1,37 +0,0 @@
-// Copyright 2014 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <type_traits>
-#include "common/bit_field.h"
-#include "common/common_types.h"
-
-namespace Pica::CommandProcessor {
-
-union CommandHeader {
-    u32 hex;
-
-    BitField<0, 16, u32> cmd_id;
-
-    // parameter_mask:
-    // Mask applied to the input value to make it possible to update
-    // parts of a register without overwriting its other fields.
-    // first bit:  0x000000FF
-    // second bit: 0x0000FF00
-    // third bit:  0x00FF0000
-    // fourth bit: 0xFF000000
-    BitField<16, 4, u32> parameter_mask;
-
-    BitField<20, 11, u32> extra_data_length;
-
-    BitField<31, 1, u32> group_commands;
-};
-static_assert(std::is_standard_layout<CommandHeader>::value == true,
-              "CommandHeader does not use standard layout");
-static_assert(sizeof(CommandHeader) == sizeof(u32), "CommandHeader has incorrect size!");
-
-void ProcessCommandList(PAddr list, u32 size);
-
-} // namespace Pica::CommandProcessor

src/video_core/debug_utils/debug_utils.cpp

@@ -2,38 +2,22 @@
 // Licensed under GPLv2
 // Refer to the license.txt file included.
 
-#include <algorithm>
-#include <condition_variable>
 #include <cstdint>
 #include <cstring>
 #include <fstream>
-#include <map>
-#include <mutex>
 #include <stdexcept>
-#include <string>
-
 #include <nihstro/bit_field.h>
 #include <nihstro/float24.h>
 #include <nihstro/shader_binary.h>
 #include "common/assert.h"
 #include "common/bit_field.h"
-#include "common/color.h"
-#include "common/common_types.h"
-#include "common/logging/log.h"
-#include "common/math_util.h"
 #include "common/vector_math.h"
+#include "core/core.h"
 #include "video_core/debug_utils/debug_utils.h"
-#include "video_core/pica_state.h"
+#include "video_core/gpu.h"
-#include "video_core/pica_types.h"
+#include "video_core/pica/regs_shader.h"
-#include "video_core/rasterizer_interface.h"
+#include "video_core/pica/shader_setup.h"
-#include "video_core/regs_rasterizer.h"
-#include "video_core/regs_shader.h"
-#include "video_core/regs_texturing.h"
 #include "video_core/renderer_base.h"
-#include "video_core/shader/shader.h"
-#include "video_core/texture/texture_decode.h"
-#include "video_core/utils.h"
-#include "video_core/video_core.h"
 
 using nihstro::DVLBHeader;
 using nihstro::DVLEHeader;
@@ -41,13 +25,13 @@ using nihstro::DVLPHeader;
 
 namespace Pica {
 
-void DebugContext::DoOnEvent(Event event, void* data) {
+void DebugContext::DoOnEvent(Event event, const void* data) {
     {
         std::unique_lock lock{breakpoint_mutex};
 
         // Commit the rasterizer's caches so framebuffers, render targets, etc. will show on debug
         // widgets
-        VideoCore::g_renderer->Rasterizer()->FlushAll();
+        Core::System::GetInstance().GPU().Renderer().Rasterizer()->FlushAll();
 
         // TODO: Should stop the CPU thread here once we multithread emulation.
 
@@ -84,8 +68,7 @@ std::shared_ptr<DebugContext> g_debug_context; // TODO: Get rid of this global
 
 namespace DebugUtils {
 
-void DumpShader(const std::string& filename, const ShaderRegs& config,
+void DumpShader(const std::string& filename, const ShaderRegs& config, const ShaderSetup& setup,
-                const Shader::ShaderSetup& setup,
                 const RasterizerRegs::VSOutputAttributes* output_attributes) {
     struct StuffToWrite {
         const u8* pointer;
@@ -288,13 +271,13 @@ void StartPicaTracing() {
     g_is_pica_tracing = true;
 }
 
-void OnPicaRegWrite(PicaTrace::Write write) {
+void OnPicaRegWrite(u16 cmd_id, u16 mask, u32 value) {
     std::lock_guard lock(pica_trace_mutex);
 
     if (!g_is_pica_tracing)
         return;
 
-    pica_trace->writes.push_back(write);
+    pica_trace->writes.push_back(PicaTrace::Write{cmd_id, mask, value});
 }
 
 std::unique_ptr<PicaTrace> FinishPicaTracing() {
@@ -313,153 +296,6 @@ std::unique_ptr<PicaTrace> FinishPicaTracing() {
     return ret;
 }
 
-static std::string ReplacePattern(const std::string& input, const std::string& pattern,
-                                  const std::string& replacement) {
-    std::size_t start = input.find(pattern);
-    if (start == std::string::npos)
-        return input;
-
-    std::string ret = input;
-    ret.replace(start, pattern.length(), replacement);
-    return ret;
-}
-
-static std::string GetTevStageConfigSourceString(
-    const TexturingRegs::TevStageConfig::Source& source) {
-
-    using Source = TexturingRegs::TevStageConfig::Source;
-    static const std::map<Source, std::string> source_map = {
-        {Source::PrimaryColor, "PrimaryColor"},
-        {Source::PrimaryFragmentColor, "PrimaryFragmentColor"},
-        {Source::SecondaryFragmentColor, "SecondaryFragmentColor"},
-        {Source::Texture0, "Texture0"},
-        {Source::Texture1, "Texture1"},
-        {Source::Texture2, "Texture2"},
-        {Source::Texture3, "Texture3"},
-        {Source::PreviousBuffer, "PreviousBuffer"},
-        {Source::Constant, "Constant"},
-        {Source::Previous, "Previous"},
-    };
-
-    const auto src_it = source_map.find(source);
-    if (src_it == source_map.end())
-        return "Unknown";
-
-    return src_it->second;
-}
-
-static std::string GetTevStageConfigColorSourceString(
-    const TexturingRegs::TevStageConfig::Source& source,
-    const TexturingRegs::TevStageConfig::ColorModifier modifier) {
-
-    using ColorModifier = TexturingRegs::TevStageConfig::ColorModifier;
-    static const std::map<ColorModifier, std::string> color_modifier_map = {
-        {ColorModifier::SourceColor, "%source.rgb"},
-        {ColorModifier::OneMinusSourceColor, "(1.0 - %source.rgb)"},
-        {ColorModifier::SourceAlpha, "%source.aaa"},
-        {ColorModifier::OneMinusSourceAlpha, "(1.0 - %source.aaa)"},
-        {ColorModifier::SourceRed, "%source.rrr"},
-        {ColorModifier::OneMinusSourceRed, "(1.0 - %source.rrr)"},
-        {ColorModifier::SourceGreen, "%source.ggg"},
-        {ColorModifier::OneMinusSourceGreen, "(1.0 - %source.ggg)"},
-        {ColorModifier::SourceBlue, "%source.bbb"},
-        {ColorModifier::OneMinusSourceBlue, "(1.0 - %source.bbb)"},
-    };
-
-    auto src_str = GetTevStageConfigSourceString(source);
-    auto modifier_it = color_modifier_map.find(modifier);
-    std::string modifier_str = "%source.????";
-    if (modifier_it != color_modifier_map.end())
-        modifier_str = modifier_it->second;
-
-    return ReplacePattern(modifier_str, "%source", src_str);
-}
-
-static std::string GetTevStageConfigAlphaSourceString(
-    const TexturingRegs::TevStageConfig::Source& source,
-    const TexturingRegs::TevStageConfig::AlphaModifier modifier) {
-
-    using AlphaModifier = TexturingRegs::TevStageConfig::AlphaModifier;
-    static const std::map<AlphaModifier, std::string> alpha_modifier_map = {
-        {AlphaModifier::SourceAlpha, "%source.a"},
-        {AlphaModifier::OneMinusSourceAlpha, "(1.0 - %source.a)"},
-        {AlphaModifier::SourceRed, "%source.r"},
-        {AlphaModifier::OneMinusSourceRed, "(1.0 - %source.r)"},
-        {AlphaModifier::SourceGreen, "%source.g"},
-        {AlphaModifier::OneMinusSourceGreen, "(1.0 - %source.g)"},
-        {AlphaModifier::SourceBlue, "%source.b"},
-        {AlphaModifier::OneMinusSourceBlue, "(1.0 - %source.b)"},
-    };
-
-    auto src_str = GetTevStageConfigSourceString(source);
-    auto modifier_it = alpha_modifier_map.find(modifier);
-    std::string modifier_str = "%source.????";
-    if (modifier_it != alpha_modifier_map.end())
-        modifier_str = modifier_it->second;
-
-    return ReplacePattern(modifier_str, "%source", src_str);
-}
-
-static std::string GetTevStageConfigOperationString(
-    const TexturingRegs::TevStageConfig::Operation& operation) {
-
-    using Operation = TexturingRegs::TevStageConfig::Operation;
-    static const std::map<Operation, std::string> combiner_map = {
-        {Operation::Replace, "%source1"},
-        {Operation::Modulate, "(%source1 * %source2)"},
-        {Operation::Add, "(%source1 + %source2)"},
-        {Operation::AddSigned, "(%source1 + %source2) - 0.5"},
-        {Operation::Lerp, "lerp(%source1, %source2, %source3)"},
-        {Operation::Subtract, "(%source1 - %source2)"},
-        {Operation::Dot3_RGB, "dot(%source1, %source2)"},
-        {Operation::MultiplyThenAdd, "((%source1 * %source2) + %source3)"},
-        {Operation::AddThenMultiply, "((%source1 + %source2) * %source3)"},
-    };
-
-    const auto op_it = combiner_map.find(operation);
-    if (op_it == combiner_map.end())
-        return "Unknown op (%source1, %source2, %source3)";
-
-    return op_it->second;
-}
-
-std::string GetTevStageConfigColorCombinerString(const TexturingRegs::TevStageConfig& tev_stage) {
-    auto op_str = GetTevStageConfigOperationString(tev_stage.color_op);
-    op_str = ReplacePattern(
-        op_str, "%source1",
-        GetTevStageConfigColorSourceString(tev_stage.color_source1, tev_stage.color_modifier1));
-    op_str = ReplacePattern(
-        op_str, "%source2",
-        GetTevStageConfigColorSourceString(tev_stage.color_source2, tev_stage.color_modifier2));
-    return ReplacePattern(
-        op_str, "%source3",
-        GetTevStageConfigColorSourceString(tev_stage.color_source3, tev_stage.color_modifier3));
-}
-
-std::string GetTevStageConfigAlphaCombinerString(const TexturingRegs::TevStageConfig& tev_stage) {
-    auto op_str = GetTevStageConfigOperationString(tev_stage.alpha_op);
-    op_str = ReplacePattern(
-        op_str, "%source1",
-        GetTevStageConfigAlphaSourceString(tev_stage.alpha_source1, tev_stage.alpha_modifier1));
-    op_str = ReplacePattern(
-        op_str, "%source2",
-        GetTevStageConfigAlphaSourceString(tev_stage.alpha_source2, tev_stage.alpha_modifier2));
-    return ReplacePattern(
-        op_str, "%source3",
-        GetTevStageConfigAlphaSourceString(tev_stage.alpha_source3, tev_stage.alpha_modifier3));
-}
-
-void DumpTevStageConfig(const std::array<TexturingRegs::TevStageConfig, 6>& stages) {
-    std::string stage_info = "Tev setup:\n";
-    for (std::size_t index = 0; index < stages.size(); ++index) {
-        const auto& tev_stage = stages[index];
-        stage_info += "Stage " + std::to_string(index) + ": " +
-                      GetTevStageConfigColorCombinerString(tev_stage) + "   " +
-                      GetTevStageConfigAlphaCombinerString(tev_stage) + "\n";
-    }
-    LOG_TRACE(HW_GPU, "{}", stage_info);
-}
-
 } // namespace DebugUtils
 
 } // namespace Pica

src/video_core/debug_utils/debug_utils.h

@@ -4,22 +4,16 @@
 
 #pragma once
 
-#include <algorithm>
-#include <array>
 #include <condition_variable>
 #include <iterator>
 #include <list>
-#include <map>
 #include <memory>
 #include <mutex>
 #include <string>
 #include <utility>
 #include <vector>
 #include "common/common_types.h"
-#include "common/vector_math.h"
+#include "video_core/pica/regs_rasterizer.h"
-#include "video_core/regs_rasterizer.h"
-#include "video_core/regs_shader.h"
-#include "video_core/regs_texturing.h"
 
 namespace CiTrace {
 class Recorder;
@@ -27,9 +21,8 @@ class Recorder;
 
 namespace Pica {
 
-namespace Shader {
+struct ShaderRegs;
 struct ShaderSetup;
-}
 
 class DebugContext {
 public:
@@ -87,7 +80,7 @@ public:
          * @param data Optional data pointer (if unused, this is a nullptr)
          * @note This function will perform nothing unless it is overridden in the child class.
          */
-        virtual void OnPicaBreakPointHit(Event event, void* data) {}
+        virtual void OnPicaBreakPointHit(Event event, const void* data) {}
 
         /**
          * Action to perform when emulation is resumed from a breakpoint.
@@ -126,7 +119,7 @@ public:
      * @param data Optional data pointer (pass nullptr if unused). Needs to remain valid until
      * Resume() is called.
      */
-    void OnEvent(Event event, void* data) {
+    void OnEvent(Event event, const void* data) {
         // This check is left in the header to allow the compiler to inline it.
         if (!breakpoints[(int)event].enabled)
             return;
@@ -134,7 +127,7 @@ public:
         DoOnEvent(event, data);
     }
 
-    void DoOnEvent(Event event, void* data);
+    void DoOnEvent(Event event, const void* data);
 
     /**
      * Resume from the current breakpoint.
@@ -181,10 +174,7 @@ extern std::shared_ptr<DebugContext> g_debug_context; // TODO: Get rid of this g
 
 namespace DebugUtils {
 
-#define PICA_LOG_TEV 0
+void DumpShader(const std::string& filename, const ShaderRegs& config, const ShaderSetup& setup,
-
-void DumpShader(const std::string& filename, const ShaderRegs& config,
-                const Shader::ShaderSetup& setup,
                 const RasterizerRegs::VSOutputAttributes* output_attributes);
 
 // Utility class to log Pica commands.
@@ -203,46 +193,9 @@ void StartPicaTracing();
 inline bool IsPicaTracing() {
     return g_is_pica_tracing;
 }
-void OnPicaRegWrite(PicaTrace::Write write);
+void OnPicaRegWrite(u16 cmd_id, u16 mask, u32 value);
 std::unique_ptr<PicaTrace> FinishPicaTracing();
 
-std::string GetTevStageConfigColorCombinerString(const TexturingRegs::TevStageConfig& tev_stage);
-std::string GetTevStageConfigAlphaCombinerString(const TexturingRegs::TevStageConfig& tev_stage);
-
-/// Dumps the Tev stage config to log at trace level
-void DumpTevStageConfig(const std::array<TexturingRegs::TevStageConfig, 6>& stages);
-
-/**
- * Used in the vertex loader to merge access records. TODO: Investigate if actually useful.
- */
-class MemoryAccessTracker {
-    /// Combine overlapping and close ranges
-    void SimplifyRanges() {
-        for (auto it = ranges.begin(); it != ranges.end(); ++it) {
-            // NOTE: We add 32 to the range end address to make sure "close" ranges are combined,
-            // too
-            auto it2 = std::next(it);
-            while (it2 != ranges.end() && it->first + it->second + 32 >= it2->first) {
-                it->second = std::max(it->second, it2->first + it2->second - it->first);
-                it2 = ranges.erase(it2);
-            }
-        }
-    }
-
-public:
-    /// Record a particular memory access in the list
-    void AddAccess(u32 paddr, u32 size) {
-        // Create new range or extend existing one
-        ranges[paddr] = std::max(ranges[paddr], size);
-
-        // Simplify ranges...
-        SimplifyRanges();
-    }
-
-    /// Map of accessed ranges (mapping start address to range size)
-    std::map<u32, u32> ranges;
-};
-
 } // namespace DebugUtils
 
 } // namespace Pica

src/video_core/gpu.cpp

@@ -0,0 +1,419 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/archives.h"
+#include "common/microprofile.h"
+#include "core/core.h"
+#include "core/core_timing.h"
+#include "core/hle/service/gsp/gsp_gpu.h"
+#include "core/hle/service/plgldr/plgldr.h"
+#include "video_core/debug_utils/debug_utils.h"
+#include "video_core/gpu.h"
+#include "video_core/gpu_debugger.h"
+#include "video_core/pica/pica_core.h"
+#include "video_core/pica/regs_lcd.h"
+#include "video_core/renderer_base.h"
+#include "video_core/renderer_software/sw_blitter.h"
+#include "video_core/video_core.h"
+
+namespace VideoCore {
+
+constexpr VAddr VADDR_LCD = 0x1ED02000;
+constexpr VAddr VADDR_GPU = 0x1EF00000;
+
+static PAddr VirtualToPhysicalAddress(VAddr addr) {
+    if (addr == 0) {
+        return 0;
+    }
+
+    if (addr >= Memory::VRAM_VADDR && addr <= Memory::VRAM_VADDR_END) {
+        return addr - Memory::VRAM_VADDR + Memory::VRAM_PADDR;
+    }
+    if (addr >= Memory::LINEAR_HEAP_VADDR && addr <= Memory::LINEAR_HEAP_VADDR_END) {
+        return addr - Memory::LINEAR_HEAP_VADDR + Memory::FCRAM_PADDR;
+    }
+    if (addr >= Memory::NEW_LINEAR_HEAP_VADDR && addr <= Memory::NEW_LINEAR_HEAP_VADDR_END) {
+        return addr - Memory::NEW_LINEAR_HEAP_VADDR + Memory::FCRAM_PADDR;
+    }
+    if (addr >= Memory::PLUGIN_3GX_FB_VADDR && addr <= Memory::PLUGIN_3GX_FB_VADDR_END) {
+        return addr - Memory::PLUGIN_3GX_FB_VADDR + Service::PLGLDR::PLG_LDR::GetPluginFBAddr();
+    }
+
+    LOG_ERROR(HW_Memory, "Unknown virtual address @ 0x{:08X}", addr);
+    return addr;
+}
+
+MICROPROFILE_DEFINE(GPU_DisplayTransfer, "GPU", "DisplayTransfer", MP_RGB(100, 100, 255));
+MICROPROFILE_DEFINE(GPU_CmdlistProcessing, "GPU", "Cmdlist Processing", MP_RGB(100, 255, 100));
+
+struct GPU::Impl {
+    Core::Timing& timing;
+    Core::System& system;
+    Memory::MemorySystem& memory;
+    Pica::DebugContext& debug_context;
+    Pica::PicaCore pica;
+    GraphicsDebugger gpu_debugger;
+    std::unique_ptr<RendererBase> renderer;
+    RasterizerInterface* rasterizer;
+    std::unique_ptr<SwRenderer::SwBlitter> sw_blitter;
+    Core::TimingEventType* vblank_event;
+    Service::GSP::InterruptHandler signal_interrupt;
+
+    explicit Impl(Core::System& system, Frontend::EmuWindow& emu_window,
+                  Frontend::EmuWindow* secondary_window)
+        : timing{system.CoreTiming()}, system{system}, memory{system.Memory()},
+          debug_context{*Pica::g_debug_context}, pica{memory, debug_context},
+          renderer{VideoCore::CreateRenderer(emu_window, secondary_window, pica, system)},
+          rasterizer{renderer->Rasterizer()}, sw_blitter{std::make_unique<SwRenderer::SwBlitter>(
+                                                  memory, rasterizer)} {}
+    ~Impl() = default;
+};
+
+GPU::GPU(Core::System& system, Frontend::EmuWindow& emu_window,
+         Frontend::EmuWindow* secondary_window)
+    : impl{std::make_unique<Impl>(system, emu_window, secondary_window)} {
+    impl->vblank_event = impl->timing.RegisterEvent(
+        "GPU::VBlankCallback",
+        [this](uintptr_t user_data, s64 cycles_late) { VBlankCallback(user_data, cycles_late); });
+    impl->timing.ScheduleEvent(FRAME_TICKS, impl->vblank_event);
+
+    // Bind the rasterizer to the PICA GPU
+    impl->pica.BindRasterizer(impl->rasterizer);
+}
+
+GPU::~GPU() = default;
+
+void GPU::SetInterruptHandler(Service::GSP::InterruptHandler handler) {
+    impl->signal_interrupt = handler;
+    impl->pica.SetInterruptHandler(handler);
+}
+
+void GPU::FlushRegion(PAddr addr, u32 size) {
+    impl->rasterizer->FlushRegion(addr, size);
+}
+
+void GPU::InvalidateRegion(PAddr addr, u32 size) {
+    impl->rasterizer->InvalidateRegion(addr, size);
+}
+
+void GPU::ClearAll(bool flush) {
+    impl->rasterizer->ClearAll(flush);
+}
+
+void GPU::Execute(const Service::GSP::Command& command) {
+    using Service::GSP::CommandId;
+    auto& regs = impl->pica.regs;
+
+    switch (command.id) {
+    case CommandId::RequestDma: {
+        Memory::RasterizerFlushVirtualRegion(command.dma_request.source_address,
+                                             command.dma_request.size, Memory::FlushMode::Flush);
+        Memory::RasterizerFlushVirtualRegion(command.dma_request.dest_address,
+                                             command.dma_request.size,
+                                             Memory::FlushMode::Invalidate);
+
+        // TODO(Subv): These memory accesses should not go through the application's memory mapping.
+        // They should go through the GSP module's memory mapping.
+        const auto process = impl->system.Kernel().GetCurrentProcess();
+        impl->memory.CopyBlock(*process, command.dma_request.dest_address,
+                               command.dma_request.source_address, command.dma_request.size);
+        impl->signal_interrupt(Service::GSP::InterruptId::DMA);
+        break;
+    }
+    case CommandId::SubmitCmdList: {
+        auto& params = command.submit_gpu_cmdlist;
+        auto& cmdbuffer = regs.internal.pipeline.command_buffer;
+
+        // Write to the command buffer GPU registers
+        cmdbuffer.addr[0].Assign(VirtualToPhysicalAddress(params.address) >> 3);
+        cmdbuffer.size[0].Assign(params.size >> 3);
+        cmdbuffer.trigger[0] = 1;
+
+        // Trigger processing of the command list
+        SubmitCmdList(0);
+        break;
+    }
+    case CommandId::MemoryFill: {
+        auto& params = command.memory_fill;
+        auto& memfill = regs.memory_fill_config;
+
+        // Write to the memory fill GPU registers.
+        if (params.start1 != 0) {
+            memfill[0].address_start = VirtualToPhysicalAddress(params.start1) >> 3;
+            memfill[0].address_end = VirtualToPhysicalAddress(params.end1) >> 3;
+            memfill[0].value_32bit = params.value1;
+            memfill[0].control = params.control1;
+            MemoryFill(0);
+        }
+        if (params.start2 != 0) {
+            memfill[1].address_start = VirtualToPhysicalAddress(params.start2) >> 3;
+            memfill[1].address_end = VirtualToPhysicalAddress(params.end2) >> 3;
+            memfill[1].value_32bit = params.value2;
+            memfill[1].control = params.control2;
+            MemoryFill(1);
+        }
+        break;
+    }
+    case CommandId::DisplayTransfer: {
+        auto& params = command.display_transfer;
+        auto& display_transfer = regs.display_transfer_config;
+
+        // Write to the transfer engine GPU registers.
+        display_transfer.input_address = VirtualToPhysicalAddress(params.in_buffer_address) >> 3;
+        display_transfer.output_address = VirtualToPhysicalAddress(params.out_buffer_address) >> 3;
+        display_transfer.input_size = params.in_buffer_size;
+        display_transfer.output_size = params.out_buffer_size;
+        display_transfer.flags = params.flags;
+        display_transfer.trigger.Assign(1);
+
+        // Trigger the display transfer.
+        MemoryTransfer();
+        break;
+    }
+    case CommandId::TextureCopy: {
+        auto& params = command.texture_copy;
+        auto& texture_copy = regs.display_transfer_config;
+
+        // Write to the transfer engine GPU registers.
+        texture_copy.input_address = VirtualToPhysicalAddress(params.in_buffer_address) >> 3;
+        texture_copy.output_address = VirtualToPhysicalAddress(params.out_buffer_address) >> 3;
+        texture_copy.texture_copy.size = params.size;
+        texture_copy.texture_copy.input_size = params.in_width_gap;
+        texture_copy.texture_copy.output_size = params.out_width_gap;
+        texture_copy.flags = params.flags;
+        texture_copy.trigger.Assign(1);
+
+        // Trigger the texture copy.
+        MemoryTransfer();
+        break;
+    }
+    case CommandId::CacheFlush: {
+        // Rasterizer flushing handled elsewhere in CPU read/write and other GPU handlers
+        // Use command.cache_flush.regions to implement this handler
+        break;
+    }
+    default:
+        LOG_ERROR(HW_GPU, "Unknown command {:#08X}", command.id.Value());
+    }
+
+    // Notify debugger that a GSP command was processed.
+    impl->debug_context.OnEvent(Pica::DebugContext::Event::GSPCommandProcessed, &command);
+}
+
+void GPU::SetBufferSwap(u32 screen_id, const Service::GSP::FrameBufferInfo& info) {
+    const PAddr phys_address_left = VirtualToPhysicalAddress(info.address_left);
+    const PAddr phys_address_right = VirtualToPhysicalAddress(info.address_right);
+
+    // Update framebuffer properties.
+    auto& framebuffer = impl->pica.regs.framebuffer_config[screen_id];
+    if (info.active_fb == 0) {
+        framebuffer.address_left1 = phys_address_left;
+        framebuffer.address_right1 = phys_address_right;
+    } else {
+        framebuffer.address_left2 = phys_address_left;
+        framebuffer.address_right2 = phys_address_right;
+    }
+
+    framebuffer.stride = info.stride;
+    framebuffer.format = info.format;
+    framebuffer.active_fb = info.shown_fb;
+
+    // Notify debugger about the buffer swap.
+    impl->debug_context.OnEvent(Pica::DebugContext::Event::BufferSwapped, nullptr);
+
+    if (screen_id == 0) {
+        MicroProfileFlip();
+        impl->system.perf_stats->EndGameFrame();
+    }
+}
+
+void GPU::SetColorFill(const Pica::ColorFill& fill) {
+    impl->pica.regs_lcd.color_fill_top = fill;
+    impl->pica.regs_lcd.color_fill_bottom = fill;
+}
+
+u32 GPU::ReadReg(VAddr addr) {
+    switch (addr & 0xFFFFF000) {
+    case VADDR_LCD: {
+        const u32 offset = addr - VADDR_LCD;
+        const u32 index = offset / sizeof(u32);
+        ASSERT(addr % sizeof(u32) == 0);
+        ASSERT(index < Pica::RegsLcd::NumIds());
+        return impl->pica.regs_lcd[index];
+    }
+    case VADDR_GPU:
+    case VADDR_GPU + 0x1000: {
+        const u32 offset = addr - VADDR_GPU;
+        const u32 index = offset / sizeof(u32);
+        ASSERT(addr % sizeof(u32) == 0);
+        ASSERT(index < Pica::PicaCore::Regs::NUM_REGS);
+        return impl->pica.regs.reg_array[index];
+    }
+    default:
+        UNREACHABLE_MSG("Read from unknown GPU address {:#08X}", addr);
+    }
+}
+
+void GPU::WriteReg(VAddr addr, u32 data) {
+    switch (addr & 0xFFFFF000) {
+    case VADDR_LCD: {
+        const u32 offset = addr - VADDR_LCD;
+        const u32 index = offset / sizeof(u32);
+        ASSERT(addr % sizeof(u32) == 0);
+        ASSERT(index < Pica::RegsLcd::NumIds());
+        impl->pica.regs_lcd[index] = data;
+        break;
+    }
+    case VADDR_GPU:
+    case VADDR_GPU + 0x1000: {
+        const u32 offset = addr - VADDR_GPU;
+        const u32 index = offset / sizeof(u32);
+
+        ASSERT(addr % sizeof(u32) == 0);
+        ASSERT(index < Pica::PicaCore::Regs::NUM_REGS);
+        impl->pica.regs.reg_array[index] = data;
+
+        // Handle registers that trigger GPU actions
+        switch (index) {
+        case GPU_REG_INDEX(memory_fill_config[0].trigger):
+            MemoryFill(0);
+            break;
+        case GPU_REG_INDEX(memory_fill_config[1].trigger):
+            MemoryFill(1);
+            break;
+        case GPU_REG_INDEX(display_transfer_config.trigger):
+            MemoryTransfer();
+            break;
+        case GPU_REG_INDEX(internal.pipeline.command_buffer.trigger[0]):
+            SubmitCmdList(0);
+            break;
+        case GPU_REG_INDEX(internal.pipeline.command_buffer.trigger[1]):
+            SubmitCmdList(1);
+            break;
+        default:
+            break;
+        }
+        break;
+    }
+    default:
+        UNREACHABLE_MSG("Write to unknown GPU address {:#08X}", addr);
+    }
+}
+
+void GPU::Sync() {
+    impl->renderer->Sync();
+}
+
+VideoCore::RendererBase& GPU::Renderer() {
+    return *impl->renderer;
+}
+
+Pica::PicaCore& GPU::PicaCore() {
+    return impl->pica;
+}
+
+const Pica::PicaCore& GPU::PicaCore() const {
+    return impl->pica;
+}
+
+Pica::DebugContext& GPU::DebugContext() {
+    return *Pica::g_debug_context;
+}
+
+GraphicsDebugger& GPU::Debugger() {
+    return impl->gpu_debugger;
+}
+
+void GPU::SubmitCmdList(u32 index) {
+    // Check if a command list was triggered.
+    auto& config = impl->pica.regs.internal.pipeline.command_buffer;
+    if (!config.trigger[index]) {
+        return;
+    }
+
+    MICROPROFILE_SCOPE(GPU_CmdlistProcessing);
+
+    // Forward command list processing to the PICA core.
+    const PAddr addr = config.GetPhysicalAddress(index);
+    const u32 size = config.GetSize(index);
+    impl->pica.ProcessCmdList(addr, size);
+    config.trigger[index] = 0;
+}
+
+void GPU::MemoryFill(u32 index) {
+    // Check if a memory fill was triggered.
+    auto& config = impl->pica.regs.memory_fill_config[index];
+    if (!config.trigger) {
+        return;
+    }
+
+    // Perform memory fill.
+    if (!impl->rasterizer->AccelerateFill(config)) {
+        impl->sw_blitter->MemoryFill(config);
+    }
+
+    // It seems that it won't signal interrupt if "address_start" is zero.
+    // TODO: hwtest this
+    if (config.GetStartAddress() != 0) {
+        if (!index) {
+            impl->signal_interrupt(Service::GSP::InterruptId::PSC0);
+        } else {
+            impl->signal_interrupt(Service::GSP::InterruptId::PSC1);
+        }
+    }
+
+    // Reset "trigger" flag and set the "finish" flag
+    // This was confirmed to happen on hardware even if "address_start" is zero.
+    config.trigger.Assign(0);
+    config.finished.Assign(1);
+}
+
+void GPU::MemoryTransfer() {
+    // Check if a transfer was triggered.
+    auto& config = impl->pica.regs.display_transfer_config;
+    if (!config.trigger.Value()) {
+        return;
+    }
+
+    MICROPROFILE_SCOPE(GPU_DisplayTransfer);
+
+    // Notify debugger about the display transfer.
+    impl->debug_context.OnEvent(Pica::DebugContext::Event::IncomingDisplayTransfer, nullptr);
+
+    // Perform memory transfer
+    if (config.is_texture_copy) {
+        if (!impl->rasterizer->AccelerateTextureCopy(config)) {
+            impl->sw_blitter->TextureCopy(config);
+        }
+    } else {
+        if (!impl->rasterizer->AccelerateDisplayTransfer(config)) {
+            impl->sw_blitter->DisplayTransfer(config);
+        }
+    }
+
+    // Complete transfer.
+    config.trigger.Assign(0);
+    impl->signal_interrupt(Service::GSP::InterruptId::PPF);
+}
+
+void GPU::VBlankCallback(std::uintptr_t user_data, s64 cycles_late) {
+    // Present renderered frame.
+    impl->renderer->SwapBuffers();
+
+    // Signal to GSP that GPU interrupt has occurred
+    impl->signal_interrupt(Service::GSP::InterruptId::PDC0);
+    impl->signal_interrupt(Service::GSP::InterruptId::PDC1);
+
+    // Reschedule recurrent event
+    impl->timing.ScheduleEvent(FRAME_TICKS - cycles_late, impl->vblank_event);
+}
+
+template <class Archive>
+void GPU::serialize(Archive& ar, const u32 file_version) {
+    ar & impl->pica;
+}
+
+SERIALIZE_IMPL(GPU)
+
+} // namespace VideoCore

src/video_core/gpu.h

@@ -0,0 +1,113 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <functional>
+#include <memory>
+#include <boost/serialization/access.hpp>
+
+#include "core/hle/service/gsp/gsp_interrupt.h"
+
+namespace Service::GSP {
+struct Command;
+struct FrameBufferInfo;
+} // namespace Service::GSP
+
+namespace Core {
+class System;
+}
+
+namespace Pica {
+class DebugContext;
+class PicaCore;
+struct RegsLcd;
+union ColorFill;
+} // namespace Pica
+
+namespace Frontend {
+class EmuWindow;
+}
+
+namespace VideoCore {
+
+/// Measured on hardware to be 2240568 timer cycles or 4481136 ARM11 cycles
+constexpr u64 FRAME_TICKS = 4481136ull;
+
+class GraphicsDebugger;
+class RendererBase;
+
+/**
+ * The GPU class is the high level interface to the video_core for core services.
+ */
+class GPU {
+public:
+    explicit GPU(Core::System& system, Frontend::EmuWindow& emu_window,
+                 Frontend::EmuWindow* secondary_window);
+    ~GPU();
+
+    /// Sets the function to call for signalling GSP interrupts.
+    void SetInterruptHandler(Service::GSP::InterruptHandler handler);
+
+    /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
+    void FlushRegion(PAddr addr, u32 size);
+
+    /// Notify rasterizer that any caches of the specified region should be invalidated
+    void InvalidateRegion(PAddr addr, u32 size);
+
+    /// Flushes and invalidates all memory in the rasterizer cache and removes any leftover state.
+    void ClearAll(bool flush);
+
+    /// Executes the provided GSP command.
+    void Execute(const Service::GSP::Command& command);
+
+    /// Updates GPU display framebuffer configuration using the specified parameters.
+    void SetBufferSwap(u32 screen_id, const Service::GSP::FrameBufferInfo& info);
+
+    /// Sets the LCD color fill configuration for the top and bottom screens.
+    void SetColorFill(const Pica::ColorFill& fill);
+
+    /// Reads a word from the GPU virtual address.
+    u32 ReadReg(VAddr addr);
+
+    /// Writes the provided value to the GPU virtual address.
+    void WriteReg(VAddr addr, u32 data);
+
+    /// Synchronizes fixed function renderer state with PICA registers.
+    void Sync();
+
+    /// Returns a mutable reference to the renderer.
+    [[nodiscard]] VideoCore::RendererBase& Renderer();
+
+    /// Returns a mutable reference to the PICA GPU.
+    [[nodiscard]] Pica::PicaCore& PicaCore();
+
+    /// Returns an immutable reference to the PICA GPU.
+    [[nodiscard]] const Pica::PicaCore& PicaCore() const;
+
+    /// Returns a mutable reference to the pica debugging context.
+    [[nodiscard]] Pica::DebugContext& DebugContext();
+
+    /// Returns a mutable reference to the GSP command debugger.
+    [[nodiscard]] GraphicsDebugger& Debugger();
+
+private:
+    void SubmitCmdList(u32 index);
+
+    void MemoryFill(u32 index);
+
+    void MemoryTransfer();
+
+    void VBlankCallback(uintptr_t user_data, s64 cycles_late);
+
+    friend class boost::serialization::access;
+    template <class Archive>
+    void serialize(Archive& ar, const u32 file_version);
+
+private:
+    struct Impl;
+    std::unique_ptr<Impl> impl;
+};
+
+} // namespace VideoCore

src/video_core/gpu_debugger.h

@@ -7,18 +7,22 @@
 #include <algorithm>
 #include <functional>
 #include <vector>
-#include "core/hle/service/gsp/gsp.h"
+#include "core/hle/service/gsp/gsp_command.h"
+
+namespace VideoCore {
 
 class GraphicsDebugger {
 public:
     // Base class for all objects which need to be notified about GPU events
     class DebuggerObserver {
-    public:
+        friend class GraphicsDebugger;
-        DebuggerObserver() : observed(nullptr) {}
 
+    public:
+        DebuggerObserver() = default;
         virtual ~DebuggerObserver() {
-            if (observed)
+            if (observed) {
                 observed->UnregisterObserver(this);
+            }
         }
 
         /**
@@ -39,20 +43,15 @@ public:
         }
 
     private:
-        GraphicsDebugger* observed;
+        GraphicsDebugger* observed{};
-
-        friend class GraphicsDebugger;
     };
 
-    void GXCommandProcessed(u8* command_data) {
+    void GXCommandProcessed(Service::GSP::Command& command_data) {
-        if (observers.empty())
+        if (observers.empty()) {
             return;
+        }
 
-        gx_command_history.emplace_back();
+        gx_command_history.emplace_back(command_data);
-        Service::GSP::Command& cmd = gx_command_history.back();
-
-        std::memcpy(&cmd, command_data, sizeof(Service::GSP::Command));
-
         ForEachObserver([this](DebuggerObserver* observer) {
             observer->GXCommandProcessed(static_cast<int>(this->gx_command_history.size()));
         });
@@ -80,6 +79,7 @@ private:
     }
 
     std::vector<DebuggerObserver*> observers;
-
     std::vector<Service::GSP::Command> gx_command_history;
 };
+
+} // namespace VideoCore

src/video_core/pica.cpp

@@ -1,70 +0,0 @@
-// Copyright 2015 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <cstring>
-#include <type_traits>
-#include "core/global.h"
-#include "video_core/geometry_pipeline.h"
-#include "video_core/pica.h"
-#include "video_core/pica_state.h"
-#include "video_core/renderer_base.h"
-#include "video_core/video_core.h"
-
-namespace Core {
-template <>
-Pica::State& Global() {
-    return Pica::g_state;
-}
-} // namespace Core
-
-namespace Pica {
-
-State g_state;
-
-void Init() {
-    g_state.Reset();
-}
-
-void Shutdown() {
-    Shader::Shutdown();
-}
-
-template <typename T>
-void Zero(T& o) {
-    static_assert(std::is_trivial_v<T>, "It's undefined behavior to memset a non-trivial type");
-    std::memset(&o, 0, sizeof(o));
-}
-
-State::State() : geometry_pipeline(*this) {
-    auto SubmitVertex = [this](const Shader::AttributeBuffer& vertex) {
-        using Pica::Shader::OutputVertex;
-        auto AddTriangle = [](const OutputVertex& v0, const OutputVertex& v1,
-                              const OutputVertex& v2) {
-            VideoCore::g_renderer->Rasterizer()->AddTriangle(v0, v1, v2);
-        };
-        primitive_assembler.SubmitVertex(
-            Shader::OutputVertex::FromAttributeBuffer(regs.rasterizer, vertex), AddTriangle);
-    };
-
-    auto SetWinding = [this]() { primitive_assembler.SetWinding(); };
-
-    g_state.gs_unit.SetVertexHandler(SubmitVertex, SetWinding);
-    g_state.geometry_pipeline.SetVertexHandler(SubmitVertex);
-}
-
-void State::Reset() {
-    Zero(regs);
-    vs = {};
-    gs = {};
-    Zero(cmd_list);
-    immediate = {};
-    primitive_assembler.Reconfigure(PipelineRegs::TriangleTopology::List);
-    vs_float_regs_counter = 0;
-    vs_uniform_write_buffer.fill(0);
-    gs_float_regs_counter = 0;
-    gs_uniform_write_buffer.fill(0);
-    default_attr_counter = 0;
-    default_attr_write_buffer.fill(0);
-}
-} // namespace Pica

src/video_core/pica.h

@@ -1,16 +0,0 @@
-// Copyright 2014 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include "video_core/regs_texturing.h"
-namespace Pica {
-
-/// Initialize Pica state
-void Init();
-
-/// Shutdown Pica state
-void Shutdown();
-
-} // namespace Pica

src/video_core/pica/geometry_pipeline.cpp

@@ -6,11 +6,13 @@
 #include <boost/serialization/export.hpp>
 #include <boost/serialization/unique_ptr.hpp>
 #include "common/archives.h"
-#include "video_core/geometry_pipeline.h"
+#include "core/core.h"
-#include "video_core/pica_state.h"
+#include "video_core/gpu.h"
-#include "video_core/regs.h"
+#include "video_core/pica/geometry_pipeline.h"
-#include "video_core/renderer_base.h"
+#include "video_core/pica/pica_core.h"
-#include "video_core/video_core.h"
+#include "video_core/pica/shader_setup.h"
+#include "video_core/pica/shader_unit.h"
+#include "video_core/shader/shader.h"
 
 namespace Pica {
 
@@ -33,7 +35,7 @@ public:
      * @param input attributes of a vertex output from vertex shader
      * @return if the buffer is full and the geometry shader should be invoked
      */
-    virtual bool SubmitVertex(const Shader::AttributeBuffer& input) = 0;
+    virtual bool SubmitVertex(const AttributeBuffer& input) = 0;
 
 private:
     template <class Archive>
@@ -49,32 +51,33 @@ private:
 // TODO: what happens when the input size is not divisible by the output size?
 class GeometryPipeline_Point : public GeometryPipelineBackend {
 public:
-    GeometryPipeline_Point(const Regs& regs, Shader::GSUnitState& unit) : regs(regs), unit(unit) {
+    GeometryPipeline_Point(const RegsInternal& regs, GeometryShaderUnit& unit)
+        : regs(regs), unit(unit) {
         ASSERT(regs.pipeline.variable_primitive == 0);
         ASSERT(regs.gs.input_to_uniform == 0);
         vs_output_num = regs.pipeline.vs_outmap_total_minus_1_a + 1;
         std::size_t gs_input_num = regs.gs.max_input_attribute_index + 1;
         ASSERT(gs_input_num % vs_output_num == 0);
-        buffer_cur = attribute_buffer.attr;
+        buffer_cur = attribute_buffer.data();
-        buffer_end = attribute_buffer.attr + gs_input_num;
+        buffer_end = attribute_buffer.data() + gs_input_num;
     }
 
     bool IsEmpty() const override {
-        return buffer_cur == attribute_buffer.attr;
+        return buffer_cur == attribute_buffer.data();
     }
 
     bool NeedIndexInput() const override {
         return false;
     }
 
-    void SubmitIndex(unsigned int val) override {
+    void SubmitIndex(u32 val) override {
         UNREACHABLE();
     }
 
-    bool SubmitVertex(const Shader::AttributeBuffer& input) override {
+    bool SubmitVertex(const AttributeBuffer& input) override {
-        buffer_cur = std::copy(input.attr, input.attr + vs_output_num, buffer_cur);
+        buffer_cur = std::copy(input.data(), input.data() + vs_output_num, buffer_cur);
         if (buffer_cur == buffer_end) {
-            buffer_cur = attribute_buffer.attr;
+            buffer_cur = attribute_buffer.data();
             unit.LoadInput(regs.gs, attribute_buffer);
             return true;
         }
@@ -82,14 +85,17 @@ public:
     }
 
 private:
-    const Regs& regs;
+    const RegsInternal& regs;
-    Shader::GSUnitState& unit;
+    GeometryShaderUnit& unit;
-    Shader::AttributeBuffer attribute_buffer;
+    AttributeBuffer attribute_buffer;
     Common::Vec4<f24>* buffer_cur;
     Common::Vec4<f24>* buffer_end;
-    unsigned int vs_output_num;
+    u32 vs_output_num;
 
-    GeometryPipeline_Point() : regs(g_state.regs), unit(g_state.gs_unit) {}
+    // TODO: REMOVE THIS
+    GeometryPipeline_Point()
+        : regs(Core::System::GetInstance().GPU().PicaCore().regs.internal),
+          unit(Core::System::GetInstance().GPU().PicaCore().gs_unit) {}
 
     template <typename Class, class Archive>
     static void serialize_common(Class* self, Archive& ar, const unsigned int version) {
@@ -101,8 +107,8 @@ private:
     template <class Archive>
     void save(Archive& ar, const unsigned int version) const {
         serialize_common(this, ar, version);
-        auto buffer_idx = static_cast<u32>(buffer_cur - attribute_buffer.attr);
+        auto buffer_idx = static_cast<u32>(buffer_cur - attribute_buffer.data());
-        auto buffer_size = static_cast<u32>(buffer_end - attribute_buffer.attr);
+        auto buffer_size = static_cast<u32>(buffer_end - attribute_buffer.data());
         ar << buffer_idx;
         ar << buffer_size;
     }
@@ -113,8 +119,8 @@ private:
         u32 buffer_idx, buffer_size;
         ar >> buffer_idx;
         ar >> buffer_size;
-        buffer_cur = attribute_buffer.attr + buffer_idx;
+        buffer_cur = attribute_buffer.data() + buffer_idx;
-        buffer_end = attribute_buffer.attr + buffer_size;
+        buffer_end = attribute_buffer.data() + buffer_size;
     }
 
     BOOST_SERIALIZATION_SPLIT_MEMBER()
@@ -127,7 +133,7 @@ private:
 // value in the batch. This mode is usually used for subdivision.
 class GeometryPipeline_VariablePrimitive : public GeometryPipelineBackend {
 public:
-    GeometryPipeline_VariablePrimitive(const Regs& regs, Shader::ShaderSetup& setup)
+    GeometryPipeline_VariablePrimitive(const RegsInternal& regs, ShaderSetup& setup)
         : regs(regs), setup(setup) {
         ASSERT(regs.pipeline.variable_primitive == 1);
         ASSERT(regs.gs.input_to_uniform == 1);
@@ -142,7 +148,7 @@ public:
         return need_index;
     }
 
-    void SubmitIndex(unsigned int val) override {
+    void SubmitIndex(u32 val) override {
         DEBUG_ASSERT(need_index);
 
         // The number of vertex input is put to the uniform register
@@ -157,15 +163,15 @@ public:
         need_index = false;
     }
 
-    bool SubmitVertex(const Shader::AttributeBuffer& input) override {
+    bool SubmitVertex(const AttributeBuffer& input) override {
         DEBUG_ASSERT(!need_index);
         if (main_vertex_num != 0) {
             // For main vertices, receive all attributes
-            buffer_cur = std::copy(input.attr, input.attr + vs_output_num, buffer_cur);
+            buffer_cur = std::copy(input.data(), input.data() + vs_output_num, buffer_cur);
             --main_vertex_num;
         } else {
             // For other vertices, only receive the first attribute (usually the position)
-            *(buffer_cur++) = input.attr[0];
+            *(buffer_cur++) = input[0];
         }
         --total_vertex_num;
 
@@ -179,14 +185,17 @@ public:
 
 private:
     bool need_index = true;
-    const Regs& regs;
+    const RegsInternal& regs;
-    Shader::ShaderSetup& setup;
+    ShaderSetup& setup;
-    unsigned int main_vertex_num;
+    u32 main_vertex_num;
-    unsigned int total_vertex_num;
+    u32 total_vertex_num;
     Common::Vec4<f24>* buffer_cur;
-    unsigned int vs_output_num;
+    u32 vs_output_num;
 
-    GeometryPipeline_VariablePrimitive() : regs(g_state.regs), setup(g_state.gs) {}
+    // TODO: REMOVE THIS
+    GeometryPipeline_VariablePrimitive()
+        : regs(Core::System::GetInstance().GPU().PicaCore().regs.internal),
+          setup(Core::System::GetInstance().GPU().PicaCore().gs_setup) {}
 
     template <typename Class, class Archive>
     static void serialize_common(Class* self, Archive& ar, const unsigned int version) {
@@ -222,8 +231,7 @@ private:
 // particle system.
 class GeometryPipeline_FixedPrimitive : public GeometryPipelineBackend {
 public:
-    GeometryPipeline_FixedPrimitive(const Regs& regs, Shader::ShaderSetup& setup)
+    GeometryPipeline_FixedPrimitive(const RegsInternal& regs, ShaderSetup& setup) : setup(setup) {
-        : regs(regs), setup(setup) {
         ASSERT(regs.pipeline.variable_primitive == 0);
         ASSERT(regs.gs.input_to_uniform == 1);
         vs_output_num = regs.pipeline.vs_outmap_total_minus_1_a + 1;
@@ -241,12 +249,12 @@ public:
         return false;
     }
 
-    void SubmitIndex(unsigned int val) override {
+    void SubmitIndex(u32 val) override {
         UNREACHABLE();
     }
 
-    bool SubmitVertex(const Shader::AttributeBuffer& input) override {
+    bool SubmitVertex(const AttributeBuffer& input) override {
-        buffer_cur = std::copy(input.attr, input.attr + vs_output_num, buffer_cur);
+        buffer_cur = std::copy(input.data(), input.data() + vs_output_num, buffer_cur);
         if (buffer_cur == buffer_end) {
             buffer_cur = buffer_begin;
             return true;
@@ -255,14 +263,15 @@ public:
     }
 
 private:
-    [[maybe_unused]] const Regs& regs;
+    ShaderSetup& setup;
-    Shader::ShaderSetup& setup;
     Common::Vec4<f24>* buffer_begin;
     Common::Vec4<f24>* buffer_cur;
     Common::Vec4<f24>* buffer_end;
-    unsigned int vs_output_num;
+    u32 vs_output_num;
 
-    GeometryPipeline_FixedPrimitive() : regs(g_state.regs), setup(g_state.gs) {}
+    // TODO: REMOVE THIS
+    GeometryPipeline_FixedPrimitive()
+        : setup(Core::System::GetInstance().GPU().PicaCore().gs_setup) {}
 
     template <typename Class, class Archive>
     static void serialize_common(Class* self, Archive& ar, const unsigned int version) {
@@ -298,52 +307,53 @@ private:
     friend class boost::serialization::access;
 };
 
-GeometryPipeline::GeometryPipeline(State& state) : state(state) {}
+GeometryPipeline::GeometryPipeline(RegsInternal& regs_, GeometryShaderUnit& gs_unit_,
+                                   ShaderSetup& gs_)
+    : regs(regs_), gs_unit(gs_unit_), gs(gs_) {}
 
 GeometryPipeline::~GeometryPipeline() = default;
 
-void GeometryPipeline::SetVertexHandler(Shader::VertexHandler vertex_handler) {
+void GeometryPipeline::SetVertexHandler(VertexHandler vertex_handler) {
     this->vertex_handler = std::move(vertex_handler);
 }
 
-void GeometryPipeline::Setup(Shader::ShaderEngine* shader_engine) {
+void GeometryPipeline::Setup(ShaderEngine* shader_engine) {
-    if (!backend)
+    if (!backend) {
         return;
+    }
 
     this->shader_engine = shader_engine;
-    shader_engine->SetupBatch(state.gs, state.regs.gs.main_offset);
+    shader_engine->SetupBatch(gs, regs.gs.main_offset);
 }
 
 void GeometryPipeline::Reconfigure() {
     ASSERT(!backend || backend->IsEmpty());
 
-    if (state.regs.pipeline.use_gs == PipelineRegs::UseGS::No) {
+    if (regs.pipeline.use_gs == PipelineRegs::UseGS::No) {
         backend = nullptr;
         return;
     }
 
-    ASSERT(state.regs.pipeline.use_gs == PipelineRegs::UseGS::Yes);
-
     // The following assumes that when geometry shader is in use, the shader unit 3 is configured as
     // a geometry shader unit.
     // TODO: what happens if this is not true?
-    ASSERT(state.regs.pipeline.gs_unit_exclusive_configuration == 1);
+    ASSERT(regs.pipeline.gs_unit_exclusive_configuration == 1);
-    ASSERT(state.regs.gs.shader_mode == ShaderRegs::ShaderMode::GS);
+    ASSERT(regs.gs.shader_mode == ShaderRegs::ShaderMode::GS);
+    ASSERT(regs.pipeline.use_gs == PipelineRegs::UseGS::Yes);
 
-    state.gs_unit.ConfigOutput(state.regs.gs);
+    gs_unit.ConfigOutput(regs.gs);
 
-    ASSERT(state.regs.pipeline.vs_outmap_total_minus_1_a ==
+    ASSERT(regs.pipeline.vs_outmap_total_minus_1_a == regs.pipeline.vs_outmap_total_minus_1_b);
-           state.regs.pipeline.vs_outmap_total_minus_1_b);
 
-    switch (state.regs.pipeline.gs_config.mode) {
+    switch (regs.pipeline.gs_config.mode) {
     case PipelineRegs::GSMode::Point:
-        backend = std::make_unique<GeometryPipeline_Point>(state.regs, state.gs_unit);
+        backend = std::make_unique<GeometryPipeline_Point>(regs, gs_unit);
         break;
     case PipelineRegs::GSMode::VariablePrimitive:
-        backend = std::make_unique<GeometryPipeline_VariablePrimitive>(state.regs, state.gs);
+        backend = std::make_unique<GeometryPipeline_VariablePrimitive>(regs, gs);
         break;
     case PipelineRegs::GSMode::FixedPrimitive:
-        backend = std::make_unique<GeometryPipeline_FixedPrimitive>(state.regs, state.gs);
+        backend = std::make_unique<GeometryPipeline_FixedPrimitive>(regs, gs);
         break;
     default:
         UNREACHABLE();
@@ -351,8 +361,9 @@ void GeometryPipeline::Reconfigure() {
 }
 
 bool GeometryPipeline::NeedIndexInput() const {
-    if (!backend)
+    if (!backend) {
         return false;
+    }
     return backend->NeedIndexInput();
 }
 
@@ -360,19 +371,19 @@ void GeometryPipeline::SubmitIndex(unsigned int val) {
     backend->SubmitIndex(val);
 }
 
-void GeometryPipeline::SubmitVertex(const Shader::AttributeBuffer& input) {
+void GeometryPipeline::SubmitVertex(const AttributeBuffer& input) {
     if (!backend) {
         // No backend means the geometry shader is disabled, so we send the vertex shader output
         // directly to the primitive assembler.
         vertex_handler(input);
     } else {
         if (backend->SubmitVertex(input)) {
-            shader_engine->Run(state.gs, state.gs_unit);
+            shader_engine->Run(gs, gs_unit);
 
             // The uniform b15 is set to true after every geometry shader invocation. This is useful
             // for the shader to know if this is the first invocation in a batch, if the program set
             // b15 to false first.
-            state.gs.uniforms.b[15] = true;
+            gs.uniforms.b[15] = true;
         }
     }
 }

src/video_core/pica/geometry_pipeline.h

@@ -6,11 +6,14 @@
 
 #include <memory>
 #include <boost/serialization/export.hpp>
-#include "video_core/shader/shader.h"
+#include "video_core/pica/shader_unit.h"
 
 namespace Pica {
 
-struct State;
+struct RegsInternal;
+struct GeometryShaderUnit;
+struct ShaderSetup;
+class ShaderEngine;
 
 class GeometryPipelineBackend;
 class GeometryPipeline_Point;
@@ -20,17 +23,14 @@ class GeometryPipeline_FixedPrimitive;
 /// A pipeline receiving from vertex shader and sending to geometry shader and primitive assembler
 class GeometryPipeline {
 public:
-    explicit GeometryPipeline(State& state);
+    explicit GeometryPipeline(RegsInternal& regs, GeometryShaderUnit& gs_unit, ShaderSetup& gs);
     ~GeometryPipeline();
 
     /// Sets the handler for receiving vertex outputs from vertex shader
-    void SetVertexHandler(Shader::VertexHandler vertex_handler);
+    void SetVertexHandler(VertexHandler vertex_handler);
 
-    /**
+    /// Setup the geometry shader unit if it is in use
-     * Setup the geometry shader unit if it is in use
+    void Setup(ShaderEngine* shader_engine);
-     * @param shader_engine the shader engine for the geometry shader to run
-     */
-    void Setup(Shader::ShaderEngine* shader_engine);
 
     /// Reconfigures the pipeline according to current register settings
     void Reconfigure();
@@ -42,13 +42,15 @@ public:
     void SubmitIndex(unsigned int val);
 
     /// Submits vertex attributes output from vertex shader
-    void SubmitVertex(const Shader::AttributeBuffer& input);
+    void SubmitVertex(const AttributeBuffer& input);
 
 private:
-    Shader::VertexHandler vertex_handler;
+    VertexHandler vertex_handler;
-    Shader::ShaderEngine* shader_engine;
+    ShaderEngine* shader_engine;
     std::unique_ptr<GeometryPipelineBackend> backend;
-    State& state;
+    RegsInternal& regs;
+    GeometryShaderUnit& gs_unit;
+    ShaderSetup& gs;
 
     template <class Archive>
     void serialize(Archive& ar, const unsigned int version);

src/video_core/pica/output_vertex.cpp

@@ -0,0 +1,50 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "video_core/pica/output_vertex.h"
+#include "video_core/pica/regs_rasterizer.h"
+
+namespace Pica {
+
+OutputVertex::OutputVertex(const RasterizerRegs& regs, const AttributeBuffer& output) {
+    // Attributes can be used without being set in GPUREG_SH_OUTMAP_Oi
+    // Hardware tests have shown that they are initialized to 1 in this case.
+    std::array<f24, 32> vertex_slots_overflow;
+    vertex_slots_overflow.fill(f24::One());
+
+    const u32 num_attributes = regs.vs_output_total & 7;
+    for (std::size_t attrib = 0; attrib < num_attributes; ++attrib) {
+        const auto output_register_map = regs.vs_output_attributes[attrib];
+        vertex_slots_overflow[output_register_map.map_x] = output[attrib][0];
+        vertex_slots_overflow[output_register_map.map_y] = output[attrib][1];
+        vertex_slots_overflow[output_register_map.map_z] = output[attrib][2];
+        vertex_slots_overflow[output_register_map.map_w] = output[attrib][3];
+    }
+
+    // Copy to result
+    std::memcpy(this, vertex_slots_overflow.data(), sizeof(OutputVertex));
+
+    // The hardware takes the absolute and saturates vertex colors, *before* doing interpolation
+    for (u32 i = 0; i < 4; ++i) {
+        const f32 c = std::fabs(color[i].ToFloat32());
+        color[i] = f24::FromFloat32(c < 1.0f ? c : 1.0f);
+    }
+}
+
+#define ASSERT_POS(var, pos)                                                                       \
+    static_assert(offsetof(OutputVertex, var) == pos * sizeof(f24), "Semantic at wrong "           \
+                                                                    "offset.")
+
+ASSERT_POS(pos, RasterizerRegs::VSOutputAttributes::POSITION_X);
+ASSERT_POS(quat, RasterizerRegs::VSOutputAttributes::QUATERNION_X);
+ASSERT_POS(color, RasterizerRegs::VSOutputAttributes::COLOR_R);
+ASSERT_POS(tc0, RasterizerRegs::VSOutputAttributes::TEXCOORD0_U);
+ASSERT_POS(tc1, RasterizerRegs::VSOutputAttributes::TEXCOORD1_U);
+ASSERT_POS(tc0_w, RasterizerRegs::VSOutputAttributes::TEXCOORD0_W);
+ASSERT_POS(view, RasterizerRegs::VSOutputAttributes::VIEW_X);
+ASSERT_POS(tc2, RasterizerRegs::VSOutputAttributes::TEXCOORD2_U);
+
+#undef ASSERT_POS
+
+} // namespace Pica

src/video_core/pica/output_vertex.h

@@ -0,0 +1,48 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/vector_math.h"
+#include "video_core/pica_types.h"
+
+namespace Pica {
+
+struct RasterizerRegs;
+
+using AttributeBuffer = std::array<Common::Vec4<f24>, 16>;
+
+struct OutputVertex {
+    OutputVertex() = default;
+    explicit OutputVertex(const RasterizerRegs& regs, const AttributeBuffer& output);
+
+    Common::Vec4<f24> pos;
+    Common::Vec4<f24> quat;
+    Common::Vec4<f24> color;
+    Common::Vec2<f24> tc0;
+    Common::Vec2<f24> tc1;
+    f24 tc0_w;
+    INSERT_PADDING_WORDS(1);
+    Common::Vec3<f24> view;
+    INSERT_PADDING_WORDS(1);
+    Common::Vec2<f24> tc2;
+
+private:
+    template <class Archive>
+    void serialize(Archive& ar, const u32) {
+        ar& pos;
+        ar& quat;
+        ar& color;
+        ar& tc0;
+        ar& tc1;
+        ar& tc0_w;
+        ar& view;
+        ar& tc2;
+    }
+    friend class boost::serialization::access;
+};
+static_assert(std::is_trivial_v<OutputVertex>, "Structure is not POD");
+static_assert(sizeof(OutputVertex) == 24 * sizeof(f32), "OutputVertex has invalid size");
+
+} // namespace Pica

src/video_core/pica/packed_attribute.h

@@ -0,0 +1,74 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <boost/serialization/binary_object.hpp>
+
+#include "common/vector_math.h"
+#include "video_core/pica_types.h"
+
+namespace Pica {
+
+/**
+ * Uniforms and fixed attributes are written in a packed format such that four float24 values are
+ * encoded in three 32-bit numbers. Uniforms can also encode four float32 values in four 32-bit
+ * numbers. We write to internal memory once a full vector is written.
+ */
+struct PackedAttribute {
+    std::array<u32, 4> buffer{};
+    u32 index{};
+
+    /// Places a word to the queue and returns true if the queue becomes full.
+    constexpr bool Push(u32 word, bool is_float32 = false) {
+        buffer[index++] = word;
+        return (index >= 4 && is_float32) || (index >= 3 && !is_float32);
+    }
+
+    /// Resets the queue discarding previous entries.
+    constexpr void Reset() {
+        index = 0;
+    }
+
+    /// Returns the queue contents with either float24 or float32 interpretation.
+    constexpr Common::Vec4<f24> Get(bool is_float32 = false) {
+        Reset();
+        if (is_float32) {
+            return AsFloat32();
+        } else {
+            return AsFloat24();
+        }
+    }
+
+private:
+    /// Decodes the queue contents with float24 transfer mode.
+    constexpr Common::Vec4<f24> AsFloat24() const {
+        const u32 x = buffer[2] & 0xFFFFFF;
+        const u32 y = ((buffer[1] & 0xFFFF) << 8) | ((buffer[2] >> 24) & 0xFF);
+        const u32 z = ((buffer[0] & 0xFF) << 16) | ((buffer[1] >> 16) & 0xFFFF);
+        const u32 w = buffer[0] >> 8;
+        return Common::Vec4<f24>{f24::FromRaw(x), f24::FromRaw(y), f24::FromRaw(z),
+                                 f24::FromRaw(w)};
+    }
+
+    /// Decodes the queue contents with float32 transfer mode.
+    constexpr Common::Vec4<f24> AsFloat32() const {
+        Common::Vec4<f24> uniform;
+        for (u32 i = 0; i < 4; i++) {
+            const f32 buffer_value = std::bit_cast<f32>(buffer[i]);
+            uniform[3 - i] = f24::FromFloat32(buffer_value);
+        }
+        return uniform;
+    }
+
+private:
+    template <class Archive>
+    void serialize(Archive& ar, const u32) {
+        ar& buffer;
+        ar& index;
+    }
+    friend class boost::serialization::access;
+};
+
+} // namespace Pica

src/video_core/pica/pica_core.cpp

@@ -0,0 +1,592 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/arch.h"
+#include "common/archives.h"
+#include "common/microprofile.h"
+#include "common/scope_exit.h"
+#include "common/settings.h"
+#include "core/core.h"
+#include "core/memory.h"
+#include "video_core/debug_utils/debug_utils.h"
+#include "video_core/pica/pica_core.h"
+#include "video_core/pica/vertex_loader.h"
+#include "video_core/rasterizer_interface.h"
+#include "video_core/shader/shader.h"
+
+namespace Pica {
+
+MICROPROFILE_DEFINE(GPU_Drawing, "GPU", "Drawing", MP_RGB(50, 50, 240));
+
+using namespace DebugUtils;
+
+union CommandHeader {
+    u32 hex;
+    BitField<0, 16, u32> cmd_id;
+    BitField<16, 4, u32> parameter_mask;
+    BitField<20, 8, u32> extra_data_length;
+    BitField<31, 1, u32> group_commands;
+};
+static_assert(sizeof(CommandHeader) == sizeof(u32), "CommandHeader has incorrect size!");
+
+PicaCore::PicaCore(Memory::MemorySystem& memory_, DebugContext& debug_context_)
+    : memory{memory_}, debug_context{debug_context_}, geometry_pipeline{regs.internal, gs_unit,
+                                                                        gs_setup},
+      shader_engine{CreateEngine(Settings::values.use_shader_jit.GetValue())} {
+    SetFramebufferDefaults();
+
+    const auto submit_vertex = [this](const AttributeBuffer& buffer) {
+        const auto add_triangle = [this](const OutputVertex& v0, const OutputVertex& v1,
+                                         const OutputVertex& v2) {
+            rasterizer->AddTriangle(v0, v1, v2);
+        };
+        const auto vertex = OutputVertex(regs.internal.rasterizer, buffer);
+        primitive_assembler.SubmitVertex(vertex, add_triangle);
+    };
+
+    gs_unit.SetVertexHandlers(submit_vertex, [this]() { primitive_assembler.SetWinding(); });
+    geometry_pipeline.SetVertexHandler(submit_vertex);
+
+    primitive_assembler.Reconfigure(PipelineRegs::TriangleTopology::List);
+}
+
+PicaCore::~PicaCore() = default;
+
+void PicaCore::SetFramebufferDefaults() {
+    auto& framebuffer_top = regs.framebuffer_config[0];
+    auto& framebuffer_sub = regs.framebuffer_config[1];
+
+    // Set framebuffer defaults from nn::gx::Initialize
+    framebuffer_top.address_left1 = 0x181E6000;
+    framebuffer_top.address_left2 = 0x1822C800;
+    framebuffer_top.address_right1 = 0x18273000;
+    framebuffer_top.address_right2 = 0x182B9800;
+    framebuffer_sub.address_left1 = 0x1848F000;
+    framebuffer_sub.address_left2 = 0x184C7800;
+
+    framebuffer_top.width.Assign(240);
+    framebuffer_top.height.Assign(400);
+    framebuffer_top.stride = 3 * 240;
+    framebuffer_top.color_format.Assign(PixelFormat::RGB8);
+    framebuffer_top.active_fb = 0;
+
+    framebuffer_sub.width.Assign(240);
+    framebuffer_sub.height.Assign(320);
+    framebuffer_sub.stride = 3 * 240;
+    framebuffer_sub.color_format.Assign(PixelFormat::RGB8);
+    framebuffer_sub.active_fb = 0;
+}
+
+void PicaCore::BindRasterizer(VideoCore::RasterizerInterface* rasterizer) {
+    this->rasterizer = rasterizer;
+}
+
+void PicaCore::SetInterruptHandler(Service::GSP::InterruptHandler& signal_interrupt) {
+    this->signal_interrupt = signal_interrupt;
+}
+
+void PicaCore::ProcessCmdList(PAddr list, u32 size) {
+    // Initialize command list tracking.
+    const u8* head = memory.GetPhysicalPointer(list);
+    cmd_list.Reset(list, head, size);
+
+    while (cmd_list.current_index < cmd_list.length) {
+        // Align read pointer to 8 bytes
+        if (cmd_list.current_index % 2 != 0) {
+            cmd_list.current_index++;
+        }
+
+        // Read the header and the value to write.
+        const u32 value = cmd_list.head[cmd_list.current_index++];
+        const CommandHeader header{cmd_list.head[cmd_list.current_index++]};
+
+        // Write to the requested PICA register.
+        WriteInternalReg(header.cmd_id, value, header.parameter_mask);
+
+        // Write any extra paramters as well.
+        for (u32 i = 0; i < header.extra_data_length; ++i) {
+            const u32 cmd = header.cmd_id + (header.group_commands ? i + 1 : 0);
+            const u32 extra_value = cmd_list.head[cmd_list.current_index++];
+            WriteInternalReg(cmd, extra_value, header.parameter_mask);
+        }
+    }
+}
+
+void PicaCore::WriteInternalReg(u32 id, u32 value, u32 mask) {
+    if (id >= RegsInternal::NUM_REGS) {
+        LOG_ERROR(
+            HW_GPU,
+            "Commandlist tried to write to invalid register 0x{:03X} (value: {:08X}, mask: {:X})",
+            id, value, mask);
+        return;
+    }
+
+    // Expand a 4-bit mask to 4-byte mask, e.g. 0b0101 -> 0x00FF00FF
+    constexpr std::array<u32, 16> ExpandBitsToBytes = {
+        0x00000000, 0x000000ff, 0x0000ff00, 0x0000ffff, 0x00ff0000, 0x00ff00ff,
+        0x00ffff00, 0x00ffffff, 0xff000000, 0xff0000ff, 0xff00ff00, 0xff00ffff,
+        0xffff0000, 0xffff00ff, 0xffffff00, 0xffffffff,
+    };
+
+    // TODO: Figure out how register masking acts on e.g. vs.uniform_setup.set_value
+    const u32 old_value = regs.internal.reg_array[id];
+    const u32 write_mask = ExpandBitsToBytes[mask];
+    regs.internal.reg_array[id] = (old_value & ~write_mask) | (value & write_mask);
+
+    // Track register write.
+    DebugUtils::OnPicaRegWrite(id, mask, regs.internal.reg_array[id]);
+
+    // Track events.
+    debug_context.OnEvent(DebugContext::Event::PicaCommandLoaded, &id);
+    SCOPE_EXIT({ debug_context.OnEvent(DebugContext::Event::PicaCommandProcessed, &id); });
+
+    switch (id) {
+    // Trigger IRQ
+    case PICA_REG_INDEX(trigger_irq):
+        signal_interrupt(Service::GSP::InterruptId::P3D);
+        break;
+
+    case PICA_REG_INDEX(pipeline.triangle_topology):
+        primitive_assembler.Reconfigure(regs.internal.pipeline.triangle_topology);
+        break;
+
+    case PICA_REG_INDEX(pipeline.restart_primitive):
+        primitive_assembler.Reset();
+        break;
+
+    case PICA_REG_INDEX(pipeline.vs_default_attributes_setup.index):
+        immediate.Reset();
+        break;
+
+    // Load default vertex input attributes
+    case PICA_REG_INDEX(pipeline.vs_default_attributes_setup.set_value[0]):
+    case PICA_REG_INDEX(pipeline.vs_default_attributes_setup.set_value[1]):
+    case PICA_REG_INDEX(pipeline.vs_default_attributes_setup.set_value[2]):
+        SubmitImmediate(value);
+        break;
+
+    case PICA_REG_INDEX(pipeline.gpu_mode):
+        // This register likely just enables vertex processing and doesn't need any special handling
+        break;
+
+    case PICA_REG_INDEX(pipeline.command_buffer.trigger[0]):
+    case PICA_REG_INDEX(pipeline.command_buffer.trigger[1]): {
+        const u32 index = static_cast<u32>(id - PICA_REG_INDEX(pipeline.command_buffer.trigger[0]));
+        const PAddr addr = regs.internal.pipeline.command_buffer.GetPhysicalAddress(index);
+        const u32 size = regs.internal.pipeline.command_buffer.GetSize(index);
+        const u8* head = memory.GetPhysicalPointer(addr);
+        cmd_list.Reset(addr, head, size);
+        break;
+    }
+
+    // It seems like these trigger vertex rendering
+    case PICA_REG_INDEX(pipeline.trigger_draw):
+    case PICA_REG_INDEX(pipeline.trigger_draw_indexed): {
+        const bool is_indexed = (id == PICA_REG_INDEX(pipeline.trigger_draw_indexed));
+        DrawArrays(is_indexed);
+        break;
+    }
+
+    case PICA_REG_INDEX(gs.bool_uniforms):
+        gs_setup.WriteUniformBoolReg(regs.internal.gs.bool_uniforms.Value());
+        break;
+
+    case PICA_REG_INDEX(gs.int_uniforms[0]):
+    case PICA_REG_INDEX(gs.int_uniforms[1]):
+    case PICA_REG_INDEX(gs.int_uniforms[2]):
+    case PICA_REG_INDEX(gs.int_uniforms[3]): {
+        const u32 index = (id - PICA_REG_INDEX(gs.int_uniforms[0]));
+        gs_setup.WriteUniformIntReg(index, regs.internal.gs.GetIntUniform(index));
+        break;
+    }
+
+    case PICA_REG_INDEX(gs.uniform_setup.set_value[0]):
+    case PICA_REG_INDEX(gs.uniform_setup.set_value[1]):
+    case PICA_REG_INDEX(gs.uniform_setup.set_value[2]):
+    case PICA_REG_INDEX(gs.uniform_setup.set_value[3]):
+    case PICA_REG_INDEX(gs.uniform_setup.set_value[4]):
+    case PICA_REG_INDEX(gs.uniform_setup.set_value[5]):
+    case PICA_REG_INDEX(gs.uniform_setup.set_value[6]):
+    case PICA_REG_INDEX(gs.uniform_setup.set_value[7]): {
+        gs_setup.WriteUniformFloatReg(regs.internal.gs, value);
+        break;
+    }
+
+    case PICA_REG_INDEX(gs.program.set_word[0]):
+    case PICA_REG_INDEX(gs.program.set_word[1]):
+    case PICA_REG_INDEX(gs.program.set_word[2]):
+    case PICA_REG_INDEX(gs.program.set_word[3]):
+    case PICA_REG_INDEX(gs.program.set_word[4]):
+    case PICA_REG_INDEX(gs.program.set_word[5]):
+    case PICA_REG_INDEX(gs.program.set_word[6]):
+    case PICA_REG_INDEX(gs.program.set_word[7]): {
+        u32& offset = regs.internal.gs.program.offset;
+        if (offset >= 4096) {
+            LOG_ERROR(HW_GPU, "Invalid GS program offset {}", offset);
+        } else {
+            gs_setup.program_code[offset] = value;
+            gs_setup.MarkProgramCodeDirty();
+            offset++;
+        }
+        break;
+    }
+
+    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[0]):
+    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[1]):
+    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[2]):
+    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[3]):
+    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[4]):
+    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[5]):
+    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[6]):
+    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[7]): {
+        u32& offset = regs.internal.gs.swizzle_patterns.offset;
+        if (offset >= gs_setup.swizzle_data.size()) {
+            LOG_ERROR(HW_GPU, "Invalid GS swizzle pattern offset {}", offset);
+        } else {
+            gs_setup.swizzle_data[offset] = value;
+            gs_setup.MarkSwizzleDataDirty();
+            offset++;
+        }
+        break;
+    }
+
+    case PICA_REG_INDEX(vs.bool_uniforms):
+        vs_setup.WriteUniformBoolReg(regs.internal.vs.bool_uniforms.Value());
+        break;
+
+    case PICA_REG_INDEX(vs.int_uniforms[0]):
+    case PICA_REG_INDEX(vs.int_uniforms[1]):
+    case PICA_REG_INDEX(vs.int_uniforms[2]):
+    case PICA_REG_INDEX(vs.int_uniforms[3]): {
+        const u32 index = (id - PICA_REG_INDEX(vs.int_uniforms[0]));
+        vs_setup.WriteUniformIntReg(index, regs.internal.vs.GetIntUniform(index));
+        break;
+    }
+
+    case PICA_REG_INDEX(vs.uniform_setup.set_value[0]):
+    case PICA_REG_INDEX(vs.uniform_setup.set_value[1]):
+    case PICA_REG_INDEX(vs.uniform_setup.set_value[2]):
+    case PICA_REG_INDEX(vs.uniform_setup.set_value[3]):
+    case PICA_REG_INDEX(vs.uniform_setup.set_value[4]):
+    case PICA_REG_INDEX(vs.uniform_setup.set_value[5]):
+    case PICA_REG_INDEX(vs.uniform_setup.set_value[6]):
+    case PICA_REG_INDEX(vs.uniform_setup.set_value[7]): {
+        vs_setup.WriteUniformFloatReg(regs.internal.vs, value);
+        break;
+    }
+
+    case PICA_REG_INDEX(vs.program.set_word[0]):
+    case PICA_REG_INDEX(vs.program.set_word[1]):
+    case PICA_REG_INDEX(vs.program.set_word[2]):
+    case PICA_REG_INDEX(vs.program.set_word[3]):
+    case PICA_REG_INDEX(vs.program.set_word[4]):
+    case PICA_REG_INDEX(vs.program.set_word[5]):
+    case PICA_REG_INDEX(vs.program.set_word[6]):
+    case PICA_REG_INDEX(vs.program.set_word[7]): {
+        u32& offset = regs.internal.vs.program.offset;
+        if (offset >= 512) {
+            LOG_ERROR(HW_GPU, "Invalid VS program offset {}", offset);
+        } else {
+            vs_setup.program_code[offset] = value;
+            vs_setup.MarkProgramCodeDirty();
+            if (!regs.internal.pipeline.gs_unit_exclusive_configuration) {
+                gs_setup.program_code[offset] = value;
+                gs_setup.MarkProgramCodeDirty();
+            }
+            offset++;
+        }
+        break;
+    }
+
+    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[0]):
+    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[1]):
+    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[2]):
+    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[3]):
+    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[4]):
+    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[5]):
+    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[6]):
+    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[7]): {
+        u32& offset = regs.internal.vs.swizzle_patterns.offset;
+        if (offset >= vs_setup.swizzle_data.size()) {
+            LOG_ERROR(HW_GPU, "Invalid VS swizzle pattern offset {}", offset);
+        } else {
+            vs_setup.swizzle_data[offset] = value;
+            vs_setup.MarkSwizzleDataDirty();
+            if (!regs.internal.pipeline.gs_unit_exclusive_configuration) {
+                gs_setup.swizzle_data[offset] = value;
+                gs_setup.MarkSwizzleDataDirty();
+            }
+            offset++;
+        }
+        break;
+    }
+
+    case PICA_REG_INDEX(lighting.lut_data[0]):
+    case PICA_REG_INDEX(lighting.lut_data[1]):
+    case PICA_REG_INDEX(lighting.lut_data[2]):
+    case PICA_REG_INDEX(lighting.lut_data[3]):
+    case PICA_REG_INDEX(lighting.lut_data[4]):
+    case PICA_REG_INDEX(lighting.lut_data[5]):
+    case PICA_REG_INDEX(lighting.lut_data[6]):
+    case PICA_REG_INDEX(lighting.lut_data[7]): {
+        auto& lut_config = regs.internal.lighting.lut_config;
+        ASSERT_MSG(lut_config.index < 256, "lut_config.index exceeded maximum value of 255!");
+
+        lighting.luts[lut_config.type][lut_config.index].raw = value;
+        lut_config.index.Assign(lut_config.index + 1);
+        break;
+    }
+
+    case PICA_REG_INDEX(texturing.fog_lut_data[0]):
+    case PICA_REG_INDEX(texturing.fog_lut_data[1]):
+    case PICA_REG_INDEX(texturing.fog_lut_data[2]):
+    case PICA_REG_INDEX(texturing.fog_lut_data[3]):
+    case PICA_REG_INDEX(texturing.fog_lut_data[4]):
+    case PICA_REG_INDEX(texturing.fog_lut_data[5]):
+    case PICA_REG_INDEX(texturing.fog_lut_data[6]):
+    case PICA_REG_INDEX(texturing.fog_lut_data[7]): {
+        fog.lut[regs.internal.texturing.fog_lut_offset % 128].raw = value;
+        regs.internal.texturing.fog_lut_offset.Assign(regs.internal.texturing.fog_lut_offset + 1);
+        break;
+    }
+
+    case PICA_REG_INDEX(texturing.proctex_lut_data[0]):
+    case PICA_REG_INDEX(texturing.proctex_lut_data[1]):
+    case PICA_REG_INDEX(texturing.proctex_lut_data[2]):
+    case PICA_REG_INDEX(texturing.proctex_lut_data[3]):
+    case PICA_REG_INDEX(texturing.proctex_lut_data[4]):
+    case PICA_REG_INDEX(texturing.proctex_lut_data[5]):
+    case PICA_REG_INDEX(texturing.proctex_lut_data[6]):
+    case PICA_REG_INDEX(texturing.proctex_lut_data[7]): {
+        auto& index = regs.internal.texturing.proctex_lut_config.index;
+
+        switch (regs.internal.texturing.proctex_lut_config.ref_table.Value()) {
+        case TexturingRegs::ProcTexLutTable::Noise:
+            proctex.noise_table[index % proctex.noise_table.size()].raw = value;
+            break;
+        case TexturingRegs::ProcTexLutTable::ColorMap:
+            proctex.color_map_table[index % proctex.color_map_table.size()].raw = value;
+            break;
+        case TexturingRegs::ProcTexLutTable::AlphaMap:
+            proctex.alpha_map_table[index % proctex.alpha_map_table.size()].raw = value;
+            break;
+        case TexturingRegs::ProcTexLutTable::Color:
+            proctex.color_table[index % proctex.color_table.size()].raw = value;
+            break;
+        case TexturingRegs::ProcTexLutTable::ColorDiff:
+            proctex.color_diff_table[index % proctex.color_diff_table.size()].raw = value;
+            break;
+        }
+        index.Assign(index + 1);
+        break;
+    }
+    default:
+        break;
+    }
+
+    // Notify the rasterizer an internal register was updated.
+    rasterizer->NotifyPicaRegisterChanged(id);
+}
+
+void PicaCore::SubmitImmediate(u32 value) {
+    // Push to word to the queue. This returns true when a full attribute is formed.
+    if (!immediate.queue.Push(value)) {
+        return;
+    }
+
+    constexpr size_t IMMEDIATE_MODE_INDEX = 0xF;
+
+    auto& setup = regs.internal.pipeline.vs_default_attributes_setup;
+    if (setup.index > IMMEDIATE_MODE_INDEX) {
+        LOG_ERROR(HW_GPU, "Invalid VS default attribute index {}", setup.index);
+        return;
+    }
+
+    // Retrieve the attribute and place it in the default attribute buffer.
+    const auto attribute = immediate.queue.Get();
+    if (setup.index < IMMEDIATE_MODE_INDEX) {
+        input_default_attributes[setup.index] = attribute;
+        setup.index++;
+        return;
+    }
+
+    // When index is 0xF the attribute is used for immediate mode drawing.
+    immediate.input_vertex[immediate.current_attribute] = attribute;
+    if (immediate.current_attribute < regs.internal.pipeline.max_input_attrib_index) {
+        immediate.current_attribute++;
+        return;
+    }
+
+    // We formed a vertex, flush.
+    DrawImmediate();
+}
+
+void PicaCore::DrawImmediate() {
+    // Compile the vertex shader.
+    shader_engine->SetupBatch(vs_setup, regs.internal.vs.main_offset);
+
+    // Track vertex in the debug recorder.
+    debug_context.OnEvent(DebugContext::Event::VertexShaderInvocation,
+                          std::addressof(immediate.input_vertex));
+    SCOPE_EXIT({ debug_context.OnEvent(DebugContext::Event::FinishedPrimitiveBatch, nullptr); });
+
+    ShaderUnit shader_unit;
+    AttributeBuffer output{};
+
+    // Invoke the vertex shader for the vertex.
+    shader_unit.LoadInput(regs.internal.vs, immediate.input_vertex);
+    shader_engine->Run(vs_setup, shader_unit);
+    shader_unit.WriteOutput(regs.internal.vs, output);
+
+    // Reconfigure geometry pipeline if needed.
+    if (immediate.reset_geometry_pipeline) {
+        geometry_pipeline.Reconfigure();
+        immediate.reset_geometry_pipeline = false;
+    }
+
+    // Send to geometry pipeline.
+    ASSERT(!geometry_pipeline.NeedIndexInput());
+    geometry_pipeline.Setup(shader_engine.get());
+    geometry_pipeline.SubmitVertex(output);
+
+    // Flush the immediate triangle.
+    rasterizer->DrawTriangles();
+    immediate.current_attribute = 0;
+}
+
+void PicaCore::DrawArrays(bool is_indexed) {
+    MICROPROFILE_SCOPE(GPU_Drawing);
+
+    // Track vertex in the debug recorder.
+    debug_context.OnEvent(DebugContext::Event::IncomingPrimitiveBatch, nullptr);
+    SCOPE_EXIT({ debug_context.OnEvent(DebugContext::Event::FinishedPrimitiveBatch, nullptr); });
+
+    const bool accelerate_draw = [this] {
+        // Geometry shaders cannot be accelerated due to register preservation.
+        if (regs.internal.pipeline.use_gs == PipelineRegs::UseGS::Yes) {
+            return false;
+        }
+
+        // TODO (wwylele): for Strip/Fan topology, if the primitive assember is not restarted
+        // after this draw call, the buffered vertex from this draw should "leak" to the next
+        // draw, in which case we should buffer the vertex into the software primitive assember,
+        // or disable accelerate draw completely. However, there is not game found yet that does
+        // this, so this is left unimplemented for now. Revisit this when an issue is found in
+        // games.
+
+        bool accelerate_draw = Settings::values.use_hw_shader && primitive_assembler.IsEmpty();
+        const auto topology = primitive_assembler.GetTopology();
+        if (topology == PipelineRegs::TriangleTopology::Shader ||
+            topology == PipelineRegs::TriangleTopology::List) {
+            accelerate_draw = accelerate_draw && (regs.internal.pipeline.num_vertices % 3) == 0;
+        }
+        return accelerate_draw;
+    }();
+
+    // Attempt to use hardware vertex shaders if possible.
+    if (accelerate_draw && rasterizer->AccelerateDrawBatch(is_indexed)) {
+        return;
+    }
+
+    // We cannot accelerate the draw, so load and execute the vertex shader for each vertex.
+    LoadVertices(is_indexed);
+
+    // Draw emitted triangles.
+    rasterizer->DrawTriangles();
+}
+
+void PicaCore::LoadVertices(bool is_indexed) {
+    // Read and validate vertex information from the loaders
+    const auto& pipeline = regs.internal.pipeline;
+    const PAddr base_address = pipeline.vertex_attributes.GetPhysicalBaseAddress();
+    const auto loader = VertexLoader(memory, pipeline);
+    regs.internal.rasterizer.ValidateSemantics();
+
+    // Locate index buffer.
+    const auto& index_info = pipeline.index_array;
+    const u8* index_address_8 = memory.GetPhysicalPointer(base_address + index_info.offset);
+    const u16* index_address_16 = reinterpret_cast<const u16*>(index_address_8);
+    const bool index_u16 = index_info.format != 0;
+
+    // Simple circular-replacement vertex cache
+    const std::size_t VERTEX_CACHE_SIZE = 64;
+    std::array<bool, VERTEX_CACHE_SIZE> vertex_cache_valid{};
+    std::array<u16, VERTEX_CACHE_SIZE> vertex_cache_ids;
+    std::array<AttributeBuffer, VERTEX_CACHE_SIZE> vertex_cache;
+    u32 vertex_cache_pos = 0;
+
+    // Compile the vertex shader for this batch.
+    ShaderUnit shader_unit;
+    AttributeBuffer vs_output;
+    shader_engine->SetupBatch(vs_setup, regs.internal.vs.main_offset);
+
+    // Setup geometry pipeline in case we are using a geometry shader.
+    geometry_pipeline.Reconfigure();
+    geometry_pipeline.Setup(shader_engine.get());
+    ASSERT(!geometry_pipeline.NeedIndexInput() || is_indexed);
+
+    for (u32 index = 0; index < pipeline.num_vertices; ++index) {
+        // Indexed rendering doesn't use the start offset
+        const u32 vertex = is_indexed
+                               ? (index_u16 ? index_address_16[index] : index_address_8[index])
+                               : (index + pipeline.vertex_offset);
+
+        bool vertex_cache_hit = false;
+        if (is_indexed) {
+            if (geometry_pipeline.NeedIndexInput()) {
+                geometry_pipeline.SubmitIndex(vertex);
+                continue;
+            }
+
+            for (u32 i = 0; i < VERTEX_CACHE_SIZE; ++i) {
+                if (vertex_cache_valid[i] && vertex == vertex_cache_ids[i]) {
+                    vs_output = vertex_cache[i];
+                    vertex_cache_hit = true;
+                    break;
+                }
+            }
+        }
+
+        if (!vertex_cache_hit) {
+            // Initialize data for the current vertex
+            AttributeBuffer input;
+            loader.LoadVertex(base_address, index, vertex, input, input_default_attributes);
+
+            // Record vertex processing to the debugger.
+            debug_context.OnEvent(DebugContext::Event::VertexShaderInvocation,
+                                  std::addressof(input));
+
+            // Invoke the vertex shader for this vertex.
+            shader_unit.LoadInput(regs.internal.vs, input);
+            shader_engine->Run(vs_setup, shader_unit);
+            shader_unit.WriteOutput(regs.internal.vs, vs_output);
+
+            // Cache the vertex when doing indexed rendering.
+            if (is_indexed) {
+                vertex_cache[vertex_cache_pos] = vs_output;
+                vertex_cache_valid[vertex_cache_pos] = true;
+                vertex_cache_ids[vertex_cache_pos] = vertex;
+                vertex_cache_pos = (vertex_cache_pos + 1) % VERTEX_CACHE_SIZE;
+            }
+        }
+
+        // Send to geometry pipeline
+        geometry_pipeline.SubmitVertex(vs_output);
+    }
+}
+
+template <class Archive>
+void PicaCore::CommandList::serialize(Archive& ar, const u32 file_version) {
+    ar& addr;
+    ar& length;
+    ar& current_index;
+    if (Archive::is_loading::value) {
+        const u8* ptr = Core::System::GetInstance().Memory().GetPhysicalPointer(addr);
+        head = reinterpret_cast<const u32*>(ptr);
+    }
+}
+
+SERIALIZE_IMPL(PicaCore::CommandList)
+
+} // namespace Pica

src/video_core/pica/pica_core.h

@@ -0,0 +1,287 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "core/hle/service/gsp/gsp_interrupt.h"
+#include "video_core/pica/geometry_pipeline.h"
+#include "video_core/pica/packed_attribute.h"
+#include "video_core/pica/primitive_assembly.h"
+#include "video_core/pica/regs_external.h"
+#include "video_core/pica/regs_internal.h"
+#include "video_core/pica/regs_lcd.h"
+#include "video_core/pica/shader_setup.h"
+#include "video_core/pica/shader_unit.h"
+
+namespace Memory {
+class MemorySystem;
+}
+
+namespace VideoCore {
+class RasterizerInterface;
+}
+
+namespace Pica {
+
+class DebugContext;
+class ShaderEngine;
+
+class PicaCore {
+public:
+    explicit PicaCore(Memory::MemorySystem& memory, DebugContext& debug_context_);
+    ~PicaCore();
+
+    void BindRasterizer(VideoCore::RasterizerInterface* rasterizer);
+
+    void SetInterruptHandler(Service::GSP::InterruptHandler& signal_interrupt);
+
+    void ProcessCmdList(PAddr list, u32 size);
+
+private:
+    void SetFramebufferDefaults();
+
+    void WriteInternalReg(u32 id, u32 value, u32 mask);
+
+    void SubmitImmediate(u32 data);
+
+    void DrawImmediate();
+
+    void DrawArrays(bool is_indexed);
+
+    void LoadVertices(bool is_indexed);
+
+public:
+    union Regs {
+        static constexpr size_t NUM_REGS = 0x732;
+
+        struct {
+            u32 hardware_id;
+            INSERT_PADDING_WORDS(0x3);
+            MemoryFillConfig memory_fill_config[2];
+            u32 vram_bank_control;
+            u32 gpu_busy;
+            INSERT_PADDING_WORDS(0x22);
+            u32 backlight_control;
+            INSERT_PADDING_WORDS(0xCF);
+            FramebufferConfig framebuffer_config[2];
+            INSERT_PADDING_WORDS(0x180);
+            DisplayTransferConfig display_transfer_config;
+            INSERT_PADDING_WORDS(0xF5);
+            RegsInternal internal;
+        };
+        std::array<u32, NUM_REGS> reg_array;
+    };
+    static_assert(sizeof(Regs) == Regs::NUM_REGS * sizeof(u32));
+
+    struct CommandList {
+        PAddr addr;
+        const u32* head;
+        u32 current_index;
+        u32 length;
+
+        void Reset(PAddr addr, const u8* head, u32 size) {
+            this->addr = addr;
+            this->head = reinterpret_cast<const u32*>(head);
+            this->length = size / sizeof(u32);
+            current_index = 0;
+        }
+
+    private:
+        friend class boost::serialization::access;
+        template <class Archive>
+        void serialize(Archive& ar, const u32 file_version);
+    };
+
+    struct ImmediateModeState {
+        AttributeBuffer input_vertex{};
+        u32 current_attribute{};
+        bool reset_geometry_pipeline{true};
+        PackedAttribute queue;
+
+        void Reset() {
+            current_attribute = 0;
+            reset_geometry_pipeline = true;
+            queue.Reset();
+        }
+
+    private:
+        friend class boost::serialization::access;
+        template <class Archive>
+        void serialize(Archive& ar, const u32 file_version) {
+            ar& input_vertex;
+            ar& current_attribute;
+            ar& reset_geometry_pipeline;
+            ar& queue;
+        }
+    };
+
+    struct ProcTex {
+        union ValueEntry {
+            u32 raw;
+
+            // LUT value, encoded as 12-bit fixed point, with 12 fraction bits
+            BitField<0, 12, u32> value; // 0.0.12 fixed point
+
+            // Difference between two entry values. Used for efficient interpolation.
+            // 0.0.12 fixed point with two's complement. The range is [-0.5, 0.5).
+            // Note: the type of this is different from the one of lighting LUT
+            BitField<12, 12, s32> difference;
+
+            f32 ToFloat() const {
+                return static_cast<f32>(value) / 4095.f;
+            }
+
+            f32 DiffToFloat() const {
+                return static_cast<f32>(difference) / 4095.f;
+            }
+        };
+
+        union ColorEntry {
+            u32 raw;
+            BitField<0, 8, u32> r;
+            BitField<8, 8, u32> g;
+            BitField<16, 8, u32> b;
+            BitField<24, 8, u32> a;
+
+            Common::Vec4<u8> ToVector() const {
+                return {static_cast<u8>(r), static_cast<u8>(g), static_cast<u8>(b),
+                        static_cast<u8>(a)};
+            }
+        };
+
+        union ColorDifferenceEntry {
+            u32 raw;
+            BitField<0, 8, s32> r; // half of the difference between two ColorEntry
+            BitField<8, 8, s32> g;
+            BitField<16, 8, s32> b;
+            BitField<24, 8, s32> a;
+
+            Common::Vec4<s32> ToVector() const {
+                return Common::Vec4<s32>{r, g, b, a} * 2;
+            }
+        };
+
+        std::array<ValueEntry, 128> noise_table;
+        std::array<ValueEntry, 128> color_map_table;
+        std::array<ValueEntry, 128> alpha_map_table;
+        std::array<ColorEntry, 256> color_table;
+        std::array<ColorDifferenceEntry, 256> color_diff_table;
+
+    private:
+        friend class boost::serialization::access;
+        template <class Archive>
+        void serialize(Archive& ar, const u32 file_version) {
+            ar& boost::serialization::make_binary_object(this, sizeof(ProcTex));
+        }
+    };
+
+    struct Lighting {
+        union LutEntry {
+            // Used for raw access
+            u32 raw;
+
+            // LUT value, encoded as 12-bit fixed point, with 12 fraction bits
+            BitField<0, 12, u32> value; // 0.0.12 fixed point
+
+            // Used for efficient interpolation.
+            BitField<12, 11, u32> difference; // 0.0.11 fixed point
+            BitField<23, 1, u32> neg_difference;
+
+            f32 ToFloat() const {
+                return static_cast<f32>(value) / 4095.f;
+            }
+
+            f32 DiffToFloat() const {
+                const f32 diff = static_cast<f32>(difference) / 2047.f;
+                return neg_difference ? -diff : diff;
+            }
+
+            template <class Archive>
+            void serialize(Archive& ar, const u32 file_version) {
+                ar& raw;
+            }
+        };
+
+        std::array<std::array<LutEntry, 256>, 24> luts;
+
+    private:
+        friend class boost::serialization::access;
+        template <class Archive>
+        void serialize(Archive& ar, const u32 file_version) {
+            ar& boost::serialization::make_binary_object(this, sizeof(Lighting));
+        }
+    };
+
+    struct Fog {
+        union LutEntry {
+            // Used for raw access
+            u32 raw;
+
+            BitField<0, 13, s32> difference; // 1.1.11 fixed point
+            BitField<13, 11, u32> value;     // 0.0.11 fixed point
+
+            f32 ToFloat() const {
+                return static_cast<f32>(value) / 2047.0f;
+            }
+
+            f32 DiffToFloat() const {
+                return static_cast<f32>(difference) / 2047.0f;
+            }
+        };
+
+        std::array<LutEntry, 128> lut;
+
+    private:
+        friend class boost::serialization::access;
+        template <class Archive>
+        void serialize(Archive& ar, const u32 file_version) {
+            ar& boost::serialization::make_binary_object(this, sizeof(Fog));
+        }
+    };
+
+    RegsLcd regs_lcd{};
+    Regs regs{};
+    // TODO: Move these to a separate shader scheduler class
+    GeometryShaderUnit gs_unit;
+    ShaderSetup vs_setup;
+    ShaderSetup gs_setup;
+    ProcTex proctex{};
+    Lighting lighting{};
+    Fog fog{};
+    AttributeBuffer input_default_attributes{};
+    ImmediateModeState immediate{};
+
+private:
+    friend class boost::serialization::access;
+    template <class Archive>
+    void serialize(Archive& ar, const u32 file_version) {
+        ar& regs_lcd;
+        ar& regs.reg_array;
+        ar& gs_unit;
+        ar& vs_setup;
+        ar& gs_setup;
+        ar& proctex;
+        ar& lighting;
+        ar& fog;
+        ar& input_default_attributes;
+        ar& immediate;
+        ar& geometry_pipeline;
+        ar& primitive_assembler;
+        ar& cmd_list;
+    }
+
+private:
+    Memory::MemorySystem& memory;
+    VideoCore::RasterizerInterface* rasterizer;
+    DebugContext& debug_context;
+    Service::GSP::InterruptHandler signal_interrupt;
+    GeometryPipeline geometry_pipeline;
+    PrimitiveAssembler primitive_assembler;
+    CommandList cmd_list;
+    std::unique_ptr<ShaderEngine> shader_engine;
+};
+
+#define GPU_REG_INDEX(field_name) (offsetof(Pica::PicaCore::Regs, field_name) / sizeof(u32))
+
+} // namespace Pica

src/video_core/pica/primitive_assembly.cpp

@@ -0,0 +1,53 @@
+// Copyright 2014 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/logging/log.h"
+#include "video_core/pica/primitive_assembly.h"
+
+namespace Pica {
+
+PrimitiveAssembler::PrimitiveAssembler(PipelineRegs::TriangleTopology topology)
+    : topology(topology) {}
+
+void PrimitiveAssembler::SubmitVertex(const OutputVertex& vtx,
+                                      const TriangleHandler& triangle_handler) {
+    switch (topology) {
+    case PipelineRegs::TriangleTopology::List:
+    case PipelineRegs::TriangleTopology::Shader:
+        if (buffer_index < 2) {
+            buffer[buffer_index++] = vtx;
+        } else {
+            buffer_index = 0;
+            if (topology == PipelineRegs::TriangleTopology::Shader && winding) {
+                triangle_handler(buffer[1], buffer[0], vtx);
+                winding = false;
+            } else {
+                triangle_handler(buffer[0], buffer[1], vtx);
+            }
+        }
+        break;
+
+    case PipelineRegs::TriangleTopology::Strip:
+    case PipelineRegs::TriangleTopology::Fan:
+        if (strip_ready) {
+            triangle_handler(buffer[0], buffer[1], vtx);
+        }
+
+        buffer[buffer_index] = vtx;
+        strip_ready |= (buffer_index == 1);
+
+        if (topology == PipelineRegs::TriangleTopology::Strip) {
+            buffer_index = !buffer_index;
+        } else if (topology == PipelineRegs::TriangleTopology::Fan) {
+            buffer_index = 1;
+        }
+        break;
+
+    default:
+        LOG_ERROR(HW_GPU, "Unknown triangle topology {:x}:", (int)topology);
+        break;
+    }
+}
+
+} // namespace Pica

src/video_core/pica/primitive_assembly.h

@@ -8,61 +8,73 @@
 #include <functional>
 #include <boost/serialization/access.hpp>
 #include <boost/serialization/array.hpp>
-#include "video_core/regs_pipeline.h"
+#include "video_core/pica/output_vertex.h"
+#include "video_core/pica/regs_pipeline.h"
 
 namespace Pica {
 
-/*
+/**
  * Utility class to build triangles from a series of vertices,
  * according to a given triangle topology.
  */
-template <typename VertexType>
 struct PrimitiveAssembler {
     using TriangleHandler =
-        std::function<void(const VertexType& v0, const VertexType& v1, const VertexType& v2)>;
+        std::function<void(const OutputVertex&, const OutputVertex&, const OutputVertex&)>;
 
     explicit PrimitiveAssembler(
         PipelineRegs::TriangleTopology topology = PipelineRegs::TriangleTopology::List);
 
-    /*
+    /**
      * Queues a vertex, builds primitives from the vertex queue according to the given
      * triangle topology, and calls triangle_handler for each generated primitive.
      * NOTE: We could specify the triangle handler in the constructor, but this way we can
      * keep event and handler code next to each other.
      */
-    void SubmitVertex(const VertexType& vtx, const TriangleHandler& triangle_handler);
+    void SubmitVertex(const OutputVertex& vtx, const TriangleHandler& triangle_handler);
 
     /**
      * Invert the vertex order of the next triangle. Called by geometry shader emitter.
      * This only takes effect for TriangleTopology::Shader.
      */
-    void SetWinding();
+    void SetWinding() noexcept {
+        winding = true;
+    }
 
     /**
      * Resets the internal state of the PrimitiveAssembler.
      */
-    void Reset();
+    void Reset() {
+        buffer_index = 0;
+        strip_ready = false;
+        winding = false;
+    }
 
     /**
      * Reconfigures the PrimitiveAssembler to use a different triangle topology.
      */
-    void Reconfigure(PipelineRegs::TriangleTopology topology);
+    void Reconfigure(PipelineRegs::TriangleTopology topology) {
+        Reset();
+        this->topology = topology;
+    }
 
     /**
      * Returns whether the PrimitiveAssembler has an empty internal buffer.
      */
-    bool IsEmpty() const;
+    bool IsEmpty() const {
+        return buffer_index == 0 && !strip_ready;
+    }
 
     /**
      * Returns the current topology.
      */
-    PipelineRegs::TriangleTopology GetTopology() const;
+    PipelineRegs::TriangleTopology GetTopology() const {
+        return topology;
+    }
 
 private:
     PipelineRegs::TriangleTopology topology;
-
     int buffer_index = 0;
-    std::array<VertexType, 2> buffer;
+    std::array<OutputVertex, 2> buffer;
     bool strip_ready = false;
     bool winding = false;
 
@@ -70,7 +82,7 @@ private:
     void serialize(Archive& ar, const unsigned int version) {
         ar& topology;
         ar& buffer_index;
-        ar& boost::serialization::make_array(buffer.data(), buffer.size());
+        ar& buffer;
         ar& strip_ready;
         ar& winding;
     }

src/video_core/pica/regs_external.h

@@ -0,0 +1,217 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/bit_field.h"
+
+namespace Pica {
+
+/**
+ * Most physical addresses which GPU registers refer to are 8-byte aligned.
+ * This function should be used to get the address from a raw register value.
+ */
+constexpr u32 DecodeAddressRegister(u32 register_value) {
+    return register_value * 8;
+}
+
+/// Components are laid out in reverse byte order, most significant bits first.
+enum class PixelFormat : u32 {
+    RGBA8 = 0,
+    RGB8 = 1,
+    RGB565 = 2,
+    RGB5A1 = 3,
+    RGBA4 = 4,
+};
+
+constexpr u32 BytesPerPixel(Pica::PixelFormat format) {
+    switch (format) {
+    case Pica::PixelFormat::RGBA8:
+        return 4;
+    case Pica::PixelFormat::RGB8:
+        return 3;
+    case Pica::PixelFormat::RGB565:
+    case Pica::PixelFormat::RGB5A1:
+    case Pica::PixelFormat::RGBA4:
+        return 2;
+    default:
+        UNREACHABLE();
+    }
+
+    return 0;
+}
+
+struct MemoryFillConfig {
+    u32 address_start;
+    u32 address_end;
+
+    union {
+        u32 value_32bit;
+
+        BitField<0, 16, u32> value_16bit;
+
+        // TODO: Verify component order
+        BitField<0, 8, u32> value_24bit_r;
+        BitField<8, 8, u32> value_24bit_g;
+        BitField<16, 8, u32> value_24bit_b;
+    };
+
+    union {
+        u32 control;
+
+        // Setting this field to 1 triggers the memory fill.
+        // This field also acts as a status flag, and gets reset to 0 upon completion.
+        BitField<0, 1, u32> trigger;
+        // Set to 1 upon completion.
+        BitField<1, 1, u32> finished;
+        // If both of these bits are unset, then it will fill the memory with a 16 bit value
+        // 1: fill with 24-bit wide values
+        BitField<8, 1, u32> fill_24bit;
+        // 1: fill with 32-bit wide values
+        BitField<9, 1, u32> fill_32bit;
+    };
+
+    inline u32 GetStartAddress() const {
+        return DecodeAddressRegister(address_start);
+    }
+
+    inline u32 GetEndAddress() const {
+        return DecodeAddressRegister(address_end);
+    }
+
+    inline std::string DebugName() const {
+        return fmt::format("from {:#X} to {:#X} with {}-bit value {:#X}", GetStartAddress(),
+                           GetEndAddress(), fill_32bit ? "32" : (fill_24bit ? "24" : "16"),
+                           value_32bit);
+    }
+};
+static_assert(sizeof(MemoryFillConfig) == 0x10);
+
+struct FramebufferConfig {
+    INSERT_PADDING_WORDS(0x17);
+
+    union {
+        u32 size;
+
+        BitField<0, 16, u32> width;
+        BitField<16, 16, u32> height;
+    };
+
+    INSERT_PADDING_WORDS(0x2);
+
+    u32 address_left1;
+    u32 address_left2;
+
+    union {
+        u32 format;
+
+        BitField<0, 3, PixelFormat> color_format;
+    };
+
+    INSERT_PADDING_WORDS(0x1);
+
+    union {
+        u32 active_fb;
+
+        // 0: Use parameters ending with "1"
+        // 1: Use parameters ending with "2"
+        BitField<0, 1, u32> second_fb_active;
+    };
+
+    INSERT_PADDING_WORDS(0x5);
+
+    // Distance between two pixel rows, in bytes
+    u32 stride;
+
+    u32 address_right1;
+    u32 address_right2;
+
+    INSERT_PADDING_WORDS(0x19);
+};
+static_assert(sizeof(FramebufferConfig) == 0x100);
+
+struct DisplayTransferConfig {
+    u32 input_address;
+    u32 output_address;
+
+    inline u32 GetPhysicalInputAddress() const {
+        return DecodeAddressRegister(input_address);
+    }
+
+    inline u32 GetPhysicalOutputAddress() const {
+        return DecodeAddressRegister(output_address);
+    }
+
+    inline std::string DebugName() const noexcept {
+        return fmt::format("from {:#x} to {:#x} with {} scaling and stride {}, width {}",
+                           GetPhysicalInputAddress(), GetPhysicalOutputAddress(),
+                           scaling == NoScale ? "no" : (scaling == ScaleX ? "X" : "XY"),
+                           input_width.Value(), output_width.Value());
+    }
+
+    union {
+        u32 output_size;
+
+        BitField<0, 16, u32> output_width;
+        BitField<16, 16, u32> output_height;
+    };
+
+    union {
+        u32 input_size;
+
+        BitField<0, 16, u32> input_width;
+        BitField<16, 16, u32> input_height;
+    };
+
+    enum ScalingMode : u32 {
+        NoScale = 0, // Doesn't scale the image
+        ScaleX = 1,  // Downscales the image in half in the X axis and applies a box filter
+        ScaleXY =
+            2, // Downscales the image in half in both the X and Y axes and applies a box filter
+    };
+
+    union {
+        u32 flags;
+
+        BitField<0, 1, u32> flip_vertically; // flips input data vertically
+        BitField<1, 1, u32> input_linear;    // Converts from linear to tiled format
+        BitField<2, 1, u32> crop_input_lines;
+        BitField<3, 1, u32> is_texture_copy; // Copies the data without performing any
+                                             // processing and respecting texture copy fields
+        BitField<5, 1, u32> dont_swizzle;
+        BitField<8, 3, PixelFormat> input_format;
+        BitField<12, 3, PixelFormat> output_format;
+        /// Uses some kind of 32x32 block swizzling mode, instead of the usual 8x8 one.
+        BitField<16, 1, u32> block_32;        // TODO(yuriks): unimplemented
+        BitField<24, 2, ScalingMode> scaling; // Determines the scaling mode of the transfer
+    };
+
+    INSERT_PADDING_WORDS(0x1);
+
+    // it seems that writing to this field triggers the display transfer
+    BitField<0, 1, u32> trigger;
+
+    INSERT_PADDING_WORDS(0x1);
+
+    struct {
+        u32 size; // The lower 4 bits are ignored
+
+        union {
+            u32 input_size;
+
+            BitField<0, 16, u32> input_width;
+            BitField<16, 16, u32> input_gap;
+        };
+
+        union {
+            u32 output_size;
+
+            BitField<0, 16, u32> output_width;
+            BitField<16, 16, u32> output_gap;
+        };
+    } texture_copy;
+};
+static_assert(sizeof(DisplayTransferConfig) == 0x2c);
+
+} // namespace Pica

src/video_core/pica/regs_internal.cpp

@@ -7,11 +7,11 @@
 #include <utility>
 
 #include "common/common_types.h"
-#include "video_core/regs.h"
+#include "video_core/pica/regs_internal.h"
 
 namespace Pica {
 
-static const std::pair<u16, const char*> register_names[] = {
+static constexpr std::pair<u16, const char*> register_names[] = {
     {0x010, "GPUREG_FINALIZE"},
 
     {0x040, "GPUREG_FACECULLING_CONFIG"},
@@ -474,15 +474,15 @@ static const std::pair<u16, const char*> register_names[] = {
     {0x2DD, "GPUREG_VSH_OPDESCS_DATA7"},
 };
 
-const char* Regs::GetRegisterName(u16 index) {
+const char* RegsInternal::GetRegisterName(u16 index) {
-    auto found = std::lower_bound(std::begin(register_names), std::end(register_names), index,
+    const auto it = std::lower_bound(std::begin(register_names), std::end(register_names), index,
-                                  [](auto p, auto i) { return p.first < i; });
+                                     [](auto p, auto i) { return p.first < i; });
-    if (found->first == index) {
+    if (it->first == index) {
-        return found->second;
+        return it->second;
-    } else {
-        // Return empty string if no match is found
-        return "";
     }
+
+    // Return empty string if no match is found
+    return "";
 }
 
 } // namespace Pica

src/video_core/pica/regs_internal.h

@@ -3,18 +3,19 @@
 // Refer to the license.txt file included.
 
 #pragma once
-#include "video_core/regs_framebuffer.h"
+
-#include "video_core/regs_lighting.h"
+#include "video_core/pica/regs_framebuffer.h"
-#include "video_core/regs_pipeline.h"
+#include "video_core/pica/regs_lighting.h"
-#include "video_core/regs_rasterizer.h"
+#include "video_core/pica/regs_pipeline.h"
-#include "video_core/regs_shader.h"
+#include "video_core/pica/regs_rasterizer.h"
-#include "video_core/regs_texturing.h"
+#include "video_core/pica/regs_shader.h"
+#include "video_core/pica/regs_texturing.h"
 
 namespace Pica {
 
-#define PICA_REG_INDEX(field_name) (offsetof(Pica::Regs, field_name) / sizeof(u32))
+#define PICA_REG_INDEX(field_name) (offsetof(Pica::RegsInternal, field_name) / sizeof(u32))
 
-struct Regs {
+struct RegsInternal {
     static constexpr std::size_t NUM_REGS = 0x300;
 
     union {
@@ -38,10 +39,11 @@ struct Regs {
     static const char* GetRegisterName(u16 index);
 };
 
-static_assert(sizeof(Regs) == Regs::NUM_REGS * sizeof(u32), "Regs struct has wrong size");
+static_assert(sizeof(RegsInternal) == RegsInternal::NUM_REGS * sizeof(u32),
+              "Regs struct has wrong size");
 
 #define ASSERT_REG_POSITION(field_name, position)                                                  \
-    static_assert(offsetof(Regs, field_name) == position * 4,                                      \
+    static_assert(offsetof(RegsInternal, field_name) == position * 4,                              \
                   "Field " #field_name " has invalid position")
 
 ASSERT_REG_POSITION(trigger_irq, 0x10);
@@ -110,4 +112,5 @@ ASSERT_REG_POSITION(gs, 0x280);
 ASSERT_REG_POSITION(vs, 0x2b0);
 
 #undef ASSERT_REG_POSITION
+
 } // namespace Pica

src/video_core/pica/regs_lcd.h

@@ -1,45 +1,46 @@
-// Copyright 2015 Citra Emulator Project
+// Copyright 2023 Citra Emulator Project
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
 #pragma once
 
-#include <cstddef>
-#include <type_traits>
 #include <boost/serialization/access.hpp>
+
 #include "common/bit_field.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"
+#include "common/vector_math.h"
 
-#define LCD_REG_INDEX(field_name) (offsetof(LCD::Regs, field_name) / sizeof(u32))
+#define LCD_REG_INDEX(field_name) (offsetof(Pica::RegsLcd, field_name) / sizeof(u32))
 
-namespace LCD {
+namespace Pica {
 
-struct Regs {
+union ColorFill {
+    u32 raw;
 
-    union ColorFill {
+    BitField<0, 8, u32> color_r;
-        u32 raw;
+    BitField<8, 8, u32> color_g;
+    BitField<16, 8, u32> color_b;
+    BitField<24, 1, u32> is_enabled;
 
-        BitField<0, 8, u32> color_r;
+    Common::Vec3<u8> AsVector() const noexcept {
-        BitField<8, 8, u32> color_g;
+        return Common::MakeVec<u8>(color_r, color_g, color_b);
-        BitField<16, 8, u32> color_b;
+    }
-        BitField<24, 1, u32> is_enabled;
+};
-    };
 
+struct RegsLcd {
     INSERT_PADDING_WORDS(0x81);
     ColorFill color_fill_top;
     INSERT_PADDING_WORDS(0xE);
     u32 backlight_top;
-
     INSERT_PADDING_WORDS(0x1F0);
-
     ColorFill color_fill_bottom;
     INSERT_PADDING_WORDS(0xE);
     u32 backlight_bottom;
     INSERT_PADDING_WORDS(0x16F);
 
     static constexpr std::size_t NumIds() {
-        return sizeof(Regs) / sizeof(u32);
+        return sizeof(RegsLcd) / sizeof(u32);
     }
 
     const u32& operator[](int index) const {
@@ -62,28 +63,16 @@ private:
     }
     friend class boost::serialization::access;
 };
-static_assert(std::is_standard_layout<Regs>::value, "Structure does not use standard layout");
+static_assert(std::is_standard_layout_v<RegsLcd>, "Structure does not use standard layout");
 
 #define ASSERT_REG_POSITION(field_name, position)                                                  \
-    static_assert(offsetof(Regs, field_name) == position * 4,                                      \
+    static_assert(offsetof(RegsLcd, field_name) == position * 4,                                   \
                   "Field " #field_name " has invalid position")
 ASSERT_REG_POSITION(color_fill_top, 0x81);
 ASSERT_REG_POSITION(backlight_top, 0x90);
 ASSERT_REG_POSITION(color_fill_bottom, 0x281);
 ASSERT_REG_POSITION(backlight_bottom, 0x290);
 
-extern Regs g_regs;
+#undef ASSERT_REG_POSITION
 
-template <typename T>
+} // namespace Pica
-void Read(T& var, const u32 addr);
-
-template <typename T>
-void Write(u32 addr, const T data);
-
-/// Initialize hardware
-void Init();
-
-/// Shutdown hardware
-void Shutdown();
-
-} // namespace LCD

src/video_core/pica/regs_lighting.h

@@ -26,16 +26,16 @@ struct LightingRegs {
         DistanceAttenuation = 16,
     };
 
-    static constexpr unsigned NumLightingSampler = 24;
+    static constexpr u32 NumLightingSampler = 24;
 
-    static LightingSampler SpotlightAttenuationSampler(unsigned index) {
+    static LightingSampler SpotlightAttenuationSampler(u32 index) {
         return static_cast<LightingSampler>(
-            static_cast<unsigned>(LightingSampler::SpotlightAttenuation) + index);
+            static_cast<u32>(LightingSampler::SpotlightAttenuation) + index);
     }
 
-    static LightingSampler DistanceAttenuationSampler(unsigned index) {
+    static LightingSampler DistanceAttenuationSampler(u32 index) {
-        return static_cast<LightingSampler>(
+        return static_cast<LightingSampler>(static_cast<u32>(LightingSampler::DistanceAttenuation) +
-            static_cast<unsigned>(LightingSampler::DistanceAttenuation) + index);
+                                            index);
     }
 
     /**

src/video_core/pica/regs_pipeline.h

@@ -20,6 +20,20 @@ struct PipelineRegs {
         FLOAT = 3,
     };
 
+    static u32 GetFormatBytes(VertexAttributeFormat format) {
+        switch (format) {
+        case VertexAttributeFormat::FLOAT:
+            return 4;
+        case VertexAttributeFormat::SHORT:
+            return 2;
+        case VertexAttributeFormat::BYTE:
+        case VertexAttributeFormat::UBYTE:
+            return 1;
+        default:
+            UNREACHABLE();
+        }
+    }
+
     struct {
         BitField<1, 28, u32> base_address;
 
@@ -194,14 +208,14 @@ struct PipelineRegs {
         BitField<0, 28, u32> addr[2]; ///< Physical address / 8 of each channel's command buffer
         u32 trigger[2]; ///< Triggers execution of the channel's command buffer when written to
 
-        unsigned GetSize(unsigned index) const {
+        u32 GetSize(u32 index) const {
             ASSERT(index < 2);
             return 8 * size[index];
         }
 
-        PAddr GetPhysicalAddress(unsigned index) const {
+        PAddr GetPhysicalAddress(u32 index) const {
             ASSERT(index < 2);
-            return (PAddr)(8 * addr[index]);
+            return 8 * addr[index];
         }
     } command_buffer;
 

src/video_core/pica/regs_rasterizer.h

@@ -104,6 +104,17 @@ struct RasterizerRegs {
         u32 raw;
     } vs_output_attributes[7];
 
+    void ValidateSemantics() {
+        for (std::size_t attrib = 0; attrib < vs_output_total; ++attrib) {
+            const u32 output_register_map = vs_output_attributes[attrib].raw;
+            for (std::size_t comp = 0; comp < 4; ++comp) {
+                const u32 semantic = (output_register_map >> (8 * comp)) & 0x1F;
+                ASSERT_MSG(semantic < 24 || semantic == VSOutputAttributes::INVALID,
+                           "Invalid/unknown semantic id: {}", semantic);
+            }
+        }
+    }
+
     INSERT_PADDING_WORDS(0xe);
 
     enum class ScissorMode : u32 {

src/video_core/pica/regs_shader.h

@@ -4,11 +4,10 @@
 
 #pragma once
 
-#include <array>
-
 #include "common/bit_field.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"
+#include "common/vector_math.h"
 
 namespace Pica {
 
@@ -22,6 +21,11 @@ struct ShaderRegs {
         BitField<24, 8, u32> w;
     } int_uniforms[4];
 
+    Common::Vec4<u8> GetIntUniform(u32 index) const {
+        const auto& values = int_uniforms[index];
+        return Common::MakeVec<u8>(values.x, values.y, values.z, values.w);
+    }
+
     INSERT_PADDING_WORDS(0x4);
 
     enum ShaderMode {
@@ -55,13 +59,13 @@ struct ShaderRegs {
     INSERT_PADDING_WORDS(0x2);
 
     struct {
-        enum Format : u32 {
+        enum class Format : u32 {
-            FLOAT24 = 0,
+            Float24 = 0,
-            FLOAT32 = 1,
+            Float32 = 1,
         };
 
         bool IsFloat32() const {
-            return format == FLOAT32;
+            return format == Format::Float32;
         }
 
         union {
@@ -70,7 +74,6 @@ struct ShaderRegs {
             // indices
             // TODO: Maybe the uppermost index is for the geometry shader? Investigate!
             BitField<0, 7, u32> index;
-
             BitField<31, 1, Format> format;
         };
 

src/video_core/pica/shader_setup.cpp

@@ -0,0 +1,61 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/bit_set.h"
+#include "common/hash.h"
+#include "video_core/pica/regs_shader.h"
+#include "video_core/pica/shader_setup.h"
+
+namespace Pica {
+
+ShaderSetup::ShaderSetup() = default;
+
+ShaderSetup::~ShaderSetup() = default;
+
+void ShaderSetup::WriteUniformBoolReg(u32 value) {
+    const auto bits = BitSet32(value);
+    for (u32 i = 0; i < uniforms.b.size(); ++i) {
+        uniforms.b[i] = bits[i];
+    }
+}
+
+void ShaderSetup::WriteUniformIntReg(u32 index, const Common::Vec4<u8> values) {
+    ASSERT(index < uniforms.i.size());
+    uniforms.i[index] = values;
+}
+
+void ShaderSetup::WriteUniformFloatReg(ShaderRegs& config, u32 value) {
+    auto& uniform_setup = config.uniform_setup;
+    const bool is_float32 = uniform_setup.IsFloat32();
+    if (!uniform_queue.Push(value, is_float32)) {
+        return;
+    }
+
+    const auto uniform = uniform_queue.Get(is_float32);
+    if (uniform_setup.index >= uniforms.f.size()) {
+        LOG_ERROR(HW_GPU, "Invalid float uniform index {}", uniform_setup.index.Value());
+        return;
+    }
+
+    uniforms.f[uniform_setup.index] = uniform;
+    uniform_setup.index.Assign(uniform_setup.index + 1);
+}
+
+u64 ShaderSetup::GetProgramCodeHash() {
+    if (program_code_hash_dirty) {
+        program_code_hash = Common::ComputeHash64(&program_code, sizeof(program_code));
+        program_code_hash_dirty = false;
+    }
+    return program_code_hash;
+}
+
+u64 ShaderSetup::GetSwizzleDataHash() {
+    if (swizzle_data_hash_dirty) {
+        swizzle_data_hash = Common::ComputeHash64(&swizzle_data, sizeof(swizzle_data));
+        swizzle_data_hash_dirty = false;
+    }
+    return swizzle_data_hash;
+}
+
+} // namespace Pica

src/video_core/pica/shader_setup.h

@@ -0,0 +1,103 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/vector_math.h"
+#include "video_core/pica/packed_attribute.h"
+#include "video_core/pica_types.h"
+
+namespace Pica {
+
+constexpr u32 MAX_PROGRAM_CODE_LENGTH = 4096;
+constexpr u32 MAX_SWIZZLE_DATA_LENGTH = 4096;
+
+using ProgramCode = std::array<u32, MAX_PROGRAM_CODE_LENGTH>;
+using SwizzleData = std::array<u32, MAX_SWIZZLE_DATA_LENGTH>;
+
+struct Uniforms {
+    alignas(16) std::array<Common::Vec4<f24>, 96> f;
+    std::array<bool, 16> b;
+    std::array<Common::Vec4<u8>, 4> i;
+
+    static size_t GetFloatUniformOffset(u32 index) {
+        return offsetof(Uniforms, f) + index * sizeof(Common::Vec4<f24>);
+    }
+
+    static size_t GetBoolUniformOffset(u32 index) {
+        return offsetof(Uniforms, b) + index * sizeof(bool);
+    }
+
+    static size_t GetIntUniformOffset(u32 index) {
+        return offsetof(Uniforms, i) + index * sizeof(Common::Vec4<u8>);
+    }
+
+private:
+    friend class boost::serialization::access;
+    template <class Archive>
+    void serialize(Archive& ar, const u32 file_version) {
+        ar& f;
+        ar& b;
+        ar& i;
+    }
+};
+
+struct ShaderRegs;
+
+/**
+ * This structure contains the state information common for all shader units such as uniforms.
+ * The geometry shaders has a unique configuration so when enabled it has its own setup.
+ */
+struct ShaderSetup {
+public:
+    explicit ShaderSetup();
+    ~ShaderSetup();
+
+    void WriteUniformBoolReg(u32 value);
+
+    void WriteUniformIntReg(u32 index, const Common::Vec4<u8> values);
+
+    void WriteUniformFloatReg(ShaderRegs& config, u32 value);
+
+    u64 GetProgramCodeHash();
+
+    u64 GetSwizzleDataHash();
+
+    void MarkProgramCodeDirty() {
+        program_code_hash_dirty = true;
+    }
+
+    void MarkSwizzleDataDirty() {
+        swizzle_data_hash_dirty = true;
+    }
+
+public:
+    Uniforms uniforms;
+    PackedAttribute uniform_queue;
+    ProgramCode program_code;
+    SwizzleData swizzle_data;
+    u32 entry_point;
+    const void* cached_shader{};
+
+private:
+    bool program_code_hash_dirty{true};
+    bool swizzle_data_hash_dirty{true};
+    u64 program_code_hash{0xDEADC0DE};
+    u64 swizzle_data_hash{0xDEADC0DE};
+
+    friend class boost::serialization::access;
+    template <class Archive>
+    void serialize(Archive& ar, const u32 file_version) {
+        ar& uniforms;
+        ar& uniform_queue;
+        ar& program_code;
+        ar& swizzle_data;
+        ar& program_code_hash_dirty;
+        ar& swizzle_data_hash_dirty;
+        ar& program_code_hash;
+        ar& swizzle_data_hash;
+    }
+};
+
+} // namespace Pica

src/video_core/pica/shader_unit.cpp

@@ -0,0 +1,63 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/bit_set.h"
+#include "video_core/pica/regs_shader.h"
+#include "video_core/pica/shader_unit.h"
+
+namespace Pica {
+
+ShaderUnit::ShaderUnit(GeometryEmitter* emitter) : emitter_ptr{emitter} {}
+
+ShaderUnit::~ShaderUnit() = default;
+
+void ShaderUnit::LoadInput(const ShaderRegs& config, const AttributeBuffer& buffer) {
+    const u32 max_attribute = config.max_input_attribute_index;
+    for (u32 attr = 0; attr <= max_attribute; ++attr) {
+        const u32 reg = config.GetRegisterForAttribute(attr);
+        input[reg] = buffer[attr];
+    }
+}
+
+void ShaderUnit::WriteOutput(const ShaderRegs& config, AttributeBuffer& buffer) {
+    u32 output_index{};
+    for (u32 reg : Common::BitSet<u32>(config.output_mask)) {
+        buffer[output_index++] = output[reg];
+    }
+}
+
+void GeometryEmitter::Emit(std::span<Common::Vec4<f24>, 16> output_regs) {
+    ASSERT(vertex_id < 3);
+
+    u32 output_index{};
+    for (u32 reg : Common::BitSet<u32>(output_mask)) {
+        buffer[vertex_id][output_index++] = output_regs[reg];
+    }
+
+    if (prim_emit) {
+        if (winding) {
+            handlers->winding_setter();
+        }
+        for (std::size_t i = 0; i < buffer.size(); ++i) {
+            handlers->vertex_handler(buffer[i]);
+        }
+    }
+}
+
+GeometryShaderUnit::GeometryShaderUnit() : ShaderUnit{&emitter} {}
+
+GeometryShaderUnit::~GeometryShaderUnit() = default;
+
+void GeometryShaderUnit::SetVertexHandlers(VertexHandler vertex_handler,
+                                           WindingSetter winding_setter) {
+    emitter.handlers = new Handlers;
+    emitter.handlers->vertex_handler = vertex_handler;
+    emitter.handlers->winding_setter = winding_setter;
+}
+
+void GeometryShaderUnit::ConfigOutput(const ShaderRegs& config) {
+    emitter.output_mask = config.output_mask;
+}
+
+} // namespace Pica

src/video_core/pica/shader_unit.h

@@ -0,0 +1,120 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <functional>
+#include <span>
+#include <boost/serialization/base_object.hpp>
+
+#include "video_core/pica/output_vertex.h"
+
+namespace Pica {
+
+/// Handler type for receiving vertex outputs from vertex shader or geometry shader
+using VertexHandler = std::function<void(const AttributeBuffer&)>;
+
+/// Handler type for signaling to invert the vertex order of the next triangle
+using WindingSetter = std::function<void()>;
+
+struct ShaderRegs;
+struct GeometryEmitter;
+
+/**
+ * This structure contains the state information that needs to be unique for a shader unit. The 3DS
+ * has four shader units that process shaders in parallel.
+ */
+struct ShaderUnit {
+    explicit ShaderUnit(GeometryEmitter* emitter = nullptr);
+    ~ShaderUnit();
+
+    void LoadInput(const ShaderRegs& config, const AttributeBuffer& input);
+
+    void WriteOutput(const ShaderRegs& config, AttributeBuffer& output);
+
+    static constexpr size_t InputOffset(s32 register_index) {
+        return offsetof(ShaderUnit, input) + register_index * sizeof(Common::Vec4<f24>);
+    }
+
+    static constexpr size_t OutputOffset(s32 register_index) {
+        return offsetof(ShaderUnit, output) + register_index * sizeof(Common::Vec4<f24>);
+    }
+
+    static constexpr size_t TemporaryOffset(s32 register_index) {
+        return offsetof(ShaderUnit, temporary) + register_index * sizeof(Common::Vec4<f24>);
+    }
+
+public:
+    s32 address_registers[3];
+    bool conditional_code[2];
+    alignas(16) std::array<Common::Vec4<f24>, 16> input;
+    alignas(16) std::array<Common::Vec4<f24>, 16> temporary;
+    alignas(16) std::array<Common::Vec4<f24>, 16> output;
+    GeometryEmitter* emitter_ptr;
+
+private:
+    friend class boost::serialization::access;
+    template <class Archive>
+    void serialize(Archive& ar, const u32 file_version) {
+        ar& input;
+        ar& temporary;
+        ar& output;
+        ar& conditional_code;
+        ar& address_registers;
+    }
+};
+
+struct Handlers {
+    VertexHandler vertex_handler;
+    WindingSetter winding_setter;
+};
+
+/// This structure contains state information for primitive emitting in geometry shader.
+struct GeometryEmitter {
+    void Emit(std::span<Common::Vec4<f24>, 16> output_regs);
+
+public:
+    std::array<AttributeBuffer, 3> buffer;
+    u8 vertex_id;
+    bool prim_emit;
+    bool winding;
+    u32 output_mask;
+    Handlers* handlers;
+
+private:
+    friend class boost::serialization::access;
+    template <class Archive>
+    void serialize(Archive& ar, const u32 file_version) {
+        ar& buffer;
+        ar& vertex_id;
+        ar& prim_emit;
+        ar& winding;
+        ar& output_mask;
+    }
+};
+
+/**
+ * This is an extended shader unit state that represents the special unit that can run both vertex
+ * shader and geometry shader. It contains an additional primitive emitter and utilities for
+ * geometry shader.
+ */
+struct GeometryShaderUnit : public ShaderUnit {
+    GeometryShaderUnit();
+    ~GeometryShaderUnit();
+
+    void SetVertexHandlers(VertexHandler vertex_handler, WindingSetter winding_setter);
+    void ConfigOutput(const ShaderRegs& config);
+
+    GeometryEmitter emitter;
+
+private:
+    friend class boost::serialization::access;
+    template <class Archive>
+    void serialize(Archive& ar, const u32 file_version) {
+        ar& boost::serialization::base_object<ShaderUnit>(*this);
+        ar& emitter;
+    }
+};
+
+} // namespace Pica

src/video_core/pica/vertex_loader.cpp

@@ -0,0 +1,109 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/alignment.h"
+#include "common/logging/log.h"
+#include "video_core/pica/vertex_loader.h"
+
+namespace Pica {
+
+VertexLoader::VertexLoader(Memory::MemorySystem& memory_, const PipelineRegs& regs)
+    : memory{memory_} {
+    const auto& attribute_config = regs.vertex_attributes;
+    num_total_attributes = attribute_config.GetNumTotalAttributes();
+
+    vertex_attribute_sources.fill(0xdeadbeef);
+
+    for (u32 i = 0; i < 16; i++) {
+        vertex_attribute_is_default[i] = attribute_config.IsDefaultAttribute(i);
+    }
+
+    // Setup attribute data from loaders
+    for (u32 loader = 0; loader < 12; ++loader) {
+        const auto& loader_config = attribute_config.attribute_loaders[loader];
+
+        u32 offset = 0;
+
+        // TODO: What happens if a loader overwrites a previous one's data?
+        for (u32 component = 0; component < loader_config.component_count; ++component) {
+            if (component >= 12) {
+                LOG_ERROR(HW_GPU,
+                          "Overflow in the vertex attribute loader {} trying to load component {}",
+                          loader, component);
+                continue;
+            }
+
+            u32 attribute_index = loader_config.GetComponent(component);
+            if (attribute_index < 12) {
+                offset = Common::AlignUp(offset,
+                                         attribute_config.GetElementSizeInBytes(attribute_index));
+                vertex_attribute_sources[attribute_index] = loader_config.data_offset + offset;
+                vertex_attribute_strides[attribute_index] =
+                    static_cast<u32>(loader_config.byte_count);
+                vertex_attribute_formats[attribute_index] =
+                    attribute_config.GetFormat(attribute_index);
+                vertex_attribute_elements[attribute_index] =
+                    attribute_config.GetNumElements(attribute_index);
+                offset += attribute_config.GetStride(attribute_index);
+            } else if (attribute_index < 16) {
+                // Attribute ids 12, 13, 14 and 15 signify 4, 8, 12 and 16-byte paddings,
+                // respectively
+                offset = Common::AlignUp(offset, 4);
+                offset += (attribute_index - 11) * 4;
+            } else {
+                UNREACHABLE(); // This is truly unreachable due to the number of bits for each
+                               // component
+            }
+        }
+    }
+}
+
+VertexLoader::~VertexLoader() = default;
+
+void VertexLoader::LoadVertex(PAddr base_address, u32 index, u32 vertex, AttributeBuffer& input,
+                              AttributeBuffer& input_default_attributes) const {
+    for (s32 i = 0; i < num_total_attributes; ++i) {
+        // Load the default attribute if we're configured to do so
+        if (vertex_attribute_is_default[i]) {
+            input[i] = input_default_attributes[i];
+            continue;
+        }
+
+        // TODO(yuriks): In this case, no data gets loaded and the vertex
+        // remains with the last value it had. This isn't currently maintained
+        // as global state, however, and so won't work in Citra yet.
+        if (vertex_attribute_elements[i] == 0) {
+            LOG_ERROR(HW_GPU, "Vertex retension unimplemented");
+            continue;
+        }
+
+        // Load per-vertex data from the loader arrays
+        const PAddr source_addr =
+            base_address + vertex_attribute_sources[i] + vertex_attribute_strides[i] * vertex;
+
+        switch (vertex_attribute_formats[i]) {
+        case PipelineRegs::VertexAttributeFormat::BYTE:
+            LoadAttribute<s8>(source_addr, i, input);
+            break;
+        case PipelineRegs::VertexAttributeFormat::UBYTE:
+            LoadAttribute<u8>(source_addr, i, input);
+            break;
+        case PipelineRegs::VertexAttributeFormat::SHORT:
+            LoadAttribute<s16>(source_addr, i, input);
+            break;
+        case PipelineRegs::VertexAttributeFormat::FLOAT:
+            LoadAttribute<f32>(source_addr, i, input);
+            break;
+        }
+
+        // Default attribute values set if array elements have < 4 components. This
+        // is *not* carried over from the default attribute settings even if they're
+        // enabled for this attribute.
+        for (u32 comp = vertex_attribute_elements[i]; comp < 4; comp++) {
+            input[i][comp] = comp == 3 ? f24::One() : f24::Zero();
+        }
+    }
+}
+
+} // namespace Pica

src/video_core/pica/vertex_loader.h

@@ -0,0 +1,47 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "core/memory.h"
+#include "video_core/pica/output_vertex.h"
+#include "video_core/pica/regs_pipeline.h"
+
+namespace Memory {
+class MemorySystem;
+}
+
+namespace Pica {
+
+class VertexLoader {
+public:
+    explicit VertexLoader(Memory::MemorySystem& memory_, const PipelineRegs& regs);
+    ~VertexLoader();
+
+    void LoadVertex(PAddr base_address, u32 index, u32 vertex, AttributeBuffer& input,
+                    AttributeBuffer& input_default_attributes) const;
+
+    template <typename T>
+    void LoadAttribute(PAddr source_addr, u32 attrib, AttributeBuffer& out) const {
+        const T* data = reinterpret_cast<const T*>(memory.GetPhysicalPointer(source_addr));
+        for (u32 comp = 0; comp < vertex_attribute_elements[attrib]; ++comp) {
+            out[attrib][comp] = f24::FromFloat32(data[comp]);
+        }
+    }
+
+    int GetNumTotalAttributes() const {
+        return num_total_attributes;
+    }
+
+private:
+    Memory::MemorySystem& memory;
+    std::array<u32, 16> vertex_attribute_sources;
+    std::array<u32, 16> vertex_attribute_strides{};
+    std::array<PipelineRegs::VertexAttributeFormat, 16> vertex_attribute_formats;
+    std::array<u32, 16> vertex_attribute_elements{};
+    std::array<bool, 16> vertex_attribute_is_default;
+    int num_total_attributes = 0;
+};
+
+} // namespace Pica

src/video_core/pica_state.h

@@ -1,255 +0,0 @@
-// Copyright 2016 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <array>
-#include <boost/serialization/array.hpp>
-#include <boost/serialization/split_member.hpp>
-#include "common/bit_field.h"
-#include "common/common_types.h"
-#include "common/vector_math.h"
-#include "core/memory.h"
-#include "video_core/geometry_pipeline.h"
-#include "video_core/primitive_assembly.h"
-#include "video_core/regs.h"
-#include "video_core/shader/shader.h"
-#include "video_core/video_core.h"
-
-// Boost::serialization doesn't like union types for some reason,
-// so we need to mark arrays of union values with a special serialization method
-template <typename Value, size_t Size>
-struct UnionArray : public std::array<Value, Size> {
-private:
-    template <class Archive>
-    void serialize(Archive& ar, const unsigned int) {
-        static_assert(sizeof(Value) == sizeof(u32));
-        ar&* static_cast<u32(*)[Size]>(static_cast<void*>(this->data()));
-    }
-    friend class boost::serialization::access;
-};
-
-namespace Pica {
-
-/// Struct used to describe current Pica state
-struct State {
-    State();
-    void Reset();
-
-    /// Pica registers
-    Regs regs;
-
-    Shader::ShaderSetup vs;
-    Shader::ShaderSetup gs;
-
-    Shader::AttributeBuffer input_default_attributes;
-
-    struct ProcTex {
-        union ValueEntry {
-            u32 raw;
-
-            // LUT value, encoded as 12-bit fixed point, with 12 fraction bits
-            BitField<0, 12, u32> value; // 0.0.12 fixed point
-
-            // Difference between two entry values. Used for efficient interpolation.
-            // 0.0.12 fixed point with two's complement. The range is [-0.5, 0.5).
-            // Note: the type of this is different from the one of lighting LUT
-            BitField<12, 12, s32> difference;
-
-            float ToFloat() const {
-                return static_cast<float>(value) / 4095.f;
-            }
-
-            float DiffToFloat() const {
-                return static_cast<float>(difference) / 4095.f;
-            }
-        };
-
-        union ColorEntry {
-            u32 raw;
-            BitField<0, 8, u32> r;
-            BitField<8, 8, u32> g;
-            BitField<16, 8, u32> b;
-            BitField<24, 8, u32> a;
-
-            Common::Vec4<u8> ToVector() const {
-                return {static_cast<u8>(r), static_cast<u8>(g), static_cast<u8>(b),
-                        static_cast<u8>(a)};
-            }
-        };
-
-        union ColorDifferenceEntry {
-            u32 raw;
-            BitField<0, 8, s32> r; // half of the difference between two ColorEntry
-            BitField<8, 8, s32> g;
-            BitField<16, 8, s32> b;
-            BitField<24, 8, s32> a;
-
-            Common::Vec4<s32> ToVector() const {
-                return Common::Vec4<s32>{r, g, b, a} * 2;
-            }
-        };
-
-        UnionArray<ValueEntry, 128> noise_table;
-        UnionArray<ValueEntry, 128> color_map_table;
-        UnionArray<ValueEntry, 128> alpha_map_table;
-        UnionArray<ColorEntry, 256> color_table;
-        UnionArray<ColorDifferenceEntry, 256> color_diff_table;
-
-    private:
-        friend class boost::serialization::access;
-        template <class Archive>
-        void serialize(Archive& ar, const unsigned int file_version) {
-            ar& noise_table;
-            ar& color_map_table;
-            ar& alpha_map_table;
-            ar& color_table;
-            ar& color_diff_table;
-        }
-    } proctex;
-
-    struct Lighting {
-        union LutEntry {
-            // Used for raw access
-            u32 raw;
-
-            // LUT value, encoded as 12-bit fixed point, with 12 fraction bits
-            BitField<0, 12, u32> value; // 0.0.12 fixed point
-
-            // Used for efficient interpolation.
-            BitField<12, 11, u32> difference; // 0.0.11 fixed point
-            BitField<23, 1, u32> neg_difference;
-
-            float ToFloat() const {
-                return static_cast<float>(value) / 4095.f;
-            }
-
-            float DiffToFloat() const {
-                float diff = static_cast<float>(difference) / 2047.f;
-                return neg_difference ? -diff : diff;
-            }
-
-            template <class Archive>
-            void serialize(Archive& ar, const unsigned int file_version) {
-                ar& raw;
-            }
-        };
-
-        std::array<UnionArray<LutEntry, 256>, 24> luts;
-    } lighting;
-
-    struct {
-        union LutEntry {
-            // Used for raw access
-            u32 raw;
-
-            BitField<0, 13, s32> difference; // 1.1.11 fixed point
-            BitField<13, 11, u32> value;     // 0.0.11 fixed point
-
-            float ToFloat() const {
-                return static_cast<float>(value) / 2047.0f;
-            }
-
-            float DiffToFloat() const {
-                return static_cast<float>(difference) / 2047.0f;
-            }
-        };
-
-        UnionArray<LutEntry, 128> lut;
-    } fog;
-
-    /// Current Pica command list
-    struct {
-        PAddr addr; // This exists only for serialization
-        const u32* head_ptr;
-        const u32* current_ptr;
-        u32 length;
-    } cmd_list;
-
-    /// Struct used to describe immediate mode rendering state
-    struct ImmediateModeState {
-        // Used to buffer partial vertices for immediate-mode rendering.
-        Shader::AttributeBuffer input_vertex;
-        // Index of the next attribute to be loaded into `input_vertex`.
-        u32 current_attribute = 0;
-        // Indicates the immediate mode just started and the geometry pipeline needs to reconfigure
-        bool reset_geometry_pipeline = true;
-
-    private:
-        friend class boost::serialization::access;
-        template <class Archive>
-        void serialize(Archive& ar, const unsigned int file_version) {
-            ar& input_vertex;
-            ar& current_attribute;
-            ar& reset_geometry_pipeline;
-        }
-
-    } immediate;
-
-    // the geometry shader needs to be kept in the global state because some shaders relie on
-    // preserved register value across shader invocation.
-    // TODO: also bring the three vertex shader units here and implement the shader scheduler.
-    Shader::GSUnitState gs_unit;
-
-    GeometryPipeline geometry_pipeline;
-
-    // This is constructed with a dummy triangle topology
-    PrimitiveAssembler<Shader::OutputVertex> primitive_assembler;
-
-    int vs_float_regs_counter = 0;
-    std::array<u32, 4> vs_uniform_write_buffer{};
-
-    int gs_float_regs_counter = 0;
-    std::array<u32, 4> gs_uniform_write_buffer{};
-
-    int default_attr_counter = 0;
-    std::array<u32, 3> default_attr_write_buffer{};
-
-private:
-    friend class boost::serialization::access;
-    template <class Archive>
-    void serialize(Archive& ar, const unsigned int file_version) {
-        ar& regs.reg_array;
-        ar& vs;
-        ar& gs;
-        ar& input_default_attributes;
-        ar& proctex;
-        ar& lighting.luts;
-        ar& fog.lut;
-        ar& cmd_list.addr;
-        ar& cmd_list.length;
-        ar& immediate;
-        ar& gs_unit;
-        ar& geometry_pipeline;
-        ar& primitive_assembler;
-        ar& vs_float_regs_counter;
-        ar& boost::serialization::make_array(vs_uniform_write_buffer.data(),
-                                             vs_uniform_write_buffer.size());
-        ar& gs_float_regs_counter;
-        ar& boost::serialization::make_array(gs_uniform_write_buffer.data(),
-                                             gs_uniform_write_buffer.size());
-        ar& default_attr_counter;
-        ar& boost::serialization::make_array(default_attr_write_buffer.data(),
-                                             default_attr_write_buffer.size());
-        boost::serialization::split_member(ar, *this, file_version);
-    }
-
-    template <class Archive>
-    void save(Archive& ar, const unsigned int file_version) const {
-        ar << static_cast<u32>(cmd_list.current_ptr - cmd_list.head_ptr);
-    }
-
-    template <class Archive>
-    void load(Archive& ar, const unsigned int file_version) {
-        u32 offset{};
-        ar >> offset;
-        cmd_list.head_ptr =
-            reinterpret_cast<u32*>(VideoCore::g_memory->GetPhysicalPointer(cmd_list.addr));
-        cmd_list.current_ptr = cmd_list.head_ptr + offset;
-    }
-};
-
-extern State g_state; ///< Current Pica state
-
-} // namespace Pica

src/video_core/primitive_assembly.cpp

@@ -1,87 +0,0 @@
-// Copyright 2014 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include "common/logging/log.h"
-#include "video_core/primitive_assembly.h"
-#include "video_core/regs_pipeline.h"
-#include "video_core/shader/shader.h"
-
-namespace Pica {
-
-template <typename VertexType>
-PrimitiveAssembler<VertexType>::PrimitiveAssembler(PipelineRegs::TriangleTopology topology)
-    : topology(topology) {}
-
-template <typename VertexType>
-void PrimitiveAssembler<VertexType>::SubmitVertex(const VertexType& vtx,
-                                                  const TriangleHandler& triangle_handler) {
-    switch (topology) {
-    case PipelineRegs::TriangleTopology::List:
-    case PipelineRegs::TriangleTopology::Shader:
-        if (buffer_index < 2) {
-            buffer[buffer_index++] = vtx;
-        } else {
-            buffer_index = 0;
-            if (topology == PipelineRegs::TriangleTopology::Shader && winding) {
-                triangle_handler(buffer[1], buffer[0], vtx);
-                winding = false;
-            } else {
-                triangle_handler(buffer[0], buffer[1], vtx);
-            }
-        }
-        break;
-
-    case PipelineRegs::TriangleTopology::Strip:
-    case PipelineRegs::TriangleTopology::Fan:
-        if (strip_ready)
-            triangle_handler(buffer[0], buffer[1], vtx);
-
-        buffer[buffer_index] = vtx;
-
-        strip_ready |= (buffer_index == 1);
-
-        if (topology == PipelineRegs::TriangleTopology::Strip)
-            buffer_index = !buffer_index;
-        else if (topology == PipelineRegs::TriangleTopology::Fan)
-            buffer_index = 1;
-        break;
-
-    default:
-        LOG_ERROR(HW_GPU, "Unknown triangle topology {:x}:", (int)topology);
-        break;
-    }
-}
-
-template <typename VertexType>
-void PrimitiveAssembler<VertexType>::SetWinding() {
-    winding = true;
-}
-
-template <typename VertexType>
-void PrimitiveAssembler<VertexType>::Reset() {
-    buffer_index = 0;
-    strip_ready = false;
-    winding = false;
-}
-
-template <typename VertexType>
-void PrimitiveAssembler<VertexType>::Reconfigure(PipelineRegs::TriangleTopology topology) {
-    Reset();
-    this->topology = topology;
-}
-
-template <typename VertexType>
-bool PrimitiveAssembler<VertexType>::IsEmpty() const {
-    return buffer_index == 0 && strip_ready == false;
-}
-
-template <typename VertexType>
-PipelineRegs::TriangleTopology PrimitiveAssembler<VertexType>::GetTopology() const {
-    return topology;
-}
-
-// explicitly instantiate use cases
-template struct PrimitiveAssembler<Shader::OutputVertex>;
-
-} // namespace Pica

src/video_core/rasterizer_accelerated.cpp

@@ -2,10 +2,9 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <limits>
 #include "common/alignment.h"
 #include "core/memory.h"
-#include "video_core/pica_state.h"
+#include "video_core/pica/pica_core.h"
 #include "video_core/rasterizer_accelerated.h"
 
 namespace VideoCore {
@@ -22,7 +21,7 @@ static Common::Vec3f LightColor(const Pica::LightingRegs::LightColor& color) {
     return Common::Vec3u{color.r, color.g, color.b} / 255.0f;
 }
 
-RasterizerAccelerated::HardwareVertex::HardwareVertex(const Pica::Shader::OutputVertex& v,
+RasterizerAccelerated::HardwareVertex::HardwareVertex(const Pica::OutputVertex& v,
                                                       bool flip_quaternion) {
     position[0] = v.pos.x.ToFloat32();
     position[1] = v.pos.y.ToFloat32();
@@ -52,8 +51,8 @@ RasterizerAccelerated::HardwareVertex::HardwareVertex(const Pica::Shader::Output
     }
 }
 
-RasterizerAccelerated::RasterizerAccelerated(Memory::MemorySystem& memory_)
+RasterizerAccelerated::RasterizerAccelerated(Memory::MemorySystem& memory_, Pica::PicaCore& pica_)
-    : memory{memory_}, regs{Pica::g_state.regs} {
+    : memory{memory_}, pica{pica_}, regs{pica.regs.internal} {
     fs_uniform_block_data.lighting_lut_dirty.fill(true);
 }
 
@@ -82,9 +81,8 @@ static bool AreQuaternionsOpposite(Common::Vec4<f24> qa, Common::Vec4<f24> qb) {
     return (Common::Dot(a, b) < 0.f);
 }
 
-void RasterizerAccelerated::AddTriangle(const Pica::Shader::OutputVertex& v0,
+void RasterizerAccelerated::AddTriangle(const Pica::OutputVertex& v0, const Pica::OutputVertex& v1,
-                                        const Pica::Shader::OutputVertex& v1,
+                                        const Pica::OutputVertex& v2) {
-                                        const Pica::Shader::OutputVertex& v2) {
     vertex_batch.emplace_back(v0, false);
     vertex_batch.emplace_back(v1, AreQuaternionsOpposite(v0.quat, v1.quat));
     vertex_batch.emplace_back(v2, AreQuaternionsOpposite(v0.quat, v2.quat));
@@ -146,7 +144,7 @@ void RasterizerAccelerated::SyncEntireState() {
     }
 
     SyncGlobalAmbient();
-    for (unsigned light_index = 0; light_index < 8; light_index++) {
+    for (u32 light_index = 0; light_index < 8; light_index++) {
         SyncLightSpecular0(light_index);
         SyncLightSpecular1(light_index);
         SyncLightDiffuse(light_index);
@@ -162,7 +160,7 @@ void RasterizerAccelerated::SyncEntireState() {
     SyncShadowBias();
     SyncShadowTextureBias();
 
-    for (unsigned tex_index = 0; tex_index < 3; tex_index++) {
+    for (u32 tex_index = 0; tex_index < 3; tex_index++) {
         SyncTextureLodBias(tex_index);
     }
 }

src/video_core/rasterizer_accelerated.h

@@ -6,7 +6,6 @@
 
 #include "common/vector_math.h"
 #include "video_core/rasterizer_interface.h"
-#include "video_core/regs_texturing.h"
 #include "video_core/shader/generator/pica_fs_config.h"
 #include "video_core/shader/generator/shader_uniforms.h"
 
@@ -15,19 +14,21 @@ class MemorySystem;
 }
 
 namespace Pica {
-struct Regs;
+class PicaCore;
 }
 
 namespace VideoCore {
 
 class RasterizerAccelerated : public RasterizerInterface {
 public:
-    RasterizerAccelerated(Memory::MemorySystem& memory);
+    explicit RasterizerAccelerated(Memory::MemorySystem& memory, Pica::PicaCore& pica);
     virtual ~RasterizerAccelerated() = default;
 
-    void AddTriangle(const Pica::Shader::OutputVertex& v0, const Pica::Shader::OutputVertex& v1,
+    void AddTriangle(const Pica::OutputVertex& v0, const Pica::OutputVertex& v1,
-                     const Pica::Shader::OutputVertex& v2) override;
+                     const Pica::OutputVertex& v2) override;
+
     void NotifyPicaRegisterChanged(u32 id) override;
+
     void SyncEntireState() override;
 
 protected:
@@ -128,7 +129,7 @@ protected:
     /// Structure that the hardware rendered vertices are composed of
     struct HardwareVertex {
         HardwareVertex() = default;
-        HardwareVertex(const Pica::Shader::OutputVertex& v, bool flip_quaternion);
+        HardwareVertex(const Pica::OutputVertex& v, bool flip_quaternion);
 
         Common::Vec4f position;
         Common::Vec4f color;
@@ -151,7 +152,8 @@ protected:
 
 protected:
     Memory::MemorySystem& memory;
-    Pica::Regs& regs;
+    Pica::PicaCore& pica;
+    Pica::RegsInternal& regs;
 
     std::vector<HardwareVertex> vertex_batch;
     Pica::Shader::UserConfig user_config{};
@@ -159,8 +161,8 @@ protected:
 
     VSUniformBlockData vs_uniform_block_data{};
     FSUniformBlockData fs_uniform_block_data{};
-    std::array<std::array<Common::Vec2f, 256>, Pica::LightingRegs::NumLightingSampler>
+    using LightLUT = std::array<Common::Vec2f, 256>;
-        lighting_lut_data{};
+    std::array<LightLUT, Pica::LightingRegs::NumLightingSampler> lighting_lut_data{};
     std::array<Common::Vec2f, 128> fog_lut_data{};
     std::array<Common::Vec2f, 128> proctex_noise_lut_data{};
     std::array<Common::Vec2f, 128> proctex_color_map_data{};

src/video_core/rasterizer_cache/framebuffer_base.h

@@ -6,9 +6,9 @@
 
 #include "common/hash.h"
 #include "common/math_util.h"
+#include "video_core/pica/regs_rasterizer.h"
 #include "video_core/rasterizer_cache/slot_id.h"
 #include "video_core/rasterizer_cache/surface_params.h"
-#include "video_core/regs_rasterizer.h"
 
 namespace VideoCore {
 

src/video_core/rasterizer_cache/pixel_format.cpp

@@ -2,6 +2,7 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include "video_core/pica/regs_external.h"
 #include "video_core/rasterizer_cache/pixel_format.h"
 
 namespace VideoCore {
@@ -134,17 +135,17 @@ PixelFormat PixelFormatFromDepthFormat(Pica::FramebufferRegs::DepthFormat format
     }
 }
 
-PixelFormat PixelFormatFromGPUPixelFormat(GPU::Regs::PixelFormat format) {
+PixelFormat PixelFormatFromGPUPixelFormat(Pica::PixelFormat format) {
     switch (format) {
-    case GPU::Regs::PixelFormat::RGBA8:
+    case Pica::PixelFormat::RGBA8:
         return PixelFormat::RGBA8;
-    case GPU::Regs::PixelFormat::RGB8:
+    case Pica::PixelFormat::RGB8:
         return PixelFormat::RGB8;
-    case GPU::Regs::PixelFormat::RGB565:
+    case Pica::PixelFormat::RGB565:
         return PixelFormat::RGB565;
-    case GPU::Regs::PixelFormat::RGB5A1:
+    case Pica::PixelFormat::RGB5A1:
         return PixelFormat::RGB5A1;
-    case GPU::Regs::PixelFormat::RGBA4:
+    case Pica::PixelFormat::RGBA4:
         return PixelFormat::RGBA4;
     default:
         return PixelFormat::Invalid;

src/video_core/rasterizer_cache/pixel_format.h

@@ -6,9 +6,12 @@
 
 #include <limits>
 #include <string_view>
-#include "core/hw/gpu.h"
+#include "video_core/pica/regs_framebuffer.h"
-#include "video_core/regs_framebuffer.h"
+#include "video_core/pica/regs_texturing.h"
-#include "video_core/regs_texturing.h"
+
+namespace Pica {
+enum class PixelFormat : u32;
+}
 
 namespace VideoCore {
 
@@ -109,6 +112,6 @@ PixelFormat PixelFormatFromColorFormat(Pica::FramebufferRegs::ColorFormat format
 
 PixelFormat PixelFormatFromDepthFormat(Pica::FramebufferRegs::DepthFormat format);
 
-PixelFormat PixelFormatFromGPUPixelFormat(GPU::Regs::PixelFormat format);
+PixelFormat PixelFormatFromGPUPixelFormat(Pica::PixelFormat format);
 
 } // namespace VideoCore

src/video_core/rasterizer_cache/rasterizer_cache.h

@@ -14,9 +14,10 @@
 #include "common/settings.h"
 #include "core/memory.h"
 #include "video_core/custom_textures/custom_tex_manager.h"
+#include "video_core/pica/regs_external.h"
+#include "video_core/pica/regs_internal.h"
 #include "video_core/rasterizer_cache/rasterizer_cache_base.h"
 #include "video_core/rasterizer_cache/surface_base.h"
-#include "video_core/regs.h"
 #include "video_core/renderer_base.h"
 #include "video_core/texture/texture_decode.h"
 
@@ -34,7 +35,7 @@ constexpr auto RangeFromInterval(const auto& map, const auto& interval) {
 template <class T>
 RasterizerCache<T>::RasterizerCache(Memory::MemorySystem& memory_,
                                     CustomTexManager& custom_tex_manager_, Runtime& runtime_,
-                                    Pica::Regs& regs_, RendererBase& renderer_)
+                                    Pica::RegsInternal& regs_, RendererBase& renderer_)
     : memory{memory_}, custom_tex_manager{custom_tex_manager_}, runtime{runtime_}, regs{regs_},
       renderer{renderer_}, resolution_scale_factor{renderer.GetResolutionScaleFactor()},
       filter{Settings::values.texture_filter.GetValue()},
@@ -151,7 +152,7 @@ void RasterizerCache<T>::RemoveTextureCubeFace(SurfaceId surface_id) {
 }
 
 template <class T>
-bool RasterizerCache<T>::AccelerateTextureCopy(const GPU::Regs::DisplayTransferConfig& config) {
+bool RasterizerCache<T>::AccelerateTextureCopy(const Pica::DisplayTransferConfig& config) {
     const DebugScope scope{runtime, Common::Vec4f{0.f, 0.f, 1.f, 1.f},
                            "RasterizerCache::AccelerateTextureCopy ({})", config.DebugName()};
 
@@ -249,7 +250,7 @@ bool RasterizerCache<T>::AccelerateTextureCopy(const GPU::Regs::DisplayTransferC
 }
 
 template <class T>
-bool RasterizerCache<T>::AccelerateDisplayTransfer(const GPU::Regs::DisplayTransferConfig& config) {
+bool RasterizerCache<T>::AccelerateDisplayTransfer(const Pica::DisplayTransferConfig& config) {
     const DebugScope scope{runtime, Common::Vec4f{0.f, 0.f, 1.f, 1.f},
                            "RasterizerCache::AccelerateDisplayTransfer ({})", config.DebugName()};
 
@@ -274,10 +275,9 @@ bool RasterizerCache<T>::AccelerateDisplayTransfer(const GPU::Regs::DisplayTrans
 
     // Using flip_vertically alongside crop_input_lines produces skewed output on hardware.
     // We have to emulate this because some games rely on this behaviour to render correctly.
-    if (config.flip_vertically && config.crop_input_lines &&
+    if (config.flip_vertically && config.crop_input_lines) {
-        config.input_width > config.output_width) {
         dst_params.addr += (config.input_width - config.output_width) * (config.output_height - 1) *
-                           GPU::Regs::BytesPerPixel(config.output_format);
+                           Pica::BytesPerPixel(config.output_format);
     }
 
     auto [src_surface_id, src_rect] = GetSurfaceSubRect(src_params, ScaleMatch::Ignore, true);
@@ -320,7 +320,7 @@ bool RasterizerCache<T>::AccelerateDisplayTransfer(const GPU::Regs::DisplayTrans
 }
 
 template <class T>
-bool RasterizerCache<T>::AccelerateFill(const GPU::Regs::MemoryFillConfig& config) {
+bool RasterizerCache<T>::AccelerateFill(const Pica::MemoryFillConfig& config) {
     const DebugScope scope{runtime, Common::Vec4f{1.f, 0.f, 1.f, 1.f},
                            "RasterizerCache::AccelerateFill ({})", config.DebugName()};
 

src/video_core/rasterizer_cache/rasterizer_cache_base.h

@@ -12,6 +12,7 @@
 #include <vector>
 #include <boost/icl/interval_map.hpp>
 #include <tsl/robin_map.h>
+
 #include "video_core/rasterizer_cache/framebuffer_base.h"
 #include "video_core/rasterizer_cache/sampler_params.h"
 #include "video_core/rasterizer_cache/surface_params.h"
@@ -22,8 +23,10 @@ class MemorySystem;
 }
 
 namespace Pica {
-struct Regs;
+struct RegsInternal;
-}
+struct DisplayTransferConfig;
+struct MemoryFillConfig;
+} // namespace Pica
 
 namespace Pica::Texture {
 struct TextureInfo;
@@ -74,20 +77,20 @@ class RasterizerCache {
 
 public:
     explicit RasterizerCache(Memory::MemorySystem& memory, CustomTexManager& custom_tex_manager,
-                             Runtime& runtime, Pica::Regs& regs, RendererBase& renderer);
+                             Runtime& runtime, Pica::RegsInternal& regs, RendererBase& renderer);
     ~RasterizerCache();
 
     /// Notify the cache that a new frame has been queued
     void TickFrame();
 
     /// Perform hardware accelerated texture copy according to the provided configuration
-    bool AccelerateTextureCopy(const GPU::Regs::DisplayTransferConfig& config);
+    bool AccelerateTextureCopy(const Pica::DisplayTransferConfig& config);
 
     /// Perform hardware accelerated display transfer according to the provided configuration
-    bool AccelerateDisplayTransfer(const GPU::Regs::DisplayTransferConfig& config);
+    bool AccelerateDisplayTransfer(const Pica::DisplayTransferConfig& config);
 
     /// Perform hardware accelerated memory fill according to the provided configuration
-    bool AccelerateFill(const GPU::Regs::MemoryFillConfig& config);
+    bool AccelerateFill(const Pica::MemoryFillConfig& config);
 
     /// Returns a reference to the surface object assigned to surface_id
     Surface& GetSurface(SurfaceId surface_id);
@@ -212,7 +215,7 @@ private:
     Memory::MemorySystem& memory;
     CustomTexManager& custom_tex_manager;
     Runtime& runtime;
-    Pica::Regs& regs;
+    Pica::RegsInternal& regs;
     RendererBase& renderer;
     std::unordered_map<TextureCubeConfig, TextureCube> texture_cube_cache;
     tsl::robin_pg_map<u64, std::vector<SurfaceId>, Common::IdentityHash<u64>> page_table;