Merge pull request #3281 from jroweboy/texcache-pt2

Texture Cache Rework

src/citra/config.cpp

@@ -89,7 +89,7 @@ void Config::ReadValues() {
     Settings::values.use_hw_renderer = sdl2_config->GetBoolean("Renderer", "use_hw_renderer", true);
     Settings::values.use_shader_jit = sdl2_config->GetBoolean("Renderer", "use_shader_jit", true);
     Settings::values.resolution_factor =
-        (float)sdl2_config->GetReal("Renderer", "resolution_factor", 1.0);
+        static_cast<u16>(sdl2_config->GetInteger("Renderer", "resolution_factor", 1));
     Settings::values.use_vsync = sdl2_config->GetBoolean("Renderer", "use_vsync", false);
     Settings::values.use_frame_limit = sdl2_config->GetBoolean("Renderer", "use_frame_limit", true);
     Settings::values.frame_limit =

src/citra_qt/configuration/config.cpp

@@ -75,7 +75,8 @@ void Config::ReadValues() {
     qt_config->beginGroup("Renderer");
     Settings::values.use_hw_renderer = qt_config->value("use_hw_renderer", true).toBool();
     Settings::values.use_shader_jit = qt_config->value("use_shader_jit", true).toBool();
-    Settings::values.resolution_factor = qt_config->value("resolution_factor", 1.0).toFloat();
+    Settings::values.resolution_factor =
+        static_cast<u16>(qt_config->value("resolution_factor", 1).toInt());
     Settings::values.use_vsync = qt_config->value("use_vsync", false).toBool();
     Settings::values.use_frame_limit = qt_config->value("use_frame_limit", true).toBool();
     Settings::values.frame_limit = qt_config->value("frame_limit", 100).toInt();
@@ -240,7 +241,7 @@ void Config::SaveValues() {
     qt_config->beginGroup("Renderer");
     qt_config->setValue("use_hw_renderer", Settings::values.use_hw_renderer);
     qt_config->setValue("use_shader_jit", Settings::values.use_shader_jit);
-    qt_config->setValue("resolution_factor", (double)Settings::values.resolution_factor);
+    qt_config->setValue("resolution_factor", Settings::values.resolution_factor);
     qt_config->setValue("use_vsync", Settings::values.use_vsync);
     qt_config->setValue("use_frame_limit", Settings::values.use_frame_limit);
     qt_config->setValue("frame_limit", Settings::values.frame_limit);

src/citra_qt/configuration/configure_graphics.cpp

@@ -23,81 +23,11 @@ ConfigureGraphics::ConfigureGraphics(QWidget* parent)
 
 ConfigureGraphics::~ConfigureGraphics() {}
 
-enum class Resolution : int {
-    Auto,
-    Scale1x,
-    Scale2x,
-    Scale3x,
-    Scale4x,
-    Scale5x,
-    Scale6x,
-    Scale7x,
-    Scale8x,
-    Scale9x,
-    Scale10x,
-};
-
-float ToResolutionFactor(Resolution option) {
-    switch (option) {
-    case Resolution::Auto:
-        return 0.f;
-    case Resolution::Scale1x:
-        return 1.f;
-    case Resolution::Scale2x:
-        return 2.f;
-    case Resolution::Scale3x:
-        return 3.f;
-    case Resolution::Scale4x:
-        return 4.f;
-    case Resolution::Scale5x:
-        return 5.f;
-    case Resolution::Scale6x:
-        return 6.f;
-    case Resolution::Scale7x:
-        return 7.f;
-    case Resolution::Scale8x:
-        return 8.f;
-    case Resolution::Scale9x:
-        return 9.f;
-    case Resolution::Scale10x:
-        return 10.f;
-    }
-    return 0.f;
-}
-
-Resolution FromResolutionFactor(float factor) {
-    if (factor == 0.f) {
-        return Resolution::Auto;
-    } else if (factor == 1.f) {
-        return Resolution::Scale1x;
-    } else if (factor == 2.f) {
-        return Resolution::Scale2x;
-    } else if (factor == 3.f) {
-        return Resolution::Scale3x;
-    } else if (factor == 4.f) {
-        return Resolution::Scale4x;
-    } else if (factor == 5.f) {
-        return Resolution::Scale5x;
-    } else if (factor == 6.f) {
-        return Resolution::Scale6x;
-    } else if (factor == 7.f) {
-        return Resolution::Scale7x;
-    } else if (factor == 8.f) {
-        return Resolution::Scale8x;
-    } else if (factor == 9.f) {
-        return Resolution::Scale9x;
-    } else if (factor == 10.f) {
-        return Resolution::Scale10x;
-    }
-    return Resolution::Auto;
-}
-
 void ConfigureGraphics::setConfiguration() {
     ui->toggle_hw_renderer->setChecked(Settings::values.use_hw_renderer);
     ui->resolution_factor_combobox->setEnabled(Settings::values.use_hw_renderer);
     ui->toggle_shader_jit->setChecked(Settings::values.use_shader_jit);
-    ui->resolution_factor_combobox->setCurrentIndex(
-        static_cast<int>(FromResolutionFactor(Settings::values.resolution_factor)));
+    ui->resolution_factor_combobox->setCurrentIndex(Settings::values.resolution_factor);
     ui->toggle_vsync->setChecked(Settings::values.use_vsync);
     ui->toggle_frame_limit->setChecked(Settings::values.use_frame_limit);
     ui->frame_limit->setValue(Settings::values.frame_limit);
@@ -109,7 +39,7 @@ void ConfigureGraphics::applyConfiguration() {
     Settings::values.use_hw_renderer = ui->toggle_hw_renderer->isChecked();
     Settings::values.use_shader_jit = ui->toggle_shader_jit->isChecked();
     Settings::values.resolution_factor =
-        ToResolutionFactor(static_cast<Resolution>(ui->resolution_factor_combobox->currentIndex()));
+        static_cast<u16>(ui->resolution_factor_combobox->currentIndex());
     Settings::values.use_vsync = ui->toggle_vsync->isChecked();
     Settings::values.use_frame_limit = ui->toggle_frame_limit->isChecked();
     Settings::values.frame_limit = ui->frame_limit->value();

src/common/math_util.h

@@ -29,7 +29,7 @@ struct Rectangle {
     T right;
     T bottom;
 
-    Rectangle() {}
+    Rectangle() = default;
 
     Rectangle(T left, T top, T right, T bottom)
         : left(left), top(top), right(right), bottom(bottom) {}

src/core/frontend/framebuffer_layout.cpp

@@ -16,8 +16,8 @@ static const float TOP_SCREEN_ASPECT_RATIO =
 static const float BOT_SCREEN_ASPECT_RATIO =
     static_cast<float>(Core::kScreenBottomHeight) / Core::kScreenBottomWidth;
 
-float FramebufferLayout::GetScalingRatio() const {
-    return static_cast<float>(top_screen.GetWidth()) / Core::kScreenTopWidth;
+u16 FramebufferLayout::GetScalingRatio() const {
+    return static_cast<u16>(((top_screen.GetWidth() - 1) / Core::kScreenTopWidth) + 1);
 }
 
 // Finds the largest size subrectangle contained in window area that is confined to the aspect ratio

src/core/frontend/framebuffer_layout.h

@@ -21,7 +21,7 @@ struct FramebufferLayout {
      * Returns the ration of pixel size of the top screen, compared to the native size of the 3DS
      * screen.
      */
-    float GetScalingRatio() const;
+    u16 GetScalingRatio() const;
 };
 
 /**

src/core/hle/kernel/vm_manager.cpp

@@ -58,7 +58,6 @@ void VMManager::Reset() {
 
     page_table.pointers.fill(nullptr);
     page_table.attributes.fill(Memory::PageType::Unmapped);
-    page_table.cached_res_count.fill(0);
 
     UpdatePageTableForVMA(initial_vma);
 }

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

@@ -465,7 +465,7 @@ static void ExecuteCommand(const Command& command, u32 thread_id) {
                                              command.dma_request.size, Memory::FlushMode::Flush);
         Memory::RasterizerFlushVirtualRegion(command.dma_request.dest_address,
                                              command.dma_request.size,
-                                             Memory::FlushMode::FlushAndInvalidate);
+                                             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.

src/core/hw/gpu.cpp

@@ -96,20 +96,11 @@ static void MemoryFill(const Regs::MemoryFillConfig& config) {
     u8* start = Memory::GetPhysicalPointer(start_addr);
     u8* end = Memory::GetPhysicalPointer(end_addr);
 
-    // TODO: Consider always accelerating and returning vector of
-    //       regions that the accelerated fill did not cover to
-    //       reduce/eliminate the fill that the cpu has to do.
-    //       This would also mean that the flush below is not needed.
-    //       Fill should first flush all surfaces that touch but are
-    //       not completely within the fill range.
-    //       Then fill all completely covered surfaces, and return the
-    //       regions that were between surfaces or within the touching
-    //       ones for cpu to manually fill here.
     if (VideoCore::g_renderer->Rasterizer()->AccelerateFill(config))
         return;
 
-    Memory::RasterizerFlushAndInvalidateRegion(config.GetStartAddress(),
-                                               config.GetEndAddress() - config.GetStartAddress());
+    Memory::RasterizerInvalidateRegion(config.GetStartAddress(),
+                                       config.GetEndAddress() - config.GetStartAddress());
 
     if (config.fill_24bit) {
         // fill with 24-bit values
@@ -199,7 +190,7 @@ static void DisplayTransfer(const Regs::DisplayTransferConfig& config) {
     u32 output_size = output_width * output_height * GPU::Regs::BytesPerPixel(config.output_format);
 
     Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(), input_size);
-    Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(), output_size);
+    Memory::RasterizerInvalidateRegion(config.GetPhysicalOutputAddress(), output_size);
 
     for (u32 y = 0; y < output_height; ++y) {
         for (u32 x = 0; x < output_width; ++x) {
@@ -363,8 +354,10 @@ static void TextureCopy(const Regs::DisplayTransferConfig& config) {
 
     size_t contiguous_output_size =
         config.texture_copy.size / output_width * (output_width + output_gap);
-    Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(),
-                                               static_cast<u32>(contiguous_output_size));
+    // 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;
@@ -570,4 +563,4 @@ void Shutdown() {
     LOG_DEBUG(HW_GPU, "shutdown OK");
 }
 
-} // namespace
+} // namespace GPU

src/core/memory.cpp

@@ -50,7 +50,6 @@ static void MapPages(PageTable& page_table, u32 base, u32 size, u8* memory, Page
 
         page_table.attributes[base] = type;
         page_table.pointers[base] = memory;
-        page_table.cached_res_count[base] = 0;
 
         base += 1;
         if (memory != nullptr)
@@ -200,7 +199,7 @@ void Write(const VAddr vaddr, const T data) {
         ASSERT_MSG(false, "Mapped memory page without a pointer @ %08X", vaddr);
         break;
     case PageType::RasterizerCachedMemory: {
-        RasterizerFlushVirtualRegion(vaddr, sizeof(T), FlushMode::FlushAndInvalidate);
+        RasterizerFlushVirtualRegion(vaddr, sizeof(T), FlushMode::Invalidate);
         std::memcpy(GetPointerFromVMA(vaddr), &data, sizeof(T));
         break;
     }
@@ -208,7 +207,7 @@ void Write(const VAddr vaddr, const T data) {
         WriteMMIO<T>(GetMMIOHandler(vaddr), vaddr, data);
         break;
     case PageType::RasterizerCachedSpecial: {
-        RasterizerFlushVirtualRegion(vaddr, sizeof(T), FlushMode::FlushAndInvalidate);
+        RasterizerFlushVirtualRegion(vaddr, sizeof(T), FlushMode::Invalidate);
         WriteMMIO<T>(GetMMIOHandler(vaddr), vaddr, data);
         break;
     }
@@ -334,7 +333,7 @@ u8* GetPhysicalPointer(PAddr address) {
     return target_pointer;
 }
 
-void RasterizerMarkRegionCached(PAddr start, u32 size, int count_delta) {
+void RasterizerMarkRegionCached(PAddr start, u32 size, bool cached) {
     if (start == 0) {
         return;
     }
@@ -355,14 +354,10 @@ void RasterizerMarkRegionCached(PAddr start, u32 size, int count_delta) {
         }
         VAddr vaddr = *maybe_vaddr;
 
-        u8& res_count = current_page_table->cached_res_count[vaddr >> PAGE_BITS];
-        ASSERT_MSG(count_delta <= UINT8_MAX - res_count,
-                   "Rasterizer resource cache counter overflow!");
-        ASSERT_MSG(count_delta >= -res_count, "Rasterizer resource cache counter underflow!");
+        PageType& page_type = current_page_table->attributes[vaddr >> PAGE_BITS];
 
-        // Switch page type to cached if now cached
-        if (res_count == 0) {
-            PageType& page_type = current_page_table->attributes[vaddr >> PAGE_BITS];
+        if (cached) {
+            // Switch page type to cached if now cached
             switch (page_type) {
             case PageType::Unmapped:
                 // It is not necessary for a process to have this region mapped into its address
@@ -378,13 +373,8 @@ void RasterizerMarkRegionCached(PAddr start, u32 size, int count_delta) {
             default:
                 UNREACHABLE();
             }
-        }
-
-        res_count += count_delta;
-
-        // Switch page type to uncached if now uncached
-        if (res_count == 0) {
-            PageType& page_type = current_page_table->attributes[vaddr >> PAGE_BITS];
+        } else {
+            // Switch page type to uncached if now uncached
             switch (page_type) {
             case PageType::Unmapped:
                 // It is not necessary for a process to have this region mapped into its address
@@ -414,52 +404,69 @@ void RasterizerMarkRegionCached(PAddr start, u32 size, int count_delta) {
 }
 
 void RasterizerFlushRegion(PAddr start, u32 size) {
-    if (VideoCore::g_renderer != nullptr) {
-        VideoCore::g_renderer->Rasterizer()->FlushRegion(start, 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) {
-        VideoCore::g_renderer->Rasterizer()->FlushAndInvalidateRegion(start, size);
+    if (VideoCore::g_renderer == nullptr) {
+        return;
     }
+
+    VideoCore::g_renderer->Rasterizer()->FlushAndInvalidateRegion(start, size);
 }
 
 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) {
-        VAddr end = start + size;
-
-        auto CheckRegion = [&](VAddr region_start, VAddr region_end) {
-            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 = TryVirtualToPhysicalAddress(overlap_start).value();
-            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::FlushAndInvalidate:
-                rasterizer->FlushAndInvalidateRegion(physical_start, overlap_size);
-                break;
-            }
-        };
-
-        CheckRegion(LINEAR_HEAP_VADDR, LINEAR_HEAP_VADDR_END);
-        CheckRegion(NEW_LINEAR_HEAP_VADDR, NEW_LINEAR_HEAP_VADDR_END);
-        CheckRegion(VRAM_VADDR, VRAM_VADDR_END);
+    if (VideoCore::g_renderer == nullptr) {
+        return;
     }
+
+    VAddr end = start + size;
+
+    auto CheckRegion = [&](VAddr region_start, VAddr region_end) {
+        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 = TryVirtualToPhysicalAddress(overlap_start).value();
+        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);
+    CheckRegion(NEW_LINEAR_HEAP_VADDR, NEW_LINEAR_HEAP_VADDR_END);
+    CheckRegion(VRAM_VADDR, VRAM_VADDR_END);
 }
 
 u8 Read8(const VAddr addr) {
@@ -588,7 +595,7 @@ void WriteBlock(const Kernel::Process& process, const VAddr dest_addr, const voi
         }
         case PageType::RasterizerCachedMemory: {
             RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
-                                         FlushMode::FlushAndInvalidate);
+                                         FlushMode::Invalidate);
             std::memcpy(GetPointerFromVMA(process, current_vaddr), src_buffer, copy_amount);
             break;
         }
@@ -596,7 +603,7 @@ void WriteBlock(const Kernel::Process& process, const VAddr dest_addr, const voi
             MMIORegionPointer handler = GetMMIOHandler(page_table, current_vaddr);
             DEBUG_ASSERT(handler);
             RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
-                                         FlushMode::FlushAndInvalidate);
+                                         FlushMode::Invalidate);
             handler->WriteBlock(current_vaddr, src_buffer, copy_amount);
             break;
         }
@@ -648,7 +655,7 @@ void ZeroBlock(const Kernel::Process& process, const VAddr dest_addr, const size
         }
         case PageType::RasterizerCachedMemory: {
             RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
-                                         FlushMode::FlushAndInvalidate);
+                                         FlushMode::Invalidate);
             std::memset(GetPointerFromVMA(process, current_vaddr), 0, copy_amount);
             break;
         }
@@ -656,7 +663,7 @@ void ZeroBlock(const Kernel::Process& process, const VAddr dest_addr, const size
             MMIORegionPointer handler = GetMMIOHandler(page_table, current_vaddr);
             DEBUG_ASSERT(handler);
             RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
-                                         FlushMode::FlushAndInvalidate);
+                                         FlushMode::Invalidate);
             handler->WriteBlock(current_vaddr, zeros.data(), copy_amount);
             break;
         }

src/core/memory.h

@@ -72,12 +72,6 @@ struct PageTable {
      * the corresponding entry in `pointers` MUST be set to null.
      */
     std::array<PageType, PAGE_TABLE_NUM_ENTRIES> attributes;
-
-    /**
-     * Indicates the number of externally cached resources touching a page that should be
-     * flushed before the memory is accessed
-     */
-    std::array<u8, PAGE_TABLE_NUM_ENTRIES> cached_res_count;
 };
 
 /// Physical memory regions as seen from the ARM11
@@ -245,16 +239,20 @@ boost::optional<VAddr> PhysicalToVirtualAddress(PAddr addr);
 u8* GetPhysicalPointer(PAddr address);
 
 /**
- * Adds the supplied value to the rasterizer resource cache counter of each
- * page touching the region.
+ * Mark each page touching the region as cached.
  */
-void RasterizerMarkRegionCached(PAddr start, u32 size, int count_delta);
+void RasterizerMarkRegionCached(PAddr start, u32 size, bool cached);
 
 /**
  * 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.
  */
@@ -263,6 +261,8 @@ void RasterizerFlushAndInvalidateRegion(PAddr start, u32 size);
 enum class FlushMode {
     /// Write back modified surfaces to RAM
     Flush,
+    /// Remove region from the cache
+    Invalidate,
     /// Write back modified surfaces to RAM, and also remove them from the cache
     FlushAndInvalidate,
 };

src/core/settings.h

@@ -95,7 +95,7 @@ struct Values {
     // Renderer
     bool use_hw_renderer;
     bool use_shader_jit;
-    float resolution_factor;
+    u16 resolution_factor;
     bool use_vsync;
     bool use_frame_limit;
     u16 frame_limit;

src/tests/core/arm/arm_test_common.cpp

@@ -20,7 +20,6 @@ TestEnvironment::TestEnvironment(bool mutable_memory_)
 
     page_table->pointers.fill(nullptr);
     page_table->attributes.fill(Memory::PageType::Unmapped);
-    page_table->cached_res_count.fill(0);
 
     Memory::MapIoRegion(*page_table, 0x00000000, 0x80000000, test_memory);
     Memory::MapIoRegion(*page_table, 0x80000000, 0x80000000, test_memory);

src/video_core/rasterizer_interface.h

@@ -38,6 +38,9 @@ public:
     /// Notify rasterizer that any caches of the specified region should be flushed to 3DS memory
     virtual void FlushRegion(PAddr addr, u32 size) = 0;
 
+    /// Notify rasterizer that any caches of the specified region should be invalidated
+    virtual void InvalidateRegion(PAddr addr, u32 size) = 0;
+
     /// Notify rasterizer that any caches of the specified region should be flushed to 3DS memory
     /// and invalidated
     virtual void FlushAndInvalidateRegion(PAddr addr, u32 size) = 0;

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -7,8 +7,8 @@
 #include <tuple>
 #include <utility>
 #include <glad/glad.h>
+#include "common/alignment.h"
 #include "common/assert.h"
-#include "common/color.h"
 #include "common/logging/log.h"
 #include "common/math_util.h"
 #include "common/microprofile.h"
@@ -23,6 +23,9 @@
 #include "video_core/renderer_opengl/pica_to_gl.h"
 #include "video_core/renderer_opengl/renderer_opengl.h"
 
+using PixelFormat = SurfaceParams::PixelFormat;
+using SurfaceType = SurfaceParams::SurfaceType;
+
 MICROPROFILE_DEFINE(OpenGL_Drawing, "OpenGL", "Drawing", MP_RGB(128, 128, 192));
 MICROPROFILE_DEFINE(OpenGL_Blits, "OpenGL", "Blits", MP_RGB(100, 100, 255));
 MICROPROFILE_DEFINE(OpenGL_CacheManagement, "OpenGL", "Cache Mgmt", MP_RGB(100, 255, 100));
@@ -227,21 +230,68 @@ void RasterizerOpenGL::DrawTriangles() {
     MICROPROFILE_SCOPE(OpenGL_Drawing);
     const auto& regs = Pica::g_state.regs;
 
-    // Sync and bind the framebuffer surfaces
-    CachedSurface* color_surface;
-    CachedSurface* depth_surface;
-    MathUtil::Rectangle<int> rect;
-    std::tie(color_surface, depth_surface, rect) =
-        res_cache.GetFramebufferSurfaces(regs.framebuffer.framebuffer);
+    const bool has_stencil =
+        regs.framebuffer.framebuffer.depth_format == Pica::FramebufferRegs::DepthFormat::D24S8;
 
+    const bool write_color_fb =
+        state.color_mask.red_enabled == GL_TRUE || state.color_mask.green_enabled == GL_TRUE ||
+        state.color_mask.blue_enabled == GL_TRUE || state.color_mask.alpha_enabled == GL_TRUE;
+
+    const bool write_depth_fb =
+        (state.depth.test_enabled && state.depth.write_mask == GL_TRUE) ||
+        (has_stencil && state.stencil.test_enabled && state.stencil.write_mask != 0);
+
+    const bool using_color_fb =
+        regs.framebuffer.framebuffer.GetColorBufferPhysicalAddress() != 0 && write_color_fb;
+    const bool using_depth_fb =
+        regs.framebuffer.framebuffer.GetDepthBufferPhysicalAddress() != 0 &&
+        (write_depth_fb || regs.framebuffer.output_merger.depth_test_enable != 0 ||
+         (has_stencil && state.stencil.test_enabled));
+
+    MathUtil::Rectangle<s32> viewport_rect_unscaled{
+        // These registers hold half-width and half-height, so must be multiplied by 2
+        regs.rasterizer.viewport_corner.x,  // left
+        regs.rasterizer.viewport_corner.y + // top
+            static_cast<s32>(Pica::float24::FromRaw(regs.rasterizer.viewport_size_y).ToFloat32() *
+                             2),
+        regs.rasterizer.viewport_corner.x + // right
+            static_cast<s32>(Pica::float24::FromRaw(regs.rasterizer.viewport_size_x).ToFloat32() *
+                             2),
+        regs.rasterizer.viewport_corner.y // bottom
+    };
+
+    Surface color_surface;
+    Surface depth_surface;
+    MathUtil::Rectangle<u32> surfaces_rect;
+    std::tie(color_surface, depth_surface, surfaces_rect) =
+        res_cache.GetFramebufferSurfaces(using_color_fb, using_depth_fb, viewport_rect_unscaled);
+
+    const u16 res_scale = color_surface != nullptr
+                              ? color_surface->res_scale
+                              : (depth_surface == nullptr ? 1u : depth_surface->res_scale);
+
+    MathUtil::Rectangle<u32> draw_rect{
+        static_cast<u32>(MathUtil::Clamp<s32>(static_cast<s32>(surfaces_rect.left) +
+                                                  viewport_rect_unscaled.left * res_scale,
+                                              surfaces_rect.left, surfaces_rect.right)), // Left
+        static_cast<u32>(MathUtil::Clamp<s32>(static_cast<s32>(surfaces_rect.bottom) +
+                                                  viewport_rect_unscaled.top * res_scale,
+                                              surfaces_rect.bottom, surfaces_rect.top)), // Top
+        static_cast<u32>(MathUtil::Clamp<s32>(static_cast<s32>(surfaces_rect.left) +
+                                                  viewport_rect_unscaled.right * res_scale,
+                                              surfaces_rect.left, surfaces_rect.right)), // Right
+        static_cast<u32>(MathUtil::Clamp<s32>(static_cast<s32>(surfaces_rect.bottom) +
+                                                  viewport_rect_unscaled.bottom * res_scale,
+                                              surfaces_rect.bottom, surfaces_rect.top))}; // Bottom
+
+    // Bind the framebuffer surfaces
     state.draw.draw_framebuffer = framebuffer.handle;
     state.Apply();
 
     glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D,
                            color_surface != nullptr ? color_surface->texture.handle : 0, 0);
     if (depth_surface != nullptr) {
-        if (regs.framebuffer.framebuffer.depth_format ==
-            Pica::FramebufferRegs::DepthFormat::D24S8) {
+        if (has_stencil) {
             // attach both depth and stencil
             glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
                                    depth_surface->texture.handle, 0);
@@ -259,38 +309,30 @@ void RasterizerOpenGL::DrawTriangles() {
     }
 
     // Sync the viewport
-    // These registers hold half-width and half-height, so must be multiplied by 2
-    GLsizei viewport_width =
-        (GLsizei)Pica::float24::FromRaw(regs.rasterizer.viewport_size_x).ToFloat32() * 2;
-    GLsizei viewport_height =
-        (GLsizei)Pica::float24::FromRaw(regs.rasterizer.viewport_size_y).ToFloat32() * 2;
+    state.viewport.x =
+        static_cast<GLint>(surfaces_rect.left) + viewport_rect_unscaled.left * res_scale;
+    state.viewport.y =
+        static_cast<GLint>(surfaces_rect.bottom) + viewport_rect_unscaled.bottom * res_scale;
+    state.viewport.width = static_cast<GLsizei>(viewport_rect_unscaled.GetWidth() * res_scale);
+    state.viewport.height = static_cast<GLsizei>(viewport_rect_unscaled.GetHeight() * res_scale);
 
-    glViewport(
-        (GLint)(rect.left + regs.rasterizer.viewport_corner.x * color_surface->res_scale_width),
-        (GLint)(rect.bottom + regs.rasterizer.viewport_corner.y * color_surface->res_scale_height),
-        (GLsizei)(viewport_width * color_surface->res_scale_width),
-        (GLsizei)(viewport_height * color_surface->res_scale_height));
-
-    if (uniform_block_data.data.framebuffer_scale[0] != color_surface->res_scale_width ||
-        uniform_block_data.data.framebuffer_scale[1] != color_surface->res_scale_height) {
-
-        uniform_block_data.data.framebuffer_scale[0] = color_surface->res_scale_width;
-        uniform_block_data.data.framebuffer_scale[1] = color_surface->res_scale_height;
+    if (uniform_block_data.data.framebuffer_scale != res_scale) {
+        uniform_block_data.data.framebuffer_scale = res_scale;
         uniform_block_data.dirty = true;
     }
 
     // Scissor checks are window-, not viewport-relative, which means that if the cached texture
     // sub-rect changes, the scissor bounds also need to be updated.
-    GLint scissor_x1 = static_cast<GLint>(
-        rect.left + regs.rasterizer.scissor_test.x1 * color_surface->res_scale_width);
-    GLint scissor_y1 = static_cast<GLint>(
-        rect.bottom + regs.rasterizer.scissor_test.y1 * color_surface->res_scale_height);
+    GLint scissor_x1 =
+        static_cast<GLint>(surfaces_rect.left + regs.rasterizer.scissor_test.x1 * res_scale);
+    GLint scissor_y1 =
+        static_cast<GLint>(surfaces_rect.bottom + regs.rasterizer.scissor_test.y1 * res_scale);
     // x2, y2 have +1 added to cover the entire pixel area, otherwise you might get cracks when
     // scaling or doing multisampling.
-    GLint scissor_x2 = static_cast<GLint>(
-        rect.left + (regs.rasterizer.scissor_test.x2 + 1) * color_surface->res_scale_width);
-    GLint scissor_y2 = static_cast<GLint>(
-        rect.bottom + (regs.rasterizer.scissor_test.y2 + 1) * color_surface->res_scale_height);
+    GLint scissor_x2 =
+        static_cast<GLint>(surfaces_rect.left + (regs.rasterizer.scissor_test.x2 + 1) * res_scale);
+    GLint scissor_y2 = static_cast<GLint>(surfaces_rect.bottom +
+                                          (regs.rasterizer.scissor_test.y2 + 1) * res_scale);
 
     if (uniform_block_data.data.scissor_x1 != scissor_x1 ||
         uniform_block_data.data.scissor_x2 != scissor_x2 ||
@@ -311,7 +353,7 @@ void RasterizerOpenGL::DrawTriangles() {
 
         if (texture.enabled) {
             texture_samplers[texture_index].SyncWithConfig(texture.config);
-            CachedSurface* surface = res_cache.GetTextureSurface(texture);
+            Surface surface = res_cache.GetTextureSurface(texture);
             if (surface != nullptr) {
                 state.texture_units[texture_index].texture_2d = surface->texture.handle;
             } else {
@@ -380,6 +422,15 @@ void RasterizerOpenGL::DrawTriangles() {
         uniform_block_data.dirty = false;
     }
 
+    // Viewport can have negative offsets or larger
+    // dimensions than our framebuffer sub-rect.
+    // Enable scissor test to prevent drawing
+    // outside of the framebuffer region
+    state.scissor.enabled = true;
+    state.scissor.x = draw_rect.left;
+    state.scissor.y = draw_rect.bottom;
+    state.scissor.width = draw_rect.GetWidth();
+    state.scissor.height = draw_rect.GetHeight();
     state.Apply();
 
     // Draw the vertex batch
@@ -387,16 +438,8 @@ void RasterizerOpenGL::DrawTriangles() {
                  GL_STREAM_DRAW);
     glDrawArrays(GL_TRIANGLES, 0, (GLsizei)vertex_batch.size());
 
-    // Mark framebuffer surfaces as dirty
-    // TODO: Restrict invalidation area to the viewport
-    if (color_surface != nullptr) {
-        color_surface->dirty = true;
-        res_cache.FlushRegion(color_surface->addr, color_surface->size, color_surface, true);
-    }
-    if (depth_surface != nullptr) {
-        depth_surface->dirty = true;
-        res_cache.FlushRegion(depth_surface->addr, depth_surface->size, depth_surface, true);
-    }
+    // Disable scissor test
+    state.scissor.enabled = false;
 
     vertex_batch.clear();
 
@@ -405,6 +448,22 @@ void RasterizerOpenGL::DrawTriangles() {
         state.texture_units[texture_index].texture_2d = 0;
     }
     state.Apply();
+
+    // Mark framebuffer surfaces as dirty
+    MathUtil::Rectangle<u32> draw_rect_unscaled{
+        draw_rect.left / res_scale, draw_rect.top / res_scale, draw_rect.right / res_scale,
+        draw_rect.bottom / res_scale};
+
+    if (color_surface != nullptr && write_color_fb) {
+        auto interval = color_surface->GetSubRectInterval(draw_rect_unscaled);
+        res_cache.InvalidateRegion(boost::icl::first(interval), boost::icl::length(interval),
+                                   color_surface);
+    }
+    if (depth_surface != nullptr && write_depth_fb) {
+        auto interval = depth_surface->GetSubRectInterval(draw_rect_unscaled);
+        res_cache.InvalidateRegion(boost::icl::first(interval), boost::icl::length(interval),
+                                   depth_surface);
+    }
 }
 
 void RasterizerOpenGL::NotifyPicaRegisterChanged(u32 id) {
@@ -893,227 +952,164 @@ void RasterizerOpenGL::FlushAll() {
 
 void RasterizerOpenGL::FlushRegion(PAddr addr, u32 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
-    res_cache.FlushRegion(addr, size, nullptr, false);
+    res_cache.FlushRegion(addr, size);
+}
+
+void RasterizerOpenGL::InvalidateRegion(PAddr addr, u32 size) {
+    MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+    res_cache.InvalidateRegion(addr, size, nullptr);
 }
 
 void RasterizerOpenGL::FlushAndInvalidateRegion(PAddr addr, u32 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
-    res_cache.FlushRegion(addr, size, nullptr, true);
+    res_cache.FlushRegion(addr, size);
+    res_cache.InvalidateRegion(addr, size, nullptr);
 }
 
 bool RasterizerOpenGL::AccelerateDisplayTransfer(const GPU::Regs::DisplayTransferConfig& config) {
     MICROPROFILE_SCOPE(OpenGL_Blits);
 
-    CachedSurface src_params;
+    SurfaceParams src_params;
     src_params.addr = config.GetPhysicalInputAddress();
-    // It's important to use the correct source input width to properly skip over parts of the input
-    // image which will be cropped from the output but still affect the stride of the input image.
-    src_params.width = config.input_width;
-    // Using the output's height is fine because we don't read or skip over the remaining part of
-    // the image, and it allows for smaller texture cache lookup rectangles.
+    src_params.width = config.output_width;
+    src_params.stride = config.input_width;
     src_params.height = config.output_height;
     src_params.is_tiled = !config.input_linear;
-    src_params.pixel_format = CachedSurface::PixelFormatFromGPUPixelFormat(config.input_format);
+    src_params.pixel_format = SurfaceParams::PixelFormatFromGPUPixelFormat(config.input_format);
+    src_params.UpdateParams();
 
-    CachedSurface dst_params;
+    SurfaceParams dst_params;
     dst_params.addr = config.GetPhysicalOutputAddress();
-    dst_params.width =
-        config.scaling != config.NoScale ? config.output_width / 2 : config.output_width.Value();
-    dst_params.height =
-        config.scaling == config.ScaleXY ? config.output_height / 2 : config.output_height.Value();
+    dst_params.width = config.scaling != config.NoScale ? config.output_width.Value() / 2
+                                                        : config.output_width.Value();
+    dst_params.height = config.scaling == config.ScaleXY ? config.output_height.Value() / 2
+                                                         : config.output_height.Value();
     dst_params.is_tiled = config.input_linear != config.dont_swizzle;
-    dst_params.pixel_format = CachedSurface::PixelFormatFromGPUPixelFormat(config.output_format);
+    dst_params.pixel_format = SurfaceParams::PixelFormatFromGPUPixelFormat(config.output_format);
+    dst_params.UpdateParams();
 
-    MathUtil::Rectangle<int> src_rect;
-    CachedSurface* src_surface = res_cache.GetSurfaceRect(src_params, false, true, src_rect);
-
-    if (src_surface == nullptr) {
+    MathUtil::Rectangle<u32> src_rect;
+    Surface src_surface;
+    std::tie(src_surface, src_rect) =
+        res_cache.GetSurfaceSubRect(src_params, ScaleMatch::Ignore, true);
+    if (src_surface == nullptr)
         return false;
-    }
 
-    // Adjust the source rectangle to take into account parts of the input lines being cropped
-    if (config.input_width > config.output_width) {
-        src_rect.right -= static_cast<int>((config.input_width - config.output_width) *
-                                           src_surface->res_scale_width);
-    }
+    dst_params.res_scale = src_surface->res_scale;
 
-    // Require destination surface to have same resolution scale as source to preserve scaling
-    dst_params.res_scale_width = src_surface->res_scale_width;
-    dst_params.res_scale_height = src_surface->res_scale_height;
-
-    MathUtil::Rectangle<int> dst_rect;
-    CachedSurface* dst_surface = res_cache.GetSurfaceRect(dst_params, true, false, dst_rect);
-
-    if (dst_surface == nullptr) {
+    MathUtil::Rectangle<u32> dst_rect;
+    Surface dst_surface;
+    std::tie(dst_surface, dst_rect) =
+        res_cache.GetSurfaceSubRect(dst_params, ScaleMatch::Upscale, false);
+    if (dst_surface == nullptr)
         return false;
-    }
 
-    // Don't accelerate if the src and dst surfaces are the same
-    if (src_surface == dst_surface) {
+    if (src_surface->is_tiled != dst_surface->is_tiled)
+        std::swap(src_rect.top, src_rect.bottom);
+
+    if (config.flip_vertically)
+        std::swap(src_rect.top, src_rect.bottom);
+
+    if (!res_cache.BlitSurfaces(src_surface, src_rect, dst_surface, dst_rect))
         return false;
-    }
 
-    if (config.flip_vertically) {
-        std::swap(dst_rect.top, dst_rect.bottom);
-    }
-
-    if (!res_cache.TryBlitSurfaces(src_surface, src_rect, dst_surface, dst_rect)) {
-        return false;
-    }
-
-    u32 dst_size = dst_params.width * dst_params.height *
-                   CachedSurface::GetFormatBpp(dst_params.pixel_format) / 8;
-    dst_surface->dirty = true;
-    res_cache.FlushRegion(config.GetPhysicalOutputAddress(), dst_size, dst_surface, true);
+    res_cache.InvalidateRegion(dst_params.addr, dst_params.size, dst_surface);
     return true;
 }
 
 bool RasterizerOpenGL::AccelerateTextureCopy(const GPU::Regs::DisplayTransferConfig& config) {
-    // TODO(tfarley): Try to hardware accelerate this
-    return false;
-}
-
-bool RasterizerOpenGL::AccelerateFill(const GPU::Regs::MemoryFillConfig& config) {
-    MICROPROFILE_SCOPE(OpenGL_Blits);
-    using PixelFormat = CachedSurface::PixelFormat;
-    using SurfaceType = CachedSurface::SurfaceType;
-
-    CachedSurface* dst_surface = res_cache.TryGetFillSurface(config);
-
-    if (dst_surface == nullptr) {
+    u32 copy_size = Common::AlignDown(config.texture_copy.size, 16);
+    if (copy_size == 0) {
         return false;
     }
 
-    OpenGLState cur_state = OpenGLState::GetCurState();
-
-    SurfaceType dst_type = CachedSurface::GetFormatType(dst_surface->pixel_format);
-
-    GLuint old_fb = cur_state.draw.draw_framebuffer;
-    cur_state.draw.draw_framebuffer = framebuffer.handle;
-    // TODO: When scissor test is implemented, need to disable scissor test in cur_state here so
-    // Clear call isn't affected
-    cur_state.Apply();
-
-    if (dst_type == SurfaceType::Color || dst_type == SurfaceType::Texture) {
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D,
-                               dst_surface->texture.handle, 0);
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
-                               0);
-
-        GLfloat color_values[4] = {0.0f, 0.0f, 0.0f, 0.0f};
-
-        // TODO: Handle additional pixel format and fill value size combinations to accelerate more
-        // cases
-        //       For instance, checking if fill value's bytes/bits repeat to allow filling
-        //       I8/A8/I4/A4/...
-        //       Currently only handles formats that are multiples of the fill value size
-
-        if (config.fill_24bit) {
-            switch (dst_surface->pixel_format) {
-            case PixelFormat::RGB8:
-                color_values[0] = config.value_24bit_r / 255.0f;
-                color_values[1] = config.value_24bit_g / 255.0f;
-                color_values[2] = config.value_24bit_b / 255.0f;
-                break;
-            default:
-                return false;
-            }
-        } else if (config.fill_32bit) {
-            u32 value = config.value_32bit;
-
-            switch (dst_surface->pixel_format) {
-            case PixelFormat::RGBA8:
-                color_values[0] = (value >> 24) / 255.0f;
-                color_values[1] = ((value >> 16) & 0xFF) / 255.0f;
-                color_values[2] = ((value >> 8) & 0xFF) / 255.0f;
-                color_values[3] = (value & 0xFF) / 255.0f;
-                break;
-            default:
-                return false;
-            }
-        } else {
-            u16 value_16bit = config.value_16bit.Value();
-            Math::Vec4<u8> color;
-
-            switch (dst_surface->pixel_format) {
-            case PixelFormat::RGBA8:
-                color_values[0] = (value_16bit >> 8) / 255.0f;
-                color_values[1] = (value_16bit & 0xFF) / 255.0f;
-                color_values[2] = color_values[0];
-                color_values[3] = color_values[1];
-                break;
-            case PixelFormat::RGB5A1:
-                color = Color::DecodeRGB5A1((const u8*)&value_16bit);
-                color_values[0] = color[0] / 31.0f;
-                color_values[1] = color[1] / 31.0f;
-                color_values[2] = color[2] / 31.0f;
-                color_values[3] = color[3];
-                break;
-            case PixelFormat::RGB565:
-                color = Color::DecodeRGB565((const u8*)&value_16bit);
-                color_values[0] = color[0] / 31.0f;
-                color_values[1] = color[1] / 63.0f;
-                color_values[2] = color[2] / 31.0f;
-                break;
-            case PixelFormat::RGBA4:
-                color = Color::DecodeRGBA4((const u8*)&value_16bit);
-                color_values[0] = color[0] / 15.0f;
-                color_values[1] = color[1] / 15.0f;
-                color_values[2] = color[2] / 15.0f;
-                color_values[3] = color[3] / 15.0f;
-                break;
-            case PixelFormat::IA8:
-            case PixelFormat::RG8:
-                color_values[0] = (value_16bit >> 8) / 255.0f;
-                color_values[1] = (value_16bit & 0xFF) / 255.0f;
-                break;
-            default:
-                return false;
-            }
-        }
-
-        cur_state.color_mask.red_enabled = GL_TRUE;
-        cur_state.color_mask.green_enabled = GL_TRUE;
-        cur_state.color_mask.blue_enabled = GL_TRUE;
-        cur_state.color_mask.alpha_enabled = GL_TRUE;
-        cur_state.Apply();
-        glClearBufferfv(GL_COLOR, 0, color_values);
-    } else if (dst_type == SurfaceType::Depth) {
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D,
-                               dst_surface->texture.handle, 0);
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
-
-        GLfloat value_float;
-        if (dst_surface->pixel_format == CachedSurface::PixelFormat::D16) {
-            value_float = config.value_32bit / 65535.0f; // 2^16 - 1
-        } else if (dst_surface->pixel_format == CachedSurface::PixelFormat::D24) {
-            value_float = config.value_32bit / 16777215.0f; // 2^24 - 1
-        }
-
-        cur_state.depth.write_mask = GL_TRUE;
-        cur_state.Apply();
-        glClearBufferfv(GL_DEPTH, 0, &value_float);
-    } else if (dst_type == SurfaceType::DepthStencil) {
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
-                               dst_surface->texture.handle, 0);
-
-        GLfloat value_float = (config.value_32bit & 0xFFFFFF) / 16777215.0f; // 2^24 - 1
-        GLint value_int = (config.value_32bit >> 24);
-
-        cur_state.depth.write_mask = GL_TRUE;
-        cur_state.stencil.write_mask = 0xFF;
-        cur_state.Apply();
-        glClearBufferfi(GL_DEPTH_STENCIL, 0, value_float, value_int);
+    u32 input_gap = config.texture_copy.input_gap * 16;
+    u32 input_width = config.texture_copy.input_width * 16;
+    if (input_width == 0 && input_gap != 0) {
+        return false;
+    }
+    if (input_gap == 0 || input_width >= copy_size) {
+        input_width = copy_size;
+        input_gap = 0;
+    }
+    if (copy_size % input_width != 0) {
+        return false;
     }
 
-    cur_state.draw.draw_framebuffer = old_fb;
-    // TODO: Return scissor test to previous value when scissor test is implemented
-    cur_state.Apply();
+    u32 output_gap = config.texture_copy.output_gap * 16;
+    u32 output_width = config.texture_copy.output_width * 16;
+    if (output_width == 0 && output_gap != 0) {
+        return false;
+    }
+    if (output_gap == 0 || output_width >= copy_size) {
+        output_width = copy_size;
+        output_gap = 0;
+    }
+    if (copy_size % output_width != 0) {
+        return false;
+    }
 
-    dst_surface->dirty = true;
-    res_cache.FlushRegion(dst_surface->addr, dst_surface->size, dst_surface, true);
+    SurfaceParams src_params;
+    src_params.addr = config.GetPhysicalInputAddress();
+    src_params.stride = input_width + input_gap; // stride in bytes
+    src_params.width = input_width;              // width in bytes
+    src_params.height = copy_size / input_width;
+    src_params.size = ((src_params.height - 1) * src_params.stride) + src_params.width;
+    src_params.end = src_params.addr + src_params.size;
+
+    MathUtil::Rectangle<u32> src_rect;
+    Surface src_surface;
+    std::tie(src_surface, src_rect) = res_cache.GetTexCopySurface(src_params);
+    if (src_surface == nullptr) {
+        return false;
+    }
+
+    if (output_gap != 0 &&
+        (output_width !=
+             src_surface->BytesInPixels(src_rect.GetWidth() / src_surface->res_scale) *
+                 (src_surface->is_tiled ? 8 : 1) ||
+         output_gap % src_surface->BytesInPixels(src_surface->is_tiled ? 64 : 1) != 0)) {
+        return false;
+    }
+
+    SurfaceParams dst_params = *src_surface;
+    dst_params.addr = config.GetPhysicalOutputAddress();
+    dst_params.width = src_rect.GetWidth() / src_surface->res_scale;
+    dst_params.stride =
+        dst_params.width +
+        src_surface->PixelsInBytes(src_surface->is_tiled ? output_gap / 8 : output_gap);
+    dst_params.height = src_rect.GetHeight() / src_surface->res_scale;
+    dst_params.res_scale = src_surface->res_scale;
+    dst_params.UpdateParams();
+
+    // Since we are going to invalidate the gap if there is one, we will have to load it first
+    const bool load_gap = output_gap != 0;
+    MathUtil::Rectangle<u32> dst_rect;
+    Surface dst_surface;
+    std::tie(dst_surface, dst_rect) =
+        res_cache.GetSurfaceSubRect(dst_params, ScaleMatch::Upscale, load_gap);
+    if (src_surface == nullptr) {
+        return false;
+    }
+
+    if (dst_surface->type == SurfaceType::Texture) {
+        return false;
+    }
+
+    if (!res_cache.BlitSurfaces(src_surface, src_rect, dst_surface, dst_rect)) {
+        return false;
+    }
+
+    res_cache.InvalidateRegion(dst_params.addr, dst_params.size, dst_surface);
+    return true;
+}
+
+bool RasterizerOpenGL::AccelerateFill(const GPU::Regs::MemoryFillConfig& config) {
+    Surface dst_surface = res_cache.GetFillSurface(config);
+    if (dst_surface == nullptr)
+        return false;
+
+    res_cache.InvalidateRegion(dst_surface->addr, dst_surface->size, dst_surface);
     return true;
 }
 
@@ -1125,16 +1121,19 @@ bool RasterizerOpenGL::AccelerateDisplay(const GPU::Regs::FramebufferConfig& con
     }
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
 
-    CachedSurface src_params;
+    SurfaceParams src_params;
     src_params.addr = framebuffer_addr;
-    src_params.width = config.width;
+    src_params.width = std::min(config.width.Value(), pixel_stride);
     src_params.height = config.height;
-    src_params.pixel_stride = pixel_stride;
+    src_params.stride = pixel_stride;
     src_params.is_tiled = false;
-    src_params.pixel_format = CachedSurface::PixelFormatFromGPUPixelFormat(config.color_format);
+    src_params.pixel_format = SurfaceParams::PixelFormatFromGPUPixelFormat(config.color_format);
+    src_params.UpdateParams();
 
-    MathUtil::Rectangle<int> src_rect;
-    CachedSurface* src_surface = res_cache.GetSurfaceRect(src_params, false, true, src_rect);
+    MathUtil::Rectangle<u32> src_rect;
+    Surface src_surface;
+    std::tie(src_surface, src_rect) =
+        res_cache.GetSurfaceSubRect(src_params, ScaleMatch::Ignore, true);
 
     if (src_surface == nullptr) {
         return false;
@@ -1144,8 +1143,8 @@ bool RasterizerOpenGL::AccelerateDisplay(const GPU::Regs::FramebufferConfig& con
     u32 scaled_height = src_surface->GetScaledHeight();
 
     screen_info.display_texcoords = MathUtil::Rectangle<float>(
-        (float)src_rect.top / (float)scaled_height, (float)src_rect.left / (float)scaled_width,
-        (float)src_rect.bottom / (float)scaled_height, (float)src_rect.right / (float)scaled_width);
+        (float)src_rect.bottom / (float)scaled_height, (float)src_rect.left / (float)scaled_width,
+        (float)src_rect.top / (float)scaled_height, (float)src_rect.right / (float)scaled_width);
 
     screen_info.display_texture = src_surface->texture.handle;
 
@@ -1158,8 +1157,8 @@ void RasterizerOpenGL::SamplerInfo::Create() {
     wrap_s = wrap_t = TextureConfig::Repeat;
     border_color = 0;
 
-    glSamplerParameteri(sampler.handle, GL_TEXTURE_MIN_FILTER,
-                        GL_LINEAR); // default is GL_LINEAR_MIPMAP_LINEAR
+    // default is GL_LINEAR_MIPMAP_LINEAR
+    glSamplerParameteri(sampler.handle, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
     // Other attributes have correct defaults
 }
 

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -43,6 +43,7 @@ public:
     void NotifyPicaRegisterChanged(u32 id) override;
     void FlushAll() override;
     void FlushRegion(PAddr addr, u32 size) override;
+    void InvalidateRegion(PAddr addr, u32 size) override;
     void FlushAndInvalidateRegion(PAddr addr, u32 size) override;
     bool AccelerateDisplayTransfer(const GPU::Regs::DisplayTransferConfig& config) override;
     bool AccelerateTextureCopy(const GPU::Regs::DisplayTransferConfig& config) override;
@@ -135,7 +136,7 @@ private:
     //       the end of a uniform block is included in UNIFORM_BLOCK_DATA_SIZE or not.
     //       Not following that rule will cause problems on some AMD drivers.
     struct UniformData {
-        alignas(8) GLvec2 framebuffer_scale;
+        GLint framebuffer_scale;
         GLint alphatest_ref;
         GLfloat depth_scale;
         GLfloat depth_offset;
@@ -155,7 +156,7 @@ private:
     };
 
     static_assert(
-        sizeof(UniformData) == 0x470,
+        sizeof(UniformData) == 0x460,
         "The size of the UniformData structure has changed, update the structure in the shader");
     static_assert(sizeof(UniformData) < 16384,
                   "UniformData structure must be less than 16kb as per the OpenGL spec");

src/video_core/renderer_opengl/gl_rasterizer_cache.h

@@ -13,6 +13,7 @@
 #pragma GCC diagnostic ignored "-Wunused-local-typedefs"
 #endif
 #include <boost/icl/interval_map.hpp>
+#include <boost/icl/interval_set.hpp>
 #ifdef __GNUC__
 #pragma GCC diagnostic pop
 #endif
@@ -20,21 +21,37 @@
 #include "common/assert.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"
+#include "common/math_util.h"
 #include "core/hw/gpu.h"
 #include "video_core/regs_framebuffer.h"
 #include "video_core/regs_texturing.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 
-namespace MathUtil {
-template <class T>
-struct Rectangle;
-}
-
 struct CachedSurface;
+using Surface = std::shared_ptr<CachedSurface>;
+using SurfaceSet = std::set<Surface>;
 
-using SurfaceCache = boost::icl::interval_map<PAddr, std::set<std::shared_ptr<CachedSurface>>>;
+using SurfaceRegions = boost::icl::interval_set<PAddr>;
+using SurfaceMap = boost::icl::interval_map<PAddr, Surface>;
+using SurfaceCache = boost::icl::interval_map<PAddr, SurfaceSet>;
 
-struct CachedSurface {
+using SurfaceInterval = SurfaceCache::interval_type;
+static_assert(std::is_same<SurfaceRegions::interval_type, SurfaceCache::interval_type>() &&
+                  std::is_same<SurfaceMap::interval_type, SurfaceCache::interval_type>(),
+              "incorrect interval types");
+
+using SurfaceRect_Tuple = std::tuple<Surface, MathUtil::Rectangle<u32>>;
+using SurfaceSurfaceRect_Tuple = std::tuple<Surface, Surface, MathUtil::Rectangle<u32>>;
+
+using PageMap = boost::icl::interval_map<u32, int>;
+
+enum class ScaleMatch {
+    Exact,   // only accept same res scale
+    Upscale, // only allow higher scale than params
+    Ignore   // accept every scaled res
+};
+
+struct SurfaceParams {
     enum class PixelFormat {
         // First 5 formats are shared between textures and color buffers
         RGBA8 = 0,
@@ -68,11 +85,12 @@ struct CachedSurface {
         Texture = 1,
         Depth = 2,
         DepthStencil = 3,
-        Invalid = 4,
+        Fill = 4,
+        Invalid = 5
     };
 
-    static unsigned int GetFormatBpp(CachedSurface::PixelFormat format) {
-        static const std::array<unsigned int, 18> bpp_table = {
+    static constexpr unsigned int GetFormatBpp(PixelFormat format) {
+        constexpr std::array<unsigned int, 18> bpp_table = {
             32, // RGBA8
             24, // RGB8
             16, // RGB5A1
@@ -93,8 +111,11 @@ struct CachedSurface {
             32, // D24S8
         };
 
-        ASSERT((unsigned int)format < ARRAY_SIZE(bpp_table));
-        return bpp_table[(unsigned int)format];
+        assert(static_cast<size_t>(format) < bpp_table.size());
+        return bpp_table[static_cast<size_t>(format)];
+    }
+    unsigned int GetFormatBpp() const {
+        return GetFormatBpp(pixel_format);
     }
 
     static PixelFormat PixelFormatFromTextureFormat(Pica::TexturingRegs::TextureFormat format) {
@@ -142,7 +163,7 @@ struct CachedSurface {
         return false;
     }
 
-    static SurfaceType GetFormatType(PixelFormat pixel_format) {
+    static constexpr SurfaceType GetFormatType(PixelFormat pixel_format) {
         if ((unsigned int)pixel_format < 5) {
             return SurfaceType::Color;
         }
@@ -162,31 +183,117 @@ struct CachedSurface {
         return SurfaceType::Invalid;
     }
 
+    /// Update the params "size", "end" and "type" from the already set "addr", "width", "height"
+    /// and "pixel_format"
+    void UpdateParams() {
+        if (stride == 0) {
+            stride = width;
+        }
+        type = GetFormatType(pixel_format);
+        size = !is_tiled ? BytesInPixels(stride * (height - 1) + width)
+                         : BytesInPixels(stride * 8 * (height / 8 - 1) + width * 8);
+        end = addr + size;
+    }
+
+    SurfaceInterval GetInterval() const {
+        return SurfaceInterval::right_open(addr, end);
+    }
+
+    // Returns the outer rectangle containing "interval"
+    SurfaceParams FromInterval(SurfaceInterval interval) const;
+
+    SurfaceInterval GetSubRectInterval(MathUtil::Rectangle<u32> unscaled_rect) const;
+
+    // Returns the region of the biggest valid rectange within interval
+    SurfaceInterval GetCopyableInterval(const Surface& src_surface) const;
+
     u32 GetScaledWidth() const {
-        return (u32)(width * res_scale_width);
+        return width * res_scale;
     }
 
     u32 GetScaledHeight() const {
-        return (u32)(height * res_scale_height);
+        return height * res_scale;
     }
 
-    PAddr addr;
-    u32 size;
+    MathUtil::Rectangle<u32> GetRect() const {
+        return {0, height, width, 0};
+    }
 
-    PAddr min_valid;
-    PAddr max_valid;
+    MathUtil::Rectangle<u32> GetScaledRect() const {
+        return {0, GetScaledHeight(), GetScaledWidth(), 0};
+    }
+
+    u32 PixelsInBytes(u32 size) const {
+        return size * CHAR_BIT / GetFormatBpp(pixel_format);
+    }
+
+    u32 BytesInPixels(u32 pixels) const {
+        return pixels * GetFormatBpp(pixel_format) / CHAR_BIT;
+    }
+
+    bool ExactMatch(const SurfaceParams& other_surface) const;
+    bool CanSubRect(const SurfaceParams& sub_surface) const;
+    bool CanExpand(const SurfaceParams& expanded_surface) const;
+    bool CanTexCopy(const SurfaceParams& texcopy_params) const;
+
+    MathUtil::Rectangle<u32> GetSubRect(const SurfaceParams& sub_surface) const;
+    MathUtil::Rectangle<u32> GetScaledSubRect(const SurfaceParams& sub_surface) const;
+
+    PAddr addr = 0;
+    PAddr end = 0;
+    u32 size = 0;
+
+    u32 width = 0;
+    u32 height = 0;
+    u32 stride = 0;
+    u16 res_scale = 1;
+
+    bool is_tiled = false;
+    PixelFormat pixel_format = PixelFormat::Invalid;
+    SurfaceType type = SurfaceType::Invalid;
+};
+
+struct CachedSurface : SurfaceParams {
+    bool CanFill(const SurfaceParams& dest_surface, SurfaceInterval fill_interval) const;
+    bool CanCopy(const SurfaceParams& dest_surface, SurfaceInterval copy_interval) const;
+
+    bool IsRegionValid(SurfaceInterval interval) const {
+        return (invalid_regions.find(interval) == invalid_regions.end());
+    }
+
+    bool IsSurfaceFullyInvalid() const {
+        return (invalid_regions & GetInterval()) == SurfaceRegions(GetInterval());
+    }
+
+    bool registered = false;
+    SurfaceRegions invalid_regions;
+
+    u32 fill_size = 0; /// Number of bytes to read from fill_data
+    std::array<u8, 4> fill_data;
 
     OGLTexture texture;
-    u32 width;
-    u32 height;
-    /// Stride between lines, in pixels. Only valid for images in linear format.
-    u32 pixel_stride = 0;
-    float res_scale_width = 1.f;
-    float res_scale_height = 1.f;
 
-    bool is_tiled;
-    PixelFormat pixel_format;
-    bool dirty;
+    static constexpr unsigned int GetGLBytesPerPixel(PixelFormat format) {
+        // OpenGL needs 4 bpp alignment for D24 since using GL_UNSIGNED_INT as type
+        return format == PixelFormat::Invalid
+                   ? 0
+                   : (format == PixelFormat::D24 || GetFormatType(format) == SurfaceType::Texture)
+                         ? 4
+                         : SurfaceParams::GetFormatBpp(format) / 8;
+    }
+
+    std::unique_ptr<u8[]> gl_buffer;
+    size_t gl_buffer_size = 0;
+
+    // Read/Write data in 3DS memory to/from gl_buffer
+    void LoadGLBuffer(PAddr load_start, PAddr load_end);
+    void FlushGLBuffer(PAddr flush_start, PAddr flush_end);
+
+    // Upload/Download data in gl_buffer in/to this surface's texture
+    void UploadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint read_fb_handle,
+                         GLuint draw_fb_handle);
+    void DownloadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint read_fb_handle,
+                           GLuint draw_fb_handle);
 };
 
 class RasterizerCacheOpenGL : NonCopyable {
@@ -194,46 +301,78 @@ public:
     RasterizerCacheOpenGL();
     ~RasterizerCacheOpenGL();
 
-    /// Blits one texture to another
-    void BlitTextures(GLuint src_tex, GLuint dst_tex, CachedSurface::SurfaceType type,
-                      const MathUtil::Rectangle<int>& src_rect,
-                      const MathUtil::Rectangle<int>& dst_rect);
+    /// Blit one surface's texture to another
+    bool BlitSurfaces(const Surface& src_surface, const MathUtil::Rectangle<u32>& src_rect,
+                      const Surface& dst_surface, const MathUtil::Rectangle<u32>& dst_rect);
 
-    /// Attempt to blit one surface's texture to another
-    bool TryBlitSurfaces(CachedSurface* src_surface, const MathUtil::Rectangle<int>& src_rect,
-                         CachedSurface* dst_surface, const MathUtil::Rectangle<int>& dst_rect);
+    void ConvertD24S8toABGR(GLuint src_tex, const MathUtil::Rectangle<u32>& src_rect,
+                            GLuint dst_tex, const MathUtil::Rectangle<u32>& dst_rect);
 
-    /// Loads a texture from 3DS memory to OpenGL and caches it (if not already cached)
-    CachedSurface* GetSurface(const CachedSurface& params, bool match_res_scale,
-                              bool load_if_create);
+    /// Copy one surface's region to another
+    void CopySurface(const Surface& src_surface, const Surface& dst_surface,
+                     SurfaceInterval copy_interval);
+
+    /// Load a texture from 3DS memory to OpenGL and cache it (if not already cached)
+    Surface GetSurface(const SurfaceParams& params, ScaleMatch match_res_scale,
+                       bool load_if_create);
 
     /// Attempt to find a subrect (resolution scaled) of a surface, otherwise loads a texture from
     /// 3DS memory to OpenGL and caches it (if not already cached)
-    CachedSurface* GetSurfaceRect(const CachedSurface& params, bool match_res_scale,
-                                  bool load_if_create, MathUtil::Rectangle<int>& out_rect);
+    SurfaceRect_Tuple GetSurfaceSubRect(const SurfaceParams& params, ScaleMatch match_res_scale,
+                                        bool load_if_create);
 
-    /// Gets a surface based on the texture configuration
-    CachedSurface* GetTextureSurface(const Pica::TexturingRegs::FullTextureConfig& config);
+    /// Get a surface based on the texture configuration
+    Surface GetTextureSurface(const Pica::TexturingRegs::FullTextureConfig& config);
 
-    /// Gets the color and depth surfaces and rect (resolution scaled) based on the framebuffer
-    /// configuration
-    std::tuple<CachedSurface*, CachedSurface*, MathUtil::Rectangle<int>> GetFramebufferSurfaces(
-        const Pica::FramebufferRegs::FramebufferConfig& config);
+    /// Get the color and depth surfaces based on the framebuffer configuration
+    SurfaceSurfaceRect_Tuple GetFramebufferSurfaces(bool using_color_fb, bool using_depth_fb,
+                                                    const MathUtil::Rectangle<s32>& viewport_rect);
 
-    /// Attempt to get a surface that exactly matches the fill region and format
-    CachedSurface* TryGetFillSurface(const GPU::Regs::MemoryFillConfig& config);
+    /// Get a surface that matches the fill config
+    Surface GetFillSurface(const GPU::Regs::MemoryFillConfig& config);
 
-    /// Write the surface back to memory
-    void FlushSurface(CachedSurface* surface);
+    /// Get a surface that matches a "texture copy" display transfer config
+    SurfaceRect_Tuple GetTexCopySurface(const SurfaceParams& params);
 
-    /// Write any cached resources overlapping the region back to memory (if dirty) and optionally
-    /// invalidate them in the cache
-    void FlushRegion(PAddr addr, u32 size, const CachedSurface* skip_surface, bool invalidate);
+    /// Write any cached resources overlapping the region back to memory (if dirty)
+    void FlushRegion(PAddr addr, u32 size, Surface flush_surface = nullptr);
+
+    /// Mark region as being invalidated by region_owner (nullptr if 3DS memory)
+    void InvalidateRegion(PAddr addr, u32 size, const Surface& region_owner);
 
     /// Flush all cached resources tracked by this cache manager
     void FlushAll();
 
 private:
+    void DuplicateSurface(const Surface& src_surface, const Surface& dest_surface);
+
+    /// Update surface's texture for given region when necessary
+    void ValidateSurface(const Surface& surface, PAddr addr, u32 size);
+
+    /// Create a new surface
+    Surface CreateSurface(const SurfaceParams& params);
+
+    /// Register surface into the cache
+    void RegisterSurface(const Surface& surface);
+
+    /// Remove surface from the cache
+    void UnregisterSurface(const Surface& surface);
+
+    /// Increase/decrease the number of surface in pages touching the specified region
+    void UpdatePagesCachedCount(PAddr addr, u32 size, int delta);
+
     SurfaceCache surface_cache;
-    OGLFramebuffer transfer_framebuffers[2];
+    PageMap cached_pages;
+    SurfaceMap dirty_regions;
+    SurfaceSet remove_surfaces;
+
+    OGLFramebuffer read_framebuffer;
+    OGLFramebuffer draw_framebuffer;
+
+    OGLVertexArray attributeless_vao;
+    OGLBuffer d24s8_abgr_buffer;
+    GLsizeiptr d24s8_abgr_buffer_size;
+    OGLShader d24s8_abgr_shader;
+    GLint d24s8_abgr_tbo_size_u_id;
+    GLint d24s8_abgr_viewport_u_id;
 };

src/video_core/renderer_opengl/gl_resource_manager.h

@@ -36,7 +36,7 @@ public:
         if (handle == 0)
             return;
         glDeleteTextures(1, &handle);
-        OpenGLState::ResetTexture(handle);
+        OpenGLState::GetCurState().ResetTexture(handle).Apply();
         handle = 0;
     }
 
@@ -69,7 +69,7 @@ public:
         if (handle == 0)
             return;
         glDeleteSamplers(1, &handle);
-        OpenGLState::ResetSampler(handle);
+        OpenGLState::GetCurState().ResetSampler(handle).Apply();
         handle = 0;
     }
 
@@ -102,7 +102,7 @@ public:
         if (handle == 0)
             return;
         glDeleteProgram(handle);
-        OpenGLState::ResetProgram(handle);
+        OpenGLState::GetCurState().ResetProgram(handle).Apply();
         handle = 0;
     }
 
@@ -135,7 +135,7 @@ public:
         if (handle == 0)
             return;
         glDeleteBuffers(1, &handle);
-        OpenGLState::ResetBuffer(handle);
+        OpenGLState::GetCurState().ResetBuffer(handle).Apply();
         handle = 0;
     }
 
@@ -168,7 +168,7 @@ public:
         if (handle == 0)
             return;
         glDeleteVertexArrays(1, &handle);
-        OpenGLState::ResetVertexArray(handle);
+        OpenGLState::GetCurState().ResetVertexArray(handle).Apply();
         handle = 0;
     }
 
@@ -201,7 +201,7 @@ public:
         if (handle == 0)
             return;
         glDeleteFramebuffers(1, &handle);
-        OpenGLState::ResetFramebuffer(handle);
+        OpenGLState::GetCurState().ResetFramebuffer(handle).Apply();
         handle = 0;
     }
 

src/video_core/renderer_opengl/gl_shader_gen.cpp

@@ -41,7 +41,7 @@ struct LightSrc {
 };
 
 layout (std140) uniform shader_data {
-    vec2 framebuffer_scale;
+    int framebuffer_scale;
     int alphatest_ref;
     float depth_scale;
     float depth_offset;

src/video_core/renderer_opengl/gl_state.cpp

@@ -69,6 +69,17 @@ OpenGLState::OpenGLState() {
     draw.uniform_buffer = 0;
     draw.shader_program = 0;
 
+    scissor.enabled = false;
+    scissor.x = 0;
+    scissor.y = 0;
+    scissor.width = 0;
+    scissor.height = 0;
+
+    viewport.x = 0;
+    viewport.y = 0;
+    viewport.width = 0;
+    viewport.height = 0;
+
     clip_distance = {};
 }
 
@@ -193,7 +204,7 @@ void OpenGLState::Apply() const {
     // Lighting LUTs
     if (lighting_lut.texture_buffer != cur_state.lighting_lut.texture_buffer) {
         glActiveTexture(TextureUnits::LightingLUT.Enum());
-        glBindTexture(GL_TEXTURE_BUFFER, cur_state.lighting_lut.texture_buffer);
+        glBindTexture(GL_TEXTURE_BUFFER, lighting_lut.texture_buffer);
     }
 
     // Fog LUT
@@ -260,6 +271,26 @@ void OpenGLState::Apply() const {
         glUseProgram(draw.shader_program);
     }
 
+    // Scissor test
+    if (scissor.enabled != cur_state.scissor.enabled) {
+        if (scissor.enabled) {
+            glEnable(GL_SCISSOR_TEST);
+        } else {
+            glDisable(GL_SCISSOR_TEST);
+        }
+    }
+
+    if (scissor.x != cur_state.scissor.x || scissor.y != cur_state.scissor.y ||
+        scissor.width != cur_state.scissor.width || scissor.height != cur_state.scissor.height) {
+        glScissor(scissor.x, scissor.y, scissor.width, scissor.height);
+    }
+
+    if (viewport.x != cur_state.viewport.x || viewport.y != cur_state.viewport.y ||
+        viewport.width != cur_state.viewport.width ||
+        viewport.height != cur_state.viewport.height) {
+        glViewport(viewport.x, viewport.y, viewport.width, viewport.height);
+    }
+
     // Clip distance
     for (size_t i = 0; i < clip_distance.size(); ++i) {
         if (clip_distance[i] != cur_state.clip_distance[i]) {
@@ -274,62 +305,68 @@ void OpenGLState::Apply() const {
     cur_state = *this;
 }
 
-void OpenGLState::ResetTexture(GLuint handle) {
-    for (auto& unit : cur_state.texture_units) {
+OpenGLState& OpenGLState::ResetTexture(GLuint handle) {
+    for (auto& unit : texture_units) {
         if (unit.texture_2d == handle) {
             unit.texture_2d = 0;
         }
     }
-    if (cur_state.lighting_lut.texture_buffer == handle)
-        cur_state.lighting_lut.texture_buffer = 0;
-    if (cur_state.fog_lut.texture_buffer == handle)
-        cur_state.fog_lut.texture_buffer = 0;
-    if (cur_state.proctex_noise_lut.texture_buffer == handle)
-        cur_state.proctex_noise_lut.texture_buffer = 0;
-    if (cur_state.proctex_color_map.texture_buffer == handle)
-        cur_state.proctex_color_map.texture_buffer = 0;
-    if (cur_state.proctex_alpha_map.texture_buffer == handle)
-        cur_state.proctex_alpha_map.texture_buffer = 0;
-    if (cur_state.proctex_lut.texture_buffer == handle)
-        cur_state.proctex_lut.texture_buffer = 0;
-    if (cur_state.proctex_diff_lut.texture_buffer == handle)
-        cur_state.proctex_diff_lut.texture_buffer = 0;
+    if (lighting_lut.texture_buffer == handle)
+        lighting_lut.texture_buffer = 0;
+    if (fog_lut.texture_buffer == handle)
+        fog_lut.texture_buffer = 0;
+    if (proctex_noise_lut.texture_buffer == handle)
+        proctex_noise_lut.texture_buffer = 0;
+    if (proctex_color_map.texture_buffer == handle)
+        proctex_color_map.texture_buffer = 0;
+    if (proctex_alpha_map.texture_buffer == handle)
+        proctex_alpha_map.texture_buffer = 0;
+    if (proctex_lut.texture_buffer == handle)
+        proctex_lut.texture_buffer = 0;
+    if (proctex_diff_lut.texture_buffer == handle)
+        proctex_diff_lut.texture_buffer = 0;
+    return *this;
 }
 
-void OpenGLState::ResetSampler(GLuint handle) {
-    for (auto& unit : cur_state.texture_units) {
+OpenGLState& OpenGLState::ResetSampler(GLuint handle) {
+    for (auto& unit : texture_units) {
         if (unit.sampler == handle) {
             unit.sampler = 0;
         }
     }
+    return *this;
 }
 
-void OpenGLState::ResetProgram(GLuint handle) {
-    if (cur_state.draw.shader_program == handle) {
-        cur_state.draw.shader_program = 0;
+OpenGLState& OpenGLState::ResetProgram(GLuint handle) {
+    if (draw.shader_program == handle) {
+        draw.shader_program = 0;
     }
+    return *this;
 }
 
-void OpenGLState::ResetBuffer(GLuint handle) {
-    if (cur_state.draw.vertex_buffer == handle) {
-        cur_state.draw.vertex_buffer = 0;
+OpenGLState& OpenGLState::ResetBuffer(GLuint handle) {
+    if (draw.vertex_buffer == handle) {
+        draw.vertex_buffer = 0;
     }
-    if (cur_state.draw.uniform_buffer == handle) {
-        cur_state.draw.uniform_buffer = 0;
+    if (draw.uniform_buffer == handle) {
+        draw.uniform_buffer = 0;
     }
+    return *this;
 }
 
-void OpenGLState::ResetVertexArray(GLuint handle) {
-    if (cur_state.draw.vertex_array == handle) {
-        cur_state.draw.vertex_array = 0;
+OpenGLState& OpenGLState::ResetVertexArray(GLuint handle) {
+    if (draw.vertex_array == handle) {
+        draw.vertex_array = 0;
     }
+    return *this;
 }
 
-void OpenGLState::ResetFramebuffer(GLuint handle) {
-    if (cur_state.draw.read_framebuffer == handle) {
-        cur_state.draw.read_framebuffer = 0;
+OpenGLState& OpenGLState::ResetFramebuffer(GLuint handle) {
+    if (draw.read_framebuffer == handle) {
+        draw.read_framebuffer = 0;
     }
-    if (cur_state.draw.draw_framebuffer == handle) {
-        cur_state.draw.draw_framebuffer = 0;
+    if (draw.draw_framebuffer == handle) {
+        draw.draw_framebuffer = 0;
     }
+    return *this;
 }

src/video_core/renderer_opengl/gl_state.h

@@ -124,25 +124,40 @@ public:
         GLuint shader_program;   // GL_CURRENT_PROGRAM
     } draw;
 
+    struct {
+        bool enabled; // GL_SCISSOR_TEST
+        GLint x;
+        GLint y;
+        GLsizei width;
+        GLsizei height;
+    } scissor;
+
+    struct {
+        GLint x;
+        GLint y;
+        GLsizei width;
+        GLsizei height;
+    } viewport;
+
     std::array<bool, 2> clip_distance; // GL_CLIP_DISTANCE
 
     OpenGLState();
 
     /// Get the currently active OpenGL state
-    static const OpenGLState& GetCurState() {
+    static OpenGLState GetCurState() {
         return cur_state;
     }
 
     /// Apply this state as the current OpenGL state
     void Apply() const;
 
-    /// Resets and unbinds any references to the given resource in the current OpenGL state
-    static void ResetTexture(GLuint handle);
-    static void ResetSampler(GLuint handle);
-    static void ResetProgram(GLuint handle);
-    static void ResetBuffer(GLuint handle);
-    static void ResetVertexArray(GLuint handle);
-    static void ResetFramebuffer(GLuint handle);
+    /// Resets any references to the given resource
+    OpenGLState& ResetTexture(GLuint handle);
+    OpenGLState& ResetSampler(GLuint handle);
+    OpenGLState& ResetProgram(GLuint handle);
+    OpenGLState& ResetBuffer(GLuint handle);
+    OpenGLState& ResetVertexArray(GLuint handle);
+    OpenGLState& ResetFramebuffer(GLuint handle);
 
 private:
     static OpenGLState cur_state;

src/video_core/swrasterizer/swrasterizer.h

@@ -11,7 +11,7 @@ namespace Pica {
 namespace Shader {
 struct OutputVertex;
 }
-}
+} // namespace Pica
 
 namespace VideoCore {
 
@@ -22,6 +22,7 @@ class SWRasterizer : public RasterizerInterface {
     void NotifyPicaRegisterChanged(u32 id) override {}
     void FlushAll() override {}
     void FlushRegion(PAddr addr, u32 size) override {}
+    void InvalidateRegion(PAddr addr, u32 size) override {}
     void FlushAndInvalidateRegion(PAddr addr, u32 size) override {}
 };
 

src/video_core/utils.h

@@ -9,9 +9,9 @@
 namespace VideoCore {
 
 // 8x8 Z-Order coordinate from 2D coordinates
-static inline u32 MortonInterleave(u32 x, u32 y) {
-    static const u32 xlut[] = {0x00, 0x01, 0x04, 0x05, 0x10, 0x11, 0x14, 0x15};
-    static const u32 ylut[] = {0x00, 0x02, 0x08, 0x0a, 0x20, 0x22, 0x28, 0x2a};
+static constexpr u32 MortonInterleave(u32 x, u32 y) {
+    constexpr u32 xlut[] = {0x00, 0x01, 0x04, 0x05, 0x10, 0x11, 0x14, 0x15};
+    constexpr u32 ylut[] = {0x00, 0x02, 0x08, 0x0a, 0x20, 0x22, 0x28, 0x2a};
     return xlut[x % 8] + ylut[y % 8];
 }