Merge pull request #595 from bunnei/raster-cache

Rewrite the OpenGL rasterizer cache

src/core/settings.h

@@ -129,6 +129,7 @@ struct Values {
     // Renderer
     float resolution_factor;
     bool toggle_framelimit;
+    bool use_accurate_framebuffers;
 
     float bg_red;
     float bg_green;

src/core/telemetry_session.cpp

@@ -161,6 +161,8 @@ TelemetrySession::TelemetrySession() {
              Settings::values.resolution_factor);
     AddField(Telemetry::FieldType::UserConfig, "Renderer_ToggleFramelimit",
              Settings::values.toggle_framelimit);
+    AddField(Telemetry::FieldType::UserConfig, "Renderer_UseAccurateFramebuffers",
+             Settings::values.use_accurate_framebuffers);
     AddField(Telemetry::FieldType::UserConfig, "System_UseDockedMode",
              Settings::values.use_docked_mode);
 }

src/video_core/engines/maxwell_3d.h

@@ -321,19 +321,7 @@ public:
             INSERT_PADDING_WORDS(1);
         };
 
-        union {
+        struct RenderTargetConfig {
-            struct {
-                INSERT_PADDING_WORDS(0x45);
-
-                struct {
-                    INSERT_PADDING_WORDS(1);
-                    u32 data;
-                    u32 entry;
-                } macros;
-
-                INSERT_PADDING_WORDS(0x1B8);
-
-                struct {
             u32 address_high;
             u32 address_low;
             u32 width;
@@ -349,7 +337,21 @@ public:
                 return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
                                              address_low);
             }
-                } rt[NumRenderTargets];
+        };
+
+        union {
+            struct {
+                INSERT_PADDING_WORDS(0x45);
+
+                struct {
+                    INSERT_PADDING_WORDS(1);
+                    u32 data;
+                    u32 entry;
+                } macros;
+
+                INSERT_PADDING_WORDS(0x1B8);
+
+                RenderTargetConfig rt[NumRenderTargets];
 
                 struct {
                     f32 scale_x;

src/video_core/rasterizer_interface.h

@@ -51,9 +51,8 @@ public:
     }
 
     /// Attempt to use a faster method to display the framebuffer to screen
-    virtual bool AccelerateDisplay(const Tegra::FramebufferConfig& framebuffer,
+    virtual bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
-                                   VAddr framebuffer_addr, u32 pixel_stride,
+                                   u32 pixel_stride, ScreenInfo& screen_info) {
-                                   ScreenInfo& screen_info) {
         return false;
     }
 

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -146,7 +146,6 @@ std::pair<u8*, GLintptr> RasterizerOpenGL::SetupVertexArrays(u8* array_ptr,
         u64 size = end - start + 1;
 
         // Copy vertex array data
-        res_cache.FlushRegion(start, size, nullptr);
         Memory::ReadBlock(*memory_manager->GpuToCpuAddress(start), array_ptr, size);
 
         // Bind the vertex array to the buffer at the current offset.
@@ -325,29 +324,22 @@ void RasterizerOpenGL::DrawArrays() {
     std::tie(color_surface, depth_surface, surfaces_rect) =
         res_cache.GetFramebufferSurfaces(using_color_fb, using_depth_fb, viewport_rect);
 
-    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>(
+        static_cast<u32>(std::clamp<s32>(static_cast<s32>(surfaces_rect.left) + viewport_rect.left,
-            std::clamp<s32>(static_cast<s32>(surfaces_rect.left) + viewport_rect.left * res_scale,
                                          surfaces_rect.left, surfaces_rect.right)), // Left
-        static_cast<u32>(
+        static_cast<u32>(std::clamp<s32>(static_cast<s32>(surfaces_rect.bottom) + viewport_rect.top,
-            std::clamp<s32>(static_cast<s32>(surfaces_rect.bottom) + viewport_rect.top * res_scale,
                                          surfaces_rect.bottom, surfaces_rect.top)), // Top
-        static_cast<u32>(
+        static_cast<u32>(std::clamp<s32>(static_cast<s32>(surfaces_rect.left) + viewport_rect.right,
-            std::clamp<s32>(static_cast<s32>(surfaces_rect.left) + viewport_rect.right * res_scale,
                                          surfaces_rect.left, surfaces_rect.right)), // Right
-        static_cast<u32>(std::clamp<s32>(static_cast<s32>(surfaces_rect.bottom) +
+        static_cast<u32>(
-                                             viewport_rect.bottom * res_scale,
+            std::clamp<s32>(static_cast<s32>(surfaces_rect.bottom) + viewport_rect.bottom,
                             surfaces_rect.bottom, surfaces_rect.top))}; // Bottom
 
     // Bind the framebuffer surfaces
     BindFramebufferSurfaces(color_surface, depth_surface, has_stencil);
 
     // Sync the viewport
-    SyncViewport(surfaces_rect, res_scale);
+    SyncViewport(surfaces_rect);
 
     // Sync the blend state registers
     SyncBlendState();
@@ -442,19 +434,11 @@ void RasterizerOpenGL::DrawArrays() {
     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.MarkSurfaceAsDirty(color_surface);
-        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.MarkSurfaceAsDirty(depth_surface);
-        res_cache.InvalidateRegion(boost::icl::first(interval), boost::icl::length(interval),
-                                   depth_surface);
     }
 }
 
@@ -462,7 +446,7 @@ void RasterizerOpenGL::NotifyMaxwellRegisterChanged(u32 method) {}
 
 void RasterizerOpenGL::FlushAll() {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
-    res_cache.FlushAll();
+    res_cache.FlushRegion(0, Kernel::VMManager::MAX_ADDRESS);
 }
 
 void RasterizerOpenGL::FlushRegion(Tegra::GPUVAddr addr, u64 size) {
@@ -472,13 +456,13 @@ void RasterizerOpenGL::FlushRegion(Tegra::GPUVAddr addr, u64 size) {
 
 void RasterizerOpenGL::InvalidateRegion(Tegra::GPUVAddr addr, u64 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
-    res_cache.InvalidateRegion(addr, size, nullptr);
+    res_cache.InvalidateRegion(addr, size);
 }
 
 void RasterizerOpenGL::FlushAndInvalidateRegion(Tegra::GPUVAddr addr, u64 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
     res_cache.FlushRegion(addr, size);
-    res_cache.InvalidateRegion(addr, size, nullptr);
+    res_cache.InvalidateRegion(addr, size);
 }
 
 bool RasterizerOpenGL::AccelerateDisplayTransfer(const void* config) {
@@ -497,45 +481,28 @@ bool RasterizerOpenGL::AccelerateFill(const void* config) {
     return true;
 }
 
-bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& framebuffer,
+bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
                                          VAddr framebuffer_addr, u32 pixel_stride,
                                          ScreenInfo& screen_info) {
-    if (framebuffer_addr == 0) {
+    if (!framebuffer_addr) {
-        return false;
+        return {};
     }
+
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
 
-    SurfaceParams src_params;
+    const auto& surface{res_cache.TryFindFramebufferSurface(framebuffer_addr)};
-    src_params.cpu_addr = framebuffer_addr;
+    if (!surface) {
-    src_params.addr = res_cache.TryFindFramebufferGpuAddress(framebuffer_addr).get_value_or(0);
+        return {};
-    src_params.width = std::min(framebuffer.width, pixel_stride);
-    src_params.height = framebuffer.height;
-    src_params.stride = pixel_stride;
-    src_params.is_tiled = true;
-    src_params.block_height = Tegra::Texture::TICEntry::DefaultBlockHeight;
-    src_params.pixel_format =
-        SurfaceParams::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format);
-    src_params.component_type =
-        SurfaceParams::ComponentTypeFromGPUPixelFormat(framebuffer.pixel_format);
-    src_params.UpdateParams();
-
-    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;
     }
 
-    u32 scaled_width = src_surface->GetScaledWidth();
+    // Verify that the cached surface is the same size and format as the requested framebuffer
-    u32 scaled_height = src_surface->GetScaledHeight();
+    const auto& params{surface->GetSurfaceParams()};
+    const auto& pixel_format{SurfaceParams::PixelFormatFromGPUPixelFormat(config.pixel_format)};
+    ASSERT_MSG(params.width == config.width, "Framebuffer width is different");
+    ASSERT_MSG(params.height == config.height, "Framebuffer height is different");
+    ASSERT_MSG(params.pixel_format == pixel_format, "Framebuffer pixel_format is different");
 
-    screen_info.display_texcoords = MathUtil::Rectangle<float>(
+    screen_info.display_texture = surface->Texture().handle;
-        (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;
 
     return true;
 }
@@ -674,7 +641,7 @@ u32 RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, GLuint program,
         texture_samplers[current_bindpoint].SyncWithConfig(texture.tsc);
         Surface surface = res_cache.GetTextureSurface(texture);
         if (surface != nullptr) {
-            state.texture_units[current_bindpoint].texture_2d = surface->texture.handle;
+            state.texture_units[current_bindpoint].texture_2d = surface->Texture().handle;
             state.texture_units[current_bindpoint].swizzle.r =
                 MaxwellToGL::SwizzleSource(texture.tic.x_source);
             state.texture_units[current_bindpoint].swizzle.g =
@@ -700,16 +667,16 @@ void RasterizerOpenGL::BindFramebufferSurfaces(const Surface& color_surface,
     state.Apply();
 
     glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D,
-                           color_surface != nullptr ? color_surface->texture.handle : 0, 0);
+                           color_surface != nullptr ? color_surface->Texture().handle : 0, 0);
     if (depth_surface != nullptr) {
         if (has_stencil) {
             // attach both depth and stencil
             glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
-                                   depth_surface->texture.handle, 0);
+                                   depth_surface->Texture().handle, 0);
         } else {
             // attach depth
             glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D,
-                                   depth_surface->texture.handle, 0);
+                                   depth_surface->Texture().handle, 0);
             // clear stencil attachment
             glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
         }
@@ -720,14 +687,14 @@ void RasterizerOpenGL::BindFramebufferSurfaces(const Surface& color_surface,
     }
 }
 
-void RasterizerOpenGL::SyncViewport(const MathUtil::Rectangle<u32>& surfaces_rect, u16 res_scale) {
+void RasterizerOpenGL::SyncViewport(const MathUtil::Rectangle<u32>& surfaces_rect) {
     const auto& regs = Core::System().GetInstance().GPU().Maxwell3D().regs;
     const MathUtil::Rectangle<s32> viewport_rect{regs.viewport_transform[0].GetRect()};
 
-    state.viewport.x = static_cast<GLint>(surfaces_rect.left) + viewport_rect.left * res_scale;
+    state.viewport.x = static_cast<GLint>(surfaces_rect.left) + viewport_rect.left;
-    state.viewport.y = static_cast<GLint>(surfaces_rect.bottom) + viewport_rect.bottom * res_scale;
+    state.viewport.y = static_cast<GLint>(surfaces_rect.bottom) + viewport_rect.bottom;
-    state.viewport.width = static_cast<GLsizei>(viewport_rect.GetWidth() * res_scale);
+    state.viewport.width = static_cast<GLsizei>(viewport_rect.GetWidth());
-    state.viewport.height = static_cast<GLsizei>(viewport_rect.GetHeight() * res_scale);
+    state.viewport.height = static_cast<GLsizei>(viewport_rect.GetHeight());
 }
 
 void RasterizerOpenGL::SyncClipEnabled() {

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -109,7 +109,7 @@ private:
                       u32 current_unit, const std::vector<GLShader::SamplerEntry>& entries);
 
     /// Syncs the viewport to match the guest state
-    void SyncViewport(const MathUtil::Rectangle<u32>& surfaces_rect, u16 res_scale);
+    void SyncViewport(const MathUtil::Rectangle<u32>& surfaces_rect);
 
     /// Syncs the clip enabled status to match the guest state
     void SyncClipEnabled();

src/video_core/renderer_opengl/gl_rasterizer_cache.h

@@ -1,57 +1,26 @@
-// Copyright 2015 Citra Emulator Project
+// Copyright 2018 yuzu Emulator Project
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
 #pragma once
 
 #include <array>
+#include <map>
 #include <memory>
-#include <set>
+#include <vector>
-#include <tuple>
-#ifdef __GNUC__
-#pragma GCC diagnostic push
-#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
-#include <boost/optional.hpp>
-#include <glad/glad.h>
-#include "common/assert.h"
-#include "common/common_funcs.h"
 #include "common/common_types.h"
+#include "common/hash.h"
 #include "common/math_util.h"
-#include "video_core/gpu.h"
+#include "video_core/engines/maxwell_3d.h"
-#include "video_core/memory_manager.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/textures/texture.h"
 
-struct CachedSurface;
+class CachedSurface;
 using Surface = std::shared_ptr<CachedSurface>;
-using SurfaceSet = std::set<Surface>;
-
-using SurfaceRegions = boost::icl::interval_set<Tegra::GPUVAddr>;
-using SurfaceMap = boost::icl::interval_map<Tegra::GPUVAddr, Surface>;
-using SurfaceCache = boost::icl::interval_map<Tegra::GPUVAddr, SurfaceSet>;
-
-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<u64, 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 {
         ABGR8 = 0,
@@ -93,10 +62,10 @@ struct SurfaceParams {
     /**
      * Gets the compression factor for the specified PixelFormat. This applies to just the
      * "compressed width" and "compressed height", not the overall compression factor of a
-     * compressed image. This is used for maintaining proper surface sizes for compressed texture
+     * compressed image. This is used for maintaining proper surface sizes for compressed
-     * formats.
+     * texture formats.
      */
-    static constexpr u32 GetCompresssionFactor(PixelFormat format) {
+    static constexpr u32 GetCompressionFactor(PixelFormat format) {
         if (format == PixelFormat::Invalid)
             return 0;
 
@@ -112,15 +81,12 @@ struct SurfaceParams {
             4, // DXT23
             4, // DXT45
             4, // DXN1
-            1, // ASTC_2D_4X4
+            4, // ASTC_2D_4X4
         }};
 
         ASSERT(static_cast<size_t>(format) < compression_factor_table.size());
         return compression_factor_table[static_cast<size_t>(format)];
     }
-    u32 GetCompresssionFactor() const {
-        return GetCompresssionFactor(pixel_format);
-    }
 
     static constexpr u32 GetFormatBpp(PixelFormat format) {
         if (format == PixelFormat::Invalid)
@@ -165,25 +131,6 @@ struct SurfaceParams {
         }
     }
 
-    static bool IsFormatASTC(PixelFormat format) {
-        switch (format) {
-        case PixelFormat::ASTC_2D_4X4:
-            return true;
-        default:
-            return false;
-        }
-    }
-
-    static PixelFormat PixelFormatFromGPUPixelFormat(Tegra::FramebufferConfig::PixelFormat format) {
-        switch (format) {
-        case Tegra::FramebufferConfig::PixelFormat::ABGR8:
-            return PixelFormat::ABGR8;
-        default:
-            NGLOG_CRITICAL(HW_GPU, "Unimplemented format={}", static_cast<u32>(format));
-            UNREACHABLE();
-        }
-    }
-
     static PixelFormat PixelFormatFromTextureFormat(Tegra::Texture::TextureFormat format) {
         // TODO(Subv): Properly implement this
         switch (format) {
@@ -276,36 +223,16 @@ struct SurfaceParams {
         }
     }
 
-    static ComponentType ComponentTypeFromGPUPixelFormat(
+    static PixelFormat PixelFormatFromGPUPixelFormat(Tegra::FramebufferConfig::PixelFormat format) {
-        Tegra::FramebufferConfig::PixelFormat format) {
         switch (format) {
         case Tegra::FramebufferConfig::PixelFormat::ABGR8:
-            return ComponentType::UNorm;
+            return PixelFormat::ABGR8;
         default:
             NGLOG_CRITICAL(HW_GPU, "Unimplemented format={}", static_cast<u32>(format));
             UNREACHABLE();
         }
     }
 
-    static bool CheckFormatsBlittable(PixelFormat pixel_format_a, PixelFormat pixel_format_b) {
-        SurfaceType a_type = GetFormatType(pixel_format_a);
-        SurfaceType b_type = GetFormatType(pixel_format_b);
-
-        if (a_type == SurfaceType::ColorTexture && b_type == SurfaceType::ColorTexture) {
-            return true;
-        }
-
-        if (a_type == SurfaceType::Depth && b_type == SurfaceType::Depth) {
-            return true;
-        }
-
-        if (a_type == SurfaceType::DepthStencil && b_type == SurfaceType::DepthStencil) {
-            return true;
-        }
-
-        return false;
-    }
-
     static SurfaceType GetFormatType(PixelFormat pixel_format) {
         if (static_cast<size_t>(pixel_format) < MaxPixelFormat) {
             return SurfaceType::ColorTexture;
@@ -317,168 +244,101 @@ struct SurfaceParams {
         return SurfaceType::Invalid;
     }
 
-    /// Update the params "size", "end" and "type" from the already set "addr", "width", "height"
+    /// Returns the rectangle corresponding to this surface
-    /// and "pixel_format"
+    MathUtil::Rectangle<u32> GetRect() const;
-    void UpdateParams() {
+
-        if (stride == 0) {
+    /// Returns the size of this surface in bytes, adjusted for compression
-            stride = width;
+    size_t SizeInBytes() const {
-        }
+        const u32 compression_factor{GetCompressionFactor(pixel_format)};
-        type = GetFormatType(pixel_format);
+        ASSERT(width % compression_factor == 0);
-        size = !is_tiled ? BytesInPixels(stride * (height - 1) + width)
+        ASSERT(height % compression_factor == 0);
-                         : BytesInPixels(stride * 8 * (height / 8 - 1) + width * 8);
+        return (width / compression_factor) * (height / compression_factor) *
-        end = addr + size;
+               GetFormatBpp(pixel_format) / CHAR_BIT;
-    }
-
-    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;
-
-    /**
-     * Gets the actual width (in pixels) of the surface. This is provided because `width` is used
-     * for tracking the surface region in memory, which may be compressed for certain formats. In
-     * this scenario, `width` is actually the compressed width.
-     */
-    u32 GetActualWidth() const {
-        return width * GetCompresssionFactor();
-    }
-
-    /**
-     * Gets the actual height (in pixels) of the surface. This is provided because `height` is used
-     * for tracking the surface region in memory, which may be compressed for certain formats. In
-     * this scenario, `height` is actually the compressed height.
-     */
-    u32 GetActualHeight() const {
-        return height * GetCompresssionFactor();
-    }
-
-    u32 GetScaledWidth() const {
-        return width * res_scale;
-    }
-
-    u32 GetScaledHeight() const {
-        return height * res_scale;
-    }
-
-    MathUtil::Rectangle<u32> GetRect() const {
-        return {0, height, width, 0};
-    }
-
-    MathUtil::Rectangle<u32> GetScaledRect() const {
-        return {0, GetScaledHeight(), GetScaledWidth(), 0};
-    }
-
-    u64 PixelsInBytes(u64 size) const {
-        return size * CHAR_BIT / GetFormatBpp(pixel_format);
-    }
-
-    u64 BytesInPixels(u64 pixels) const {
-        return pixels * GetFormatBpp(pixel_format) / CHAR_BIT;
     }
 
+    /// Returns the CPU virtual address for this surface
     VAddr GetCpuAddr() const;
 
-    bool ExactMatch(const SurfaceParams& other_surface) const;
+    /// Returns true if the specified region overlaps with this surface's region in Switch memory
-    bool CanSubRect(const SurfaceParams& sub_surface) const;
+    bool IsOverlappingRegion(Tegra::GPUVAddr region_addr, size_t region_size) const {
-    bool CanExpand(const SurfaceParams& expanded_surface) const;
+        return addr <= (region_addr + region_size) && region_addr <= (addr + size_in_bytes);
-    bool CanTexCopy(const SurfaceParams& texcopy_params) const;
+    }
 
-    MathUtil::Rectangle<u32> GetSubRect(const SurfaceParams& sub_surface) const;
+    /// Creates SurfaceParams from a texture configation
-    MathUtil::Rectangle<u32> GetScaledSubRect(const SurfaceParams& sub_surface) const;
+    static SurfaceParams CreateForTexture(const Tegra::Texture::FullTextureInfo& config);
 
-    Tegra::GPUVAddr addr = 0;
+    /// Creates SurfaceParams from a framebuffer configation
-    Tegra::GPUVAddr end = 0;
+    static SurfaceParams CreateForFramebuffer(
-    boost::optional<VAddr> cpu_addr;
+        const Tegra::Engines::Maxwell3D::Regs::RenderTargetConfig& config);
-    u64 size = 0;
 
-    u32 width = 0;
+    Tegra::GPUVAddr addr;
-    u32 height = 0;
+    bool is_tiled;
-    u32 stride = 0;
+    u32 block_height;
-    u32 block_height = 0;
+    PixelFormat pixel_format;
-    u16 res_scale = 1;
+    ComponentType component_type;
-
+    SurfaceType type;
-    bool is_tiled = false;
+    u32 width;
-    PixelFormat pixel_format = PixelFormat::Invalid;
+    u32 height;
-    SurfaceType type = SurfaceType::Invalid;
+    u32 unaligned_height;
-    ComponentType component_type = ComponentType::Invalid;
+    size_t size_in_bytes;
 };
 
-struct CachedSurface : SurfaceParams {
+/// Hashable variation of SurfaceParams, used for a key in the surface cache
-    bool CanFill(const SurfaceParams& dest_surface, SurfaceInterval fill_interval) const;
+struct SurfaceKey : Common::HashableStruct<SurfaceParams> {
-    bool CanCopy(const SurfaceParams& dest_surface, SurfaceInterval copy_interval) const;
+    static SurfaceKey Create(const SurfaceParams& params) {
+        SurfaceKey res;
+        res.state = params;
+        return res;
+    }
+};
 
-    bool IsRegionValid(SurfaceInterval interval) const {
+namespace std {
-        return (invalid_regions.find(interval) == invalid_regions.end());
+template <>
+struct hash<SurfaceKey> {
+    size_t operator()(const SurfaceKey& k) const {
+        return k.Hash();
+    }
+};
+} // namespace std
+
+class CachedSurface final {
+public:
+    CachedSurface(const SurfaceParams& params);
+
+    const OGLTexture& Texture() const {
+        return texture;
     }
 
-    bool IsSurfaceFullyInvalid() const {
+    static constexpr unsigned int GetGLBytesPerPixel(SurfaceParams::PixelFormat format) {
-        return (invalid_regions & GetInterval()) == SurfaceRegions(GetInterval());
+        if (format == SurfaceParams::PixelFormat::Invalid)
-    }
-
-    bool registered = false;
-    SurfaceRegions invalid_regions;
-
-    u64 fill_size = 0; /// Number of bytes to read from fill_data
-    std::array<u8, 4> fill_data;
-
-    OGLTexture texture;
-
-    static constexpr unsigned int GetGLBytesPerPixel(PixelFormat format) {
-        if (format == PixelFormat::Invalid)
             return 0;
 
         return SurfaceParams::GetFormatBpp(format) / CHAR_BIT;
     }
 
-    std::unique_ptr<u8[]> gl_buffer;
+    const SurfaceParams& GetSurfaceParams() const {
-    size_t gl_buffer_size = 0;
+        return params;
+    }
 
     // Read/Write data in Switch memory to/from gl_buffer
-    void LoadGLBuffer(Tegra::GPUVAddr load_start, Tegra::GPUVAddr load_end);
+    void LoadGLBuffer();
-    void FlushGLBuffer(Tegra::GPUVAddr flush_start, Tegra::GPUVAddr flush_end);
+    void FlushGLBuffer();
 
     // Upload/Download data in gl_buffer in/to this surface's texture
-    void UploadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint read_fb_handle,
+    void UploadGLTexture(GLuint read_fb_handle, GLuint draw_fb_handle);
-                         GLuint draw_fb_handle);
+    void DownloadGLTexture(GLuint read_fb_handle, GLuint draw_fb_handle);
-    void DownloadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint read_fb_handle,
+
-                           GLuint draw_fb_handle);
+private:
+    OGLTexture texture;
+    std::vector<u8> gl_buffer;
+    SurfaceParams params;
 };
 
-class RasterizerCacheOpenGL : NonCopyable {
+class RasterizerCacheOpenGL final : NonCopyable {
 public:
     RasterizerCacheOpenGL();
     ~RasterizerCacheOpenGL();
 
-    /// 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);
-
-    void ConvertD24S8toABGR(GLuint src_tex, const MathUtil::Rectangle<u32>& src_rect,
-                            GLuint dst_tex, const MathUtil::Rectangle<u32>& dst_rect);
-
-    /// Copy one surface's region to another
-    void CopySurface(const Surface& src_surface, const Surface& dst_surface,
-                     SurfaceInterval copy_interval);
-
-    /// Load a texture from Switch memory to OpenGL and cache it (if not already cached)
-    Surface GetSurface(const SurfaceParams& params, ScaleMatch match_res_scale,
-                       bool load_if_create);
-
-    /// Tries to find a framebuffer GPU address based on the provided CPU address
-    boost::optional<Tegra::GPUVAddr> TryFindFramebufferGpuAddress(VAddr cpu_addr) const;
-
-    /// Attempt to find a subrect (resolution scaled) of a surface, otherwise loads a texture from
-    /// Switch memory to OpenGL and caches it (if not already cached)
-    SurfaceRect_Tuple GetSurfaceSubRect(const SurfaceParams& params, ScaleMatch match_res_scale,
-                                        bool load_if_create);
-
     /// Get a surface based on the texture configuration
     Surface GetTextureSurface(const Tegra::Texture::FullTextureInfo& config);
 
@@ -486,29 +346,21 @@ public:
     SurfaceSurfaceRect_Tuple GetFramebufferSurfaces(bool using_color_fb, bool using_depth_fb,
                                                     const MathUtil::Rectangle<s32>& viewport);
 
-    /// Get a surface that matches the fill config
+    /// Marks the specified surface as "dirty", in that it is out of sync with Switch memory
-    Surface GetFillSurface(const void* config);
+    void MarkSurfaceAsDirty(const Surface& surface);
 
-    /// Get a surface that matches a "texture copy" display transfer config
+    /// Tries to find a framebuffer GPU address based on the provided CPU address
-    SurfaceRect_Tuple GetTexCopySurface(const SurfaceParams& params);
+    Surface TryFindFramebufferSurface(VAddr cpu_addr) const;
 
     /// Write any cached resources overlapping the region back to memory (if dirty)
-    void FlushRegion(Tegra::GPUVAddr addr, u64 size, Surface flush_surface = nullptr);
+    void FlushRegion(Tegra::GPUVAddr addr, size_t size);
 
-    /// Mark region as being invalidated by region_owner (nullptr if Switch memory)
+    /// Mark the specified region as being invalidated
-    void InvalidateRegion(Tegra::GPUVAddr addr, u64 size, const Surface& region_owner);
+    void InvalidateRegion(Tegra::GPUVAddr addr, size_t size);
-
-    /// Flush all cached resources tracked by this cache manager
-    void FlushAll();
 
 private:
-    void DuplicateSurface(const Surface& src_surface, const Surface& dest_surface);
+    void LoadSurface(const Surface& surface);
-
+    Surface GetSurface(const SurfaceParams& params);
-    /// Update surface's texture for given region when necessary
-    void ValidateSurface(const Surface& surface, Tegra::GPUVAddr addr, u64 size);
-
-    /// Create a new surface
-    Surface CreateSurface(const SurfaceParams& params);
 
     /// Register surface into the cache
     void RegisterSurface(const Surface& surface);
@@ -519,18 +371,9 @@ private:
     /// Increase/decrease the number of surface in pages touching the specified region
     void UpdatePagesCachedCount(Tegra::GPUVAddr addr, u64 size, int delta);
 
-    SurfaceCache surface_cache;
+    std::unordered_map<SurfaceKey, Surface> surface_cache;
     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;
-    OGLProgram d24s8_abgr_shader;
-    GLint d24s8_abgr_tbo_size_u_id;
-    GLint d24s8_abgr_viewport_u_id;
 };

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -150,7 +150,6 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf
                                          screen_info)) {
         // Reset the screen info's display texture to its own permanent texture
         screen_info.display_texture = screen_info.texture.resource.handle;
-        screen_info.display_texcoords = MathUtil::Rectangle<float>(0.f, 0.f, 1.f, 1.f);
 
         Memory::RasterizerFlushVirtualRegion(framebuffer_addr, size_in_bytes,
                                              Memory::FlushMode::Flush);

src/video_core/renderer_opengl/renderer_opengl.h

@@ -27,7 +27,7 @@ struct TextureInfo {
 /// Structure used for storing information about the display target for the Switch screen
 struct ScreenInfo {
     GLuint display_texture;
-    MathUtil::Rectangle<float> display_texcoords;
+    const MathUtil::Rectangle<float> display_texcoords{0.0f, 0.0f, 1.0f, 1.0f};
     TextureInfo texture;
 };
 

src/yuzu/configuration/config.cpp

@@ -84,6 +84,8 @@ void Config::ReadValues() {
     qt_config->beginGroup("Renderer");
     Settings::values.resolution_factor = qt_config->value("resolution_factor", 1.0).toFloat();
     Settings::values.toggle_framelimit = qt_config->value("toggle_framelimit", true).toBool();
+    Settings::values.use_accurate_framebuffers =
+        qt_config->value("use_accurate_framebuffers", false).toBool();
 
     Settings::values.bg_red = qt_config->value("bg_red", 0.0).toFloat();
     Settings::values.bg_green = qt_config->value("bg_green", 0.0).toFloat();
@@ -184,6 +186,7 @@ void Config::SaveValues() {
     qt_config->beginGroup("Renderer");
     qt_config->setValue("resolution_factor", (double)Settings::values.resolution_factor);
     qt_config->setValue("toggle_framelimit", Settings::values.toggle_framelimit);
+    qt_config->setValue("use_accurate_framebuffers", Settings::values.use_accurate_framebuffers);
 
     // Cast to double because Qt's written float values are not human-readable
     qt_config->setValue("bg_red", (double)Settings::values.bg_red);

src/yuzu/configuration/configure_graphics.cpp

@@ -59,11 +59,13 @@ void ConfigureGraphics::setConfiguration() {
     ui->resolution_factor_combobox->setCurrentIndex(
         static_cast<int>(FromResolutionFactor(Settings::values.resolution_factor)));
     ui->toggle_framelimit->setChecked(Settings::values.toggle_framelimit);
+    ui->use_accurate_framebuffers->setChecked(Settings::values.use_accurate_framebuffers);
 }
 
 void ConfigureGraphics::applyConfiguration() {
     Settings::values.resolution_factor =
         ToResolutionFactor(static_cast<Resolution>(ui->resolution_factor_combobox->currentIndex()));
     Settings::values.toggle_framelimit = ui->toggle_framelimit->isChecked();
+    Settings::values.use_accurate_framebuffers = ui->use_accurate_framebuffers->isChecked();
     Settings::Apply();
 }

src/yuzu/configuration/configure_graphics.ui

@@ -29,6 +29,13 @@
           </property>
          </widget>
         </item>
+        <item>
+         <widget class="QCheckBox" name="use_accurate_framebuffers">
+          <property name="text">
+           <string>Use accurate framebuffers (slow)</string>
+          </property>
+         </widget>
+        </item>
         <item>
          <layout class="QHBoxLayout" name="horizontalLayout">
           <item>

src/yuzu_cmd/config.cpp

@@ -98,6 +98,8 @@ void Config::ReadValues() {
         (float)sdl2_config->GetReal("Renderer", "resolution_factor", 1.0);
     Settings::values.toggle_framelimit =
         sdl2_config->GetBoolean("Renderer", "toggle_framelimit", true);
+    Settings::values.use_accurate_framebuffers =
+        sdl2_config->GetBoolean("Renderer", "use_accurate_framebuffers", false);
 
     Settings::values.bg_red = (float)sdl2_config->GetReal("Renderer", "bg_red", 0.0);
     Settings::values.bg_green = (float)sdl2_config->GetReal("Renderer", "bg_green", 0.0);

src/yuzu_cmd/default_ini.h

@@ -102,6 +102,10 @@ resolution_factor =
 # 0 (default): Off, 1: On
 use_vsync =
 
+# Whether to use accurate framebuffers
+# 0 (default): Off (fast), 1 : On (slow)
+use_accurate_framebuffers =
+
 # The clear color for the renderer. What shows up on the sides of the bottom screen.
 # Must be in range of 0.0-1.0. Defaults to 1.0 for all.
 bg_red =