Merge pull request #254 from bunnei/port-citra-renderer

Port Citra OpenGL rasterizer code

externals/glad/Readme.md

@@ -1,5 +1,5 @@
 These files were generated by the [glad](https://github.com/Dav1dde/glad) OpenGL loader generator and have been checked in as-is. You can re-generate them using glad with the following command:
 
 ```
-python -m glad --profile core --out-path glad/ --api gl=3.3,gles=3.0
+python -m glad --profile core --out-path glad/ --api gl=3.3 --generator=c
 ```

externals/glad/include/KHR/khrplatform.h

@@ -26,7 +26,7 @@
 
 /* Khronos platform-specific types and definitions.
  *
- * $Revision: 23298 $ on $Date: 2013-09-30 17:07:13 -0700 (Mon, 30 Sep 2013) $
+ * $Revision: 32517 $ on $Date: 2016-03-11 02:41:19 -0800 (Fri, 11 Mar 2016) $
  *
  * Adopters may modify this file to suit their platform. Adopters are
  * encouraged to submit platform specific modifications to the Khronos
@@ -101,6 +101,8 @@
 #   define KHRONOS_APICALL __declspec(dllimport)
 #elif defined (__SYMBIAN32__)
 #   define KHRONOS_APICALL IMPORT_C
+#elif defined(__ANDROID__)
+#   define KHRONOS_APICALL __attribute__((visibility("default")))
 #else
 #   define KHRONOS_APICALL
 #endif
@@ -223,7 +225,7 @@ typedef signed   short int     khronos_int16_t;
 typedef unsigned short int     khronos_uint16_t;
 
 /*
- * Types that differ between LLP64 and LP64 architectures - in LLP64, 
+ * Types that differ between LLP64 and LP64 architectures - in LLP64,
  * pointers are 64 bits, but 'long' is still 32 bits. Win64 appears
  * to be the only LLP64 architecture in current use.
  */

src/video_core/CMakeLists.txt

@@ -11,13 +11,24 @@ add_library(video_core STATIC
     gpu.h
     memory_manager.cpp
     memory_manager.h
+    rasterizer_interface.h
     renderer_base.cpp
     renderer_base.h
+    renderer_opengl/gl_rasterizer.cpp
+    renderer_opengl/gl_rasterizer.h
+    renderer_opengl/gl_rasterizer_cache.cpp
+    renderer_opengl/gl_rasterizer_cache.h
     renderer_opengl/gl_resource_manager.h
+    renderer_opengl/gl_shader_decompiler.cpp
+    renderer_opengl/gl_shader_decompiler.h
+    renderer_opengl/gl_shader_gen.cpp
+    renderer_opengl/gl_shader_gen.h
     renderer_opengl/gl_shader_util.cpp
     renderer_opengl/gl_shader_util.h
     renderer_opengl/gl_state.cpp
     renderer_opengl/gl_state.h
+    renderer_opengl/gl_stream_buffer.cpp
+    renderer_opengl/gl_stream_buffer.h
     renderer_opengl/renderer_opengl.cpp
     renderer_opengl/renderer_opengl.h
     utils.h

src/video_core/rasterizer_interface.h

@@ -0,0 +1,61 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/common_types.h"
+
+struct ScreenInfo;
+
+namespace VideoCore {
+
+class RasterizerInterface {
+public:
+    virtual ~RasterizerInterface() {}
+
+    /// Draw the current batch of triangles
+    virtual void DrawTriangles() = 0;
+
+    /// Notify rasterizer that the specified Maxwell register has been changed
+    virtual void NotifyMaxwellRegisterChanged(u32 id) = 0;
+
+    /// Notify rasterizer that all caches should be flushed to 3DS memory
+    virtual void FlushAll() = 0;
+
+    /// 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;
+
+    /// Attempt to use a faster method to perform a display transfer with is_texture_copy = 0
+    virtual bool AccelerateDisplayTransfer(const void* config) {
+        return false;
+    }
+
+    /// Attempt to use a faster method to perform a display transfer with is_texture_copy = 1
+    virtual bool AccelerateTextureCopy(const void* config) {
+        return false;
+    }
+
+    /// Attempt to use a faster method to fill a region
+    virtual bool AccelerateFill(const void* config) {
+        return false;
+    }
+
+    /// Attempt to use a faster method to display the framebuffer to screen
+    virtual bool AccelerateDisplay(const void* config, PAddr framebuffer_addr, u32 pixel_stride,
+                                   ScreenInfo& screen_info) {
+        return false;
+    }
+
+    virtual bool AccelerateDrawBatch(bool is_indexed) {
+        return false;
+    }
+};
+} // namespace VideoCore

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -0,0 +1,269 @@
+// Copyright 2015 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <memory>
+#include <string>
+#include <tuple>
+#include <utility>
+#include <glad/glad.h>
+#include "common/alignment.h"
+#include "common/assert.h"
+#include "common/logging/log.h"
+#include "common/math_util.h"
+#include "common/microprofile.h"
+#include "common/scope_exit.h"
+#include "common/vector_math.h"
+#include "core/settings.h"
+#include "video_core/renderer_opengl/gl_rasterizer.h"
+#include "video_core/renderer_opengl/gl_shader_gen.h"
+#include "video_core/renderer_opengl/renderer_opengl.h"
+
+using PixelFormat = SurfaceParams::PixelFormat;
+using SurfaceType = SurfaceParams::SurfaceType;
+
+MICROPROFILE_DEFINE(OpenGL_VAO, "OpenGL", "Vertex Array Setup", MP_RGB(128, 128, 192));
+MICROPROFILE_DEFINE(OpenGL_VS, "OpenGL", "Vertex Shader Setup", MP_RGB(128, 128, 192));
+MICROPROFILE_DEFINE(OpenGL_FS, "OpenGL", "Fragment Shader Setup", MP_RGB(128, 128, 192));
+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));
+
+enum class UniformBindings : GLuint { Common, VS, FS };
+
+static void SetShaderUniformBlockBinding(GLuint shader, const char* name, UniformBindings binding,
+                                         size_t expected_size) {
+    GLuint ub_index = glGetUniformBlockIndex(shader, name);
+    if (ub_index != GL_INVALID_INDEX) {
+        GLint ub_size = 0;
+        glGetActiveUniformBlockiv(shader, ub_index, GL_UNIFORM_BLOCK_DATA_SIZE, &ub_size);
+        ASSERT_MSG(ub_size == expected_size,
+                   "Uniform block size did not match! Got %d, expected %zu",
+                   static_cast<int>(ub_size), expected_size);
+        glUniformBlockBinding(shader, ub_index, static_cast<GLuint>(binding));
+    }
+}
+
+static void SetShaderUniformBlockBindings(GLuint shader) {
+    SetShaderUniformBlockBinding(shader, "shader_data", UniformBindings::Common,
+                                 sizeof(RasterizerOpenGL::UniformData));
+    SetShaderUniformBlockBinding(shader, "vs_config", UniformBindings::VS,
+                                 sizeof(RasterizerOpenGL::VSUniformData));
+    SetShaderUniformBlockBinding(shader, "fs_config", UniformBindings::FS,
+                                 sizeof(RasterizerOpenGL::FSUniformData));
+}
+
+RasterizerOpenGL::RasterizerOpenGL() {
+    has_ARB_buffer_storage = false;
+    has_ARB_direct_state_access = false;
+    has_ARB_separate_shader_objects = false;
+    has_ARB_vertex_attrib_binding = false;
+
+    GLint ext_num;
+    glGetIntegerv(GL_NUM_EXTENSIONS, &ext_num);
+    for (GLint i = 0; i < ext_num; i++) {
+        std::string extension{reinterpret_cast<const char*>(glGetStringi(GL_EXTENSIONS, i))};
+
+        if (extension == "GL_ARB_buffer_storage") {
+            has_ARB_buffer_storage = true;
+        } else if (extension == "GL_ARB_direct_state_access") {
+            has_ARB_direct_state_access = true;
+        } else if (extension == "GL_ARB_separate_shader_objects") {
+            has_ARB_separate_shader_objects = true;
+        } else if (extension == "GL_ARB_vertex_attrib_binding") {
+            has_ARB_vertex_attrib_binding = true;
+        }
+    }
+
+    // Clipping plane 0 is always enabled for PICA fixed clip plane z <= 0
+    state.clip_distance[0] = true;
+
+    // Generate VBO, VAO and UBO
+    vertex_buffer = OGLStreamBuffer::MakeBuffer(GLAD_GL_ARB_buffer_storage, GL_ARRAY_BUFFER);
+    vertex_buffer->Create(VERTEX_BUFFER_SIZE, VERTEX_BUFFER_SIZE / 2);
+    sw_vao.Create();
+    uniform_buffer.Create();
+
+    state.draw.vertex_array = sw_vao.handle;
+    state.draw.vertex_buffer = vertex_buffer->GetHandle();
+    state.draw.uniform_buffer = uniform_buffer.handle;
+    state.Apply();
+
+    glBufferData(GL_UNIFORM_BUFFER, sizeof(UniformData), nullptr, GL_STATIC_DRAW);
+    glBindBufferBase(GL_UNIFORM_BUFFER, 0, uniform_buffer.handle);
+
+    uniform_block_data.dirty = true;
+
+    // Create render framebuffer
+    framebuffer.Create();
+
+    if (has_ARB_separate_shader_objects) {
+        hw_vao.Create();
+        hw_vao_enabled_attributes.fill(false);
+
+        stream_buffer = OGLStreamBuffer::MakeBuffer(has_ARB_buffer_storage, GL_ARRAY_BUFFER);
+        stream_buffer->Create(STREAM_BUFFER_SIZE, STREAM_BUFFER_SIZE / 2);
+        state.draw.vertex_buffer = stream_buffer->GetHandle();
+
+        pipeline.Create();
+        vs_input_index_min = 0;
+        vs_input_index_max = 0;
+        state.draw.program_pipeline = pipeline.handle;
+        state.draw.shader_program = 0;
+        state.draw.vertex_array = hw_vao.handle;
+        state.Apply();
+
+        glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, stream_buffer->GetHandle());
+
+        vs_uniform_buffer.Create();
+        glBindBuffer(GL_UNIFORM_BUFFER, vs_uniform_buffer.handle);
+        glBufferData(GL_UNIFORM_BUFFER, sizeof(VSUniformData), nullptr, GL_STREAM_COPY);
+        glBindBufferBase(GL_UNIFORM_BUFFER, 1, vs_uniform_buffer.handle);
+    } else {
+        UNIMPLEMENTED();
+    }
+
+    accelerate_draw = AccelDraw::Disabled;
+
+    glEnable(GL_BLEND);
+
+    // Sync fixed function OpenGL state
+    SyncClipEnabled();
+    SyncClipCoef();
+    SyncCullMode();
+    SyncBlendEnabled();
+    SyncBlendFuncs();
+    SyncBlendColor();
+}
+
+RasterizerOpenGL::~RasterizerOpenGL() {
+    if (stream_buffer != nullptr) {
+        state.draw.vertex_buffer = stream_buffer->GetHandle();
+        state.Apply();
+        stream_buffer->Release();
+    }
+}
+
+static constexpr std::array<GLenum, 4> vs_attrib_types{
+    GL_BYTE,          // VertexAttributeFormat::BYTE
+    GL_UNSIGNED_BYTE, // VertexAttributeFormat::UBYTE
+    GL_SHORT,         // VertexAttributeFormat::SHORT
+    GL_FLOAT          // VertexAttributeFormat::FLOAT
+};
+
+void RasterizerOpenGL::AnalyzeVertexArray(bool is_indexed) {
+    UNIMPLEMENTED();
+}
+
+void RasterizerOpenGL::SetupVertexArray(u8* array_ptr, GLintptr buffer_offset) {
+    MICROPROFILE_SCOPE(OpenGL_VAO);
+    UNIMPLEMENTED();
+}
+
+void RasterizerOpenGL::SetupVertexShader(VSUniformData* ub_ptr, GLintptr buffer_offset) {
+    MICROPROFILE_SCOPE(OpenGL_VS);
+    UNIMPLEMENTED();
+}
+
+void RasterizerOpenGL::SetupFragmentShader(FSUniformData* ub_ptr, GLintptr buffer_offset) {
+    MICROPROFILE_SCOPE(OpenGL_FS);
+    UNIMPLEMENTED();
+}
+
+bool RasterizerOpenGL::AccelerateDrawBatch(bool is_indexed) {
+    if (!has_ARB_separate_shader_objects) {
+        UNIMPLEMENTED();
+        return false;
+    }
+
+    accelerate_draw = is_indexed ? AccelDraw::Indexed : AccelDraw::Arrays;
+    DrawTriangles();
+
+    return true;
+}
+
+void RasterizerOpenGL::DrawTriangles() {
+    MICROPROFILE_SCOPE(OpenGL_Drawing);
+    UNIMPLEMENTED();
+}
+
+void RasterizerOpenGL::NotifyMaxwellRegisterChanged(u32 id) {}
+
+void RasterizerOpenGL::FlushAll() {
+    MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+    res_cache.FlushAll();
+}
+
+void RasterizerOpenGL::FlushRegion(PAddr addr, u32 size) {
+    MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+    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);
+    res_cache.InvalidateRegion(addr, size, nullptr);
+}
+
+bool RasterizerOpenGL::AccelerateDisplayTransfer(const void* config) {
+    MICROPROFILE_SCOPE(OpenGL_Blits);
+    UNIMPLEMENTED();
+    return true;
+}
+
+bool RasterizerOpenGL::AccelerateTextureCopy(const void* config) {
+    UNIMPLEMENTED();
+    return true;
+}
+
+bool RasterizerOpenGL::AccelerateFill(const void* config) {
+    UNIMPLEMENTED();
+    return true;
+}
+
+bool RasterizerOpenGL::AccelerateDisplay(const void* config, PAddr framebuffer_addr,
+                                         u32 pixel_stride, ScreenInfo& screen_info) {
+    UNIMPLEMENTED();
+    return true;
+}
+
+void RasterizerOpenGL::SetShader() {
+    UNIMPLEMENTED();
+}
+
+void RasterizerOpenGL::SyncClipEnabled() {
+    UNIMPLEMENTED();
+}
+
+void RasterizerOpenGL::SyncClipCoef() {
+    UNIMPLEMENTED();
+}
+
+void RasterizerOpenGL::SyncCullMode() {
+    UNIMPLEMENTED();
+}
+
+void RasterizerOpenGL::SyncDepthScale() {
+    UNIMPLEMENTED();
+}
+
+void RasterizerOpenGL::SyncDepthOffset() {
+    UNIMPLEMENTED();
+}
+
+void RasterizerOpenGL::SyncBlendEnabled() {
+    UNIMPLEMENTED();
+}
+
+void RasterizerOpenGL::SyncBlendFuncs() {
+    UNIMPLEMENTED();
+}
+
+void RasterizerOpenGL::SyncBlendColor() {
+    UNIMPLEMENTED();
+}

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -0,0 +1,162 @@
+// Copyright 2015 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <cstddef>
+#include <cstring>
+#include <memory>
+#include <unordered_map>
+#include <vector>
+#include <glad/glad.h>
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "common/hash.h"
+#include "common/vector_math.h"
+#include "video_core/rasterizer_interface.h"
+#include "video_core/renderer_opengl/gl_rasterizer_cache.h"
+#include "video_core/renderer_opengl/gl_resource_manager.h"
+#include "video_core/renderer_opengl/gl_shader_gen.h"
+#include "video_core/renderer_opengl/gl_state.h"
+#include "video_core/renderer_opengl/gl_stream_buffer.h"
+
+struct ScreenInfo;
+
+class RasterizerOpenGL : public VideoCore::RasterizerInterface {
+public:
+    RasterizerOpenGL();
+    ~RasterizerOpenGL() override;
+
+    void DrawTriangles() override;
+    void NotifyMaxwellRegisterChanged(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 void* config) override;
+    bool AccelerateTextureCopy(const void* config) override;
+    bool AccelerateFill(const void* config) override;
+    bool AccelerateDisplay(const void* config, PAddr framebuffer_addr, u32 pixel_stride,
+                           ScreenInfo& screen_info) override;
+    bool AccelerateDrawBatch(bool is_indexed) override;
+
+    struct VertexShader {
+        OGLShader shader;
+    };
+
+    struct FragmentShader {
+        OGLShader shader;
+    };
+
+    /// Uniform structure for the Uniform Buffer Object, all vectors must be 16-byte aligned
+    // NOTE: Always keep a vec4 at the end. The GL spec is not clear wether the alignment at
+    //       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 {};
+
+    // static_assert(
+    //    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");
+
+    struct VSUniformData {};
+    // static_assert(
+    //    sizeof(VSUniformData) == 1856,
+    //    "The size of the VSUniformData structure has changed, update the structure in the
+    //    shader");
+    static_assert(sizeof(VSUniformData) < 16384,
+                  "VSUniformData structure must be less than 16kb as per the OpenGL spec");
+
+    struct FSUniformData {};
+    // static_assert(
+    //    sizeof(FSUniformData) == 1856,
+    //    "The size of the FSUniformData structure has changed, update the structure in the
+    //    shader");
+    static_assert(sizeof(FSUniformData) < 16384,
+                  "FSUniformData structure must be less than 16kb as per the OpenGL spec");
+
+private:
+    struct SamplerInfo {};
+
+    /// Syncs the clip enabled status to match the guest state
+    void SyncClipEnabled();
+
+    /// Syncs the clip coefficients to match the guest state
+    void SyncClipCoef();
+
+    /// Sets the OpenGL shader in accordance with the current guest state
+    void SetShader();
+
+    /// Syncs the cull mode to match the guest state
+    void SyncCullMode();
+
+    /// Syncs the depth scale to match the guest state
+    void SyncDepthScale();
+
+    /// Syncs the depth offset to match the guest state
+    void SyncDepthOffset();
+
+    /// Syncs the blend enabled status to match the guest state
+    void SyncBlendEnabled();
+
+    /// Syncs the blend functions to match the guest state
+    void SyncBlendFuncs();
+
+    /// Syncs the blend color to match the guest state
+    void SyncBlendColor();
+
+    bool has_ARB_buffer_storage;
+    bool has_ARB_direct_state_access;
+    bool has_ARB_separate_shader_objects;
+    bool has_ARB_vertex_attrib_binding;
+
+    OpenGLState state;
+
+    RasterizerCacheOpenGL res_cache;
+
+    struct {
+        UniformData data;
+        bool dirty;
+    } uniform_block_data = {};
+
+    OGLPipeline pipeline;
+    OGLVertexArray sw_vao;
+    OGLVertexArray hw_vao;
+    std::array<bool, 16> hw_vao_enabled_attributes;
+
+    std::array<SamplerInfo, 3> texture_samplers;
+    static constexpr size_t VERTEX_BUFFER_SIZE = 128 * 1024 * 1024;
+    std::unique_ptr<OGLStreamBuffer> vertex_buffer;
+    OGLBuffer uniform_buffer;
+    OGLFramebuffer framebuffer;
+
+    static constexpr size_t STREAM_BUFFER_SIZE = 4 * 1024 * 1024;
+    std::unique_ptr<OGLStreamBuffer> stream_buffer;
+
+    GLint vs_input_index_min;
+    GLint vs_input_index_max;
+    GLsizeiptr vs_input_size;
+
+    void AnalyzeVertexArray(bool is_indexed);
+    void SetupVertexArray(u8* array_ptr, GLintptr buffer_offset);
+
+    OGLBuffer vs_uniform_buffer;
+    std::unordered_map<GLShader::MaxwellVSConfig, VertexShader*> vs_shader_map;
+    std::unordered_map<std::string, VertexShader> vs_shader_cache;
+    OGLShader vs_default_shader;
+
+    void SetupVertexShader(VSUniformData* ub_ptr, GLintptr buffer_offset);
+
+    OGLBuffer fs_uniform_buffer;
+    std::unordered_map<GLShader::MaxwellFSConfig, FragmentShader*> fs_shader_map;
+    std::unordered_map<std::string, FragmentShader> fs_shader_cache;
+    OGLShader fs_default_shader;
+
+    void SetupFragmentShader(FSUniformData* ub_ptr, GLintptr buffer_offset);
+
+    enum class AccelDraw { Disabled, Arrays, Indexed };
+    AccelDraw accelerate_draw;
+};

src/video_core/renderer_opengl/gl_rasterizer_cache.cpp

@@ -0,0 +1,1361 @@
+// Copyright 2015 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <atomic>
+#include <cstring>
+#include <iterator>
+#include <memory>
+#include <unordered_set>
+#include <utility>
+#include <vector>
+#include <boost/optional.hpp>
+#include <boost/range/iterator_range.hpp>
+#include <glad/glad.h>
+#include "common/alignment.h"
+#include "common/bit_field.h"
+#include "common/color.h"
+#include "common/logging/log.h"
+#include "common/math_util.h"
+#include "common/microprofile.h"
+#include "common/scope_exit.h"
+#include "common/vector_math.h"
+#include "core/frontend/emu_window.h"
+#include "core/memory.h"
+#include "core/settings.h"
+#include "video_core/renderer_opengl/gl_rasterizer_cache.h"
+#include "video_core/renderer_opengl/gl_state.h"
+#include "video_core/utils.h"
+#include "video_core/video_core.h"
+
+using SurfaceType = SurfaceParams::SurfaceType;
+using PixelFormat = SurfaceParams::PixelFormat;
+
+struct FormatTuple {
+    GLint internal_format;
+    GLenum format;
+    GLenum type;
+};
+
+static constexpr std::array<FormatTuple, 5> fb_format_tuples = {{
+    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8},     // RGBA8
+    {GL_RGB8, GL_BGR, GL_UNSIGNED_BYTE},              // RGB8
+    {GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_5_5_5_1}, // RGB5A1
+    {GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5},     // RGB565
+    {GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4},   // RGBA4
+}};
+
+static constexpr std::array<FormatTuple, 4> depth_format_tuples = {{
+    {GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // D16
+    {},
+    {GL_DEPTH_COMPONENT24, GL_DEPTH_COMPONENT, GL_UNSIGNED_INT},   // D24
+    {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // D24S8
+}};
+
+static constexpr FormatTuple tex_tuple = {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE};
+
+static const FormatTuple& GetFormatTuple(PixelFormat pixel_format) {
+    const SurfaceType type = SurfaceParams::GetFormatType(pixel_format);
+    if (type == SurfaceType::Color) {
+        ASSERT(static_cast<size_t>(pixel_format) < fb_format_tuples.size());
+        return fb_format_tuples[static_cast<unsigned int>(pixel_format)];
+    } else if (type == SurfaceType::Depth || type == SurfaceType::DepthStencil) {
+        size_t tuple_idx = static_cast<size_t>(pixel_format) - 14;
+        ASSERT(tuple_idx < depth_format_tuples.size());
+        return depth_format_tuples[tuple_idx];
+    }
+    return tex_tuple;
+}
+
+template <typename Map, typename Interval>
+constexpr auto RangeFromInterval(Map& map, const Interval& interval) {
+    return boost::make_iterator_range(map.equal_range(interval));
+}
+
+static u16 GetResolutionScaleFactor() {
+    return static_cast<u16>(!Settings::values.resolution_factor
+                                ? VideoCore::g_emu_window->GetFramebufferLayout().GetScalingRatio()
+                                : Settings::values.resolution_factor);
+}
+
+template <bool morton_to_gl, PixelFormat format>
+static void MortonCopyTile(u32 stride, u8* tile_buffer, u8* gl_buffer) {
+    constexpr u32 bytes_per_pixel = SurfaceParams::GetFormatBpp(format) / 8;
+    constexpr u32 gl_bytes_per_pixel = CachedSurface::GetGLBytesPerPixel(format);
+    for (u32 y = 0; y < 8; ++y) {
+        for (u32 x = 0; x < 8; ++x) {
+            u8* tile_ptr = tile_buffer + VideoCore::MortonInterleave(x, y) * bytes_per_pixel;
+            u8* gl_ptr = gl_buffer + ((7 - y) * stride + x) * gl_bytes_per_pixel;
+            if (morton_to_gl) {
+                if (format == PixelFormat::D24S8) {
+                    gl_ptr[0] = tile_ptr[3];
+                    std::memcpy(gl_ptr + 1, tile_ptr, 3);
+                } else {
+                    std::memcpy(gl_ptr, tile_ptr, bytes_per_pixel);
+                }
+            } else {
+                if (format == PixelFormat::D24S8) {
+                    std::memcpy(tile_ptr, gl_ptr + 1, 3);
+                    tile_ptr[3] = gl_ptr[0];
+                } else {
+                    std::memcpy(tile_ptr, gl_ptr, bytes_per_pixel);
+                }
+            }
+        }
+    }
+}
+
+template <bool morton_to_gl, PixelFormat format>
+static void MortonCopy(u32 stride, u32 height, u8* gl_buffer, PAddr base, PAddr start, PAddr end) {
+    constexpr u32 bytes_per_pixel = SurfaceParams::GetFormatBpp(format) / 8;
+    constexpr u32 tile_size = bytes_per_pixel * 64;
+
+    constexpr u32 gl_bytes_per_pixel = CachedSurface::GetGLBytesPerPixel(format);
+    static_assert(gl_bytes_per_pixel >= bytes_per_pixel, "");
+    gl_buffer += gl_bytes_per_pixel - bytes_per_pixel;
+
+    const PAddr aligned_down_start = base + Common::AlignDown(start - base, tile_size);
+    const PAddr aligned_start = base + Common::AlignUp(start - base, tile_size);
+    const PAddr aligned_end = base + Common::AlignDown(end - base, tile_size);
+
+    ASSERT(!morton_to_gl || (aligned_start == start && aligned_end == end));
+
+    const u64 begin_pixel_index = (aligned_down_start - base) / bytes_per_pixel;
+    u32 x = static_cast<u32>((begin_pixel_index % (stride * 8)) / 8);
+    u32 y = static_cast<u32>((begin_pixel_index / (stride * 8)) * 8);
+
+    gl_buffer += ((height - 8 - y) * stride + x) * gl_bytes_per_pixel;
+
+    auto glbuf_next_tile = [&] {
+        x = (x + 8) % stride;
+        gl_buffer += 8 * gl_bytes_per_pixel;
+        if (!x) {
+            y += 8;
+            gl_buffer -= stride * 9 * gl_bytes_per_pixel;
+        }
+    };
+
+    u8* tile_buffer = Memory::GetPhysicalPointer(start);
+
+    if (start < aligned_start && !morton_to_gl) {
+        std::array<u8, tile_size> tmp_buf;
+        MortonCopyTile<morton_to_gl, format>(stride, &tmp_buf[0], gl_buffer);
+        std::memcpy(tile_buffer, &tmp_buf[start - aligned_down_start],
+                    std::min(aligned_start, end) - start);
+
+        tile_buffer += aligned_start - start;
+        glbuf_next_tile();
+    }
+
+    const u8* const buffer_end = tile_buffer + aligned_end - aligned_start;
+    while (tile_buffer < buffer_end) {
+        MortonCopyTile<morton_to_gl, format>(stride, tile_buffer, gl_buffer);
+        tile_buffer += tile_size;
+        glbuf_next_tile();
+    }
+
+    if (end > std::max(aligned_start, aligned_end) && !morton_to_gl) {
+        std::array<u8, tile_size> tmp_buf;
+        MortonCopyTile<morton_to_gl, format>(stride, &tmp_buf[0], gl_buffer);
+        std::memcpy(tile_buffer, &tmp_buf[0], end - aligned_end);
+    }
+}
+
+static constexpr std::array<void (*)(u32, u32, u8*, PAddr, PAddr, PAddr), 18> morton_to_gl_fns = {
+    MortonCopy<true, PixelFormat::RGBA8>,  // 0
+    MortonCopy<true, PixelFormat::RGB8>,   // 1
+    MortonCopy<true, PixelFormat::RGB5A1>, // 2
+    MortonCopy<true, PixelFormat::RGB565>, // 3
+    MortonCopy<true, PixelFormat::RGBA4>,  // 4
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,                             // 5 - 13
+    MortonCopy<true, PixelFormat::D16>,  // 14
+    nullptr,                             // 15
+    MortonCopy<true, PixelFormat::D24>,  // 16
+    MortonCopy<true, PixelFormat::D24S8> // 17
+};
+
+static constexpr std::array<void (*)(u32, u32, u8*, PAddr, PAddr, PAddr), 18> gl_to_morton_fns = {
+    MortonCopy<false, PixelFormat::RGBA8>,  // 0
+    MortonCopy<false, PixelFormat::RGB8>,   // 1
+    MortonCopy<false, PixelFormat::RGB5A1>, // 2
+    MortonCopy<false, PixelFormat::RGB565>, // 3
+    MortonCopy<false, PixelFormat::RGBA4>,  // 4
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,                              // 5 - 13
+    MortonCopy<false, PixelFormat::D16>,  // 14
+    nullptr,                              // 15
+    MortonCopy<false, PixelFormat::D24>,  // 16
+    MortonCopy<false, PixelFormat::D24S8> // 17
+};
+
+// Allocate an uninitialized texture of appropriate size and format for the surface
+static void AllocateSurfaceTexture(GLuint texture, const FormatTuple& format_tuple, u32 width,
+                                   u32 height) {
+    OpenGLState cur_state = OpenGLState::GetCurState();
+
+    // Keep track of previous texture bindings
+    GLuint old_tex = cur_state.texture_units[0].texture_2d;
+    cur_state.texture_units[0].texture_2d = texture;
+    cur_state.Apply();
+    glActiveTexture(GL_TEXTURE0);
+
+    glTexImage2D(GL_TEXTURE_2D, 0, format_tuple.internal_format, width, height, 0,
+                 format_tuple.format, format_tuple.type, nullptr);
+
+    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, 0);
+    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
+
+    // Restore previous texture bindings
+    cur_state.texture_units[0].texture_2d = old_tex;
+    cur_state.Apply();
+}
+
+static bool BlitTextures(GLuint src_tex, const MathUtil::Rectangle<u32>& src_rect, GLuint dst_tex,
+                         const MathUtil::Rectangle<u32>& dst_rect, SurfaceType type,
+                         GLuint read_fb_handle, GLuint draw_fb_handle) {
+    OpenGLState state = OpenGLState::GetCurState();
+
+    OpenGLState prev_state = state;
+    SCOPE_EXIT({ prev_state.Apply(); });
+
+    // Make sure textures aren't bound to texture units, since going to bind them to framebuffer
+    // components
+    state.ResetTexture(src_tex);
+    state.ResetTexture(dst_tex);
+
+    state.draw.read_framebuffer = read_fb_handle;
+    state.draw.draw_framebuffer = draw_fb_handle;
+    state.Apply();
+
+    u32 buffers = 0;
+
+    if (type == SurfaceType::Color || type == SurfaceType::Texture) {
+        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, src_tex,
+                               0);
+        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
+                               0);
+
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, dst_tex,
+                               0);
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
+                               0);
+
+        buffers = GL_COLOR_BUFFER_BIT;
+    } else if (type == SurfaceType::Depth) {
+        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
+        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, src_tex, 0);
+        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
+
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, dst_tex, 0);
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
+
+        buffers = GL_DEPTH_BUFFER_BIT;
+    } else if (type == SurfaceType::DepthStencil) {
+        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
+        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
+                               src_tex, 0);
+
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
+                               dst_tex, 0);
+
+        buffers = GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT;
+    }
+
+    glBlitFramebuffer(src_rect.left, src_rect.bottom, src_rect.right, src_rect.top, dst_rect.left,
+                      dst_rect.bottom, dst_rect.right, dst_rect.top, buffers,
+                      buffers == GL_COLOR_BUFFER_BIT ? GL_LINEAR : GL_NEAREST);
+
+    return true;
+}
+
+static bool FillSurface(const Surface& surface, const u8* fill_data,
+                        const MathUtil::Rectangle<u32>& fill_rect, GLuint draw_fb_handle) {
+    UNIMPLEMENTED();
+    return true;
+}
+
+SurfaceParams SurfaceParams::FromInterval(SurfaceInterval interval) const {
+    SurfaceParams params = *this;
+    const u32 tiled_size = is_tiled ? 8 : 1;
+    const u64 stride_tiled_bytes = BytesInPixels(stride * tiled_size);
+    PAddr aligned_start =
+        addr + Common::AlignDown(boost::icl::first(interval) - addr, stride_tiled_bytes);
+    PAddr aligned_end =
+        addr + Common::AlignUp(boost::icl::last_next(interval) - addr, stride_tiled_bytes);
+
+    if (aligned_end - aligned_start > stride_tiled_bytes) {
+        params.addr = aligned_start;
+        params.height = static_cast<u32>((aligned_end - aligned_start) / BytesInPixels(stride));
+    } else {
+        // 1 row
+        ASSERT(aligned_end - aligned_start == stride_tiled_bytes);
+        const u64 tiled_alignment = BytesInPixels(is_tiled ? 8 * 8 : 1);
+        aligned_start =
+            addr + Common::AlignDown(boost::icl::first(interval) - addr, tiled_alignment);
+        aligned_end =
+            addr + Common::AlignUp(boost::icl::last_next(interval) - addr, tiled_alignment);
+        params.addr = aligned_start;
+        params.width = static_cast<u32>(PixelsInBytes(aligned_end - aligned_start) / tiled_size);
+        params.stride = params.width;
+        params.height = tiled_size;
+    }
+    params.UpdateParams();
+
+    return params;
+}
+
+SurfaceInterval SurfaceParams::GetSubRectInterval(MathUtil::Rectangle<u32> unscaled_rect) const {
+    if (unscaled_rect.GetHeight() == 0 || unscaled_rect.GetWidth() == 0) {
+        return {};
+    }
+
+    if (is_tiled) {
+        unscaled_rect.left = Common::AlignDown(unscaled_rect.left, 8) * 8;
+        unscaled_rect.bottom = Common::AlignDown(unscaled_rect.bottom, 8) / 8;
+        unscaled_rect.right = Common::AlignUp(unscaled_rect.right, 8) * 8;
+        unscaled_rect.top = Common::AlignUp(unscaled_rect.top, 8) / 8;
+    }
+
+    const u32 stride_tiled = !is_tiled ? stride : stride * 8;
+
+    const u32 pixel_offset =
+        stride_tiled * (!is_tiled ? unscaled_rect.bottom : (height / 8) - unscaled_rect.top) +
+        unscaled_rect.left;
+
+    const u32 pixels = (unscaled_rect.GetHeight() - 1) * stride_tiled + unscaled_rect.GetWidth();
+
+    return {addr + BytesInPixels(pixel_offset), addr + BytesInPixels(pixel_offset + pixels)};
+}
+
+MathUtil::Rectangle<u32> SurfaceParams::GetSubRect(const SurfaceParams& sub_surface) const {
+    const u32 begin_pixel_index = static_cast<u32>(PixelsInBytes(sub_surface.addr - addr));
+
+    if (is_tiled) {
+        const int x0 = (begin_pixel_index % (stride * 8)) / 8;
+        const int y0 = (begin_pixel_index / (stride * 8)) * 8;
+        // Top to bottom
+        return MathUtil::Rectangle<u32>(x0, height - y0, x0 + sub_surface.width,
+                                        height - (y0 + sub_surface.height));
+    }
+
+    const int x0 = begin_pixel_index % stride;
+    const int y0 = begin_pixel_index / stride;
+    // Bottom to top
+    return MathUtil::Rectangle<u32>(x0, y0 + sub_surface.height, x0 + sub_surface.width, y0);
+}
+
+MathUtil::Rectangle<u32> SurfaceParams::GetScaledSubRect(const SurfaceParams& sub_surface) const {
+    auto rect = GetSubRect(sub_surface);
+    rect.left = rect.left * res_scale;
+    rect.right = rect.right * res_scale;
+    rect.top = rect.top * res_scale;
+    rect.bottom = rect.bottom * res_scale;
+    return rect;
+}
+
+bool SurfaceParams::ExactMatch(const SurfaceParams& other_surface) const {
+    return std::tie(other_surface.addr, other_surface.width, other_surface.height,
+                    other_surface.stride, other_surface.pixel_format, other_surface.is_tiled) ==
+               std::tie(addr, width, height, stride, pixel_format, is_tiled) &&
+           pixel_format != PixelFormat::Invalid;
+}
+
+bool SurfaceParams::CanSubRect(const SurfaceParams& sub_surface) const {
+    return sub_surface.addr >= addr && sub_surface.end <= end &&
+           sub_surface.pixel_format == pixel_format && pixel_format != PixelFormat::Invalid &&
+           sub_surface.is_tiled == is_tiled &&
+           (sub_surface.addr - addr) % BytesInPixels(is_tiled ? 64 : 1) == 0 &&
+           (sub_surface.stride == stride || sub_surface.height <= (is_tiled ? 8u : 1u)) &&
+           GetSubRect(sub_surface).left + sub_surface.width <= stride;
+}
+
+bool SurfaceParams::CanExpand(const SurfaceParams& expanded_surface) const {
+    return pixel_format != PixelFormat::Invalid && pixel_format == expanded_surface.pixel_format &&
+           addr <= expanded_surface.end && expanded_surface.addr <= end &&
+           is_tiled == expanded_surface.is_tiled && stride == expanded_surface.stride &&
+           (std::max(expanded_surface.addr, addr) - std::min(expanded_surface.addr, addr)) %
+                   BytesInPixels(stride * (is_tiled ? 8 : 1)) ==
+               0;
+}
+
+bool SurfaceParams::CanTexCopy(const SurfaceParams& texcopy_params) const {
+    if (pixel_format == PixelFormat::Invalid || addr > texcopy_params.addr ||
+        end < texcopy_params.end) {
+        return false;
+    }
+    if (texcopy_params.width != texcopy_params.stride) {
+        const u32 tile_stride = static_cast<u32>(BytesInPixels(stride * (is_tiled ? 8 : 1)));
+        return (texcopy_params.addr - addr) % BytesInPixels(is_tiled ? 64 : 1) == 0 &&
+               texcopy_params.width % BytesInPixels(is_tiled ? 64 : 1) == 0 &&
+               (texcopy_params.height == 1 || texcopy_params.stride == tile_stride) &&
+               ((texcopy_params.addr - addr) % tile_stride) + texcopy_params.width <= tile_stride;
+    }
+    return FromInterval(texcopy_params.GetInterval()).GetInterval() == texcopy_params.GetInterval();
+}
+
+bool CachedSurface::CanFill(const SurfaceParams& dest_surface,
+                            SurfaceInterval fill_interval) const {
+    if (type == SurfaceType::Fill && IsRegionValid(fill_interval) &&
+        boost::icl::first(fill_interval) >= addr &&
+        boost::icl::last_next(fill_interval) <= end && // dest_surface is within our fill range
+        dest_surface.FromInterval(fill_interval).GetInterval() ==
+            fill_interval) { // make sure interval is a rectangle in dest surface
+        if (fill_size * 8 != dest_surface.GetFormatBpp()) {
+            // Check if bits repeat for our fill_size
+            const u32 dest_bytes_per_pixel = std::max(dest_surface.GetFormatBpp() / 8, 1u);
+            std::vector<u8> fill_test(fill_size * dest_bytes_per_pixel);
+
+            for (u32 i = 0; i < dest_bytes_per_pixel; ++i)
+                std::memcpy(&fill_test[i * fill_size], &fill_data[0], fill_size);
+
+            for (u32 i = 0; i < fill_size; ++i)
+                if (std::memcmp(&fill_test[dest_bytes_per_pixel * i], &fill_test[0],
+                                dest_bytes_per_pixel) != 0)
+                    return false;
+
+            if (dest_surface.GetFormatBpp() == 4 && (fill_test[0] & 0xF) != (fill_test[0] >> 4))
+                return false;
+        }
+        return true;
+    }
+    return false;
+}
+
+bool CachedSurface::CanCopy(const SurfaceParams& dest_surface,
+                            SurfaceInterval copy_interval) const {
+    SurfaceParams subrect_params = dest_surface.FromInterval(copy_interval);
+    ASSERT(subrect_params.GetInterval() == copy_interval);
+    if (CanSubRect(subrect_params))
+        return true;
+
+    if (CanFill(dest_surface, copy_interval))
+        return true;
+
+    return false;
+}
+
+SurfaceInterval SurfaceParams::GetCopyableInterval(const Surface& src_surface) const {
+    SurfaceInterval result{};
+    const auto valid_regions =
+        SurfaceRegions(GetInterval() & src_surface->GetInterval()) - src_surface->invalid_regions;
+    for (auto& valid_interval : valid_regions) {
+        const SurfaceInterval aligned_interval{
+            addr + Common::AlignUp(boost::icl::first(valid_interval) - addr,
+                                   BytesInPixels(is_tiled ? 8 * 8 : 1)),
+            addr + Common::AlignDown(boost::icl::last_next(valid_interval) - addr,
+                                     BytesInPixels(is_tiled ? 8 * 8 : 1))};
+
+        if (BytesInPixels(is_tiled ? 8 * 8 : 1) > boost::icl::length(valid_interval) ||
+            boost::icl::length(aligned_interval) == 0) {
+            continue;
+        }
+
+        // Get the rectangle within aligned_interval
+        const u32 stride_bytes = static_cast<u32>(BytesInPixels(stride)) * (is_tiled ? 8 : 1);
+        SurfaceInterval rect_interval{
+            addr + Common::AlignUp(boost::icl::first(aligned_interval) - addr, stride_bytes),
+            addr + Common::AlignDown(boost::icl::last_next(aligned_interval) - addr, stride_bytes),
+        };
+        if (boost::icl::first(rect_interval) > boost::icl::last_next(rect_interval)) {
+            // 1 row
+            rect_interval = aligned_interval;
+        } else if (boost::icl::length(rect_interval) == 0) {
+            // 2 rows that do not make a rectangle, return the larger one
+            const SurfaceInterval row1{boost::icl::first(aligned_interval),
+                                       boost::icl::first(rect_interval)};
+            const SurfaceInterval row2{boost::icl::first(rect_interval),
+                                       boost::icl::last_next(aligned_interval)};
+            rect_interval = (boost::icl::length(row1) > boost::icl::length(row2)) ? row1 : row2;
+        }
+
+        if (boost::icl::length(rect_interval) > boost::icl::length(result)) {
+            result = rect_interval;
+        }
+    }
+    return result;
+}
+
+void RasterizerCacheOpenGL::CopySurface(const Surface& src_surface, const Surface& dst_surface,
+                                        SurfaceInterval copy_interval) {
+    SurfaceParams subrect_params = dst_surface->FromInterval(copy_interval);
+    ASSERT(subrect_params.GetInterval() == copy_interval);
+
+    ASSERT(src_surface != dst_surface);
+
+    // This is only called when CanCopy is true, no need to run checks here
+    if (src_surface->type == SurfaceType::Fill) {
+        // FillSurface needs a 4 bytes buffer
+        const u64 fill_offset =
+            (boost::icl::first(copy_interval) - src_surface->addr) % src_surface->fill_size;
+        std::array<u8, 4> fill_buffer;
+
+        u64 fill_buff_pos = fill_offset;
+        for (int i : {0, 1, 2, 3})
+            fill_buffer[i] = src_surface->fill_data[fill_buff_pos++ % src_surface->fill_size];
+
+        FillSurface(dst_surface, &fill_buffer[0], dst_surface->GetScaledSubRect(subrect_params),
+                    draw_framebuffer.handle);
+        return;
+    }
+    if (src_surface->CanSubRect(subrect_params)) {
+        BlitTextures(src_surface->texture.handle, src_surface->GetScaledSubRect(subrect_params),
+                     dst_surface->texture.handle, dst_surface->GetScaledSubRect(subrect_params),
+                     src_surface->type, read_framebuffer.handle, draw_framebuffer.handle);
+        return;
+    }
+    UNREACHABLE();
+}
+
+MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 64, 192));
+void CachedSurface::LoadGLBuffer(PAddr load_start, PAddr load_end) {
+    ASSERT(type != SurfaceType::Fill);
+
+    const u8* const texture_src_data = Memory::GetPhysicalPointer(addr);
+    if (texture_src_data == nullptr)
+        return;
+
+    if (gl_buffer == nullptr) {
+        gl_buffer_size = width * height * GetGLBytesPerPixel(pixel_format);
+        gl_buffer.reset(new u8[gl_buffer_size]);
+    }
+
+    // TODO: Should probably be done in ::Memory:: and check for other regions too
+    if (load_start < Memory::VRAM_VADDR_END && load_end > Memory::VRAM_VADDR_END)
+        load_end = Memory::VRAM_VADDR_END;
+
+    if (load_start < Memory::VRAM_VADDR && load_end > Memory::VRAM_VADDR)
+        load_start = Memory::VRAM_VADDR;
+
+    MICROPROFILE_SCOPE(OpenGL_SurfaceLoad);
+
+    ASSERT(load_start >= addr && load_end <= end);
+    const u32 start_offset = load_start - addr;
+
+    if (!is_tiled) {
+        ASSERT(type == SurfaceType::Color);
+        std::memcpy(&gl_buffer[start_offset], texture_src_data + start_offset,
+                    load_end - load_start);
+    } else {
+        if (type == SurfaceType::Texture) {
+            UNIMPLEMENTED();
+        } else {
+            morton_to_gl_fns[static_cast<size_t>(pixel_format)](stride, height, &gl_buffer[0], addr,
+                                                                load_start, load_end);
+        }
+    }
+}
+
+MICROPROFILE_DEFINE(OpenGL_SurfaceFlush, "OpenGL", "Surface Flush", MP_RGB(128, 192, 64));
+void CachedSurface::FlushGLBuffer(PAddr flush_start, PAddr flush_end) {
+    u8* const dst_buffer = Memory::GetPhysicalPointer(addr);
+    if (dst_buffer == nullptr)
+        return;
+
+    ASSERT(gl_buffer_size == width * height * GetGLBytesPerPixel(pixel_format));
+
+    // TODO: Should probably be done in ::Memory:: and check for other regions too
+    // same as loadglbuffer()
+    if (flush_start < Memory::VRAM_VADDR_END && flush_end > Memory::VRAM_VADDR_END)
+        flush_end = Memory::VRAM_VADDR_END;
+
+    if (flush_start < Memory::VRAM_VADDR && flush_end > Memory::VRAM_VADDR)
+        flush_start = Memory::VRAM_VADDR;
+
+    MICROPROFILE_SCOPE(OpenGL_SurfaceFlush);
+
+    ASSERT(flush_start >= addr && flush_end <= end);
+    const u64 start_offset = flush_start - addr;
+    const u64 end_offset = flush_end - addr;
+
+    if (type == SurfaceType::Fill) {
+        const u64 coarse_start_offset = start_offset - (start_offset % fill_size);
+        const u64 backup_bytes = start_offset % fill_size;
+        std::array<u8, 4> backup_data;
+        if (backup_bytes)
+            std::memcpy(&backup_data[0], &dst_buffer[coarse_start_offset], backup_bytes);
+
+        for (u64 offset = coarse_start_offset; offset < end_offset; offset += fill_size) {
+            std::memcpy(&dst_buffer[offset], &fill_data[0],
+                        std::min(fill_size, end_offset - offset));
+        }
+
+        if (backup_bytes)
+            std::memcpy(&dst_buffer[coarse_start_offset], &backup_data[0], backup_bytes);
+    } else if (!is_tiled) {
+        ASSERT(type == SurfaceType::Color);
+        std::memcpy(dst_buffer + start_offset, &gl_buffer[start_offset], flush_end - flush_start);
+    } else {
+        gl_to_morton_fns[static_cast<size_t>(pixel_format)](stride, height, &gl_buffer[0], addr,
+                                                            flush_start, flush_end);
+    }
+}
+
+MICROPROFILE_DEFINE(OpenGL_TextureUL, "OpenGL", "Texture Upload", MP_RGB(128, 64, 192));
+void CachedSurface::UploadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint read_fb_handle,
+                                    GLuint draw_fb_handle) {
+    if (type == SurfaceType::Fill)
+        return;
+
+    MICROPROFILE_SCOPE(OpenGL_TextureUL);
+
+    ASSERT(gl_buffer_size == width * height * GetGLBytesPerPixel(pixel_format));
+
+    // Load data from memory to the surface
+    GLint x0 = static_cast<GLint>(rect.left);
+    GLint y0 = static_cast<GLint>(rect.bottom);
+    size_t buffer_offset = (y0 * stride + x0) * GetGLBytesPerPixel(pixel_format);
+
+    const FormatTuple& tuple = GetFormatTuple(pixel_format);
+    GLuint target_tex = texture.handle;
+
+    // If not 1x scale, create 1x texture that we will blit from to replace texture subrect in
+    // surface
+    OGLTexture unscaled_tex;
+    if (res_scale != 1) {
+        x0 = 0;
+        y0 = 0;
+
+        unscaled_tex.Create();
+        AllocateSurfaceTexture(unscaled_tex.handle, tuple, rect.GetWidth(), rect.GetHeight());
+        target_tex = unscaled_tex.handle;
+    }
+
+    OpenGLState cur_state = OpenGLState::GetCurState();
+
+    GLuint old_tex = cur_state.texture_units[0].texture_2d;
+    cur_state.texture_units[0].texture_2d = target_tex;
+    cur_state.Apply();
+
+    // Ensure no bad interactions with GL_UNPACK_ALIGNMENT
+    ASSERT(stride * GetGLBytesPerPixel(pixel_format) % 4 == 0);
+    glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast<GLint>(stride));
+
+    glActiveTexture(GL_TEXTURE0);
+    glTexSubImage2D(GL_TEXTURE_2D, 0, x0, y0, static_cast<GLsizei>(rect.GetWidth()),
+                    static_cast<GLsizei>(rect.GetHeight()), tuple.format, tuple.type,
+                    &gl_buffer[buffer_offset]);
+
+    glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
+
+    cur_state.texture_units[0].texture_2d = old_tex;
+    cur_state.Apply();
+
+    if (res_scale != 1) {
+        auto scaled_rect = rect;
+        scaled_rect.left *= res_scale;
+        scaled_rect.top *= res_scale;
+        scaled_rect.right *= res_scale;
+        scaled_rect.bottom *= res_scale;
+
+        BlitTextures(unscaled_tex.handle, {0, rect.GetHeight(), rect.GetWidth(), 0}, texture.handle,
+                     scaled_rect, type, read_fb_handle, draw_fb_handle);
+    }
+}
+
+MICROPROFILE_DEFINE(OpenGL_TextureDL, "OpenGL", "Texture Download", MP_RGB(128, 192, 64));
+void CachedSurface::DownloadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint read_fb_handle,
+                                      GLuint draw_fb_handle) {
+    if (type == SurfaceType::Fill)
+        return;
+
+    MICROPROFILE_SCOPE(OpenGL_TextureDL);
+
+    if (gl_buffer == nullptr) {
+        gl_buffer_size = width * height * GetGLBytesPerPixel(pixel_format);
+        gl_buffer.reset(new u8[gl_buffer_size]);
+    }
+
+    OpenGLState state = OpenGLState::GetCurState();
+    OpenGLState prev_state = state;
+    SCOPE_EXIT({ prev_state.Apply(); });
+
+    const FormatTuple& tuple = GetFormatTuple(pixel_format);
+
+    // Ensure no bad interactions with GL_PACK_ALIGNMENT
+    ASSERT(stride * GetGLBytesPerPixel(pixel_format) % 4 == 0);
+    glPixelStorei(GL_PACK_ROW_LENGTH, static_cast<GLint>(stride));
+    size_t buffer_offset = (rect.bottom * stride + rect.left) * GetGLBytesPerPixel(pixel_format);
+
+    // If not 1x scale, blit scaled texture to a new 1x texture and use that to flush
+    if (res_scale != 1) {
+        auto scaled_rect = rect;
+        scaled_rect.left *= res_scale;
+        scaled_rect.top *= res_scale;
+        scaled_rect.right *= res_scale;
+        scaled_rect.bottom *= res_scale;
+
+        OGLTexture unscaled_tex;
+        unscaled_tex.Create();
+
+        MathUtil::Rectangle<u32> unscaled_tex_rect{0, rect.GetHeight(), rect.GetWidth(), 0};
+        AllocateSurfaceTexture(unscaled_tex.handle, tuple, rect.GetWidth(), rect.GetHeight());
+        BlitTextures(texture.handle, scaled_rect, unscaled_tex.handle, unscaled_tex_rect, type,
+                     read_fb_handle, draw_fb_handle);
+
+        state.texture_units[0].texture_2d = unscaled_tex.handle;
+        state.Apply();
+
+        glActiveTexture(GL_TEXTURE0);
+        glGetTexImage(GL_TEXTURE_2D, 0, tuple.format, tuple.type, &gl_buffer[buffer_offset]);
+    } else {
+        state.ResetTexture(texture.handle);
+        state.draw.read_framebuffer = read_fb_handle;
+        state.Apply();
+
+        if (type == SurfaceType::Color || type == SurfaceType::Texture) {
+            glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D,
+                                   texture.handle, 0);
+            glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
+                                   0, 0);
+        } else if (type == SurfaceType::Depth) {
+            glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
+            glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D,
+                                   texture.handle, 0);
+            glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
+        } else {
+            glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
+            glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
+                                   texture.handle, 0);
+        }
+        glReadPixels(static_cast<GLint>(rect.left), static_cast<GLint>(rect.bottom),
+                     static_cast<GLsizei>(rect.GetWidth()), static_cast<GLsizei>(rect.GetHeight()),
+                     tuple.format, tuple.type, &gl_buffer[buffer_offset]);
+    }
+
+    glPixelStorei(GL_PACK_ROW_LENGTH, 0);
+}
+
+enum MatchFlags {
+    Invalid = 1,      // Flag that can be applied to other match types, invalid matches require
+                      // validation before they can be used
+    Exact = 1 << 1,   // Surfaces perfectly match
+    SubRect = 1 << 2, // Surface encompasses params
+    Copy = 1 << 3,    // Surface we can copy from
+    Expand = 1 << 4,  // Surface that can expand params
+    TexCopy = 1 << 5  // Surface that will match a display transfer "texture copy" parameters
+};
+
+constexpr MatchFlags operator|(MatchFlags lhs, MatchFlags rhs) {
+    return static_cast<MatchFlags>(static_cast<int>(lhs) | static_cast<int>(rhs));
+}
+
+/// Get the best surface match (and its match type) for the given flags
+template <MatchFlags find_flags>
+Surface FindMatch(const SurfaceCache& surface_cache, const SurfaceParams& params,
+                  ScaleMatch match_scale_type,
+                  boost::optional<SurfaceInterval> validate_interval = boost::none) {
+    Surface match_surface = nullptr;
+    bool match_valid = false;
+    u32 match_scale = 0;
+    SurfaceInterval match_interval{};
+
+    for (auto& pair : RangeFromInterval(surface_cache, params.GetInterval())) {
+        for (auto& surface : pair.second) {
+            bool res_scale_matched = match_scale_type == ScaleMatch::Exact
+                                         ? (params.res_scale == surface->res_scale)
+                                         : (params.res_scale <= surface->res_scale);
+            // validity will be checked in GetCopyableInterval
+            bool is_valid =
+                find_flags & MatchFlags::Copy
+                    ? true
+                    : surface->IsRegionValid(validate_interval.value_or(params.GetInterval()));
+
+            if (!(find_flags & MatchFlags::Invalid) && !is_valid)
+                continue;
+
+            auto IsMatch_Helper = [&](auto check_type, auto match_fn) {
+                if (!(find_flags & check_type))
+                    return;
+
+                bool matched;
+                SurfaceInterval surface_interval;
+                std::tie(matched, surface_interval) = match_fn();
+                if (!matched)
+                    return;
+
+                if (!res_scale_matched && match_scale_type != ScaleMatch::Ignore &&
+                    surface->type != SurfaceType::Fill)
+                    return;
+
+                // Found a match, update only if this is better than the previous one
+                auto UpdateMatch = [&] {
+                    match_surface = surface;
+                    match_valid = is_valid;
+                    match_scale = surface->res_scale;
+                    match_interval = surface_interval;
+                };
+
+                if (surface->res_scale > match_scale) {
+                    UpdateMatch();
+                    return;
+                } else if (surface->res_scale < match_scale) {
+                    return;
+                }
+
+                if (is_valid && !match_valid) {
+                    UpdateMatch();
+                    return;
+                } else if (is_valid != match_valid) {
+                    return;
+                }
+
+                if (boost::icl::length(surface_interval) > boost::icl::length(match_interval)) {
+                    UpdateMatch();
+                }
+            };
+            IsMatch_Helper(std::integral_constant<MatchFlags, MatchFlags::Exact>{}, [&] {
+                return std::make_pair(surface->ExactMatch(params), surface->GetInterval());
+            });
+            IsMatch_Helper(std::integral_constant<MatchFlags, MatchFlags::SubRect>{}, [&] {
+                return std::make_pair(surface->CanSubRect(params), surface->GetInterval());
+            });
+            IsMatch_Helper(std::integral_constant<MatchFlags, MatchFlags::Copy>{}, [&] {
+                auto copy_interval =
+                    params.FromInterval(*validate_interval).GetCopyableInterval(surface);
+                bool matched = boost::icl::length(copy_interval & *validate_interval) != 0 &&
+                               surface->CanCopy(params, copy_interval);
+                return std::make_pair(matched, copy_interval);
+            });
+            IsMatch_Helper(std::integral_constant<MatchFlags, MatchFlags::Expand>{}, [&] {
+                return std::make_pair(surface->CanExpand(params), surface->GetInterval());
+            });
+            IsMatch_Helper(std::integral_constant<MatchFlags, MatchFlags::TexCopy>{}, [&] {
+                return std::make_pair(surface->CanTexCopy(params), surface->GetInterval());
+            });
+        }
+    }
+    return match_surface;
+}
+
+RasterizerCacheOpenGL::RasterizerCacheOpenGL() {
+    read_framebuffer.Create();
+    draw_framebuffer.Create();
+
+    attributeless_vao.Create();
+
+    d24s8_abgr_buffer.Create();
+    d24s8_abgr_buffer_size = 0;
+
+    const char* vs_source = R"(
+#version 330 core
+const vec2 vertices[4] = vec2[4](vec2(-1.0, -1.0), vec2(1.0, -1.0), vec2(-1.0, 1.0), vec2(1.0, 1.0));
+void main() {
+    gl_Position = vec4(vertices[gl_VertexID], 0.0, 1.0);
+}
+)";
+    const char* fs_source = R"(
+#version 330 core
+
+uniform samplerBuffer tbo;
+uniform vec2 tbo_size;
+uniform vec4 viewport;
+
+out vec4 color;
+
+void main() {
+    vec2 tbo_coord = (gl_FragCoord.xy - viewport.xy) * tbo_size / viewport.zw;
+    int tbo_offset = int(tbo_coord.y) * int(tbo_size.x) + int(tbo_coord.x);
+    color = texelFetch(tbo, tbo_offset).rabg;
+}
+)";
+    d24s8_abgr_shader.Create(vs_source, nullptr, fs_source);
+
+    OpenGLState state = OpenGLState::GetCurState();
+    GLuint old_program = state.draw.shader_program;
+    state.draw.shader_program = d24s8_abgr_shader.handle;
+    state.Apply();
+
+    GLint tbo_u_id = glGetUniformLocation(d24s8_abgr_shader.handle, "tbo");
+    ASSERT(tbo_u_id != -1);
+    glUniform1i(tbo_u_id, 0);
+
+    state.draw.shader_program = old_program;
+    state.Apply();
+
+    d24s8_abgr_tbo_size_u_id = glGetUniformLocation(d24s8_abgr_shader.handle, "tbo_size");
+    ASSERT(d24s8_abgr_tbo_size_u_id != -1);
+    d24s8_abgr_viewport_u_id = glGetUniformLocation(d24s8_abgr_shader.handle, "viewport");
+    ASSERT(d24s8_abgr_viewport_u_id != -1);
+}
+
+RasterizerCacheOpenGL::~RasterizerCacheOpenGL() {
+    FlushAll();
+    while (!surface_cache.empty())
+        UnregisterSurface(*surface_cache.begin()->second.begin());
+}
+
+bool RasterizerCacheOpenGL::BlitSurfaces(const Surface& src_surface,
+                                         const MathUtil::Rectangle<u32>& src_rect,
+                                         const Surface& dst_surface,
+                                         const MathUtil::Rectangle<u32>& dst_rect) {
+    if (!SurfaceParams::CheckFormatsBlittable(src_surface->pixel_format, dst_surface->pixel_format))
+        return false;
+
+    return BlitTextures(src_surface->texture.handle, src_rect, dst_surface->texture.handle,
+                        dst_rect, src_surface->type, read_framebuffer.handle,
+                        draw_framebuffer.handle);
+}
+
+void RasterizerCacheOpenGL::ConvertD24S8toABGR(GLuint src_tex,
+                                               const MathUtil::Rectangle<u32>& src_rect,
+                                               GLuint dst_tex,
+                                               const MathUtil::Rectangle<u32>& dst_rect) {
+    OpenGLState prev_state = OpenGLState::GetCurState();
+    SCOPE_EXIT({ prev_state.Apply(); });
+
+    OpenGLState state;
+    state.draw.read_framebuffer = read_framebuffer.handle;
+    state.draw.draw_framebuffer = draw_framebuffer.handle;
+    state.Apply();
+
+    glBindBuffer(GL_PIXEL_PACK_BUFFER, d24s8_abgr_buffer.handle);
+
+    GLsizeiptr target_pbo_size = src_rect.GetWidth() * src_rect.GetHeight() * 4;
+    if (target_pbo_size > d24s8_abgr_buffer_size) {
+        d24s8_abgr_buffer_size = target_pbo_size * 2;
+        glBufferData(GL_PIXEL_PACK_BUFFER, d24s8_abgr_buffer_size, nullptr, GL_STREAM_COPY);
+    }
+
+    glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
+    glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, src_tex,
+                           0);
+    glReadPixels(static_cast<GLint>(src_rect.left), static_cast<GLint>(src_rect.bottom),
+                 static_cast<GLsizei>(src_rect.GetWidth()),
+                 static_cast<GLsizei>(src_rect.GetHeight()), GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8,
+                 0);
+
+    glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
+
+    // PBO now contains src_tex in RABG format
+    state.draw.shader_program = d24s8_abgr_shader.handle;
+    state.draw.vertex_array = attributeless_vao.handle;
+    state.viewport.x = static_cast<GLint>(dst_rect.left);
+    state.viewport.y = static_cast<GLint>(dst_rect.bottom);
+    state.viewport.width = static_cast<GLsizei>(dst_rect.GetWidth());
+    state.viewport.height = static_cast<GLsizei>(dst_rect.GetHeight());
+    state.Apply();
+
+    OGLTexture tbo;
+    tbo.Create();
+    glActiveTexture(GL_TEXTURE0);
+    glBindTexture(GL_TEXTURE_BUFFER, tbo.handle);
+    glTexBuffer(GL_TEXTURE_BUFFER, GL_RGBA8, d24s8_abgr_buffer.handle);
+
+    glUniform2f(d24s8_abgr_tbo_size_u_id, static_cast<GLfloat>(src_rect.GetWidth()),
+                static_cast<GLfloat>(src_rect.GetHeight()));
+    glUniform4f(d24s8_abgr_viewport_u_id, static_cast<GLfloat>(state.viewport.x),
+                static_cast<GLfloat>(state.viewport.y), static_cast<GLfloat>(state.viewport.width),
+                static_cast<GLfloat>(state.viewport.height));
+
+    glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, dst_tex, 0);
+    glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
+    glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
+
+    glBindTexture(GL_TEXTURE_BUFFER, 0);
+}
+
+Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, ScaleMatch match_res_scale,
+                                          bool load_if_create) {
+    if (params.addr == 0 || params.height * params.width == 0) {
+        return nullptr;
+    }
+    // Use GetSurfaceSubRect instead
+    ASSERT(params.width == params.stride);
+
+    ASSERT(!params.is_tiled || (params.width % 8 == 0 && params.height % 8 == 0));
+
+    // Check for an exact match in existing surfaces
+    Surface surface =
+        FindMatch<MatchFlags::Exact | MatchFlags::Invalid>(surface_cache, params, match_res_scale);
+
+    if (surface == nullptr) {
+        u16 target_res_scale = params.res_scale;
+        if (match_res_scale != ScaleMatch::Exact) {
+            // This surface may have a subrect of another surface with a higher res_scale, find it
+            // to adjust our params
+            SurfaceParams find_params = params;
+            Surface expandable = FindMatch<MatchFlags::Expand | MatchFlags::Invalid>(
+                surface_cache, find_params, match_res_scale);
+            if (expandable != nullptr && expandable->res_scale > target_res_scale) {
+                target_res_scale = expandable->res_scale;
+            }
+            // Keep res_scale when reinterpreting d24s8 -> rgba8
+            if (params.pixel_format == PixelFormat::RGBA8) {
+                find_params.pixel_format = PixelFormat::D24S8;
+                expandable = FindMatch<MatchFlags::Expand | MatchFlags::Invalid>(
+                    surface_cache, find_params, match_res_scale);
+                if (expandable != nullptr && expandable->res_scale > target_res_scale) {
+                    target_res_scale = expandable->res_scale;
+                }
+            }
+        }
+        SurfaceParams new_params = params;
+        new_params.res_scale = target_res_scale;
+        surface = CreateSurface(new_params);
+        RegisterSurface(surface);
+    }
+
+    if (load_if_create) {
+        ValidateSurface(surface, params.addr, params.size);
+    }
+
+    return surface;
+}
+
+SurfaceRect_Tuple RasterizerCacheOpenGL::GetSurfaceSubRect(const SurfaceParams& params,
+                                                           ScaleMatch match_res_scale,
+                                                           bool load_if_create) {
+    if (params.addr == 0 || params.height * params.width == 0) {
+        return std::make_tuple(nullptr, MathUtil::Rectangle<u32>{});
+    }
+
+    // Attempt to find encompassing surface
+    Surface surface = FindMatch<MatchFlags::SubRect | MatchFlags::Invalid>(surface_cache, params,
+                                                                           match_res_scale);
+
+    // Check if FindMatch failed because of res scaling
+    // If that's the case create a new surface with
+    // the dimensions of the lower res_scale surface
+    // to suggest it should not be used again
+    if (surface == nullptr && match_res_scale != ScaleMatch::Ignore) {
+        surface = FindMatch<MatchFlags::SubRect | MatchFlags::Invalid>(surface_cache, params,
+                                                                       ScaleMatch::Ignore);
+        if (surface != nullptr) {
+            ASSERT(surface->res_scale < params.res_scale);
+            SurfaceParams new_params = *surface;
+            new_params.res_scale = params.res_scale;
+
+            surface = CreateSurface(new_params);
+            RegisterSurface(surface);
+        }
+    }
+
+    SurfaceParams aligned_params = params;
+    if (params.is_tiled) {
+        aligned_params.height = Common::AlignUp(params.height, 8);
+        aligned_params.width = Common::AlignUp(params.width, 8);
+        aligned_params.stride = Common::AlignUp(params.stride, 8);
+        aligned_params.UpdateParams();
+    }
+
+    // Check for a surface we can expand before creating a new one
+    if (surface == nullptr) {
+        surface = FindMatch<MatchFlags::Expand | MatchFlags::Invalid>(surface_cache, aligned_params,
+                                                                      match_res_scale);
+        if (surface != nullptr) {
+            aligned_params.width = aligned_params.stride;
+            aligned_params.UpdateParams();
+
+            SurfaceParams new_params = *surface;
+            new_params.addr = std::min(aligned_params.addr, surface->addr);
+            new_params.end = std::max(aligned_params.end, surface->end);
+            new_params.size = new_params.end - new_params.addr;
+            new_params.height = static_cast<u32>(
+                new_params.size / aligned_params.BytesInPixels(aligned_params.stride));
+            ASSERT(new_params.size % aligned_params.BytesInPixels(aligned_params.stride) == 0);
+
+            Surface new_surface = CreateSurface(new_params);
+            DuplicateSurface(surface, new_surface);
+
+            // Delete the expanded surface, this can't be done safely yet
+            // because it may still be in use
+            remove_surfaces.emplace(surface);
+
+            surface = new_surface;
+            RegisterSurface(new_surface);
+        }
+    }
+
+    // No subrect found - create and return a new surface
+    if (surface == nullptr) {
+        SurfaceParams new_params = aligned_params;
+        // Can't have gaps in a surface
+        new_params.width = aligned_params.stride;
+        new_params.UpdateParams();
+        // GetSurface will create the new surface and possibly adjust res_scale if necessary
+        surface = GetSurface(new_params, match_res_scale, load_if_create);
+    } else if (load_if_create) {
+        ValidateSurface(surface, aligned_params.addr, aligned_params.size);
+    }
+
+    return std::make_tuple(surface, surface->GetScaledSubRect(params));
+}
+
+Surface RasterizerCacheOpenGL::GetTextureSurface(const void* config) {
+    UNIMPLEMENTED();
+    return {};
+}
+
+SurfaceSurfaceRect_Tuple RasterizerCacheOpenGL::GetFramebufferSurfaces(
+    bool using_color_fb, bool using_depth_fb, const MathUtil::Rectangle<s32>& viewport_rect) {
+    UNIMPLEMENTED();
+    return {};
+}
+
+Surface RasterizerCacheOpenGL::GetFillSurface(const void* config) {
+    UNIMPLEMENTED();
+    return {};
+}
+
+SurfaceRect_Tuple RasterizerCacheOpenGL::GetTexCopySurface(const SurfaceParams& params) {
+    MathUtil::Rectangle<u32> rect{};
+
+    Surface match_surface = FindMatch<MatchFlags::TexCopy | MatchFlags::Invalid>(
+        surface_cache, params, ScaleMatch::Ignore);
+
+    if (match_surface != nullptr) {
+        ValidateSurface(match_surface, params.addr, params.size);
+
+        SurfaceParams match_subrect;
+        if (params.width != params.stride) {
+            const u32 tiled_size = match_surface->is_tiled ? 8 : 1;
+            match_subrect = params;
+            match_subrect.width =
+                static_cast<u32>(match_surface->PixelsInBytes(params.width) / tiled_size);
+            match_subrect.stride =
+                static_cast<u32>(match_surface->PixelsInBytes(params.stride) / tiled_size);
+            match_subrect.height *= tiled_size;
+        } else {
+            match_subrect = match_surface->FromInterval(params.GetInterval());
+            ASSERT(match_subrect.GetInterval() == params.GetInterval());
+        }
+
+        rect = match_surface->GetScaledSubRect(match_subrect);
+    }
+
+    return std::make_tuple(match_surface, rect);
+}
+
+void RasterizerCacheOpenGL::DuplicateSurface(const Surface& src_surface,
+                                             const Surface& dest_surface) {
+    ASSERT(dest_surface->addr <= src_surface->addr && dest_surface->end >= src_surface->end);
+
+    BlitSurfaces(src_surface, src_surface->GetScaledRect(), dest_surface,
+                 dest_surface->GetScaledSubRect(*src_surface));
+
+    dest_surface->invalid_regions -= src_surface->GetInterval();
+    dest_surface->invalid_regions += src_surface->invalid_regions;
+
+    SurfaceRegions regions;
+    for (auto& pair : RangeFromInterval(dirty_regions, src_surface->GetInterval())) {
+        if (pair.second == src_surface) {
+            regions += pair.first;
+        }
+    }
+    for (auto& interval : regions) {
+        dirty_regions.set({interval, dest_surface});
+    }
+}
+
+void RasterizerCacheOpenGL::ValidateSurface(const Surface& surface, PAddr addr, u64 size) {
+    if (size == 0)
+        return;
+
+    const SurfaceInterval validate_interval(addr, addr + size);
+
+    if (surface->type == SurfaceType::Fill) {
+        // Sanity check, fill surfaces will always be valid when used
+        ASSERT(surface->IsRegionValid(validate_interval));
+        return;
+    }
+
+    while (true) {
+        const auto it = surface->invalid_regions.find(validate_interval);
+        if (it == surface->invalid_regions.end())
+            break;
+
+        const auto interval = *it & validate_interval;
+        // Look for a valid surface to copy from
+        SurfaceParams params = surface->FromInterval(interval);
+
+        Surface copy_surface =
+            FindMatch<MatchFlags::Copy>(surface_cache, params, ScaleMatch::Ignore, interval);
+        if (copy_surface != nullptr) {
+            SurfaceInterval copy_interval = params.GetCopyableInterval(copy_surface);
+            CopySurface(copy_surface, surface, copy_interval);
+            surface->invalid_regions.erase(copy_interval);
+            continue;
+        }
+
+        // D24S8 to RGBA8
+        if (surface->pixel_format == PixelFormat::RGBA8) {
+            params.pixel_format = PixelFormat::D24S8;
+            Surface reinterpret_surface =
+                FindMatch<MatchFlags::Copy>(surface_cache, params, ScaleMatch::Ignore, interval);
+            if (reinterpret_surface != nullptr) {
+                ASSERT(reinterpret_surface->pixel_format == PixelFormat::D24S8);
+
+                SurfaceInterval convert_interval = params.GetCopyableInterval(reinterpret_surface);
+                SurfaceParams convert_params = surface->FromInterval(convert_interval);
+                auto src_rect = reinterpret_surface->GetScaledSubRect(convert_params);
+                auto dest_rect = surface->GetScaledSubRect(convert_params);
+
+                ConvertD24S8toABGR(reinterpret_surface->texture.handle, src_rect,
+                                   surface->texture.handle, dest_rect);
+
+                surface->invalid_regions.erase(convert_interval);
+                continue;
+            }
+        }
+
+        // Load data from 3DS memory
+        FlushRegion(params.addr, params.size);
+        surface->LoadGLBuffer(params.addr, params.end);
+        surface->UploadGLTexture(surface->GetSubRect(params), read_framebuffer.handle,
+                                 draw_framebuffer.handle);
+        surface->invalid_regions.erase(params.GetInterval());
+    }
+}
+
+void RasterizerCacheOpenGL::FlushRegion(PAddr addr, u64 size, Surface flush_surface) {
+    if (size == 0)
+        return;
+
+    const SurfaceInterval flush_interval(addr, addr + size);
+    SurfaceRegions flushed_intervals;
+
+    for (auto& pair : RangeFromInterval(dirty_regions, flush_interval)) {
+        // small sizes imply that this most likely comes from the cpu, flush the entire region
+        // the point is to avoid thousands of small writes every frame if the cpu decides to access
+        // that region, anything higher than 8 you're guaranteed it comes from a service
+        const auto interval = size <= 8 ? pair.first : pair.first & flush_interval;
+        auto& surface = pair.second;
+
+        if (flush_surface != nullptr && surface != flush_surface)
+            continue;
+
+        // Sanity check, this surface is the last one that marked this region dirty
+        ASSERT(surface->IsRegionValid(interval));
+
+        if (surface->type != SurfaceType::Fill) {
+            SurfaceParams params = surface->FromInterval(interval);
+            surface->DownloadGLTexture(surface->GetSubRect(params), read_framebuffer.handle,
+                                       draw_framebuffer.handle);
+        }
+        surface->FlushGLBuffer(boost::icl::first(interval), boost::icl::last_next(interval));
+        flushed_intervals += interval;
+    }
+    // Reset dirty regions
+    dirty_regions -= flushed_intervals;
+}
+
+void RasterizerCacheOpenGL::FlushAll() {
+    FlushRegion(0, 0xFFFFFFFF);
+}
+
+void RasterizerCacheOpenGL::InvalidateRegion(PAddr addr, u64 size, const Surface& region_owner) {
+    if (size == 0)
+        return;
+
+    const SurfaceInterval invalid_interval(addr, addr + size);
+
+    if (region_owner != nullptr) {
+        ASSERT(region_owner->type != SurfaceType::Texture);
+        ASSERT(addr >= region_owner->addr && addr + size <= region_owner->end);
+        // Surfaces can't have a gap
+        ASSERT(region_owner->width == region_owner->stride);
+        region_owner->invalid_regions.erase(invalid_interval);
+    }
+
+    for (auto& pair : RangeFromInterval(surface_cache, invalid_interval)) {
+        for (auto& cached_surface : pair.second) {
+            if (cached_surface == region_owner)
+                continue;
+
+            // If cpu is invalidating this region we want to remove it
+            // to (likely) mark the memory pages as uncached
+            if (region_owner == nullptr && size <= 8) {
+                FlushRegion(cached_surface->addr, cached_surface->size, cached_surface);
+                remove_surfaces.emplace(cached_surface);
+                continue;
+            }
+
+            const auto interval = cached_surface->GetInterval() & invalid_interval;
+            cached_surface->invalid_regions.insert(interval);
+
+            // Remove only "empty" fill surfaces to avoid destroying and recreating OGL textures
+            if (cached_surface->type == SurfaceType::Fill &&
+                cached_surface->IsSurfaceFullyInvalid()) {
+                remove_surfaces.emplace(cached_surface);
+            }
+        }
+    }
+
+    if (region_owner != nullptr)
+        dirty_regions.set({invalid_interval, region_owner});
+    else
+        dirty_regions.erase(invalid_interval);
+
+    for (auto& remove_surface : remove_surfaces) {
+        if (remove_surface == region_owner) {
+            Surface expanded_surface = FindMatch<MatchFlags::SubRect | MatchFlags::Invalid>(
+                surface_cache, *region_owner, ScaleMatch::Ignore);
+            ASSERT(expanded_surface);
+
+            if ((region_owner->invalid_regions - expanded_surface->invalid_regions).empty()) {
+                DuplicateSurface(region_owner, expanded_surface);
+            } else {
+                continue;
+            }
+        }
+        UnregisterSurface(remove_surface);
+    }
+
+    remove_surfaces.clear();
+}
+
+Surface RasterizerCacheOpenGL::CreateSurface(const SurfaceParams& params) {
+    Surface surface = std::make_shared<CachedSurface>();
+    static_cast<SurfaceParams&>(*surface) = params;
+
+    surface->texture.Create();
+
+    surface->gl_buffer_size = 0;
+    surface->invalid_regions.insert(surface->GetInterval());
+    AllocateSurfaceTexture(surface->texture.handle, GetFormatTuple(surface->pixel_format),
+                           surface->GetScaledWidth(), surface->GetScaledHeight());
+
+    return surface;
+}
+
+void RasterizerCacheOpenGL::RegisterSurface(const Surface& surface) {
+    if (surface->registered) {
+        return;
+    }
+    surface->registered = true;
+    surface_cache.add({surface->GetInterval(), SurfaceSet{surface}});
+    UpdatePagesCachedCount(surface->addr, surface->size, 1);
+}
+
+void RasterizerCacheOpenGL::UnregisterSurface(const Surface& surface) {
+    if (!surface->registered) {
+        return;
+    }
+    surface->registered = false;
+    UpdatePagesCachedCount(surface->addr, surface->size, -1);
+    surface_cache.subtract({surface->GetInterval(), SurfaceSet{surface}});
+}
+
+void RasterizerCacheOpenGL::UpdatePagesCachedCount(PAddr addr, u64 size, int delta) {
+    UNIMPLEMENTED();
+}

src/video_core/renderer_opengl/gl_rasterizer_cache.h

@@ -0,0 +1,350 @@
+// Copyright 2015 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <memory>
+#include <set>
+#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 <glad/glad.h>
+#include "common/assert.h"
+#include "common/common_funcs.h"
+#include "common/common_types.h"
+#include "common/math_util.h"
+#include "video_core/renderer_opengl/gl_resource_manager.h"
+
+struct CachedSurface;
+using Surface = std::shared_ptr<CachedSurface>;
+using SurfaceSet = std::set<Surface>;
+
+using SurfaceRegions = boost::icl::interval_set<PAddr>;
+using SurfaceMap = boost::icl::interval_map<PAddr, Surface>;
+using SurfaceCache = boost::icl::interval_map<PAddr, 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<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,
+        RGB8 = 1,
+        RGB5A1 = 2,
+        RGB565 = 3,
+        RGBA4 = 4,
+
+        // Texture-only formats
+        IA8 = 5,
+        RG8 = 6,
+        I8 = 7,
+        A8 = 8,
+        IA4 = 9,
+        I4 = 10,
+        A4 = 11,
+        ETC1 = 12,
+        ETC1A4 = 13,
+
+        // Depth buffer-only formats
+        D16 = 14,
+        // gap
+        D24 = 16,
+        D24S8 = 17,
+
+        Invalid = 255,
+    };
+
+    enum class SurfaceType {
+        Color = 0,
+        Texture = 1,
+        Depth = 2,
+        DepthStencil = 3,
+        Fill = 4,
+        Invalid = 5
+    };
+
+    static constexpr unsigned int GetFormatBpp(PixelFormat format) {
+        constexpr std::array<unsigned int, 18> bpp_table = {
+            32, // RGBA8
+            24, // RGB8
+            16, // RGB5A1
+            16, // RGB565
+            16, // RGBA4
+            16, // IA8
+            16, // RG8
+            8,  // I8
+            8,  // A8
+            8,  // IA4
+            4,  // I4
+            4,  // A4
+            4,  // ETC1
+            8,  // ETC1A4
+            16, // D16
+            0,
+            24, // D24
+            32, // D24S8
+        };
+
+        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 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::Color || a_type == SurfaceType::Texture) &&
+            (b_type == SurfaceType::Color || b_type == SurfaceType::Texture)) {
+            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 constexpr SurfaceType GetFormatType(PixelFormat pixel_format) {
+        if ((unsigned int)pixel_format < 5) {
+            return SurfaceType::Color;
+        }
+
+        if ((unsigned int)pixel_format < 14) {
+            return SurfaceType::Texture;
+        }
+
+        if (pixel_format == PixelFormat::D16 || pixel_format == PixelFormat::D24) {
+            return SurfaceType::Depth;
+        }
+
+        if (pixel_format == PixelFormat::D24S8) {
+            return SurfaceType::DepthStencil;
+        }
+
+        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 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;
+    }
+
+    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;
+    u64 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;
+
+    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) {
+        // 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 {
+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 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)
+    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 void* 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);
+
+    /// Get a surface that matches the fill config
+    Surface GetFillSurface(const void* config);
+
+    /// 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)
+    void FlushRegion(PAddr addr, u64 size, Surface flush_surface = nullptr);
+
+    /// Mark region as being invalidated by region_owner (nullptr if 3DS memory)
+    void InvalidateRegion(PAddr addr, u64 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, u64 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, u64 size, int delta);
+
+    SurfaceCache 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;
+    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;
     }
 
@@ -91,10 +91,13 @@ public:
     }
 
     /// Creates a new internal OpenGL resource and stores the handle
-    void Create(const char* vert_shader, const char* frag_shader) {
+    void Create(const char* vert_shader, const char* geo_shader, const char* frag_shader,
+                const std::vector<const char*>& feedback_vars = {},
+                bool separable_program = false) {
         if (handle != 0)
             return;
-        handle = GLShader::LoadProgram(vert_shader, frag_shader);
+        handle = GLShader::LoadProgram(vert_shader, geo_shader, frag_shader, feedback_vars,
+                                       separable_program);
     }
 
     /// Deletes the internal OpenGL resource
@@ -102,7 +105,40 @@ public:
         if (handle == 0)
             return;
         glDeleteProgram(handle);
-        OpenGLState::ResetProgram(handle);
+        OpenGLState::GetCurState().ResetProgram(handle).Apply();
+        handle = 0;
+    }
+
+    GLuint handle = 0;
+};
+
+class OGLPipeline : private NonCopyable {
+public:
+    OGLPipeline() = default;
+    OGLPipeline(OGLPipeline&& o) {
+        handle = std::exchange<GLuint>(o.handle, 0);
+    }
+    ~OGLPipeline() {
+        Release();
+    }
+    OGLPipeline& operator=(OGLPipeline&& o) {
+        handle = std::exchange<GLuint>(o.handle, 0);
+        return *this;
+    }
+
+    /// Creates a new internal OpenGL resource and stores the handle
+    void Create() {
+        if (handle != 0)
+            return;
+        glGenProgramPipelines(1, &handle);
+    }
+
+    /// Deletes the internal OpenGL resource
+    void Release() {
+        if (handle == 0)
+            return;
+        glDeleteProgramPipelines(1, &handle);
+        OpenGLState::GetCurState().ResetPipeline(handle).Apply();
         handle = 0;
     }
 
@@ -135,13 +171,46 @@ public:
         if (handle == 0)
             return;
         glDeleteBuffers(1, &handle);
-        OpenGLState::ResetBuffer(handle);
+        OpenGLState::GetCurState().ResetBuffer(handle).Apply();
         handle = 0;
     }
 
     GLuint handle = 0;
 };
 
+class OGLSync : private NonCopyable {
+public:
+    OGLSync() = default;
+
+    OGLSync(OGLSync&& o) : handle(std::exchange(o.handle, nullptr)) {}
+
+    ~OGLSync() {
+        Release();
+    }
+    OGLSync& operator=(OGLSync&& o) {
+        Release();
+        handle = std::exchange(o.handle, nullptr);
+        return *this;
+    }
+
+    /// Creates a new internal OpenGL resource and stores the handle
+    void Create() {
+        if (handle != 0)
+            return;
+        handle = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
+    }
+
+    /// Deletes the internal OpenGL resource
+    void Release() {
+        if (handle == 0)
+            return;
+        glDeleteSync(handle);
+        handle = 0;
+    }
+
+    GLsync handle = 0;
+};
+
 class OGLVertexArray : private NonCopyable {
 public:
     OGLVertexArray() = default;
@@ -168,7 +237,7 @@ public:
         if (handle == 0)
             return;
         glDeleteVertexArrays(1, &handle);
-        OpenGLState::ResetVertexArray(handle);
+        OpenGLState::GetCurState().ResetVertexArray(handle).Apply();
         handle = 0;
     }
 
@@ -201,7 +270,7 @@ 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_decompiler.cpp

@@ -0,0 +1,58 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <string>
+#include <queue>
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "video_core/renderer_opengl/gl_shader_decompiler.h"
+
+namespace Maxwell3D {
+namespace Shader {
+namespace Decompiler {
+
+constexpr u32 PROGRAM_END = MAX_PROGRAM_CODE_LENGTH;
+
+class Impl {
+public:
+    Impl(const std::array<u32, MAX_PROGRAM_CODE_LENGTH>& program_code,
+         const std::array<u32, MAX_SWIZZLE_DATA_LENGTH>& swizzle_data, u32 main_offset,
+         const std::function<std::string(u32)>& inputreg_getter,
+         const std::function<std::string(u32)>& outputreg_getter, bool sanitize_mul,
+         const std::string& emit_cb, const std::string& setemit_cb)
+        : program_code(program_code), swizzle_data(swizzle_data), main_offset(main_offset),
+          inputreg_getter(inputreg_getter), outputreg_getter(outputreg_getter),
+          sanitize_mul(sanitize_mul), emit_cb(emit_cb), setemit_cb(setemit_cb) {}
+
+    std::string Decompile() {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+private:
+    const std::array<u32, MAX_PROGRAM_CODE_LENGTH>& program_code;
+    const std::array<u32, MAX_SWIZZLE_DATA_LENGTH>& swizzle_data;
+    u32 main_offset;
+    const std::function<std::string(u32)>& inputreg_getter;
+    const std::function<std::string(u32)>& outputreg_getter;
+    bool sanitize_mul;
+    const std::string& emit_cb;
+    const std::string& setemit_cb;
+};
+
+std::string DecompileProgram(const std::array<u32, MAX_PROGRAM_CODE_LENGTH>& program_code,
+                             const std::array<u32, MAX_SWIZZLE_DATA_LENGTH>& swizzle_data,
+                             u32 main_offset,
+                             const std::function<std::string(u32)>& inputreg_getter,
+                             const std::function<std::string(u32)>& outputreg_getter,
+                             bool sanitize_mul, const std::string& emit_cb,
+                             const std::string& setemit_cb) {
+    Impl impl(program_code, swizzle_data, main_offset, inputreg_getter, outputreg_getter,
+              sanitize_mul, emit_cb, setemit_cb);
+    return impl.Decompile();
+}
+
+} // namespace Decompiler
+} // namespace Shader
+} // namespace Maxwell3D

src/video_core/renderer_opengl/gl_shader_decompiler.h

@@ -0,0 +1,27 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <array>
+#include <functional>
+#include <string>
+#include "common/common_types.h"
+
+namespace Maxwell3D {
+namespace Shader {
+namespace Decompiler {
+
+constexpr size_t MAX_PROGRAM_CODE_LENGTH{0x100000};
+constexpr size_t MAX_SWIZZLE_DATA_LENGTH{0x100000};
+
+std::string DecompileProgram(const std::array<u32, MAX_PROGRAM_CODE_LENGTH>& program_code,
+                             const std::array<u32, MAX_SWIZZLE_DATA_LENGTH>& swizzle_data,
+                             u32 main_offset,
+                             const std::function<std::string(u32)>& inputreg_getter,
+                             const std::function<std::string(u32)>& outputreg_getter,
+                             bool sanitize_mul, const std::string& emit_cb = "",
+                             const std::string& setemit_cb = "");
+
+} // namespace Decompiler
+} // namespace Shader
+} // namespace Maxwell3D

src/video_core/renderer_opengl/gl_shader_gen.cpp

@@ -0,0 +1,20 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/assert.h"
+#include "video_core/renderer_opengl/gl_shader_gen.h"
+
+namespace GLShader {
+
+std::string GenerateVertexShader(const MaxwellVSConfig& config) {
+    UNIMPLEMENTED();
+    return {};
+}
+
+std::string GenerateFragmentShader(const MaxwellFSConfig& config) {
+    UNIMPLEMENTED();
+    return {};
+}
+
+} // namespace GLShader

src/video_core/renderer_opengl/gl_shader_gen.h

@@ -0,0 +1,66 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <cstring>
+#include <string>
+#include <type_traits>
+#include "common/hash.h"
+
+namespace GLShader {
+
+enum Attributes {
+    ATTRIBUTE_POSITION,
+    ATTRIBUTE_COLOR,
+    ATTRIBUTE_TEXCOORD0,
+    ATTRIBUTE_TEXCOORD1,
+    ATTRIBUTE_TEXCOORD2,
+    ATTRIBUTE_TEXCOORD0_W,
+    ATTRIBUTE_NORMQUAT,
+    ATTRIBUTE_VIEW,
+};
+
+struct MaxwellShaderConfigCommon {
+    explicit MaxwellShaderConfigCommon(){};
+};
+
+struct MaxwellVSConfig : MaxwellShaderConfigCommon {
+    explicit MaxwellVSConfig() : MaxwellShaderConfigCommon() {}
+
+    bool operator==(const MaxwellVSConfig& o) const {
+        return std::memcmp(this, &o, sizeof(MaxwellVSConfig)) == 0;
+    };
+};
+
+struct MaxwellFSConfig : MaxwellShaderConfigCommon {
+    explicit MaxwellFSConfig() : MaxwellShaderConfigCommon() {}
+
+    bool operator==(const MaxwellFSConfig& o) const {
+        return std::memcmp(this, &o, sizeof(MaxwellFSConfig)) == 0;
+    };
+};
+
+std::string GenerateVertexShader(const MaxwellVSConfig& config);
+std::string GenerateFragmentShader(const MaxwellFSConfig& config);
+
+} // namespace GLShader
+
+namespace std {
+
+template <>
+struct hash<GLShader::MaxwellVSConfig> {
+    size_t operator()(const GLShader::MaxwellVSConfig& k) const {
+        return Common::ComputeHash64(&k, sizeof(GLShader::MaxwellVSConfig));
+    }
+};
+
+template <>
+struct hash<GLShader::MaxwellFSConfig> {
+    size_t operator()(const GLShader::MaxwellFSConfig& k) const {
+        return Common::ComputeHash64(&k, sizeof(GLShader::MaxwellFSConfig));
+    }
+};
+
+} // namespace std

src/video_core/renderer_opengl/gl_shader_util.cpp

@@ -10,53 +10,85 @@
 
 namespace GLShader {
 
-GLuint LoadProgram(const char* vertex_shader, const char* fragment_shader) {
-
+GLuint LoadProgram(const char* vertex_shader, const char* geometry_shader,
+                   const char* fragment_shader, const std::vector<const char*>& feedback_vars,
+                   bool separable_program) {
     // Create the shaders
-    GLuint vertex_shader_id = glCreateShader(GL_VERTEX_SHADER);
-    GLuint fragment_shader_id = glCreateShader(GL_FRAGMENT_SHADER);
+    GLuint vertex_shader_id = vertex_shader ? glCreateShader(GL_VERTEX_SHADER) : 0;
+    GLuint geometry_shader_id = geometry_shader ? glCreateShader(GL_GEOMETRY_SHADER) : 0;
+    GLuint fragment_shader_id = fragment_shader ? glCreateShader(GL_FRAGMENT_SHADER) : 0;
 
     GLint result = GL_FALSE;
     int info_log_length;
 
-    // Compile Vertex Shader
-    LOG_DEBUG(Render_OpenGL, "Compiling vertex shader...");
-
-    glShaderSource(vertex_shader_id, 1, &vertex_shader, nullptr);
-    glCompileShader(vertex_shader_id);
-
-    // Check Vertex Shader
-    glGetShaderiv(vertex_shader_id, GL_COMPILE_STATUS, &result);
-    glGetShaderiv(vertex_shader_id, GL_INFO_LOG_LENGTH, &info_log_length);
-
-    if (info_log_length > 1) {
-        std::vector<char> vertex_shader_error(info_log_length);
-        glGetShaderInfoLog(vertex_shader_id, info_log_length, nullptr, &vertex_shader_error[0]);
-        if (result == GL_TRUE) {
-            LOG_DEBUG(Render_OpenGL, "%s", &vertex_shader_error[0]);
-        } else {
-            LOG_ERROR(Render_OpenGL, "Error compiling vertex shader:\n%s", &vertex_shader_error[0]);
+    if (vertex_shader) {
+        // Compile Vertex Shader
+        LOG_DEBUG(Render_OpenGL, "Compiling vertex shader...");
+
+        glShaderSource(vertex_shader_id, 1, &vertex_shader, nullptr);
+        glCompileShader(vertex_shader_id);
+
+        // Check Vertex Shader
+        glGetShaderiv(vertex_shader_id, GL_COMPILE_STATUS, &result);
+        glGetShaderiv(vertex_shader_id, GL_INFO_LOG_LENGTH, &info_log_length);
+
+        if (info_log_length > 1) {
+            std::vector<char> vertex_shader_error(info_log_length);
+            glGetShaderInfoLog(vertex_shader_id, info_log_length, nullptr, &vertex_shader_error[0]);
+            if (result == GL_TRUE) {
+                LOG_DEBUG(Render_OpenGL, "%s", &vertex_shader_error[0]);
+            } else {
+                LOG_ERROR(Render_OpenGL, "Error compiling vertex shader:\n%s",
+                          &vertex_shader_error[0]);
+            }
         }
     }
 
-    // Compile Fragment Shader
-    LOG_DEBUG(Render_OpenGL, "Compiling fragment shader...");
-
-    glShaderSource(fragment_shader_id, 1, &fragment_shader, nullptr);
-    glCompileShader(fragment_shader_id);
-
-    // Check Fragment Shader
-    glGetShaderiv(fragment_shader_id, GL_COMPILE_STATUS, &result);
-    glGetShaderiv(fragment_shader_id, GL_INFO_LOG_LENGTH, &info_log_length);
+    if (geometry_shader) {
+        // Compile Geometry Shader
+        LOG_DEBUG(Render_OpenGL, "Compiling geometry shader...");
+
+        glShaderSource(geometry_shader_id, 1, &geometry_shader, nullptr);
+        glCompileShader(geometry_shader_id);
+
+        // Check Geometry Shader
+        glGetShaderiv(geometry_shader_id, GL_COMPILE_STATUS, &result);
+        glGetShaderiv(geometry_shader_id, GL_INFO_LOG_LENGTH, &info_log_length);
+
+        if (info_log_length > 1) {
+            std::vector<char> geometry_shader_error(info_log_length);
+            glGetShaderInfoLog(geometry_shader_id, info_log_length, nullptr,
+                               &geometry_shader_error[0]);
+            if (result == GL_TRUE) {
+                LOG_DEBUG(Render_OpenGL, "%s", &geometry_shader_error[0]);
+            } else {
+                LOG_ERROR(Render_OpenGL, "Error compiling geometry shader:\n%s",
+                          &geometry_shader_error[0]);
+            }
+        }
+    }
 
-    if (info_log_length > 1) {
-        std::vector<char> fragment_shader_error(info_log_length);
-        glGetShaderInfoLog(fragment_shader_id, info_log_length, nullptr, &fragment_shader_error[0]);
-        if (result == GL_TRUE) {
-            LOG_DEBUG(Render_OpenGL, "%s", &fragment_shader_error[0]);
-        } else {
-            LOG_ERROR(Render_OpenGL, "Error compiling fragment shader:\n%s",
-                      &fragment_shader_error[0]);
+    if (fragment_shader) {
+        // Compile Fragment Shader
+        LOG_DEBUG(Render_OpenGL, "Compiling fragment shader...");
+
+        glShaderSource(fragment_shader_id, 1, &fragment_shader, nullptr);
+        glCompileShader(fragment_shader_id);
+
+        // Check Fragment Shader
+        glGetShaderiv(fragment_shader_id, GL_COMPILE_STATUS, &result);
+        glGetShaderiv(fragment_shader_id, GL_INFO_LOG_LENGTH, &info_log_length);
+
+        if (info_log_length > 1) {
+            std::vector<char> fragment_shader_error(info_log_length);
+            glGetShaderInfoLog(fragment_shader_id, info_log_length, nullptr,
+                               &fragment_shader_error[0]);
+            if (result == GL_TRUE) {
+                LOG_DEBUG(Render_OpenGL, "%s", &fragment_shader_error[0]);
+            } else {
+                LOG_ERROR(Render_OpenGL, "Error compiling fragment shader:\n%s",
+                          &fragment_shader_error[0]);
+            }
         }
     }
 
@@ -64,8 +96,25 @@ GLuint LoadProgram(const char* vertex_shader, const char* fragment_shader) {
     LOG_DEBUG(Render_OpenGL, "Linking program...");
 
     GLuint program_id = glCreateProgram();
-    glAttachShader(program_id, vertex_shader_id);
-    glAttachShader(program_id, fragment_shader_id);
+    if (vertex_shader) {
+        glAttachShader(program_id, vertex_shader_id);
+    }
+    if (geometry_shader) {
+        glAttachShader(program_id, geometry_shader_id);
+    }
+    if (fragment_shader) {
+        glAttachShader(program_id, fragment_shader_id);
+    }
+
+    if (!feedback_vars.empty()) {
+        auto varyings = feedback_vars;
+        glTransformFeedbackVaryings(program_id, static_cast<GLsizei>(feedback_vars.size()),
+                                    &varyings[0], GL_INTERLEAVED_ATTRIBS);
+    }
+
+    if (separable_program) {
+        glProgramParameteri(program_id, GL_PROGRAM_SEPARABLE, GL_TRUE);
+    }
 
     glLinkProgram(program_id);
 
@@ -85,13 +134,30 @@ GLuint LoadProgram(const char* vertex_shader, const char* fragment_shader) {
 
     // If the program linking failed at least one of the shaders was probably bad
     if (result == GL_FALSE) {
-        LOG_ERROR(Render_OpenGL, "Vertex shader:\n%s", vertex_shader);
-        LOG_ERROR(Render_OpenGL, "Fragment shader:\n%s", fragment_shader);
+        if (vertex_shader) {
+            LOG_ERROR(Render_OpenGL, "Vertex shader:\n%s", vertex_shader);
+        }
+        if (geometry_shader) {
+            LOG_ERROR(Render_OpenGL, "Geometry shader:\n%s", geometry_shader);
+        }
+        if (fragment_shader) {
+            LOG_ERROR(Render_OpenGL, "Fragment shader:\n%s", fragment_shader);
+        }
     }
     ASSERT_MSG(result == GL_TRUE, "Shader not linked");
 
-    glDeleteShader(vertex_shader_id);
-    glDeleteShader(fragment_shader_id);
+    if (vertex_shader) {
+        glDetachShader(program_id, vertex_shader_id);
+        glDeleteShader(vertex_shader_id);
+    }
+    if (geometry_shader) {
+        glDetachShader(program_id, geometry_shader_id);
+        glDeleteShader(geometry_shader_id);
+    }
+    if (fragment_shader) {
+        glDetachShader(program_id, fragment_shader_id);
+        glDeleteShader(fragment_shader_id);
+    }
 
     return program_id;
 }

src/video_core/renderer_opengl/gl_shader_util.h

@@ -4,6 +4,7 @@
 
 #pragma once
 
+#include <vector>
 #include <glad/glad.h>
 
 namespace GLShader {
@@ -11,9 +12,12 @@ namespace GLShader {
 /**
  * Utility function to create and compile an OpenGL GLSL shader program (vertex + fragment shader)
  * @param vertex_shader String of the GLSL vertex shader program
+ * @param geometry_shader String of the GLSL geometry shader program
  * @param fragment_shader String of the GLSL fragment shader program
  * @returns Handle of the newly created OpenGL shader object
  */
-GLuint LoadProgram(const char* vertex_shader, const char* fragment_shader);
+GLuint LoadProgram(const char* vertex_shader, const char* geometry_shader,
+                   const char* fragment_shader, const std::vector<const char*>& feedback_vars = {},
+                   bool separable_program = false);
 
 } // namespace GLShader

src/video_core/renderer_opengl/gl_state.cpp

@@ -33,7 +33,7 @@ OpenGLState::OpenGLState() {
     stencil.action_depth_pass = GL_KEEP;
     stencil.action_stencil_fail = GL_KEEP;
 
-    blend.enabled = false;
+    blend.enabled = true;
     blend.rgb_equation = GL_FUNC_ADD;
     blend.a_equation = GL_FUNC_ADD;
     blend.src_rgb_func = GL_ONE;
@@ -68,6 +68,18 @@ OpenGLState::OpenGLState() {
     draw.vertex_buffer = 0;
     draw.uniform_buffer = 0;
     draw.shader_program = 0;
+    draw.program_pipeline = 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 = {};
 }
@@ -148,9 +160,6 @@ void OpenGLState::Apply() const {
     if (blend.enabled != cur_state.blend.enabled) {
         if (blend.enabled) {
             glEnable(GL_BLEND);
-
-            cur_state.logic_op = GL_COPY;
-            glLogicOp(cur_state.logic_op);
             glDisable(GL_COLOR_LOGIC_OP);
         } else {
             glDisable(GL_BLEND);
@@ -196,7 +205,7 @@ 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
@@ -263,6 +272,31 @@ void OpenGLState::Apply() const {
         glUseProgram(draw.shader_program);
     }
 
+    // Program pipeline
+    if (draw.program_pipeline != cur_state.draw.program_pipeline) {
+        glBindProgramPipeline(draw.program_pipeline);
+    }
+
+    // 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]) {
@@ -277,62 +311,75 @@ 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;
+}
+
+OpenGLState& OpenGLState::ResetProgram(GLuint handle) {
+    if (draw.shader_program == handle) {
+        draw.shader_program = 0;
+    }
+    return *this;
 }
 
-void OpenGLState::ResetProgram(GLuint handle) {
-    if (cur_state.draw.shader_program == handle) {
-        cur_state.draw.shader_program = 0;
+OpenGLState& OpenGLState::ResetPipeline(GLuint handle) {
+    if (draw.program_pipeline == handle) {
+        draw.program_pipeline = 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

@@ -122,27 +122,44 @@ public:
         GLuint vertex_buffer;    // GL_ARRAY_BUFFER_BINDING
         GLuint uniform_buffer;   // GL_UNIFORM_BUFFER_BINDING
         GLuint shader_program;   // GL_CURRENT_PROGRAM
+        GLuint program_pipeline; // GL_PROGRAM_PIPELINE_BINDING
     } 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& ResetPipeline(GLuint handle);
+    OpenGLState& ResetBuffer(GLuint handle);
+    OpenGLState& ResetVertexArray(GLuint handle);
+    OpenGLState& ResetFramebuffer(GLuint handle);
 
 private:
     static OpenGLState cur_state;

src/video_core/renderer_opengl/gl_stream_buffer.cpp

@@ -0,0 +1,182 @@
+// Copyright 2018 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <deque>
+#include <vector>
+#include "common/alignment.h"
+#include "common/assert.h"
+#include "video_core/renderer_opengl/gl_state.h"
+#include "video_core/renderer_opengl/gl_stream_buffer.h"
+
+class OrphanBuffer : public OGLStreamBuffer {
+public:
+    explicit OrphanBuffer(GLenum target) : OGLStreamBuffer(target) {}
+    ~OrphanBuffer() override;
+
+private:
+    void Create(size_t size, size_t sync_subdivide) override;
+    void Release() override;
+
+    std::pair<u8*, GLintptr> Map(size_t size, size_t alignment) override;
+    void Unmap() override;
+
+    std::vector<u8> data;
+};
+
+class StorageBuffer : public OGLStreamBuffer {
+public:
+    explicit StorageBuffer(GLenum target) : OGLStreamBuffer(target) {}
+    ~StorageBuffer() override;
+
+private:
+    void Create(size_t size, size_t sync_subdivide) override;
+    void Release() override;
+
+    std::pair<u8*, GLintptr> Map(size_t size, size_t alignment) override;
+    void Unmap() override;
+
+    struct Fence {
+        OGLSync sync;
+        size_t offset;
+    };
+    std::deque<Fence> head;
+    std::deque<Fence> tail;
+
+    u8* mapped_ptr;
+};
+
+OGLStreamBuffer::OGLStreamBuffer(GLenum target) {
+    gl_target = target;
+}
+
+GLuint OGLStreamBuffer::GetHandle() const {
+    return gl_buffer.handle;
+}
+
+std::unique_ptr<OGLStreamBuffer> OGLStreamBuffer::MakeBuffer(bool storage_buffer, GLenum target) {
+    if (storage_buffer) {
+        return std::make_unique<StorageBuffer>(target);
+    }
+    return std::make_unique<OrphanBuffer>(target);
+}
+
+OrphanBuffer::~OrphanBuffer() {
+    Release();
+}
+
+void OrphanBuffer::Create(size_t size, size_t /*sync_subdivide*/) {
+    buffer_pos = 0;
+    buffer_size = size;
+    data.resize(buffer_size);
+
+    if (gl_buffer.handle == 0) {
+        gl_buffer.Create();
+        glBindBuffer(gl_target, gl_buffer.handle);
+    }
+
+    glBufferData(gl_target, static_cast<GLsizeiptr>(buffer_size), nullptr, GL_STREAM_DRAW);
+}
+
+void OrphanBuffer::Release() {
+    gl_buffer.Release();
+}
+
+std::pair<u8*, GLintptr> OrphanBuffer::Map(size_t size, size_t alignment) {
+    buffer_pos = Common::AlignUp(buffer_pos, alignment);
+
+    if (buffer_pos + size > buffer_size) {
+        Create(std::max(buffer_size, size), 0);
+    }
+
+    mapped_size = size;
+    return std::make_pair(&data[buffer_pos], static_cast<GLintptr>(buffer_pos));
+}
+
+void OrphanBuffer::Unmap() {
+    glBufferSubData(gl_target, static_cast<GLintptr>(buffer_pos),
+                    static_cast<GLsizeiptr>(mapped_size), &data[buffer_pos]);
+    buffer_pos += mapped_size;
+}
+
+StorageBuffer::~StorageBuffer() {
+    Release();
+}
+
+void StorageBuffer::Create(size_t size, size_t sync_subdivide) {
+    if (gl_buffer.handle != 0)
+        return;
+
+    buffer_pos = 0;
+    buffer_size = size;
+    buffer_sync_subdivide = std::max<size_t>(sync_subdivide, 1);
+
+    gl_buffer.Create();
+    glBindBuffer(gl_target, gl_buffer.handle);
+
+    glBufferStorage(gl_target, static_cast<GLsizeiptr>(buffer_size), nullptr,
+                    GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT);
+    mapped_ptr = reinterpret_cast<u8*>(
+        glMapBufferRange(gl_target, 0, static_cast<GLsizeiptr>(buffer_size),
+                         GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT | GL_MAP_FLUSH_EXPLICIT_BIT));
+}
+
+void StorageBuffer::Release() {
+    if (gl_buffer.handle == 0)
+        return;
+
+    glUnmapBuffer(gl_target);
+
+    gl_buffer.Release();
+    head.clear();
+    tail.clear();
+}
+
+std::pair<u8*, GLintptr> StorageBuffer::Map(size_t size, size_t alignment) {
+    ASSERT(size <= buffer_size);
+
+    OGLSync sync;
+
+    buffer_pos = Common::AlignUp(buffer_pos, alignment);
+    size_t effective_offset = Common::AlignDown(buffer_pos, buffer_sync_subdivide);
+
+    if (!head.empty() &&
+        (effective_offset > head.back().offset || buffer_pos + size > buffer_size)) {
+        ASSERT(head.back().sync.handle == 0);
+        head.back().sync.Create();
+    }
+
+    if (buffer_pos + size > buffer_size) {
+        if (!tail.empty()) {
+            std::swap(sync, tail.back().sync);
+            tail.clear();
+        }
+        std::swap(tail, head);
+        buffer_pos = 0;
+        effective_offset = 0;
+    }
+
+    while (!tail.empty() && buffer_pos + size > tail.front().offset) {
+        std::swap(sync, tail.front().sync);
+        tail.pop_front();
+    }
+
+    if (sync.handle != 0) {
+        glClientWaitSync(sync.handle, GL_SYNC_FLUSH_COMMANDS_BIT, GL_TIMEOUT_IGNORED);
+        sync.Release();
+    }
+
+    if (head.empty() || effective_offset > head.back().offset) {
+        head.emplace_back();
+        head.back().offset = effective_offset;
+    }
+
+    mapped_size = size;
+    return std::make_pair(&mapped_ptr[buffer_pos], static_cast<GLintptr>(buffer_pos));
+}
+
+void StorageBuffer::Unmap() {
+    glFlushMappedBufferRange(gl_target, static_cast<GLintptr>(buffer_pos),
+                             static_cast<GLsizeiptr>(mapped_size));
+    buffer_pos += mapped_size;
+}

src/video_core/renderer_opengl/gl_stream_buffer.h

@@ -0,0 +1,34 @@
+// Copyright 2018 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <memory>
+#include <glad/glad.h>
+#include "common/common_types.h"
+#include "video_core/renderer_opengl/gl_resource_manager.h"
+
+class OGLStreamBuffer : private NonCopyable {
+public:
+    explicit OGLStreamBuffer(GLenum target);
+    virtual ~OGLStreamBuffer() = default;
+
+public:
+    static std::unique_ptr<OGLStreamBuffer> MakeBuffer(bool storage_buffer, GLenum target);
+
+    virtual void Create(size_t size, size_t sync_subdivide) = 0;
+    virtual void Release() {}
+
+    GLuint GetHandle() const;
+
+    virtual std::pair<u8*, GLintptr> Map(size_t size, size_t alignment) = 0;
+    virtual void Unmap() = 0;
+
+protected:
+    OGLBuffer gl_buffer;
+    GLenum gl_target;
+
+    size_t buffer_pos = 0;
+    size_t buffer_size = 0;
+    size_t buffer_sync_subdivide = 0;
+    size_t mapped_size = 0;
+};

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -318,7 +318,7 @@ void RendererOpenGL::InitOpenGLObjects() {
                  0.0f);
 
     // Link shaders and get variable locations
-    shader.Create(vertex_shader, fragment_shader);
+    shader.Create(vertex_shader, nullptr, fragment_shader);
     state.draw.shader_program = shader.handle;
     state.Apply();
     uniform_modelview_matrix = glGetUniformLocation(shader.handle, "modelview_matrix");