12 files changed, 945 insertions, 322 deletions

diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h
index 965976334..be29dc8be 100644
--- a/src/video_core/renderer_opengl/gl_buffer_cache.h
+++ b/src/video_core/renderer_opengl/gl_buffer_cache.h
@@ -15,15 +15,18 @@
 
 namespace OpenGL {
 
-struct CachedBufferEntry final {
-    VAddr GetAddr() const {
+struct CachedBufferEntry final : public RasterizerCacheObject {
+    VAddr GetAddr() const override {
         return addr;
     }
 
-    std::size_t GetSizeInBytes() const {
+    std::size_t GetSizeInBytes() const override {
         return size;
     }
 
+    // We do not have to flush this cache as things in it are never modified by us.
+    void Flush() override {}
+
     VAddr addr;
     std::size_t size;
     GLintptr offset;
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 209bdf181..468253033 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -255,7 +255,7 @@ DrawParameters RasterizerOpenGL::SetupDraw() {
     return params;
 }
 
-void RasterizerOpenGL::SetupShaders() {
+void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
     MICROPROFILE_SCOPE(OpenGL_Shader);
     const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
 
@@ -270,6 +270,11 @@ void RasterizerOpenGL::SetupShaders() {
 
         // Skip stages that are not enabled
         if (!gpu.regs.IsShaderConfigEnabled(index)) {
+            switch (program) {
+            case Maxwell::ShaderProgram::Geometry:
+                shader_program_manager->UseTrivialGeometryShader();
+                break;
+            }
             continue;
         }
 
@@ -281,18 +286,26 @@ void RasterizerOpenGL::SetupShaders() {
             &ubo, sizeof(ubo), static_cast<std::size_t>(uniform_buffer_alignment));
 
         // Bind the buffer
-        glBindBufferRange(GL_UNIFORM_BUFFER, stage, buffer_cache.GetHandle(), offset, sizeof(ubo));
+        glBindBufferRange(GL_UNIFORM_BUFFER, static_cast<GLuint>(stage), buffer_cache.GetHandle(),
+                          offset, static_cast<GLsizeiptr>(sizeof(ubo)));
 
         Shader shader{shader_cache.GetStageProgram(program)};
 
         switch (program) {
         case Maxwell::ShaderProgram::VertexA:
         case Maxwell::ShaderProgram::VertexB: {
-            shader_program_manager->UseProgrammableVertexShader(shader->GetProgramHandle());
+            shader_program_manager->UseProgrammableVertexShader(
+                shader->GetProgramHandle(primitive_mode));
+            break;
+        }
+        case Maxwell::ShaderProgram::Geometry: {
+            shader_program_manager->UseProgrammableGeometryShader(
+                shader->GetProgramHandle(primitive_mode));
             break;
         }
         case Maxwell::ShaderProgram::Fragment: {
-            shader_program_manager->UseProgrammableFragmentShader(shader->GetProgramHandle());
+            shader_program_manager->UseProgrammableFragmentShader(
+                shader->GetProgramHandle(primitive_mode));
             break;
         }
         default:
@@ -302,12 +315,13 @@ void RasterizerOpenGL::SetupShaders() {
         }
 
         // Configure the const buffers for this shader stage.
-        current_constbuffer_bindpoint = SetupConstBuffers(static_cast<Maxwell::ShaderStage>(stage),
-                                                          shader, current_constbuffer_bindpoint);
+        current_constbuffer_bindpoint =
+            SetupConstBuffers(static_cast<Maxwell::ShaderStage>(stage), shader, primitive_mode,
+                              current_constbuffer_bindpoint);
 
         // Configure the textures for this shader stage.
         current_texture_bindpoint = SetupTextures(static_cast<Maxwell::ShaderStage>(stage), shader,
-                                                  current_texture_bindpoint);
+                                                  primitive_mode, current_texture_bindpoint);
 
         // When VertexA is enabled, we have dual vertex shaders
         if (program == Maxwell::ShaderProgram::VertexA) {
@@ -317,8 +331,6 @@ void RasterizerOpenGL::SetupShaders() {
     }
 
     state.Apply();
-
-    shader_program_manager->UseTrivialGeometryShader();
 }
 
 std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
@@ -412,6 +424,13 @@ void RasterizerOpenGL::ConfigureFramebuffers(bool using_color_fb, bool using_dep
             // Used when just a single color attachment is enabled, e.g. for clearing a color buffer
             Surface color_surface =
                 res_cache.GetColorBufferSurface(*single_color_target, preserve_contents);
+
+            if (color_surface) {
+                // Assume that a surface will be written to if it is used as a framebuffer, even if
+                // the shader doesn't actually write to it.
+                color_surface->MarkAsModified(true, res_cache);
+            }
+
             glFramebufferTexture2D(
                 GL_DRAW_FRAMEBUFFER,
                 GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(*single_color_target), GL_TEXTURE_2D,
@@ -422,6 +441,13 @@ void RasterizerOpenGL::ConfigureFramebuffers(bool using_color_fb, bool using_dep
             std::array<GLenum, Maxwell::NumRenderTargets> buffers;
             for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) {
                 Surface color_surface = res_cache.GetColorBufferSurface(index, preserve_contents);
+
+                if (color_surface) {
+                    // Assume that a surface will be written to if it is used as a framebuffer, even
+                    // if the shader doesn't actually write to it.
+                    color_surface->MarkAsModified(true, res_cache);
+                }
+
                 buffers[index] = GL_COLOR_ATTACHMENT0 + regs.rt_control.GetMap(index);
                 glFramebufferTexture2D(
                     GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index),
@@ -441,6 +467,10 @@ void RasterizerOpenGL::ConfigureFramebuffers(bool using_color_fb, bool using_dep
     }
 
     if (depth_surface) {
+        // Assume that a surface will be written to if it is used as a framebuffer, even if
+        // the shader doesn't actually write to it.
+        depth_surface->MarkAsModified(true, res_cache);
+
         if (regs.stencil_enable) {
             // Attach both depth and stencil
             glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
@@ -541,6 +571,7 @@ void RasterizerOpenGL::DrawArrays() {
     SyncLogicOpState();
     SyncCullMode();
     SyncAlphaTest();
+    SyncScissorTest();
     SyncTransformFeedback();
     SyncPointState();
 
@@ -580,7 +611,7 @@ void RasterizerOpenGL::DrawArrays() {
 
     SetupVertexArrays();
     DrawParameters params = SetupDraw();
-    SetupShaders();
+    SetupShaders(params.primitive_mode);
 
     buffer_cache.Unmap();
 
@@ -604,7 +635,14 @@ void RasterizerOpenGL::DrawArrays() {
 
 void RasterizerOpenGL::FlushAll() {}
 
-void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) {}
+void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) {
+    MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+
+    if (Settings::values.use_accurate_gpu_emulation) {
+        // Only flush if use_accurate_gpu_emulation is enabled, as it incurs a performance hit
+        res_cache.FlushRegion(addr, size);
+    }
+}
 
 void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
@@ -614,6 +652,7 @@ void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
 }
 
 void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) {
+    FlushRegion(addr, size);
     InvalidateRegion(addr, size);
 }
 
@@ -719,7 +758,7 @@ void RasterizerOpenGL::SamplerInfo::SyncWithConfig(const Tegra::Texture::TSCEntr
 }
 
 u32 RasterizerOpenGL::SetupConstBuffers(Maxwell::ShaderStage stage, Shader& shader,
-                                        u32 current_bindpoint) {
+                                        GLenum primitive_mode, u32 current_bindpoint) {
     MICROPROFILE_SCOPE(OpenGL_UBO);
     const auto& gpu = Core::System::GetInstance().GPU();
     const auto& maxwell3d = gpu.Maxwell3D();
@@ -771,7 +810,7 @@ u32 RasterizerOpenGL::SetupConstBuffers(Maxwell::ShaderStage stage, Shader& shad
             buffer.address, size, static_cast<std::size_t>(uniform_buffer_alignment));
 
         // Now configure the bindpoint of the buffer inside the shader
-        glUniformBlockBinding(shader->GetProgramHandle(),
+        glUniformBlockBinding(shader->GetProgramHandle(primitive_mode),
                               shader->GetProgramResourceIndex(used_buffer),
                               current_bindpoint + bindpoint);
 
@@ -787,7 +826,8 @@ u32 RasterizerOpenGL::SetupConstBuffers(Maxwell::ShaderStage stage, Shader& shad
     return current_bindpoint + static_cast<u32>(entries.size());
 }
 
-u32 RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, Shader& shader, u32 current_unit) {
+u32 RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, Shader& shader,
+                                    GLenum primitive_mode, u32 current_unit) {
     MICROPROFILE_SCOPE(OpenGL_Texture);
     const auto& gpu = Core::System::GetInstance().GPU();
     const auto& maxwell3d = gpu.Maxwell3D();
@@ -802,8 +842,8 @@ u32 RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, Shader& shader,
 
         // Bind the uniform to the sampler.
 
-        glProgramUniform1i(shader->GetProgramHandle(), shader->GetUniformLocation(entry),
-                           current_bindpoint);
+        glProgramUniform1i(shader->GetProgramHandle(primitive_mode),
+                           shader->GetUniformLocation(entry), current_bindpoint);
 
         const auto texture = maxwell3d.GetStageTexture(entry.GetStage(), entry.GetOffset());
 
@@ -972,6 +1012,22 @@ void RasterizerOpenGL::SyncAlphaTest() {
     }
 }
 
+void RasterizerOpenGL::SyncScissorTest() {
+    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+
+    state.scissor.enabled = (regs.scissor_test.enable != 0);
+    // TODO(Blinkhawk): Figure if the hardware supports scissor testing per viewport and how it's
+    // implemented.
+    if (regs.scissor_test.enable != 0) {
+        const u32 width = regs.scissor_test.max_x - regs.scissor_test.min_x;
+        const u32 height = regs.scissor_test.max_y - regs.scissor_test.min_y;
+        state.scissor.x = regs.scissor_test.min_x;
+        state.scissor.y = regs.scissor_test.min_y;
+        state.scissor.width = width;
+        state.scissor.height = height;
+    }
+}
+
 void RasterizerOpenGL::SyncTransformFeedback() {
     const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
 
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index 0dab2018b..b1f7ccc7e 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -120,7 +120,7 @@ private:
      * @returns The next available bindpoint for use in the next shader stage.
      */
     u32 SetupConstBuffers(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage, Shader& shader,
-                          u32 current_bindpoint);
+                          GLenum primitive_mode, u32 current_bindpoint);
 
     /*
      * Configures the current textures to use for the draw command.
@@ -130,7 +130,7 @@ private:
      * @returns The next available bindpoint for use in the next shader stage.
      */
     u32 SetupTextures(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage, Shader& shader,
-                      u32 current_unit);
+                      GLenum primitive_mode, u32 current_unit);
 
     /// Syncs the viewport to match the guest state
     void SyncViewport();
@@ -165,6 +165,9 @@ private:
     /// Syncs the alpha test state to match the guest state
     void SyncAlphaTest();
 
+    /// Syncs the scissor test state to match the guest state
+    void SyncScissorTest();
+
     /// Syncs the transform feedback state to match the guest state
     void SyncTransformFeedback();
 
@@ -207,7 +210,7 @@ private:
 
     DrawParameters SetupDraw();
 
-    void SetupShaders();
+    void SetupShaders(GLenum primitive_mode);
 
     enum class AccelDraw { Disabled, Arrays, Indexed };
     AccelDraw accelerate_draw = AccelDraw::Disabled;
diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
index 56ff83eff..1cb77aaf2 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
@@ -34,18 +34,57 @@ struct FormatTuple {
     bool compressed;
 };
 
-static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) {
-    auto& gpu{Core::System::GetInstance().GPU()};
-    const auto cpu_addr{gpu.MemoryManager().GpuToCpuAddress(gpu_addr)};
-    return cpu_addr ? *cpu_addr : 0;
+static bool IsPixelFormatASTC(PixelFormat format) {
+    switch (format) {
+    case PixelFormat::ASTC_2D_4X4:
+    case PixelFormat::ASTC_2D_5X4:
+    case PixelFormat::ASTC_2D_8X8:
+    case PixelFormat::ASTC_2D_8X5:
+        return true;
+    default:
+        return false;
+    }
+}
+
+static std::pair<u32, u32> GetASTCBlockSize(PixelFormat format) {
+    switch (format) {
+    case PixelFormat::ASTC_2D_4X4:
+        return {4, 4};
+    case PixelFormat::ASTC_2D_5X4:
+        return {5, 4};
+    case PixelFormat::ASTC_2D_8X8:
+        return {8, 8};
+    case PixelFormat::ASTC_2D_8X5:
+        return {8, 5};
+    default:
+        LOG_CRITICAL(HW_GPU, "Unhandled format: {}", static_cast<u32>(format));
+        UNREACHABLE();
+    }
+}
+
+void SurfaceParams::InitCacheParameters(Tegra::GPUVAddr gpu_addr_) {
+    auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
+    const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr_)};
+
+    addr = cpu_addr ? *cpu_addr : 0;
+    gpu_addr = gpu_addr_;
+    size_in_bytes = SizeInBytesRaw();
+
+    if (IsPixelFormatASTC(pixel_format)) {
+        // ASTC is uncompressed in software, in emulated as RGBA8
+        size_in_bytes_gl = width * height * depth * 4;
+    } else {
+        size_in_bytes_gl = SizeInBytesGL();
+    }
 }
 
 /*static*/ SurfaceParams SurfaceParams::CreateForTexture(
     const Tegra::Texture::FullTextureInfo& config, const GLShader::SamplerEntry& entry) {
     SurfaceParams params{};
-    params.addr = TryGetCpuAddr(config.tic.Address());
     params.is_tiled = config.tic.IsTiled();
+    params.block_width = params.is_tiled ? config.tic.BlockWidth() : 0,
     params.block_height = params.is_tiled ? config.tic.BlockHeight() : 0,
+    params.block_depth = params.is_tiled ? config.tic.BlockDepth() : 0,
     params.pixel_format =
         PixelFormatFromTextureFormat(config.tic.format, config.tic.r_type.Value());
     params.component_type = ComponentTypeFromTexture(config.tic.r_type.Value());
@@ -85,20 +124,23 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) {
         break;
     }
 
-    params.size_in_bytes_total = params.SizeInBytesTotal();
-    params.size_in_bytes_2d = params.SizeInBytes2D();
     params.max_mip_level = config.tic.max_mip_level + 1;
     params.rt = {};
 
+    params.InitCacheParameters(config.tic.Address());
+
     return params;
 }
 
 /*static*/ SurfaceParams SurfaceParams::CreateForFramebuffer(std::size_t index) {
     const auto& config{Core::System::GetInstance().GPU().Maxwell3D().regs.rt[index]};
     SurfaceParams params{};
-    params.addr = TryGetCpuAddr(config.Address());
-    params.is_tiled = true;
-    params.block_height = Tegra::Texture::TICEntry::DefaultBlockHeight;
+
+    params.is_tiled =
+        config.memory_layout.type == Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout::BlockLinear;
+    params.block_width = 1 << config.memory_layout.block_width;
+    params.block_height = 1 << config.memory_layout.block_height;
+    params.block_depth = 1 << config.memory_layout.block_depth;
     params.pixel_format = PixelFormatFromRenderTargetFormat(config.format);
     params.component_type = ComponentTypeFromRenderTarget(config.format);
     params.type = GetFormatType(params.pixel_format);
@@ -107,8 +149,6 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) {
     params.unaligned_height = config.height;
     params.target = SurfaceTarget::Texture2D;
     params.depth = 1;
-    params.size_in_bytes_total = params.SizeInBytesTotal();
-    params.size_in_bytes_2d = params.SizeInBytes2D();
     params.max_mip_level = 0;
 
     // Render target specific parameters, not used for caching
@@ -117,16 +157,21 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) {
     params.rt.layer_stride = config.layer_stride;
     params.rt.base_layer = config.base_layer;
 
+    params.InitCacheParameters(config.Address());
+
     return params;
 }
 
-/*static*/ SurfaceParams SurfaceParams::CreateForDepthBuffer(u32 zeta_width, u32 zeta_height,
-                                                             Tegra::GPUVAddr zeta_address,
-                                                             Tegra::DepthFormat format) {
+/*static*/ SurfaceParams SurfaceParams::CreateForDepthBuffer(
+    u32 zeta_width, u32 zeta_height, Tegra::GPUVAddr zeta_address, Tegra::DepthFormat format,
+    u32 block_width, u32 block_height, u32 block_depth,
+    Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout type) {
     SurfaceParams params{};
-    params.addr = TryGetCpuAddr(zeta_address);
-    params.is_tiled = true;
-    params.block_height = Tegra::Texture::TICEntry::DefaultBlockHeight;
+
+    params.is_tiled = type == Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout::BlockLinear;
+    params.block_width = 1 << std::min(block_width, 5U);
+    params.block_height = 1 << std::min(block_height, 5U);
+    params.block_depth = 1 << std::min(block_depth, 5U);
     params.pixel_format = PixelFormatFromDepthFormat(format);
     params.component_type = ComponentTypeFromDepthFormat(format);
     params.type = GetFormatType(params.pixel_format);
@@ -135,20 +180,22 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) {
     params.unaligned_height = zeta_height;
     params.target = SurfaceTarget::Texture2D;
     params.depth = 1;
-    params.size_in_bytes_total = params.SizeInBytesTotal();
-    params.size_in_bytes_2d = params.SizeInBytes2D();
     params.max_mip_level = 0;
     params.rt = {};
 
+    params.InitCacheParameters(zeta_address);
+
     return params;
 }
 
 /*static*/ SurfaceParams SurfaceParams::CreateForFermiCopySurface(
     const Tegra::Engines::Fermi2D::Regs::Surface& config) {
     SurfaceParams params{};
-    params.addr = TryGetCpuAddr(config.Address());
+
     params.is_tiled = !config.linear;
-    params.block_height = params.is_tiled ? config.BlockHeight() : 0,
+    params.block_width = params.is_tiled ? std::min(config.BlockWidth(), 32U) : 0,
+    params.block_height = params.is_tiled ? std::min(config.BlockHeight(), 32U) : 0,
+    params.block_depth = params.is_tiled ? std::min(config.BlockDepth(), 32U) : 0,
     params.pixel_format = PixelFormatFromRenderTargetFormat(config.format);
     params.component_type = ComponentTypeFromRenderTarget(config.format);
     params.type = GetFormatType(params.pixel_format);
@@ -157,11 +204,11 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) {
     params.unaligned_height = config.height;
     params.target = SurfaceTarget::Texture2D;
     params.depth = 1;
-    params.size_in_bytes_total = params.SizeInBytesTotal();
-    params.size_in_bytes_2d = params.SizeInBytes2D();
     params.max_mip_level = 0;
     params.rt = {};
 
+    params.InitCacheParameters(config.Address());
+
     return params;
 }
 
@@ -221,6 +268,8 @@ static constexpr std::array<FormatTuple, SurfaceParams::MaxPixelFormat> tex_form
     {GL_RG32UI, GL_RG_INTEGER, GL_UNSIGNED_INT, ComponentType::UInt, false},              // RG32UI
     {GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT, ComponentType::UInt, false},              // R32UI
     {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_8X8
+    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_8X5
+    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_5X4
 
     // Depth formats
     {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT, ComponentType::Float, false}, // Z32F
@@ -264,28 +313,6 @@ static const FormatTuple& GetFormatTuple(PixelFormat pixel_format, ComponentType
     return format;
 }
 
-static bool IsPixelFormatASTC(PixelFormat format) {
-    switch (format) {
-    case PixelFormat::ASTC_2D_4X4:
-    case PixelFormat::ASTC_2D_8X8:
-        return true;
-    default:
-        return false;
-    }
-}
-
-static std::pair<u32, u32> GetASTCBlockSize(PixelFormat format) {
-    switch (format) {
-    case PixelFormat::ASTC_2D_4X4:
-        return {4, 4};
-    case PixelFormat::ASTC_2D_8X8:
-        return {8, 8};
-    default:
-        LOG_CRITICAL(HW_GPU, "Unhandled format: {}", static_cast<u32>(format));
-        UNREACHABLE();
-    }
-}
-
 MathUtil::Rectangle<u32> SurfaceParams::GetRect() const {
     u32 actual_height{unaligned_height};
     if (IsPixelFormatASTC(pixel_format)) {
@@ -313,29 +340,27 @@ static bool IsFormatBCn(PixelFormat format) {
 }
 
 template <bool morton_to_gl, PixelFormat format>
-void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, std::size_t gl_buffer_size,
-                VAddr addr) {
-    constexpr u32 bytes_per_pixel = SurfaceParams::GetFormatBpp(format) / CHAR_BIT;
-    constexpr u32 gl_bytes_per_pixel = CachedSurface::GetGLBytesPerPixel(format);
+void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth, u32 depth, u8* gl_buffer,
+                std::size_t gl_buffer_size, VAddr addr) {
+    constexpr u32 bytes_per_pixel = SurfaceParams::GetBytesPerPixel(format);
+
+    // With the BCn formats (DXT and DXN), each 4x4 tile is swizzled instead of just individual
+    // pixel values.
+    const u32 tile_size{IsFormatBCn(format) ? 4U : 1U};
 
     if (morton_to_gl) {
-        // With the BCn formats (DXT and DXN), each 4x4 tile is swizzled instead of just individual
-        // pixel values.
-        const u32 tile_size{IsFormatBCn(format) ? 4U : 1U};
         const std::vector<u8> data = Tegra::Texture::UnswizzleTexture(
-            addr, tile_size, bytes_per_pixel, stride, height, block_height);
+            addr, tile_size, bytes_per_pixel, stride, height, depth, block_height, block_depth);
         const std::size_t size_to_copy{std::min(gl_buffer_size, data.size())};
         memcpy(gl_buffer, data.data(), size_to_copy);
     } else {
-        // TODO(bunnei): Assumes the default rendering GOB size of 16 (128 lines). We should
-        // check the configuration for this and perform more generic un/swizzle
-        LOG_WARNING(Render_OpenGL, "need to use correct swizzle/GOB parameters!");
-        VideoCore::MortonCopyPixels128(stride, height, bytes_per_pixel, gl_bytes_per_pixel,
-                                       Memory::GetPointer(addr), gl_buffer, morton_to_gl);
+        Tegra::Texture::CopySwizzledData(stride / tile_size, height / tile_size, depth,
+                                         bytes_per_pixel, bytes_per_pixel, Memory::GetPointer(addr),
+                                         gl_buffer, false, block_height, block_depth);
     }
 }
 
-static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr),
+static constexpr std::array<void (*)(u32, u32, u32, u32, u32, u8*, std::size_t, VAddr),
                             SurfaceParams::MaxPixelFormat>
     morton_to_gl_fns = {
         // clang-format off
@@ -385,6 +410,8 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr),
         MortonCopy<true, PixelFormat::RG32UI>,
         MortonCopy<true, PixelFormat::R32UI>,
         MortonCopy<true, PixelFormat::ASTC_2D_8X8>,
+        MortonCopy<true, PixelFormat::ASTC_2D_8X5>,
+        MortonCopy<true, PixelFormat::ASTC_2D_5X4>,
         MortonCopy<true, PixelFormat::Z32F>,
         MortonCopy<true, PixelFormat::Z16>,
         MortonCopy<true, PixelFormat::Z24S8>,
@@ -393,7 +420,7 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr),
         // clang-format on
 };
 
-static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr),
+static constexpr std::array<void (*)(u32, u32, u32, u32, u32, u8*, std::size_t, VAddr),
                             SurfaceParams::MaxPixelFormat>
     gl_to_morton_fns = {
         // clang-format off
@@ -410,17 +437,16 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr),
         MortonCopy<false, PixelFormat::RGBA16UI>,
         MortonCopy<false, PixelFormat::R11FG11FB10F>,
         MortonCopy<false, PixelFormat::RGBA32UI>,
-        // TODO(Subv): Swizzling DXT1/DXT23/DXT45/DXN1/DXN2/BC7U/BC6H_UF16/BC6H_SF16/ASTC_2D_4X4
-        // formats are not supported
-        nullptr,
-        nullptr,
-        nullptr,
-        nullptr,
-        nullptr,
-        nullptr,
-        nullptr,
-        nullptr,
-        nullptr,
+        MortonCopy<false, PixelFormat::DXT1>,
+        MortonCopy<false, PixelFormat::DXT23>,
+        MortonCopy<false, PixelFormat::DXT45>,
+        MortonCopy<false, PixelFormat::DXN1>,
+        MortonCopy<false, PixelFormat::DXN2UNORM>,
+        MortonCopy<false, PixelFormat::DXN2SNORM>,
+        MortonCopy<false, PixelFormat::BC7U>,
+        MortonCopy<false, PixelFormat::BC6H_UF16>,
+        MortonCopy<false, PixelFormat::BC6H_SF16>,
+        // TODO(Subv): Swizzling ASTC formats are not supported
         nullptr,
         MortonCopy<false, PixelFormat::G8R8U>,
         MortonCopy<false, PixelFormat::G8R8S>,
@@ -445,6 +471,8 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr),
         MortonCopy<false, PixelFormat::RG32UI>,
         MortonCopy<false, PixelFormat::R32UI>,
         nullptr,
+        nullptr,
+        nullptr,
         MortonCopy<false, PixelFormat::Z32F>,
         MortonCopy<false, PixelFormat::Z16>,
         MortonCopy<false, PixelFormat::Z24S8>,
@@ -604,22 +632,21 @@ static void CopySurface(const Surface& src_surface, const Surface& dst_surface,
     auto source_format = GetFormatTuple(src_params.pixel_format, src_params.component_type);
     auto dest_format = GetFormatTuple(dst_params.pixel_format, dst_params.component_type);
 
-    std::size_t buffer_size =
-        std::max(src_params.size_in_bytes_total, dst_params.size_in_bytes_total);
+    std::size_t buffer_size = std::max(src_params.size_in_bytes, dst_params.size_in_bytes);
 
     glBindBuffer(GL_PIXEL_PACK_BUFFER, copy_pbo_handle);
     glBufferData(GL_PIXEL_PACK_BUFFER, buffer_size, nullptr, GL_STREAM_DRAW_ARB);
     if (source_format.compressed) {
         glGetCompressedTextureImage(src_surface->Texture().handle, src_attachment,
-                                    static_cast<GLsizei>(src_params.size_in_bytes_total), nullptr);
+                                    static_cast<GLsizei>(src_params.size_in_bytes), nullptr);
     } else {
         glGetTextureImage(src_surface->Texture().handle, src_attachment, source_format.format,
-                          source_format.type, static_cast<GLsizei>(src_params.size_in_bytes_total),
+                          source_format.type, static_cast<GLsizei>(src_params.size_in_bytes),
                           nullptr);
     }
     // If the new texture is bigger than the previous one, we need to fill in the rest with data
     // from the CPU.
-    if (src_params.size_in_bytes_total < dst_params.size_in_bytes_total) {
+    if (src_params.size_in_bytes < dst_params.size_in_bytes) {
         // Upload the rest of the memory.
         if (dst_params.is_tiled) {
             // TODO(Subv): We might have to de-tile the subtexture and re-tile it with the rest
@@ -629,12 +656,12 @@ static void CopySurface(const Surface& src_surface, const Surface& dst_surface,
             LOG_DEBUG(HW_GPU, "Trying to upload extra texture data from the CPU during "
                               "reinterpretation but the texture is tiled.");
         }
-        std::size_t remaining_size =
-            dst_params.size_in_bytes_total - src_params.size_in_bytes_total;
+        std::size_t remaining_size = dst_params.size_in_bytes - src_params.size_in_bytes;
         std::vector<u8> data(remaining_size);
-        Memory::ReadBlock(dst_params.addr + src_params.size_in_bytes_total, data.data(),
-                          data.size());
-        glBufferSubData(GL_PIXEL_PACK_BUFFER, src_params.size_in_bytes_total, remaining_size,
+        std::memcpy(data.data(), Memory::GetPointer(dst_params.addr + src_params.size_in_bytes),
+                    data.size());
+
+        glBufferSubData(GL_PIXEL_PACK_BUFFER, src_params.size_in_bytes, remaining_size,
                         data.data());
     }
 
@@ -680,7 +707,8 @@ static void CopySurface(const Surface& src_surface, const Surface& dst_surface,
 }
 
 CachedSurface::CachedSurface(const SurfaceParams& params)
-    : params(params), gl_target(SurfaceTargetToGL(params.target)) {
+    : params(params), gl_target(SurfaceTargetToGL(params.target)),
+      cached_size_in_bytes(params.size_in_bytes) {
     texture.Create();
     const auto& rect{params.GetRect()};
 
@@ -730,9 +758,21 @@ CachedSurface::CachedSurface(const SurfaceParams& params)
 
     VideoCore::LabelGLObject(GL_TEXTURE, texture.handle, params.addr,
                              SurfaceParams::SurfaceTargetName(params.target));
+
+    // Clamp size to mapped GPU memory region
+    // TODO(bunnei): Super Mario Odyssey maps a 0x40000 byte region and then uses it for a 0x80000
+    // R32F render buffer. We do not yet know if this is a game bug or something else, but this
+    // check is necessary to prevent flushing from overwriting unmapped memory.
+
+    auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
+    const u64 max_size{memory_manager.GetRegionEnd(params.gpu_addr) - params.gpu_addr};
+    if (cached_size_in_bytes > max_size) {
+        LOG_ERROR(HW_GPU, "Surface size {} exceeds region size {}", params.size_in_bytes, max_size);
+        cached_size_in_bytes = max_size;
+    }
 }
 
-static void ConvertS8Z24ToZ24S8(std::vector<u8>& data, u32 width, u32 height) {
+static void ConvertS8Z24ToZ24S8(std::vector<u8>& data, u32 width, u32 height, bool reverse) {
     union S8Z24 {
         BitField<0, 24, u32> z24;
         BitField<24, 8, u32> s8;
@@ -745,22 +785,29 @@ static void ConvertS8Z24ToZ24S8(std::vector<u8>& data, u32 width, u32 height) {
     };
     static_assert(sizeof(Z24S8) == 4, "Z24S8 is incorrect size");
 
-    S8Z24 input_pixel{};
-    Z24S8 output_pixel{};
-    constexpr auto bpp{CachedSurface::GetGLBytesPerPixel(PixelFormat::S8Z24)};
+    S8Z24 s8z24_pixel{};
+    Z24S8 z24s8_pixel{};
+    constexpr auto bpp{SurfaceParams::GetBytesPerPixel(PixelFormat::S8Z24)};
     for (std::size_t y = 0; y < height; ++y) {
         for (std::size_t x = 0; x < width; ++x) {
             const std::size_t offset{bpp * (y * width + x)};
-            std::memcpy(&input_pixel, &data[offset], sizeof(S8Z24));
-            output_pixel.s8.Assign(input_pixel.s8);
-            output_pixel.z24.Assign(input_pixel.z24);
-            std::memcpy(&data[offset], &output_pixel, sizeof(Z24S8));
+            if (reverse) {
+                std::memcpy(&z24s8_pixel, &data[offset], sizeof(Z24S8));
+                s8z24_pixel.s8.Assign(z24s8_pixel.s8);
+                s8z24_pixel.z24.Assign(z24s8_pixel.z24);
+                std::memcpy(&data[offset], &s8z24_pixel, sizeof(S8Z24));
+            } else {
+                std::memcpy(&s8z24_pixel, &data[offset], sizeof(S8Z24));
+                z24s8_pixel.s8.Assign(s8z24_pixel.s8);
+                z24s8_pixel.z24.Assign(s8z24_pixel.z24);
+                std::memcpy(&data[offset], &z24s8_pixel, sizeof(Z24S8));
+            }
         }
     }
 }
 
 static void ConvertG8R8ToR8G8(std::vector<u8>& data, u32 width, u32 height) {
-    constexpr auto bpp{CachedSurface::GetGLBytesPerPixel(PixelFormat::G8R8U)};
+    constexpr auto bpp{SurfaceParams::GetBytesPerPixel(PixelFormat::G8R8U)};
     for (std::size_t y = 0; y < height; ++y) {
         for (std::size_t x = 0; x < width; ++x) {
             const std::size_t offset{bpp * (y * width + x)};
@@ -780,7 +827,9 @@ static void ConvertFormatAsNeeded_LoadGLBuffer(std::vector<u8>& data, PixelForma
                                                u32 width, u32 height) {
     switch (pixel_format) {
     case PixelFormat::ASTC_2D_4X4:
-    case PixelFormat::ASTC_2D_8X8: {
+    case PixelFormat::ASTC_2D_8X8:
+    case PixelFormat::ASTC_2D_8X5:
+    case PixelFormat::ASTC_2D_5X4: {
         // Convert ASTC pixel formats to RGBA8, as most desktop GPUs do not support ASTC.
         u32 block_width{};
         u32 block_height{};
@@ -790,7 +839,7 @@ static void ConvertFormatAsNeeded_LoadGLBuffer(std::vector<u8>& data, PixelForma
     }
     case PixelFormat::S8Z24:
         // Convert the S8Z24 depth format to Z24S8, as OpenGL does not support S8Z24.
-        ConvertS8Z24ToZ24S8(data, width, height);
+        ConvertS8Z24ToZ24S8(data, width, height, false);
         break;
 
     case PixelFormat::G8R8U:
@@ -801,49 +850,54 @@ static void ConvertFormatAsNeeded_LoadGLBuffer(std::vector<u8>& data, PixelForma
     }
 }
 
+/**
+ * Helper function to perform software conversion (as needed) when flushing a buffer from OpenGL to
+ * Switch memory. This is for Maxwell pixel formats that cannot be represented as-is in OpenGL or
+ * with typical desktop GPUs.
+ */
+static void ConvertFormatAsNeeded_FlushGLBuffer(std::vector<u8>& data, PixelFormat pixel_format,
+                                                u32 width, u32 height) {
+    switch (pixel_format) {
+    case PixelFormat::G8R8U:
+    case PixelFormat::G8R8S:
+    case PixelFormat::ASTC_2D_4X4:
+    case PixelFormat::ASTC_2D_8X8: {
+        LOG_CRITICAL(HW_GPU, "Conversion of format {} after texture flushing is not implemented",
+                     static_cast<u32>(pixel_format));
+        UNREACHABLE();
+        break;
+    }
+    case PixelFormat::S8Z24:
+        // Convert the Z24S8 depth format to S8Z24, as OpenGL does not support S8Z24.
+        ConvertS8Z24ToZ24S8(data, width, height, true);
+        break;
+    }
+}
+
 MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 64, 192));
 void CachedSurface::LoadGLBuffer() {
-    ASSERT(params.type != SurfaceType::Fill);
-
-    const u8* const texture_src_data = Memory::GetPointer(params.addr);
-
-    ASSERT(texture_src_data);
-
-    const u32 bytes_per_pixel = GetGLBytesPerPixel(params.pixel_format);
-    const u32 copy_size = params.width * params.height * bytes_per_pixel;
-    const std::size_t total_size = copy_size * params.depth;
-
     MICROPROFILE_SCOPE(OpenGL_SurfaceLoad);
 
+    gl_buffer.resize(params.size_in_bytes_gl);
     if (params.is_tiled) {
-        gl_buffer.resize(total_size);
+        u32 depth = params.depth;
+        u32 block_depth = params.block_depth;
 
-        // TODO(bunnei): This only unswizzles and copies a 2D texture - we do not yet know how to do
-        // this for 3D textures, etc.
-        switch (params.target) {
-        case SurfaceParams::SurfaceTarget::Texture2D:
-            // Pass impl. to the fallback code below
-            break;
-        case SurfaceParams::SurfaceTarget::Texture2DArray:
-        case SurfaceParams::SurfaceTarget::TextureCubemap:
-            for (std::size_t index = 0; index < params.depth; ++index) {
-                const std::size_t offset{index * copy_size};
-                morton_to_gl_fns[static_cast<std::size_t>(params.pixel_format)](
-                    params.width, params.block_height, params.height, gl_buffer.data() + offset,
-                    copy_size, params.addr + offset);
-            }
-            break;
-        default:
-            LOG_CRITICAL(HW_GPU, "Unimplemented tiled load for target={}",
-                         static_cast<u32>(params.target));
-            UNREACHABLE();
+        ASSERT_MSG(params.block_width == 1, "Block width is defined as {} on texture type {}",
+                   params.block_width, static_cast<u32>(params.target));
+
+        if (params.target == SurfaceParams::SurfaceTarget::Texture2D) {
+            // TODO(Blinkhawk): Eliminate this condition once all texture types are implemented.
+            depth = 1U;
+            block_depth = 1U;
         }
 
         morton_to_gl_fns[static_cast<std::size_t>(params.pixel_format)](
-            params.width, params.block_height, params.height, gl_buffer.data(), copy_size,
-            params.addr);
+            params.width, params.block_height, params.height, block_depth, depth, gl_buffer.data(),
+            gl_buffer.size(), params.addr);
     } else {
-        const u8* const texture_src_data_end{texture_src_data + total_size};
+        const auto texture_src_data{Memory::GetPointer(params.addr)};
+        const auto texture_src_data_end{texture_src_data + params.size_in_bytes_gl};
         gl_buffer.assign(texture_src_data, texture_src_data_end);
     }
 
@@ -852,7 +906,44 @@ void CachedSurface::LoadGLBuffer() {
 
 MICROPROFILE_DEFINE(OpenGL_SurfaceFlush, "OpenGL", "Surface Flush", MP_RGB(128, 192, 64));
 void CachedSurface::FlushGLBuffer() {
-    ASSERT_MSG(false, "Unimplemented");
+    MICROPROFILE_SCOPE(OpenGL_SurfaceFlush);
+
+    ASSERT_MSG(!IsPixelFormatASTC(params.pixel_format), "Unimplemented");
+
+    // OpenGL temporary buffer needs to be big enough to store raw texture size
+    gl_buffer.resize(GetSizeInBytes());
+
+    const FormatTuple& tuple = GetFormatTuple(params.pixel_format, params.component_type);
+    // Ensure no bad interactions with GL_UNPACK_ALIGNMENT
+    ASSERT(params.width * SurfaceParams::GetBytesPerPixel(params.pixel_format) % 4 == 0);
+    glPixelStorei(GL_PACK_ROW_LENGTH, static_cast<GLint>(params.width));
+    ASSERT(!tuple.compressed);
+    glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
+    glGetTextureImage(texture.handle, 0, tuple.format, tuple.type, gl_buffer.size(),
+                      gl_buffer.data());
+    glPixelStorei(GL_PACK_ROW_LENGTH, 0);
+    ConvertFormatAsNeeded_FlushGLBuffer(gl_buffer, params.pixel_format, params.width,
+                                        params.height);
+    ASSERT(params.type != SurfaceType::Fill);
+    const u8* const texture_src_data = Memory::GetPointer(params.addr);
+    ASSERT(texture_src_data);
+    if (params.is_tiled) {
+        u32 depth = params.depth;
+        u32 block_depth = params.block_depth;
+
+        ASSERT_MSG(params.block_width == 1, "Block width is defined as {} on texture type {}",
+                   params.block_width, static_cast<u32>(params.target));
+
+        if (params.target == SurfaceParams::SurfaceTarget::Texture2D) {
+            // TODO(Blinkhawk): Eliminate this condition once all texture types are implemented.
+            depth = 1U;
+        }
+        gl_to_morton_fns[static_cast<size_t>(params.pixel_format)](
+            params.width, params.block_height, params.height, block_depth, depth, gl_buffer.data(),
+            gl_buffer.size(), GetAddr());
+    } else {
+        std::memcpy(Memory::GetPointer(GetAddr()), gl_buffer.data(), GetSizeInBytes());
+    }
 }
 
 MICROPROFILE_DEFINE(OpenGL_TextureUL, "OpenGL", "Texture Upload", MP_RGB(128, 64, 192));
@@ -862,9 +953,6 @@ void CachedSurface::UploadGLTexture(GLuint read_fb_handle, GLuint draw_fb_handle
 
     MICROPROFILE_SCOPE(OpenGL_TextureUL);
 
-    ASSERT(gl_buffer.size() == static_cast<std::size_t>(params.width) * params.height *
-                                   GetGLBytesPerPixel(params.pixel_format) * params.depth);
-
     const auto& rect{params.GetRect()};
 
     // Load data from memory to the surface
@@ -873,7 +961,7 @@ void CachedSurface::UploadGLTexture(GLuint read_fb_handle, GLuint draw_fb_handle
     std::size_t buffer_offset =
         static_cast<std::size_t>(static_cast<std::size_t>(y0) * params.width +
                                  static_cast<std::size_t>(x0)) *
-        GetGLBytesPerPixel(params.pixel_format);
+        SurfaceParams::GetBytesPerPixel(params.pixel_format);
 
     const FormatTuple& tuple = GetFormatTuple(params.pixel_format, params.component_type);
     const GLuint target_tex = texture.handle;
@@ -889,7 +977,7 @@ void CachedSurface::UploadGLTexture(GLuint read_fb_handle, GLuint draw_fb_handle
     cur_state.Apply();
 
     // Ensure no bad interactions with GL_UNPACK_ALIGNMENT
-    ASSERT(params.width * GetGLBytesPerPixel(params.pixel_format) % 4 == 0);
+    ASSERT(params.width * SurfaceParams::GetBytesPerPixel(params.pixel_format) % 4 == 0);
     glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast<GLint>(params.width));
 
     glActiveTexture(GL_TEXTURE0);
@@ -899,7 +987,7 @@ void CachedSurface::UploadGLTexture(GLuint read_fb_handle, GLuint draw_fb_handle
             glCompressedTexImage2D(
                 SurfaceTargetToGL(params.target), 0, tuple.internal_format,
                 static_cast<GLsizei>(params.width), static_cast<GLsizei>(params.height), 0,
-                static_cast<GLsizei>(params.size_in_bytes_2d), &gl_buffer[buffer_offset]);
+                static_cast<GLsizei>(params.size_in_bytes_gl), &gl_buffer[buffer_offset]);
             break;
         case SurfaceParams::SurfaceTarget::Texture3D:
         case SurfaceParams::SurfaceTarget::Texture2DArray:
@@ -907,16 +995,16 @@ void CachedSurface::UploadGLTexture(GLuint read_fb_handle, GLuint draw_fb_handle
                 SurfaceTargetToGL(params.target), 0, tuple.internal_format,
                 static_cast<GLsizei>(params.width), static_cast<GLsizei>(params.height),
                 static_cast<GLsizei>(params.depth), 0,
-                static_cast<GLsizei>(params.size_in_bytes_total), &gl_buffer[buffer_offset]);
+                static_cast<GLsizei>(params.size_in_bytes_gl), &gl_buffer[buffer_offset]);
             break;
         case SurfaceParams::SurfaceTarget::TextureCubemap:
             for (std::size_t face = 0; face < params.depth; ++face) {
                 glCompressedTexImage2D(static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + face),
                                        0, tuple.internal_format, static_cast<GLsizei>(params.width),
                                        static_cast<GLsizei>(params.height), 0,
-                                       static_cast<GLsizei>(params.size_in_bytes_2d),
+                                       static_cast<GLsizei>(params.SizeInBytesCubeFaceGL()),
                                        &gl_buffer[buffer_offset]);
-                buffer_offset += params.size_in_bytes_2d;
+                buffer_offset += params.SizeInBytesCubeFace();
             }
             break;
         default:
@@ -926,7 +1014,7 @@ void CachedSurface::UploadGLTexture(GLuint read_fb_handle, GLuint draw_fb_handle
             glCompressedTexImage2D(
                 GL_TEXTURE_2D, 0, tuple.internal_format, static_cast<GLsizei>(params.width),
                 static_cast<GLsizei>(params.height), 0,
-                static_cast<GLsizei>(params.size_in_bytes_2d), &gl_buffer[buffer_offset]);
+                static_cast<GLsizei>(params.size_in_bytes_gl), &gl_buffer[buffer_offset]);
         }
     } else {
 
@@ -955,7 +1043,7 @@ void CachedSurface::UploadGLTexture(GLuint read_fb_handle, GLuint draw_fb_handle
                                 y0, static_cast<GLsizei>(rect.GetWidth()),
                                 static_cast<GLsizei>(rect.GetHeight()), tuple.format, tuple.type,
                                 &gl_buffer[buffer_offset]);
-                buffer_offset += params.size_in_bytes_2d;
+                buffer_offset += params.SizeInBytesCubeFace();
             }
             break;
         default:
@@ -989,7 +1077,9 @@ Surface RasterizerCacheOpenGL::GetDepthBufferSurface(bool preserve_contents) {
     }
 
     SurfaceParams depth_params{SurfaceParams::CreateForDepthBuffer(
-        regs.zeta_width, regs.zeta_height, regs.zeta.Address(), regs.zeta.format)};
+        regs.zeta_width, regs.zeta_height, regs.zeta.Address(), regs.zeta.format,
+        regs.zeta.memory_layout.block_width, regs.zeta.memory_layout.block_height,
+        regs.zeta.memory_layout.block_depth, regs.zeta.memory_layout.type)};
 
     return GetSurface(depth_params, preserve_contents);
 }
@@ -1015,10 +1105,7 @@ Surface RasterizerCacheOpenGL::GetColorBufferSurface(std::size_t index, bool pre
 void RasterizerCacheOpenGL::LoadSurface(const Surface& surface) {
     surface->LoadGLBuffer();
     surface->UploadGLTexture(read_framebuffer.handle, draw_framebuffer.handle);
-}
-
-void RasterizerCacheOpenGL::FlushSurface(const Surface& surface) {
-    surface->FlushGLBuffer();
+    surface->MarkAsModified(false, *this);
 }
 
 Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, bool preserve_contents) {
@@ -1087,6 +1174,14 @@ void RasterizerCacheOpenGL::FermiCopySurface(
     FastCopySurface(GetSurface(src_params, true), GetSurface(dst_params, false));
 }
 
+void RasterizerCacheOpenGL::AccurateCopySurface(const Surface& src_surface,
+                                                const Surface& dst_surface) {
+    const auto& src_params{src_surface->GetSurfaceParams()};
+    const auto& dst_params{dst_surface->GetSurfaceParams()};
+    FlushRegion(src_params.addr, dst_params.size_in_bytes);
+    LoadSurface(dst_surface);
+}
+
 Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface,
                                                const SurfaceParams& new_params) {
     // Verify surface is compatible for blitting
@@ -1095,6 +1190,12 @@ Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface,
     // Get a new surface with the new parameters, and blit the previous surface to it
     Surface new_surface{GetUncachedSurface(new_params)};
 
+    // With use_accurate_gpu_emulation enabled, do an accurate surface copy
+    if (Settings::values.use_accurate_gpu_emulation) {
+        AccurateCopySurface(old_surface, new_surface);
+        return new_surface;
+    }
+
     // For compatible surfaces, we can just do fast glCopyImageSubData based copy
     if (old_params.target == new_params.target && old_params.type == new_params.type &&
         old_params.depth == new_params.depth && old_params.depth == 1 &&
@@ -1106,11 +1207,10 @@ Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface,
 
     // If the format is the same, just do a framebuffer blit. This is significantly faster than
     // using PBOs. The is also likely less accurate, as textures will be converted rather than
-    // reinterpreted. When use_accurate_framebuffers setting is enabled, perform a more accurate
+    // reinterpreted. When use_accurate_gpu_emulation setting is enabled, perform a more accurate
     // surface copy, where pixels are reinterpreted as a new format (without conversion). This
     // code path uses OpenGL PBOs and is quite slow.
-    const bool is_blit{old_params.pixel_format == new_params.pixel_format ||
-                       !Settings::values.use_accurate_framebuffers};
+    const bool is_blit{old_params.pixel_format == new_params.pixel_format};
 
     switch (new_params.target) {
     case SurfaceParams::SurfaceTarget::Texture2D:
diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h
index 0b4940b3c..7c1cb72d0 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h
@@ -18,6 +18,7 @@
 #include "video_core/rasterizer_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/renderer_opengl/gl_shader_gen.h"
+#include "video_core/textures/decoders.h"
 #include "video_core/textures/texture.h"
 
 namespace OpenGL {
@@ -74,19 +75,21 @@ struct SurfaceParams {
         RG32UI = 43,
         R32UI = 44,
         ASTC_2D_8X8 = 45,
+        ASTC_2D_8X5 = 46,
+        ASTC_2D_5X4 = 47,
 
         MaxColorFormat,
 
         // Depth formats
-        Z32F = 46,
-        Z16 = 47,
+        Z32F = 48,
+        Z16 = 49,
 
         MaxDepthFormat,
 
         // DepthStencil formats
-        Z24S8 = 48,
-        S8Z24 = 49,
-        Z32FS8 = 50,
+        Z24S8 = 50,
+        S8Z24 = 51,
+        Z32FS8 = 52,
 
         MaxDepthStencilFormat,
 
@@ -220,6 +223,8 @@ struct SurfaceParams {
             1, // RG32UI
             1, // R32UI
             4, // ASTC_2D_8X8
+            4, // ASTC_2D_8X5
+            4, // ASTC_2D_5X4
             1, // Z32F
             1, // Z16
             1, // Z24S8
@@ -282,6 +287,8 @@ struct SurfaceParams {
             64,  // RG32UI
             32,  // R32UI
             16,  // ASTC_2D_8X8
+            32,  // ASTC_2D_8X5
+            32,  // ASTC_2D_5X4
             32,  // Z32F
             16,  // Z16
             32,  // Z24S8
@@ -553,8 +560,12 @@ struct SurfaceParams {
             return PixelFormat::BC6H_SF16;
         case Tegra::Texture::TextureFormat::ASTC_2D_4X4:
             return PixelFormat::ASTC_2D_4X4;
+        case Tegra::Texture::TextureFormat::ASTC_2D_5X4:
+            return PixelFormat::ASTC_2D_5X4;
         case Tegra::Texture::TextureFormat::ASTC_2D_8X8:
             return PixelFormat::ASTC_2D_8X8;
+        case Tegra::Texture::TextureFormat::ASTC_2D_8X5:
+            return PixelFormat::ASTC_2D_8X5;
         case Tegra::Texture::TextureFormat::R16_G16:
             switch (component_type) {
             case Tegra::Texture::ComponentType::FLOAT:
@@ -691,21 +702,42 @@ struct SurfaceParams {
         return SurfaceType::Invalid;
     }
 
+    /// Returns the sizer in bytes of the specified pixel format
+    static constexpr u32 GetBytesPerPixel(PixelFormat pixel_format) {
+        if (pixel_format == SurfaceParams::PixelFormat::Invalid) {
+            return 0;
+        }
+        return GetFormatBpp(pixel_format) / CHAR_BIT;
+    }
+
     /// Returns the rectangle corresponding to this surface
     MathUtil::Rectangle<u32> GetRect() const;
 
-    /// Returns the size of this surface as a 2D texture in bytes, adjusted for compression
-    std::size_t SizeInBytes2D() const {
+    /// Returns the total size of this surface in bytes, adjusted for compression
+    std::size_t SizeInBytesRaw(bool ignore_tiled = false) const {
         const u32 compression_factor{GetCompressionFactor(pixel_format)};
-        ASSERT(width % compression_factor == 0);
-        ASSERT(height % compression_factor == 0);
-        return (width / compression_factor) * (height / compression_factor) *
-               GetFormatBpp(pixel_format) / CHAR_BIT;
+        const u32 bytes_per_pixel{GetBytesPerPixel(pixel_format)};
+        const size_t uncompressed_size{
+            Tegra::Texture::CalculateSize((ignore_tiled ? false : is_tiled), bytes_per_pixel, width,
+                                          height, depth, block_height, block_depth)};
+
+        // Divide by compression_factor^2, as height and width are factored by this
+        return uncompressed_size / (compression_factor * compression_factor);
     }
 
-    /// Returns the total size of this surface in bytes, adjusted for compression
-    std::size_t SizeInBytesTotal() const {
-        return SizeInBytes2D() * depth;
+    /// Returns the size of this surface as an OpenGL texture in bytes
+    std::size_t SizeInBytesGL() const {
+        return SizeInBytesRaw(true);
+    }
+
+    /// Returns the size of this surface as a cube face in bytes
+    std::size_t SizeInBytesCubeFace() const {
+        return size_in_bytes / 6;
+    }
+
+    /// Returns the size of this surface as an OpenGL cube face in bytes
+    std::size_t SizeInBytesCubeFaceGL() const {
+        return size_in_bytes_gl / 6;
     }
 
     /// Creates SurfaceParams from a texture configuration
@@ -716,9 +748,10 @@ struct SurfaceParams {
     static SurfaceParams CreateForFramebuffer(std::size_t index);
 
     /// Creates SurfaceParams for a depth buffer configuration
-    static SurfaceParams CreateForDepthBuffer(u32 zeta_width, u32 zeta_height,
-                                              Tegra::GPUVAddr zeta_address,
-                                              Tegra::DepthFormat format);
+    static SurfaceParams CreateForDepthBuffer(
+        u32 zeta_width, u32 zeta_height, Tegra::GPUVAddr zeta_address, Tegra::DepthFormat format,
+        u32 block_width, u32 block_height, u32 block_depth,
+        Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout type);
 
     /// Creates SurfaceParams for a Fermi2D surface copy
     static SurfaceParams CreateForFermiCopySurface(
@@ -731,9 +764,13 @@ struct SurfaceParams {
                         other.depth);
     }
 
-    VAddr addr;
+    /// Initializes parameters for caching, should be called after everything has been initialized
+    void InitCacheParameters(Tegra::GPUVAddr gpu_addr);
+
     bool is_tiled;
+    u32 block_width;
     u32 block_height;
+    u32 block_depth;
     PixelFormat pixel_format;
     ComponentType component_type;
     SurfaceType type;
@@ -741,11 +778,15 @@ struct SurfaceParams {
     u32 height;
     u32 depth;
     u32 unaligned_height;
-    std::size_t size_in_bytes_total;
-    std::size_t size_in_bytes_2d;
     SurfaceTarget target;
     u32 max_mip_level;
 
+    // Parameters used for caching
+    VAddr addr;
+    Tegra::GPUVAddr gpu_addr;
+    std::size_t size_in_bytes;
+    std::size_t size_in_bytes_gl;
+
     // Render target specific parameters, not used in caching
     struct {
         u32 index;
@@ -762,7 +803,8 @@ struct SurfaceReserveKey : Common::HashableStruct<OpenGL::SurfaceParams> {
     static SurfaceReserveKey Create(const OpenGL::SurfaceParams& params) {
         SurfaceReserveKey res;
         res.state = params;
-        res.state.rt = {}; // Ignore rt config in caching
+        res.state.gpu_addr = {}; // Ignore GPU vaddr in caching
+        res.state.rt = {};       // Ignore rt config in caching
         return res;
     }
 };
@@ -777,16 +819,20 @@ struct hash<SurfaceReserveKey> {
 
 namespace OpenGL {
 
-class CachedSurface final {
+class CachedSurface final : public RasterizerCacheObject {
 public:
     CachedSurface(const SurfaceParams& params);
 
-    VAddr GetAddr() const {
+    VAddr GetAddr() const override {
         return params.addr;
     }
 
-    std::size_t GetSizeInBytes() const {
-        return params.size_in_bytes_total;
+    std::size_t GetSizeInBytes() const override {
+        return cached_size_in_bytes;
+    }
+
+    void Flush() override {
+        FlushGLBuffer();
     }
 
     const OGLTexture& Texture() const {
@@ -797,13 +843,6 @@ public:
         return gl_target;
     }
 
-    static constexpr unsigned int GetGLBytesPerPixel(SurfaceParams::PixelFormat format) {
-        if (format == SurfaceParams::PixelFormat::Invalid)
-            return 0;
-
-        return SurfaceParams::GetFormatBpp(format) / CHAR_BIT;
-    }
-
     const SurfaceParams& GetSurfaceParams() const {
         return params;
     }
@@ -820,6 +859,7 @@ private:
     std::vector<u8> gl_buffer;
     SurfaceParams params;
     GLenum gl_target;
+    std::size_t cached_size_in_bytes;
 };
 
 class RasterizerCacheOpenGL final : public RasterizerCache<Surface> {
@@ -836,9 +876,6 @@ public:
     /// Get the color surface based on the framebuffer configuration and the specified render target
     Surface GetColorBufferSurface(std::size_t index, bool preserve_contents);
 
-    /// Flushes the surface to Switch memory
-    void FlushSurface(const Surface& surface);
-
     /// Tries to find a framebuffer using on the provided CPU address
     Surface TryFindFramebufferSurface(VAddr addr) const;
 
@@ -862,6 +899,9 @@ private:
     /// Tries to get a reserved surface for the specified parameters
     Surface TryGetReservedSurface(const SurfaceParams& params);
 
+    /// Performs a slow but accurate surface copy, flushing to RAM and reinterpreting the data
+    void AccurateCopySurface(const Surface& src_surface, const Surface& dst_surface);
+
     /// The surface reserve is a "backup" cache, this is where we put unique surfaces that have
     /// previously been used. This is to prevent surfaces from being constantly created and
     /// destroyed when used with different surface parameters.
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index 7cd8f91e4..1a03a677f 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -68,6 +68,10 @@ CachedShader::CachedShader(VAddr addr, Maxwell::ShaderProgram program_type)
         program_result = GLShader::GenerateVertexShader(setup);
         gl_type = GL_VERTEX_SHADER;
         break;
+    case Maxwell::ShaderProgram::Geometry:
+        program_result = GLShader::GenerateGeometryShader(setup);
+        gl_type = GL_GEOMETRY_SHADER;
+        break;
     case Maxwell::ShaderProgram::Fragment:
         program_result = GLShader::GenerateFragmentShader(setup);
         gl_type = GL_FRAGMENT_SHADER;
@@ -80,11 +84,16 @@ CachedShader::CachedShader(VAddr addr, Maxwell::ShaderProgram program_type)
 
     entries = program_result.second;
 
-    OGLShader shader;
-    shader.Create(program_result.first.c_str(), gl_type);
-    program.Create(true, shader.handle);
-    SetShaderUniformBlockBindings(program.handle);
-    VideoCore::LabelGLObject(GL_PROGRAM, program.handle, addr);
+    if (program_type != Maxwell::ShaderProgram::Geometry) {
+        OGLShader shader;
+        shader.Create(program_result.first.c_str(), gl_type);
+        program.Create(true, shader.handle);
+        SetShaderUniformBlockBindings(program.handle);
+        VideoCore::LabelGLObject(GL_PROGRAM, program.handle, addr);
+    } else {
+        // Store shader's code to lazily build it on draw
+        geometry_programs.code = program_result.first;
+    }
 }
 
 GLuint CachedShader::GetProgramResourceIndex(const GLShader::ConstBufferEntry& buffer) {
@@ -110,6 +119,21 @@ GLint CachedShader::GetUniformLocation(const GLShader::SamplerEntry& sampler) {
     return search->second;
 }
 
+GLuint CachedShader::LazyGeometryProgram(OGLProgram& target_program,
+                                         const std::string& glsl_topology,
+                                         const std::string& debug_name) {
+    if (target_program.handle != 0) {
+        return target_program.handle;
+    }
+    const std::string source{geometry_programs.code + "layout (" + glsl_topology + ") in;\n"};
+    OGLShader shader;
+    shader.Create(source.c_str(), GL_GEOMETRY_SHADER);
+    target_program.Create(true, shader.handle);
+    SetShaderUniformBlockBindings(target_program.handle);
+    VideoCore::LabelGLObject(GL_PROGRAM, target_program.handle, addr, debug_name);
+    return target_program.handle;
+};
+
 Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
     const VAddr program_addr{GetShaderAddress(program)};
 
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h
index 9bafe43a9..a210f1731 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.h
+++ b/src/video_core/renderer_opengl/gl_shader_cache.h
@@ -7,6 +7,7 @@
 #include <map>
 #include <memory>
 
+#include "common/assert.h"
 #include "common/common_types.h"
 #include "video_core/rasterizer_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
@@ -18,28 +19,52 @@ class CachedShader;
 using Shader = std::shared_ptr<CachedShader>;
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 
-class CachedShader final {
+class CachedShader final : public RasterizerCacheObject {
 public:
     CachedShader(VAddr addr, Maxwell::ShaderProgram program_type);
 
-    /// Gets the address of the shader in guest memory, required for cache management
-    VAddr GetAddr() const {
+    VAddr GetAddr() const override {
         return addr;
     }
 
-    /// Gets the size of the shader in guest memory, required for cache management
-    std::size_t GetSizeInBytes() const {
+    std::size_t GetSizeInBytes() const override {
         return GLShader::MAX_PROGRAM_CODE_LENGTH * sizeof(u64);
     }
 
+    // We do not have to flush this cache as things in it are never modified by us.
+    void Flush() override {}
+
     /// Gets the shader entries for the shader
     const GLShader::ShaderEntries& GetShaderEntries() const {
         return entries;
     }
 
     /// Gets the GL program handle for the shader
-    GLuint GetProgramHandle() const {
-        return program.handle;
+    GLuint GetProgramHandle(GLenum primitive_mode) {
+        if (program_type != Maxwell::ShaderProgram::Geometry) {
+            return program.handle;
+        }
+        switch (primitive_mode) {
+        case GL_POINTS:
+            return LazyGeometryProgram(geometry_programs.points, "points", "ShaderPoints");
+        case GL_LINES:
+        case GL_LINE_STRIP:
+            return LazyGeometryProgram(geometry_programs.lines, "lines", "ShaderLines");
+        case GL_LINES_ADJACENCY:
+        case GL_LINE_STRIP_ADJACENCY:
+            return LazyGeometryProgram(geometry_programs.lines_adjacency, "lines_adjacency",
+                                       "ShaderLinesAdjacency");
+        case GL_TRIANGLES:
+        case GL_TRIANGLE_STRIP:
+        case GL_TRIANGLE_FAN:
+            return LazyGeometryProgram(geometry_programs.triangles, "triangles", "ShaderTriangles");
+        case GL_TRIANGLES_ADJACENCY:
+        case GL_TRIANGLE_STRIP_ADJACENCY:
+            return LazyGeometryProgram(geometry_programs.triangles_adjacency, "triangles_adjacency",
+                                       "ShaderLines");
+        default:
+            UNREACHABLE_MSG("Unknown primitive mode.");
+        }
     }
 
     /// Gets the GL program resource location for the specified resource, caching as needed
@@ -49,12 +74,30 @@ public:
     GLint GetUniformLocation(const GLShader::SamplerEntry& sampler);
 
 private:
+    /// Generates a geometry shader or returns one that already exists.
+    GLuint LazyGeometryProgram(OGLProgram& target_program, const std::string& glsl_topology,
+                               const std::string& debug_name);
+
     VAddr addr;
     Maxwell::ShaderProgram program_type;
     GLShader::ShaderSetup setup;
     GLShader::ShaderEntries entries;
+
+    // Non-geometry program.
     OGLProgram program;
 
+    // Geometry programs. These are needed because GLSL needs an input topology but it's not
+    // declared by the hardware. Workaround this issue by generating a different shader per input
+    // topology class.
+    struct {
+        std::string code;
+        OGLProgram points;
+        OGLProgram lines;
+        OGLProgram lines_adjacency;
+        OGLProgram triangles;
+        OGLProgram triangles_adjacency;
+    } geometry_programs;
+
     std::map<u32, GLuint> resource_cache;
     std::map<u32, GLint> uniform_cache;
 };
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 85c668ca1..55c33c3a9 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -7,6 +7,7 @@
 #include <string>
 #include <string_view>
 
+#include <boost/optional.hpp>
 #include <fmt/format.h>
 
 #include "common/assert.h"
@@ -29,11 +30,32 @@ using Tegra::Shader::SubOp;
 constexpr u32 PROGRAM_END = MAX_PROGRAM_CODE_LENGTH;
 constexpr u32 PROGRAM_HEADER_SIZE = sizeof(Tegra::Shader::Header);
 
+enum : u32 { POSITION_VARYING_LOCATION = 0, GENERIC_VARYING_START_LOCATION = 1 };
+
+constexpr u32 MAX_GEOMETRY_BUFFERS = 6;
+constexpr u32 MAX_ATTRIBUTES = 0x100; // Size in vec4s, this value is untested
+
 class DecompileFail : public std::runtime_error {
 public:
     using std::runtime_error::runtime_error;
 };
 
+/// Translate topology
+static std::string GetTopologyName(Tegra::Shader::OutputTopology topology) {
+    switch (topology) {
+    case Tegra::Shader::OutputTopology::PointList:
+        return "points";
+    case Tegra::Shader::OutputTopology::LineStrip:
+        return "line_strip";
+    case Tegra::Shader::OutputTopology::TriangleStrip:
+        return "triangle_strip";
+    default:
+        LOG_CRITICAL(Render_OpenGL, "Unknown output topology {}", static_cast<u32>(topology));
+        UNREACHABLE();
+        return "points";
+    }
+}
+
 /// Describes the behaviour of code path of a given entry point and a return point.
 enum class ExitMethod {
     Undetermined, ///< Internal value. Only occur when analyzing JMP loop.
@@ -253,8 +275,9 @@ enum class InternalFlag : u64 {
 class GLSLRegisterManager {
 public:
     GLSLRegisterManager(ShaderWriter& shader, ShaderWriter& declarations,
-                        const Maxwell3D::Regs::ShaderStage& stage, const std::string& suffix)
-        : shader{shader}, declarations{declarations}, stage{stage}, suffix{suffix} {
+                        const Maxwell3D::Regs::ShaderStage& stage, const std::string& suffix,
+                        const Tegra::Shader::Header& header)
+        : shader{shader}, declarations{declarations}, stage{stage}, suffix{suffix}, header{header} {
         BuildRegisterList();
         BuildInputList();
     }
@@ -358,11 +381,13 @@ public:
      * @param reg The destination register to use.
      * @param elem The element to use for the operation.
      * @param attribute The input attribute to use as the source value.
+     * @param vertex The register that decides which vertex to read from (used in GS).
      */
     void SetRegisterToInputAttibute(const Register& reg, u64 elem, Attribute::Index attribute,
-                                    const Tegra::Shader::IpaMode& input_mode) {
+                                    const Tegra::Shader::IpaMode& input_mode,
+                                    boost::optional<Register> vertex = {}) {
         const std::string dest = GetRegisterAsFloat(reg);
-        const std::string src = GetInputAttribute(attribute, input_mode) + GetSwizzle(elem);
+        const std::string src = GetInputAttribute(attribute, input_mode, vertex) + GetSwizzle(elem);
         shader.AddLine(dest + " = " + src + ';');
     }
 
@@ -391,16 +416,29 @@ public:
      * are stored as floats, so this may require conversion.
      * @param attribute The destination output attribute.
      * @param elem The element to use for the operation.
-     * @param reg The register to use as the source value.
+     * @param val_reg The register to use as the source value.
+     * @param buf_reg The register that tells which buffer to write to (used in geometry shaders).
      */
-    void SetOutputAttributeToRegister(Attribute::Index attribute, u64 elem, const Register& reg) {
+    void SetOutputAttributeToRegister(Attribute::Index attribute, u64 elem, const Register& val_reg,
+                                      const Register& buf_reg) {
         const std::string dest = GetOutputAttribute(attribute);
-        const std::string src = GetRegisterAsFloat(reg);
+        const std::string src = GetRegisterAsFloat(val_reg);
 
         if (!dest.empty()) {
             // Can happen with unknown/unimplemented output attributes, in which case we ignore the
             // instruction for now.
-            shader.AddLine(dest + GetSwizzle(elem) + " = " + src + ';');
+            if (stage == Maxwell3D::Regs::ShaderStage::Geometry) {
+                // TODO(Rodrigo): nouveau sets some attributes after setting emitting a geometry
+                // shader. These instructions use a dirty register as buffer index. To avoid some
+                // drivers from complaining for the out of boundary writes, guard them.
+                const std::string buf_index{"min(" + GetRegisterAsInteger(buf_reg) + ", " +
+                                            std::to_string(MAX_GEOMETRY_BUFFERS - 1) + ')'};
+                shader.AddLine("amem[" + buf_index + "][" +
+                               std::to_string(static_cast<u32>(attribute)) + ']' +
+                               GetSwizzle(elem) + " = " + src + ';');
+            } else {
+                shader.AddLine(dest + GetSwizzle(elem) + " = " + src + ';');
+            }
         }
     }
 
@@ -441,41 +479,123 @@ public:
         }
     }
 
-    /// Add declarations for registers
+    /// Add declarations.
     void GenerateDeclarations(const std::string& suffix) {
+        GenerateRegisters(suffix);
+        GenerateInternalFlags();
+        GenerateInputAttrs();
+        GenerateOutputAttrs();
+        GenerateConstBuffers();
+        GenerateSamplers();
+        GenerateGeometry();
+    }
+
+    /// Returns a list of constant buffer declarations.
+    std::vector<ConstBufferEntry> GetConstBuffersDeclarations() const {
+        std::vector<ConstBufferEntry> result;
+        std::copy_if(declr_const_buffers.begin(), declr_const_buffers.end(),
+                     std::back_inserter(result), [](const auto& entry) { return entry.IsUsed(); });
+        return result;
+    }
+
+    /// Returns a list of samplers used in the shader.
+    const std::vector<SamplerEntry>& GetSamplers() const {
+        return used_samplers;
+    }
+
+    /// Returns the GLSL sampler used for the input shader sampler, and creates a new one if
+    /// necessary.
+    std::string AccessSampler(const Sampler& sampler, Tegra::Shader::TextureType type,
+                              bool is_array, bool is_shadow) {
+        const auto offset = static_cast<std::size_t>(sampler.index.Value());
+
+        // If this sampler has already been used, return the existing mapping.
+        const auto itr =
+            std::find_if(used_samplers.begin(), used_samplers.end(),
+                         [&](const SamplerEntry& entry) { return entry.GetOffset() == offset; });
+
+        if (itr != used_samplers.end()) {
+            ASSERT(itr->GetType() == type && itr->IsArray() == is_array &&
+                   itr->IsShadow() == is_shadow);
+            return itr->GetName();
+        }
+
+        // Otherwise create a new mapping for this sampler
+        const std::size_t next_index = used_samplers.size();
+        const SamplerEntry entry{stage, offset, next_index, type, is_array, is_shadow};
+        used_samplers.emplace_back(entry);
+        return entry.GetName();
+    }
+
+private:
+    /// Generates declarations for registers.
+    void GenerateRegisters(const std::string& suffix) {
         for (const auto& reg : regs) {
             declarations.AddLine(GLSLRegister::GetTypeString() + ' ' + reg.GetPrefixString() +
                                  std::to_string(reg.GetIndex()) + '_' + suffix + " = 0;");
         }
         declarations.AddNewLine();
+    }
 
+    /// Generates declarations for internal flags.
+    void GenerateInternalFlags() {
         for (u32 ii = 0; ii < static_cast<u64>(InternalFlag::Amount); ii++) {
             const InternalFlag code = static_cast<InternalFlag>(ii);
             declarations.AddLine("bool " + GetInternalFlag(code) + " = false;");
         }
         declarations.AddNewLine();
+    }
+
+    /// Generates declarations for input attributes.
+    void GenerateInputAttrs() {
+        if (stage != Maxwell3D::Regs::ShaderStage::Vertex) {
+            const std::string attr =
+                stage == Maxwell3D::Regs::ShaderStage::Geometry ? "gs_position[]" : "position";
+            declarations.AddLine("layout (location = " + std::to_string(POSITION_VARYING_LOCATION) +
+                                 ") in vec4 " + attr + ';');
+        }
 
         for (const auto element : declr_input_attribute) {
             // TODO(bunnei): Use proper number of elements for these
             u32 idx =
                 static_cast<u32>(element.first) - static_cast<u32>(Attribute::Index::Attribute_0);
-            declarations.AddLine("layout(location = " + std::to_string(idx) + ")" +
-                                 GetInputFlags(element.first) + "in vec4 " +
-                                 GetInputAttribute(element.first, element.second) + ';');
+            if (stage != Maxwell3D::Regs::ShaderStage::Vertex) {
+                // If inputs are varyings, add an offset
+                idx += GENERIC_VARYING_START_LOCATION;
+            }
+
+            std::string attr{GetInputAttribute(element.first, element.second)};
+            if (stage == Maxwell3D::Regs::ShaderStage::Geometry) {
+                attr = "gs_" + attr + "[]";
+            }
+            declarations.AddLine("layout (location = " + std::to_string(idx) + ") " +
+                                 GetInputFlags(element.first) + "in vec4 " + attr + ';');
         }
+
         declarations.AddNewLine();
+    }
 
+    /// Generates declarations for output attributes.
+    void GenerateOutputAttrs() {
+        if (stage != Maxwell3D::Regs::ShaderStage::Fragment) {
+            declarations.AddLine("layout (location = " + std::to_string(POSITION_VARYING_LOCATION) +
+                                 ") out vec4 position;");
+        }
         for (const auto& index : declr_output_attribute) {
             // TODO(bunnei): Use proper number of elements for these
-            declarations.AddLine("layout(location = " +
-                                 std::to_string(static_cast<u32>(index) -
-                                                static_cast<u32>(Attribute::Index::Attribute_0)) +
-                                 ") out vec4 " + GetOutputAttribute(index) + ';');
+            const u32 idx = static_cast<u32>(index) -
+                            static_cast<u32>(Attribute::Index::Attribute_0) +
+                            GENERIC_VARYING_START_LOCATION;
+            declarations.AddLine("layout (location = " + std::to_string(idx) + ") out vec4 " +
+                                 GetOutputAttribute(index) + ';');
         }
         declarations.AddNewLine();
+    }
 
+    /// Generates declarations for constant buffers.
+    void GenerateConstBuffers() {
         for (const auto& entry : GetConstBuffersDeclarations()) {
-            declarations.AddLine("layout(std140) uniform " + entry.GetName());
+            declarations.AddLine("layout (std140) uniform " + entry.GetName());
             declarations.AddLine('{');
             declarations.AddLine("    vec4 c" + std::to_string(entry.GetIndex()) +
                                  "[MAX_CONSTBUFFER_ELEMENTS];");
@@ -483,7 +603,10 @@ public:
             declarations.AddNewLine();
         }
         declarations.AddNewLine();
+    }
 
+    /// Generates declarations for samplers.
+    void GenerateSamplers() {
         const auto& samplers = GetSamplers();
         for (const auto& sampler : samplers) {
             declarations.AddLine("uniform " + sampler.GetTypeString() + ' ' + sampler.GetName() +
@@ -492,44 +615,42 @@ public:
         declarations.AddNewLine();
     }
 
-    /// Returns a list of constant buffer declarations
-    std::vector<ConstBufferEntry> GetConstBuffersDeclarations() const {
-        std::vector<ConstBufferEntry> result;
-        std::copy_if(declr_const_buffers.begin(), declr_const_buffers.end(),
-                     std::back_inserter(result), [](const auto& entry) { return entry.IsUsed(); });
-        return result;
-    }
-
-    /// Returns a list of samplers used in the shader
-    const std::vector<SamplerEntry>& GetSamplers() const {
-        return used_samplers;
-    }
-
-    /// Returns the GLSL sampler used for the input shader sampler, and creates a new one if
-    /// necessary.
-    std::string AccessSampler(const Sampler& sampler, Tegra::Shader::TextureType type,
-                              bool is_array, bool is_shadow) {
-        const std::size_t offset = static_cast<std::size_t>(sampler.index.Value());
+    /// Generates declarations used for geometry shaders.
+    void GenerateGeometry() {
+        if (stage != Maxwell3D::Regs::ShaderStage::Geometry)
+            return;
 
-        // If this sampler has already been used, return the existing mapping.
-        const auto itr =
-            std::find_if(used_samplers.begin(), used_samplers.end(),
-                         [&](const SamplerEntry& entry) { return entry.GetOffset() == offset; });
+        declarations.AddLine(
+            "layout (" + GetTopologyName(header.common3.output_topology) +
+            ", max_vertices = " + std::to_string(header.common4.max_output_vertices) + ") out;");
+        declarations.AddNewLine();
 
-        if (itr != used_samplers.end()) {
-            ASSERT(itr->GetType() == type && itr->IsArray() == is_array &&
-                   itr->IsShadow() == is_shadow);
-            return itr->GetName();
-        }
+        declarations.AddLine("vec4 amem[" + std::to_string(MAX_GEOMETRY_BUFFERS) + "][" +
+                             std::to_string(MAX_ATTRIBUTES) + "];");
+        declarations.AddNewLine();
 
-        // Otherwise create a new mapping for this sampler
-        const std::size_t next_index = used_samplers.size();
-        const SamplerEntry entry{stage, offset, next_index, type, is_array, is_shadow};
-        used_samplers.emplace_back(entry);
-        return entry.GetName();
+        constexpr char buffer[] = "amem[output_buffer]";
+        declarations.AddLine("void emit_vertex(uint output_buffer) {");
+        ++declarations.scope;
+        for (const auto element : declr_output_attribute) {
+            declarations.AddLine(GetOutputAttribute(element) + " = " + buffer + '[' +
+                                 std::to_string(static_cast<u32>(element)) + "];");
+        }
+
+        declarations.AddLine("position = " + std::string(buffer) + '[' +
+                             std::to_string(static_cast<u32>(Attribute::Index::Position)) + "];");
+
+        // If a geometry shader is attached, it will always flip (it's the last stage before
+        // fragment). For more info about flipping, refer to gl_shader_gen.cpp.
+        declarations.AddLine("position.xy *= viewport_flip.xy;");
+        declarations.AddLine("gl_Position = position;");
+        declarations.AddLine("position.w = 1.0;");
+        declarations.AddLine("EmitVertex();");
+        --declarations.scope;
+        declarations.AddLine('}');
+        declarations.AddNewLine();
     }
 
-private:
     /// Generates code representing a temporary (GPR) register.
     std::string GetRegister(const Register& reg, unsigned elem) {
         if (reg == Register::ZeroIndex) {
@@ -586,11 +707,19 @@ private:
 
     /// Generates code representing an input attribute register.
     std::string GetInputAttribute(Attribute::Index attribute,
-                                  const Tegra::Shader::IpaMode& input_mode) {
+                                  const Tegra::Shader::IpaMode& input_mode,
+                                  boost::optional<Register> vertex = {}) {
+        auto GeometryPass = [&](const std::string& name) {
+            if (stage == Maxwell3D::Regs::ShaderStage::Geometry && vertex) {
+                return "gs_" + name + '[' + GetRegisterAsInteger(vertex.value(), 0, false) + ']';
+            }
+            return name;
+        };
+
         switch (attribute) {
         case Attribute::Index::Position:
             if (stage != Maxwell3D::Regs::ShaderStage::Fragment) {
-                return "position";
+                return GeometryPass("position");
             } else {
                 return "vec4(gl_FragCoord.x, gl_FragCoord.y, gl_FragCoord.z, 1.0)";
             }
@@ -619,7 +748,7 @@ private:
                         UNREACHABLE();
                     }
                 }
-                return "input_attribute_" + std::to_string(index);
+                return GeometryPass("input_attribute_" + std::to_string(index));
             }
 
             LOG_CRITICAL(HW_GPU, "Unhandled input attribute: {}", static_cast<u32>(attribute));
@@ -672,7 +801,7 @@ private:
         return out;
     }
 
-    /// Generates code representing an output attribute register.
+    /// Generates code representing the declaration name of an output attribute register.
     std::string GetOutputAttribute(Attribute::Index attribute) {
         switch (attribute) {
         case Attribute::Index::Position:
@@ -708,6 +837,7 @@ private:
     std::vector<SamplerEntry> used_samplers;
     const Maxwell3D::Regs::ShaderStage& stage;
     const std::string& suffix;
+    const Tegra::Shader::Header& header;
 };
 
 class GLSLGenerator {
@@ -1103,8 +1233,8 @@ private:
             return offset + 1;
         }
 
-        shader.AddLine("// " + std::to_string(offset) + ": " + opcode->GetName() + " (" +
-                       std::to_string(instr.value) + ')');
+        shader.AddLine(
+            fmt::format("// {}: {} (0x{:016x})", offset, opcode->GetName(), instr.value));
 
         using Tegra::Shader::Pred;
         ASSERT_MSG(instr.pred.full_pred != Pred::NeverExecute,
@@ -1306,7 +1436,6 @@ private:
 
             break;
         }
-
         case OpCode::Type::Shift: {
             std::string op_a = regs.GetRegisterAsInteger(instr.gpr8, 0, true);
             std::string op_b;
@@ -1348,7 +1477,6 @@ private:
             }
             break;
         }
-
         case OpCode::Type::ArithmeticIntegerImmediate: {
             std::string op_a = regs.GetRegisterAsInteger(instr.gpr8);
             std::string op_b = std::to_string(instr.alu.imm20_32.Value());
@@ -1826,7 +1954,7 @@ private:
                 const auto LoadNextElement = [&](u32 reg_offset) {
                     regs.SetRegisterToInputAttibute(instr.gpr0.Value() + reg_offset, next_element,
                                                     static_cast<Attribute::Index>(next_index),
-                                                    input_mode);
+                                                    input_mode, instr.gpr39.Value());
 
                     // Load the next attribute element into the following register. If the element
                     // to load goes beyond the vec4 size, load the first element of the next
@@ -1890,8 +2018,8 @@ private:
 
                 const auto StoreNextElement = [&](u32 reg_offset) {
                     regs.SetOutputAttributeToRegister(static_cast<Attribute::Index>(next_index),
-                                                      next_element,
-                                                      instr.gpr0.Value() + reg_offset);
+                                                      next_element, instr.gpr0.Value() + reg_offset,
+                                                      instr.gpr39.Value());
 
                     // Load the next attribute element into the following register. If the element
                     // to load goes beyond the vec4 size, load the first element of the next
@@ -1908,9 +2036,9 @@ private:
                 break;
             }
             case OpCode::Id::TEX: {
-                ASSERT_MSG(instr.tex.array == 0, "TEX arrays unimplemented");
                 Tegra::Shader::TextureType texture_type{instr.tex.texture_type};
                 std::string coord;
+                const bool is_array = instr.tex.array != 0;
 
                 ASSERT_MSG(!instr.tex.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP),
                            "NODEP is not implemented");
@@ -1925,21 +2053,59 @@ private:
 
                 switch (num_coordinates) {
                 case 1: {
-                    const std::string x = regs.GetRegisterAsFloat(instr.gpr8);
-                    coord = "float coords = " + x + ';';
+                    if (is_array) {
+                        const std::string index = regs.GetRegisterAsInteger(instr.gpr8);
+                        const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
+                        coord = "vec2 coords = vec2(" + x + ", " + index + ");";
+                    } else {
+                        const std::string x = regs.GetRegisterAsFloat(instr.gpr8);
+                        coord = "float coords = " + x + ';';
+                    }
                     break;
                 }
                 case 2: {
-                    const std::string x = regs.GetRegisterAsFloat(instr.gpr8);
-                    const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
-                    coord = "vec2 coords = vec2(" + x + ", " + y + ");";
+                    if (is_array) {
+                        const std::string index = regs.GetRegisterAsInteger(instr.gpr8);
+                        const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
+                        const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 2);
+                        coord = "vec3 coords = vec3(" + x + ", " + y + ", " + index + ");";
+                    } else {
+                        const std::string x = regs.GetRegisterAsFloat(instr.gpr8);
+                        const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
+                        coord = "vec2 coords = vec2(" + x + ", " + y + ");";
+                    }
                     break;
                 }
                 case 3: {
-                    const std::string x = regs.GetRegisterAsFloat(instr.gpr8);
-                    const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
-                    const std::string z = regs.GetRegisterAsFloat(instr.gpr20);
-                    coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");";
+                    if (depth_compare) {
+                        if (is_array) {
+                            const std::string index = regs.GetRegisterAsInteger(instr.gpr8);
+                            const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
+                            const std::string y = regs.GetRegisterAsFloat(instr.gpr20);
+                            const std::string z = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1);
+                            coord = "vec4 coords = vec4(" + x + ", " + y + ", " + z + ", " + index +
+                                    ");";
+                        } else {
+                            const std::string x = regs.GetRegisterAsFloat(instr.gpr8);
+                            const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
+                            const std::string z = regs.GetRegisterAsFloat(instr.gpr20);
+                            coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");";
+                        }
+                    } else {
+                        if (is_array) {
+                            const std::string index = regs.GetRegisterAsInteger(instr.gpr8);
+                            const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
+                            const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 2);
+                            const std::string z = regs.GetRegisterAsFloat(instr.gpr8.Value() + 3);
+                            coord = "vec4 coords = vec4(" + x + ", " + y + ", " + z + ", " + index +
+                                    ");";
+                        } else {
+                            const std::string x = regs.GetRegisterAsFloat(instr.gpr8);
+                            const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
+                            const std::string z = regs.GetRegisterAsFloat(instr.gpr8.Value() + 2);
+                            coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");";
+                        }
+                    }
                     break;
                 }
                 default:
@@ -1958,7 +2124,7 @@ private:
                 std::string op_c;
 
                 const std::string sampler =
-                    GetSampler(instr.sampler, texture_type, false, depth_compare);
+                    GetSampler(instr.sampler, texture_type, is_array, depth_compare);
                 // Add an extra scope and declare the texture coords inside to prevent
                 // overwriting them in case they are used as outputs of the texs instruction.
 
@@ -1978,10 +2144,13 @@ private:
                 }
                 case Tegra::Shader::TextureProcessMode::LB:
                 case Tegra::Shader::TextureProcessMode::LBA: {
-                    if (num_coordinates <= 2) {
-                        op_c = regs.GetRegisterAsFloat(instr.gpr20);
+                    if (depth_compare) {
+                        if (is_array)
+                            op_c = regs.GetRegisterAsFloat(instr.gpr20.Value() + 2);
+                        else
+                            op_c = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1);
                     } else {
-                        op_c = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1);
+                        op_c = regs.GetRegisterAsFloat(instr.gpr20);
                     }
                     // TODO: Figure if A suffix changes the equation at all.
                     texture = "texture(" + sampler + ", coords, " + op_c + ')';
@@ -2124,6 +2293,8 @@ private:
                 ASSERT_MSG(!instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::MZ),
                            "MZ is not implemented");
 
+                u32 op_c_offset = 0;
+
                 switch (texture_type) {
                 case Tegra::Shader::TextureType::Texture1D: {
                     const std::string x = regs.GetRegisterAsInteger(instr.gpr8);
@@ -2138,6 +2309,7 @@ private:
                         const std::string x = regs.GetRegisterAsInteger(instr.gpr8);
                         const std::string y = regs.GetRegisterAsInteger(instr.gpr20);
                         coord = "ivec2 coords = ivec2(" + x + ", " + y + ");";
+                        op_c_offset = 1;
                     }
                     break;
                 }
@@ -2149,13 +2321,14 @@ private:
                 const std::string sampler =
                     GetSampler(instr.sampler, texture_type, is_array, false);
                 std::string texture = "texelFetch(" + sampler + ", coords, 0)";
-                const std::string op_c = regs.GetRegisterAsInteger(instr.gpr20.Value() + 1);
                 switch (instr.tlds.GetTextureProcessMode()) {
                 case Tegra::Shader::TextureProcessMode::LZ: {
                     texture = "texelFetch(" + sampler + ", coords, 0)";
                     break;
                 }
                 case Tegra::Shader::TextureProcessMode::LL: {
+                    const std::string op_c =
+                        regs.GetRegisterAsInteger(instr.gpr20.Value() + op_c_offset);
                     texture = "texelFetch(" + sampler + ", coords, " + op_c + ')';
                     break;
                 }
@@ -2496,14 +2669,14 @@ private:
                 const std::string pred =
                     GetPredicateCondition(instr.csetp.pred39, instr.csetp.neg_pred39 != 0);
                 const std::string combiner = GetPredicateCombiner(instr.csetp.op);
-                const std::string controlCode = regs.GetControlCode(instr.csetp.cc);
+                const std::string control_code = regs.GetControlCode(instr.csetp.cc);
                 if (instr.csetp.pred3 != static_cast<u64>(Pred::UnusedIndex)) {
                     SetPredicate(instr.csetp.pred3,
-                                 '(' + controlCode + ") " + combiner + " (" + pred + ')');
+                                 '(' + control_code + ") " + combiner + " (" + pred + ')');
                 }
                 if (instr.csetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) {
                     SetPredicate(instr.csetp.pred0,
-                                 "!(" + controlCode + ") " + combiner + " (" + pred + ')');
+                                 "!(" + control_code + ") " + combiner + " (" + pred + ')');
                 }
                 break;
             }
@@ -2734,6 +2907,52 @@ private:
 
                 break;
             }
+            case OpCode::Id::OUT_R: {
+                ASSERT(instr.gpr20.Value() == Register::ZeroIndex);
+                ASSERT_MSG(stage == Maxwell3D::Regs::ShaderStage::Geometry,
+                           "OUT is expected to be used in a geometry shader.");
+
+                if (instr.out.emit) {
+                    // gpr0 is used to store the next address. Hardware returns a pointer but
+                    // we just return the next index with a cyclic cap.
+                    const std::string current{regs.GetRegisterAsInteger(instr.gpr8, 0, false)};
+                    const std::string next = "((" + current + " + 1" + ") % " +
+                                             std::to_string(MAX_GEOMETRY_BUFFERS) + ')';
+                    shader.AddLine("emit_vertex(" + current + ");");
+                    regs.SetRegisterToInteger(instr.gpr0, false, 0, next, 1, 1);
+                }
+                if (instr.out.cut) {
+                    shader.AddLine("EndPrimitive();");
+                }
+
+                break;
+            }
+            case OpCode::Id::MOV_SYS: {
+                switch (instr.sys20) {
+                case Tegra::Shader::SystemVariable::InvocationInfo: {
+                    LOG_WARNING(HW_GPU, "MOV_SYS instruction with InvocationInfo is incomplete");
+                    regs.SetRegisterToInteger(instr.gpr0, false, 0, "0u", 1, 1);
+                    break;
+                }
+                default: {
+                    LOG_CRITICAL(HW_GPU, "Unhandled system move: {}",
+                                 static_cast<u32>(instr.sys20.Value()));
+                    UNREACHABLE();
+                }
+                }
+                break;
+            }
+            case OpCode::Id::ISBERD: {
+                ASSERT(instr.isberd.o == 0);
+                ASSERT(instr.isberd.skew == 0);
+                ASSERT(instr.isberd.shift == Tegra::Shader::IsberdShift::None);
+                ASSERT(instr.isberd.mode == Tegra::Shader::IsberdMode::None);
+                ASSERT_MSG(stage == Maxwell3D::Regs::ShaderStage::Geometry,
+                           "ISBERD is expected to be used in a geometry shader.");
+                LOG_WARNING(HW_GPU, "ISBERD instruction is incomplete");
+                regs.SetRegisterToFloat(instr.gpr0, 0, regs.GetRegisterAsFloat(instr.gpr8), 1, 1);
+                break;
+            }
             case OpCode::Id::BRA: {
                 ASSERT_MSG(instr.bra.constant_buffer == 0,
                            "BRA with constant buffers are not implemented");
@@ -2777,6 +2996,88 @@ private:
                 LOG_WARNING(HW_GPU, "DEPBAR instruction is stubbed");
                 break;
             }
+            case OpCode::Id::VMAD: {
+                const bool signed_a = instr.vmad.signed_a == 1;
+                const bool signed_b = instr.vmad.signed_b == 1;
+                const bool result_signed = signed_a || signed_b;
+                boost::optional<std::string> forced_result;
+
+                auto Unpack = [&](const std::string& op, bool is_chunk, bool is_signed,
+                                  Tegra::Shader::VmadType type, u64 byte_height) {
+                    const std::string value = [&]() {
+                        if (!is_chunk) {
+                            const auto offset = static_cast<u32>(byte_height * 8);
+                            return "((" + op + " >> " + std::to_string(offset) + ") & 0xff)";
+                        }
+                        const std::string zero = "0";
+
+                        switch (type) {
+                        case Tegra::Shader::VmadType::Size16_Low:
+                            return '(' + op + " & 0xffff)";
+                        case Tegra::Shader::VmadType::Size16_High:
+                            return '(' + op + " >> 16)";
+                        case Tegra::Shader::VmadType::Size32:
+                            // TODO(Rodrigo): From my hardware tests it becomes a bit "mad" when
+                            // this type is used (1 * 1 + 0 == 0x5b800000). Until a better
+                            // explanation is found: assert.
+                            UNREACHABLE_MSG("Unimplemented");
+                            return zero;
+                        case Tegra::Shader::VmadType::Invalid:
+                            // Note(Rodrigo): This flag is invalid according to nvdisasm. From my
+                            // testing (even though it's invalid) this makes the whole instruction
+                            // assign zero to target register.
+                            forced_result = boost::make_optional(zero);
+                            return zero;
+                        default:
+                            UNREACHABLE();
+                            return zero;
+                        }
+                    }();
+
+                    if (is_signed) {
+                        return "int(" + value + ')';
+                    }
+                    return value;
+                };
+
+                const std::string op_a = Unpack(regs.GetRegisterAsInteger(instr.gpr8, 0, false),
+                                                instr.vmad.is_byte_chunk_a != 0, signed_a,
+                                                instr.vmad.type_a, instr.vmad.byte_height_a);
+
+                std::string op_b;
+                if (instr.vmad.use_register_b) {
+                    op_b = Unpack(regs.GetRegisterAsInteger(instr.gpr20, 0, false),
+                                  instr.vmad.is_byte_chunk_b != 0, signed_b, instr.vmad.type_b,
+                                  instr.vmad.byte_height_b);
+                } else {
+                    op_b = '(' +
+                           std::to_string(signed_b ? static_cast<s16>(instr.alu.GetImm20_16())
+                                                   : instr.alu.GetImm20_16()) +
+                           ')';
+                }
+
+                const std::string op_c = regs.GetRegisterAsInteger(instr.gpr39, 0, result_signed);
+
+                std::string result;
+                if (forced_result) {
+                    result = *forced_result;
+                } else {
+                    result = '(' + op_a + " * " + op_b + " + " + op_c + ')';
+
+                    switch (instr.vmad.shr) {
+                    case Tegra::Shader::VmadShr::Shr7:
+                        result = '(' + result + " >> 7)";
+                        break;
+                    case Tegra::Shader::VmadShr::Shr15:
+                        result = '(' + result + " >> 15)";
+                        break;
+                    }
+                }
+                regs.SetRegisterToInteger(instr.gpr0, result_signed, 1, result, 1, 1,
+                                          instr.vmad.saturate == 1, 0, Register::Size::Word,
+                                          instr.vmad.cc);
+                break;
+            }
             default: {
                 LOG_CRITICAL(HW_GPU, "Unhandled instruction: {}", opcode->GetName());
                 UNREACHABLE();
@@ -2907,7 +3208,7 @@ private:
 
     ShaderWriter shader;
     ShaderWriter declarations;
-    GLSLRegisterManager regs{shader, declarations, stage, suffix};
+    GLSLRegisterManager regs{shader, declarations, stage, suffix, header};
 
     // Declarations
     std::set<std::string> declr_predicates;
diff --git a/src/video_core/renderer_opengl/gl_shader_gen.cpp b/src/video_core/renderer_opengl/gl_shader_gen.cpp
index b0466c18f..1e5eb32df 100644
--- a/src/video_core/renderer_opengl/gl_shader_gen.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_gen.cpp
@@ -17,7 +17,18 @@ ProgramResult GenerateVertexShader(const ShaderSetup& setup) {
     std::string out = "#version 430 core\n";
     out += "#extension GL_ARB_separate_shader_objects : enable\n\n";
     out += Decompiler::GetCommonDeclarations();
-    out += "bool exec_vertex();\n";
+
+    out += R"(
+out gl_PerVertex {
+    vec4 gl_Position;
+};
+
+layout(std140) uniform vs_config {
+    vec4 viewport_flip;
+    uvec4 instance_id;
+    uvec4 flip_stage;
+};
+)";
 
     if (setup.IsDualProgram()) {
         out += "bool exec_vertex_b();\n";
@@ -28,18 +39,17 @@ ProgramResult GenerateVertexShader(const ShaderSetup& setup) {
                                      Maxwell3D::Regs::ShaderStage::Vertex, "vertex")
             .get_value_or({});
 
-    out += R"(
-
-out gl_PerVertex {
-    vec4 gl_Position;
-};
+    out += program.first;
 
-out vec4 position;
+    if (setup.IsDualProgram()) {
+        ProgramResult program_b =
+            Decompiler::DecompileProgram(setup.program.code_b, PROGRAM_OFFSET,
+                                         Maxwell3D::Regs::ShaderStage::Vertex, "vertex_b")
+                .get_value_or({});
+        out += program_b.first;
+    }
 
-layout (std140) uniform vs_config {
-    vec4 viewport_flip;
-    uvec4 instance_id;
-};
+    out += R"(
 
 void main() {
     position = vec4(0.0, 0.0, 0.0, 0.0);
@@ -52,27 +62,52 @@ void main() {
 
     out += R"(
 
-    // Viewport can be flipped, which is unsupported by glViewport
-    position.xy *= viewport_flip.xy;
+    // Check if the flip stage is VertexB
+    if (flip_stage[0] == 1) {
+        // Viewport can be flipped, which is unsupported by glViewport
+        position.xy *= viewport_flip.xy;
+    }
     gl_Position = position;
 
     // TODO(bunnei): This is likely a hack, position.w should be interpolated as 1.0
     // For now, this is here to bring order in lieu of proper emulation
-    position.w = 1.0;
+    if (flip_stage[0] == 1) {
+        position.w = 1.0;
+    }
 }
 
 )";
 
-    out += program.first;
+    return {out, program.second};
+}
 
-    if (setup.IsDualProgram()) {
-        ProgramResult program_b =
-            Decompiler::DecompileProgram(setup.program.code_b, PROGRAM_OFFSET,
-                                         Maxwell3D::Regs::ShaderStage::Vertex, "vertex_b")
-                .get_value_or({});
-        out += program_b.first;
-    }
+ProgramResult GenerateGeometryShader(const ShaderSetup& setup) {
+    std::string out = "#version 430 core\n";
+    out += "#extension GL_ARB_separate_shader_objects : enable\n\n";
+    out += Decompiler::GetCommonDeclarations();
+    out += "bool exec_geometry();\n";
+
+    ProgramResult program =
+        Decompiler::DecompileProgram(setup.program.code, PROGRAM_OFFSET,
+                                     Maxwell3D::Regs::ShaderStage::Geometry, "geometry")
+            .get_value_or({});
+    out += R"(
+out gl_PerVertex {
+    vec4 gl_Position;
+};
 
+layout (std140) uniform gs_config {
+    vec4 viewport_flip;
+    uvec4 instance_id;
+    uvec4 flip_stage;
+};
+
+void main() {
+    exec_geometry();
+}
+
+)";
+    out += program.first;
     return {out, program.second};
 }
 
@@ -87,7 +122,6 @@ ProgramResult GenerateFragmentShader(const ShaderSetup& setup) {
                                      Maxwell3D::Regs::ShaderStage::Fragment, "fragment")
             .get_value_or({});
     out += R"(
-in vec4 position;
 layout(location = 0) out vec4 FragColor0;
 layout(location = 1) out vec4 FragColor1;
 layout(location = 2) out vec4 FragColor2;
@@ -100,6 +134,7 @@ layout(location = 7) out vec4 FragColor7;
 layout (std140) uniform fs_config {
     vec4 viewport_flip;
     uvec4 instance_id;
+    uvec4 flip_stage;
 };
 
 void main() {
@@ -110,5 +145,4 @@ void main() {
     out += program.first;
     return {out, program.second};
 }
-
-} // namespace OpenGL::GLShader
+} // namespace OpenGL::GLShader
\ No newline at end of file
diff --git a/src/video_core/renderer_opengl/gl_shader_gen.h b/src/video_core/renderer_opengl/gl_shader_gen.h
index e56f39e78..79596087a 100644
--- a/src/video_core/renderer_opengl/gl_shader_gen.h
+++ b/src/video_core/renderer_opengl/gl_shader_gen.h
@@ -196,6 +196,12 @@ private:
 ProgramResult GenerateVertexShader(const ShaderSetup& setup);
 
 /**
+ * Generates the GLSL geometry shader program source code for the given GS program
+ * @returns String of the shader source code
+ */
+ProgramResult GenerateGeometryShader(const ShaderSetup& setup);
+
+/**
  * Generates the GLSL fragment shader program source code for the given FS program
  * @returns String of the shader source code
  */
diff --git a/src/video_core/renderer_opengl/gl_shader_manager.cpp b/src/video_core/renderer_opengl/gl_shader_manager.cpp
index 022d32a86..010857ec6 100644
--- a/src/video_core/renderer_opengl/gl_shader_manager.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_manager.cpp
@@ -18,6 +18,14 @@ void MaxwellUniformData::SetFromRegs(const Maxwell3D::State::ShaderStageInfo& sh
 
     // We only assign the instance to the first component of the vector, the rest is just padding.
     instance_id[0] = state.current_instance;
+
+    // Assign in which stage the position has to be flipped
+    // (the last stage before the fragment shader).
+    if (gpu.regs.shader_config[static_cast<u32>(Maxwell3D::Regs::ShaderProgram::Geometry)].enable) {
+        flip_stage[0] = static_cast<u32>(Maxwell3D::Regs::ShaderProgram::Geometry);
+    } else {
+        flip_stage[0] = static_cast<u32>(Maxwell3D::Regs::ShaderProgram::VertexB);
+    }
 }
 
 } // namespace OpenGL::GLShader
diff --git a/src/video_core/renderer_opengl/gl_shader_manager.h b/src/video_core/renderer_opengl/gl_shader_manager.h
index 3de15ba9b..b3a191cf2 100644
--- a/src/video_core/renderer_opengl/gl_shader_manager.h
+++ b/src/video_core/renderer_opengl/gl_shader_manager.h
@@ -21,8 +21,9 @@ struct MaxwellUniformData {
     void SetFromRegs(const Maxwell3D::State::ShaderStageInfo& shader_stage);
     alignas(16) GLvec4 viewport_flip;
     alignas(16) GLuvec4 instance_id;
+    alignas(16) GLuvec4 flip_stage;
 };
-static_assert(sizeof(MaxwellUniformData) == 32, "MaxwellUniformData structure size is incorrect");
+static_assert(sizeof(MaxwellUniformData) == 48, "MaxwellUniformData structure size is incorrect");
 static_assert(sizeof(MaxwellUniformData) < 16384,
               "MaxwellUniformData structure must be less than 16kb as per the OpenGL spec");
 
@@ -36,6 +37,10 @@ public:
         vs = program;
     }
 
+    void UseProgrammableGeometryShader(GLuint program) {
+        gs = program;
+    }
+
     void UseProgrammableFragmentShader(GLuint program) {
         fs = program;
     }