14 files changed, 379 insertions, 241 deletions

diff --git a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
index 92acc57b1..21ccfe1f8 100644
--- a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
@@ -25,9 +25,9 @@ void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, u32 format, u32 width, u3
                         u32 stride, NVFlinger::BufferQueue::BufferTransformFlags transform,
                         const MathUtil::Rectangle<int>& crop_rect) {
     VAddr addr = nvmap_dev->GetObjectAddress(buffer_handle);
-    LOG_WARNING(Service,
-                "Drawing from address {:X} offset {:08X} Width {} Height {} Stride {} Format {}",
-                addr, offset, width, height, stride, format);
+    LOG_TRACE(Service,
+              "Drawing from address {:X} offset {:08X} Width {} Height {} Stride {} Format {}",
+              addr, offset, width, height, stride, format);
 
     using PixelFormat = Tegra::FramebufferConfig::PixelFormat;
     const Tegra::FramebufferConfig framebuffer{
diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp
index 9f1533263..ec1a57226 100644
--- a/src/video_core/engines/fermi_2d.cpp
+++ b/src/video_core/engines/fermi_2d.cpp
@@ -21,7 +21,9 @@ void Fermi2D::CallMethod(const GPU::MethodCall& method_call) {
     regs.reg_array[method_call.method] = method_call.argument;
 
     switch (method_call.method) {
-    case FERMI2D_REG_INDEX(trigger): {
+    // Trigger the surface copy on the last register write. This is blit_src_y, but this is 64-bit,
+    // so trigger on the second 32-bit write.
+    case FERMI2D_REG_INDEX(blit_src_y) + 1: {
         HandleSurfaceCopy();
         break;
     }
@@ -32,57 +34,23 @@ void Fermi2D::HandleSurfaceCopy() {
     LOG_WARNING(HW_GPU, "Requested a surface copy with operation {}",
                 static_cast<u32>(regs.operation));
 
-    const GPUVAddr source = regs.src.Address();
-    const GPUVAddr dest = regs.dst.Address();
-
-    // TODO(Subv): Only same-format and same-size copies are allowed for now.
-    ASSERT(regs.src.format == regs.dst.format);
-    ASSERT(regs.src.width * regs.src.height == regs.dst.width * regs.dst.height);
-
     // TODO(Subv): Only raw copies are implemented.
     ASSERT(regs.operation == Regs::Operation::SrcCopy);
 
-    const auto source_cpu = memory_manager.GpuToCpuAddress(source);
-    const auto dest_cpu = memory_manager.GpuToCpuAddress(dest);
-    ASSERT_MSG(source_cpu, "Invalid source GPU address");
-    ASSERT_MSG(dest_cpu, "Invalid destination GPU address");
-
-    u32 src_bytes_per_pixel = RenderTargetBytesPerPixel(regs.src.format);
-    u32 dst_bytes_per_pixel = RenderTargetBytesPerPixel(regs.dst.format);
-
-    if (!rasterizer.AccelerateSurfaceCopy(regs.src, regs.dst)) {
-        // All copies here update the main memory, so mark all rasterizer states as invalid.
-        Core::System::GetInstance().GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
+    const u32 src_blit_x1{static_cast<u32>(regs.blit_src_x >> 32)};
+    const u32 src_blit_y1{static_cast<u32>(regs.blit_src_y >> 32)};
+    const u32 src_blit_x2{
+        static_cast<u32>((regs.blit_src_x + (regs.blit_dst_width * regs.blit_du_dx)) >> 32)};
+    const u32 src_blit_y2{
+        static_cast<u32>((regs.blit_src_y + (regs.blit_dst_height * regs.blit_dv_dy)) >> 32)};
 
-        rasterizer.FlushRegion(*source_cpu, src_bytes_per_pixel * regs.src.width * regs.src.height);
-        // We have to invalidate the destination region to evict any outdated surfaces from the
-        // cache. We do this before actually writing the new data because the destination address
-        // might contain a dirty surface that will have to be written back to memory.
-        rasterizer.InvalidateRegion(*dest_cpu,
-                                    dst_bytes_per_pixel * regs.dst.width * regs.dst.height);
+    const MathUtil::Rectangle<u32> src_rect{src_blit_x1, src_blit_y1, src_blit_x2, src_blit_y2};
+    const MathUtil::Rectangle<u32> dst_rect{regs.blit_dst_x, regs.blit_dst_y,
+                                            regs.blit_dst_x + regs.blit_dst_width,
+                                            regs.blit_dst_y + regs.blit_dst_height};
 
-        if (regs.src.linear == regs.dst.linear) {
-            // If the input layout and the output layout are the same, just perform a raw copy.
-            ASSERT(regs.src.BlockHeight() == regs.dst.BlockHeight());
-            Memory::CopyBlock(*dest_cpu, *source_cpu,
-                              src_bytes_per_pixel * regs.dst.width * regs.dst.height);
-            return;
-        }
-        u8* src_buffer = Memory::GetPointer(*source_cpu);
-        u8* dst_buffer = Memory::GetPointer(*dest_cpu);
-        if (!regs.src.linear && regs.dst.linear) {
-            // If the input is tiled and the output is linear, deswizzle the input and copy it over.
-            Texture::CopySwizzledData(regs.src.width, regs.src.height, regs.src.depth,
-                                      src_bytes_per_pixel, dst_bytes_per_pixel, src_buffer,
-                                      dst_buffer, true, regs.src.BlockHeight(),
-                                      regs.src.BlockDepth(), 0);
-        } else {
-            // If the input is linear and the output is tiled, swizzle the input and copy it over.
-            Texture::CopySwizzledData(regs.src.width, regs.src.height, regs.src.depth,
-                                      src_bytes_per_pixel, dst_bytes_per_pixel, dst_buffer,
-                                      src_buffer, false, regs.dst.BlockHeight(),
-                                      regs.dst.BlockDepth(), 0);
-        }
+    if (!rasterizer.AccelerateSurfaceCopy(regs.src, regs.dst, src_rect, dst_rect)) {
+        UNIMPLEMENTED();
     }
 }
 
diff --git a/src/video_core/engines/fermi_2d.h b/src/video_core/engines/fermi_2d.h
index 50009bf75..c69f74cc5 100644
--- a/src/video_core/engines/fermi_2d.h
+++ b/src/video_core/engines/fermi_2d.h
@@ -94,12 +94,22 @@ public:
 
                 Operation operation;
 
-                INSERT_PADDING_WORDS(0x9);
+                INSERT_PADDING_WORDS(0x177);
 
-                // TODO(Subv): This is only a guess.
-                u32 trigger;
+                u32 blit_control;
 
-                INSERT_PADDING_WORDS(0x1A3);
+                INSERT_PADDING_WORDS(0x8);
+
+                u32 blit_dst_x;
+                u32 blit_dst_y;
+                u32 blit_dst_width;
+                u32 blit_dst_height;
+                u64 blit_du_dx;
+                u64 blit_dv_dy;
+                u64 blit_src_x;
+                u64 blit_src_y;
+
+                INSERT_PADDING_WORDS(0x21);
             };
             std::array<u32, NUM_REGS> reg_array;
         };
@@ -122,7 +132,16 @@ private:
 ASSERT_REG_POSITION(dst, 0x80);
 ASSERT_REG_POSITION(src, 0x8C);
 ASSERT_REG_POSITION(operation, 0xAB);
-ASSERT_REG_POSITION(trigger, 0xB5);
+ASSERT_REG_POSITION(blit_control, 0x223);
+ASSERT_REG_POSITION(blit_dst_x, 0x22c);
+ASSERT_REG_POSITION(blit_dst_y, 0x22d);
+ASSERT_REG_POSITION(blit_dst_width, 0x22e);
+ASSERT_REG_POSITION(blit_dst_height, 0x22f);
+ASSERT_REG_POSITION(blit_du_dx, 0x230);
+ASSERT_REG_POSITION(blit_dv_dy, 0x232);
+ASSERT_REG_POSITION(blit_src_x, 0x234);
+ASSERT_REG_POSITION(blit_src_y, 0x236);
+
 #undef ASSERT_REG_POSITION
 
 } // namespace Tegra::Engines
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index 269df9437..1f425f90b 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -186,7 +186,7 @@ enum class SubOp : u64 {
 };
 
 enum class F2iRoundingOp : u64 {
-    None = 0,
+    RoundEven = 0,
     Floor = 1,
     Ceil = 2,
     Trunc = 3,
diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h
index 77da135a0..b2a223705 100644
--- a/src/video_core/rasterizer_interface.h
+++ b/src/video_core/rasterizer_interface.h
@@ -46,7 +46,9 @@ public:
 
     /// Attempt to use a faster method to perform a surface copy
     virtual bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
-                                       const Tegra::Engines::Fermi2D::Regs::Surface& dst) {
+                                       const Tegra::Engines::Fermi2D::Regs::Surface& dst,
+                                       const MathUtil::Rectangle<u32>& src_rect,
+                                       const MathUtil::Rectangle<u32>& dst_rect) {
         return false;
     }
 
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 974ca6a20..12d876120 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -778,15 +778,11 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) {
 }
 
 bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
-                                             const Tegra::Engines::Fermi2D::Regs::Surface& dst) {
+                                             const Tegra::Engines::Fermi2D::Regs::Surface& dst,
+                                             const MathUtil::Rectangle<u32>& src_rect,
+                                             const MathUtil::Rectangle<u32>& dst_rect) {
     MICROPROFILE_SCOPE(OpenGL_Blits);
-
-    if (Settings::values.use_accurate_gpu_emulation) {
-        // Skip the accelerated copy and perform a slow but more accurate copy
-        return false;
-    }
-
-    res_cache.FermiCopySurface(src, dst);
+    res_cache.FermiCopySurface(src, dst, src_rect, dst_rect);
     return true;
 }
 
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index f3b607f4d..258d62259 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -61,7 +61,9 @@ public:
     void InvalidateRegion(VAddr addr, u64 size) override;
     void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
     bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
-                               const Tegra::Engines::Fermi2D::Regs::Surface& dst) override;
+                               const Tegra::Engines::Fermi2D::Regs::Surface& dst,
+                               const MathUtil::Rectangle<u32>& src_rect,
+                               const MathUtil::Rectangle<u32>& dst_rect) override;
     bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
                            u32 pixel_stride) override;
     bool AccelerateDrawBatch(bool is_indexed) override;
diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
index a79eee03e..59f671048 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
@@ -125,6 +125,9 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only,
 
     params.width = Common::AlignUp(config.tic.Width(), GetCompressionFactor(params.pixel_format));
     params.height = Common::AlignUp(config.tic.Height(), GetCompressionFactor(params.pixel_format));
+    if (!params.is_tiled) {
+        params.pitch = config.tic.Pitch();
+    }
     params.unaligned_height = config.tic.Height();
     params.target = SurfaceTargetFromTextureType(config.tic.texture_type);
     params.identity = SurfaceClass::Uploaded;
@@ -191,7 +194,13 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only,
                              config.format == Tegra::RenderTargetFormat::RGBA8_SRGB;
     params.component_type = ComponentTypeFromRenderTarget(config.format);
     params.type = GetFormatType(params.pixel_format);
-    params.width = config.width;
+    if (params.is_tiled) {
+        params.width = config.width;
+    } else {
+        params.pitch = config.width;
+        const u32 bpp = params.GetFormatBpp() / 8;
+        params.width = params.pitch / bpp;
+    }
     params.height = config.height;
     params.unaligned_height = config.height;
     params.target = SurfaceTarget::Texture2D;
@@ -428,7 +437,8 @@ void SwizzleFunc(const MortonSwizzleMode& mode, const SurfaceParams& params,
     }
 }
 
-static void FastCopySurface(const Surface& src_surface, const Surface& dst_surface) {
+void RasterizerCacheOpenGL::FastCopySurface(const Surface& src_surface,
+                                            const Surface& dst_surface) {
     const auto& src_params{src_surface->GetSurfaceParams()};
     const auto& dst_params{dst_surface->GetSurfaceParams()};
 
@@ -438,12 +448,15 @@ static void FastCopySurface(const Surface& src_surface, const Surface& dst_surfa
     glCopyImageSubData(src_surface->Texture().handle, SurfaceTargetToGL(src_params.target), 0, 0, 0,
                        0, dst_surface->Texture().handle, SurfaceTargetToGL(dst_params.target), 0, 0,
                        0, 0, width, height, 1);
+
+    dst_surface->MarkAsModified(true, *this);
 }
 
 MICROPROFILE_DEFINE(OpenGL_CopySurface, "OpenGL", "CopySurface", MP_RGB(128, 192, 64));
-static void CopySurface(const Surface& src_surface, const Surface& dst_surface,
-                        const GLuint copy_pbo_handle, const GLenum src_attachment = 0,
-                        const GLenum dst_attachment = 0, const std::size_t cubemap_face = 0) {
+void RasterizerCacheOpenGL::CopySurface(const Surface& src_surface, const Surface& dst_surface,
+                                        const GLuint copy_pbo_handle, const GLenum src_attachment,
+                                        const GLenum dst_attachment,
+                                        const std::size_t cubemap_face) {
     MICROPROFILE_SCOPE(OpenGL_CopySurface);
     ASSERT_MSG(dst_attachment == 0, "Unimplemented");
 
@@ -523,6 +536,8 @@ static void CopySurface(const Surface& src_surface, const Surface& dst_surface,
         }
         glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
     }
+
+    dst_surface->MarkAsModified(true, *this);
 }
 
 CachedSurface::CachedSurface(const SurfaceParams& params)
@@ -688,9 +703,20 @@ void CachedSurface::LoadGLBuffer() {
         for (u32 i = 0; i < params.max_mip_level; i++)
             SwizzleFunc(MortonSwizzleMode::MortonToLinear, params, gl_buffer[i], i);
     } else {
-        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[0].assign(texture_src_data, texture_src_data_end);
+        const u32 bpp = params.GetFormatBpp() / 8;
+        const u32 copy_size = params.width * bpp;
+        if (params.pitch == copy_size) {
+            std::memcpy(gl_buffer[0].data(), Memory::GetPointer(params.addr),
+                        params.size_in_bytes_gl);
+        } else {
+            const u8* start = Memory::GetPointer(params.addr);
+            u8* write_to = gl_buffer[0].data();
+            for (u32 h = params.height; h > 0; h--) {
+                std::memcpy(write_to, start, copy_size);
+                start += params.pitch;
+                write_to += copy_size;
+            }
+        }
     }
     for (u32 i = 0; i < params.max_mip_level; i++) {
         ConvertFormatAsNeeded_LoadGLBuffer(gl_buffer[i], params.pixel_format, params.MipWidth(i),
@@ -727,7 +753,19 @@ void CachedSurface::FlushGLBuffer() {
 
         SwizzleFunc(MortonSwizzleMode::LinearToMorton, params, gl_buffer[0], 0);
     } else {
-        std::memcpy(Memory::GetPointer(GetAddr()), gl_buffer[0].data(), GetSizeInBytes());
+        const u32 bpp = params.GetFormatBpp() / 8;
+        const u32 copy_size = params.width * bpp;
+        if (params.pitch == copy_size) {
+            std::memcpy(Memory::GetPointer(params.addr), gl_buffer[0].data(), GetSizeInBytes());
+        } else {
+            u8* start = Memory::GetPointer(params.addr);
+            const u8* read_to = gl_buffer[0].data();
+            for (u32 h = params.height; h > 0; h--) {
+                std::memcpy(start, read_to, copy_size);
+                start += params.pitch;
+                read_to += copy_size;
+            }
+        }
     }
 }
 
@@ -853,8 +891,8 @@ void CachedSurface::EnsureTextureView() {
     constexpr GLuint min_level = 0;
 
     glGenTextures(1, &texture_view.handle);
-    glTextureView(texture_view.handle, target, texture.handle, gl_internal_format, 0,
-                  params.max_mip_level, 0, 1);
+    glTextureView(texture_view.handle, target, texture.handle, gl_internal_format, min_level,
+                  params.max_mip_level, min_layer, num_layers);
     ApplyTextureDefaults(texture_view.handle, params.max_mip_level);
     glTextureParameteriv(texture_view.handle, GL_TEXTURE_SWIZZLE_RGBA,
                          reinterpret_cast<const GLint*>(swizzle.data()));
@@ -1019,26 +1057,161 @@ void RasterizerCacheOpenGL::FastLayeredCopySurface(const Surface& src_surface,
         }
         address += layer_size;
     }
+
+    dst_surface->MarkAsModified(true, *this);
+}
+
+static bool BlitSurface(const Surface& src_surface, const Surface& dst_surface,
+                        const MathUtil::Rectangle<u32>& src_rect,
+                        const MathUtil::Rectangle<u32>& dst_rect, GLuint read_fb_handle,
+                        GLuint draw_fb_handle, GLenum src_attachment = 0, GLenum dst_attachment = 0,
+                        std::size_t cubemap_face = 0) {
+
+    const auto& src_params{src_surface->GetSurfaceParams()};
+    const auto& dst_params{dst_surface->GetSurfaceParams()};
+
+    OpenGLState prev_state{OpenGLState::GetCurState()};
+    SCOPE_EXIT({ prev_state.Apply(); });
+
+    OpenGLState state;
+    state.draw.read_framebuffer = read_fb_handle;
+    state.draw.draw_framebuffer = draw_fb_handle;
+    state.Apply();
+
+    u32 buffers{};
+
+    if (src_params.type == SurfaceType::ColorTexture) {
+        switch (src_params.target) {
+        case SurfaceTarget::Texture2D:
+            glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
+                                   GL_TEXTURE_2D, src_surface->Texture().handle, 0);
+            glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
+                                   0, 0);
+            break;
+        case SurfaceTarget::TextureCubemap:
+            glFramebufferTexture2D(
+                GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
+                static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + cubemap_face),
+                src_surface->Texture().handle, 0);
+            glFramebufferTexture2D(
+                GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT,
+                static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + cubemap_face), 0, 0);
+            break;
+        case SurfaceTarget::Texture2DArray:
+            glFramebufferTextureLayer(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
+                                      src_surface->Texture().handle, 0, 0);
+            glFramebufferTextureLayer(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, 0, 0, 0);
+            break;
+        case SurfaceTarget::Texture3D:
+            glFramebufferTexture3D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
+                                   SurfaceTargetToGL(src_params.target),
+                                   src_surface->Texture().handle, 0, 0);
+            glFramebufferTexture3D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT,
+                                   SurfaceTargetToGL(src_params.target), 0, 0, 0);
+            break;
+        default:
+            glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
+                                   GL_TEXTURE_2D, src_surface->Texture().handle, 0);
+            glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
+                                   0, 0);
+            break;
+        }
+
+        switch (dst_params.target) {
+        case SurfaceTarget::Texture2D:
+            glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
+                                   GL_TEXTURE_2D, dst_surface->Texture().handle, 0);
+            glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
+                                   0, 0);
+            break;
+        case SurfaceTarget::TextureCubemap:
+            glFramebufferTexture2D(
+                GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
+                static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + cubemap_face),
+                dst_surface->Texture().handle, 0);
+            glFramebufferTexture2D(
+                GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT,
+                static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + cubemap_face), 0, 0);
+            break;
+        case SurfaceTarget::Texture2DArray:
+            glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
+                                      dst_surface->Texture().handle, 0, 0);
+            glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, 0, 0, 0);
+            break;
+
+        case SurfaceTarget::Texture3D:
+            glFramebufferTexture3D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
+                                   SurfaceTargetToGL(dst_params.target),
+                                   dst_surface->Texture().handle, 0, 0);
+            glFramebufferTexture3D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT,
+                                   SurfaceTargetToGL(dst_params.target), 0, 0, 0);
+            break;
+        default:
+            glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
+                                   GL_TEXTURE_2D, dst_surface->Texture().handle, 0);
+            glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
+                                   0, 0);
+            break;
+        }
+
+        buffers = GL_COLOR_BUFFER_BIT;
+    } else if (src_params.type == SurfaceType::Depth) {
+        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
+                               GL_TEXTURE_2D, 0, 0);
+        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D,
+                               src_surface->Texture().handle, 0);
+        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
+
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
+                               GL_TEXTURE_2D, 0, 0);
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D,
+                               dst_surface->Texture().handle, 0);
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
+
+        buffers = GL_DEPTH_BUFFER_BIT;
+    } else if (src_params.type == SurfaceType::DepthStencil) {
+        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
+                               GL_TEXTURE_2D, 0, 0);
+        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
+                               src_surface->Texture().handle, 0);
+
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
+                               GL_TEXTURE_2D, 0, 0);
+        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
+                               dst_surface->Texture().handle, 0);
+
+        buffers = GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT;
+    }
+
+    glBlitFramebuffer(src_rect.left, src_rect.top, src_rect.right, src_rect.bottom, dst_rect.left,
+                      dst_rect.top, dst_rect.right, dst_rect.bottom, buffers,
+                      buffers == GL_COLOR_BUFFER_BIT ? GL_LINEAR : GL_NEAREST);
+
+    return true;
 }
 
 void RasterizerCacheOpenGL::FermiCopySurface(
     const Tegra::Engines::Fermi2D::Regs::Surface& src_config,
-    const Tegra::Engines::Fermi2D::Regs::Surface& dst_config) {
+    const Tegra::Engines::Fermi2D::Regs::Surface& dst_config,
+    const MathUtil::Rectangle<u32>& src_rect, const MathUtil::Rectangle<u32>& dst_rect) {
 
     const auto& src_params = SurfaceParams::CreateForFermiCopySurface(src_config);
     const auto& dst_params = SurfaceParams::CreateForFermiCopySurface(dst_config);
 
-    ASSERT(src_params.width == dst_params.width);
-    ASSERT(src_params.height == dst_params.height);
     ASSERT(src_params.pixel_format == dst_params.pixel_format);
     ASSERT(src_params.block_height == dst_params.block_height);
     ASSERT(src_params.is_tiled == dst_params.is_tiled);
     ASSERT(src_params.depth == dst_params.depth);
-    ASSERT(src_params.depth == 1); // Currently, FastCopySurface only works with 2D surfaces
     ASSERT(src_params.target == dst_params.target);
     ASSERT(src_params.rt.index == dst_params.rt.index);
 
-    FastCopySurface(GetSurface(src_params, true), GetSurface(dst_params, false));
+    auto src_surface = GetSurface(src_params, true);
+    auto dst_surface = GetSurface(dst_params, true);
+
+    BlitSurface(src_surface, dst_surface, src_rect, dst_rect, read_framebuffer.handle,
+                draw_framebuffer.handle);
+
+    dst_surface->MarkAsModified(true, *this);
 }
 
 void RasterizerCacheOpenGL::AccurateCopySurface(const Surface& src_surface,
diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h
index 490b8252e..b81882d04 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h
@@ -8,6 +8,7 @@
 #include <map>
 #include <memory>
 #include <string>
+#include <unordered_set>
 #include <vector>
 
 #include "common/alignment.h"
@@ -272,6 +273,7 @@ struct SurfaceParams {
     u32 height;
     u32 depth;
     u32 unaligned_height;
+    u32 pitch;
     SurfaceTarget target;
     SurfaceClass identity;
     u32 max_mip_level;
@@ -421,7 +423,9 @@ public:
 
     /// Copies the contents of one surface to another
     void FermiCopySurface(const Tegra::Engines::Fermi2D::Regs::Surface& src_config,
-                          const Tegra::Engines::Fermi2D::Regs::Surface& dst_config);
+                          const Tegra::Engines::Fermi2D::Regs::Surface& dst_config,
+                          const MathUtil::Rectangle<u32>& src_rect,
+                          const MathUtil::Rectangle<u32>& dst_rect);
 
 private:
     void LoadSurface(const Surface& surface);
@@ -442,6 +446,10 @@ private:
     /// Performs a slow but accurate surface copy, flushing to RAM and reinterpreting the data
     void AccurateCopySurface(const Surface& src_surface, const Surface& dst_surface);
     void FastLayeredCopySurface(const Surface& src_surface, const Surface& dst_surface);
+    void FastCopySurface(const Surface& src_surface, const Surface& dst_surface);
+    void CopySurface(const Surface& src_surface, const Surface& dst_surface,
+                     const GLuint copy_pbo_handle, const GLenum src_attachment = 0,
+                     const GLenum dst_attachment = 0, const std::size_t cubemap_face = 0);
 
     /// 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
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 70e124dc4..b39bb4843 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -719,45 +719,51 @@ private:
         constexpr std::array<const char*, 4> coord_constructors = {"float", "vec2", "vec3", "vec4"};
 
         const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
-        const auto count = static_cast<u32>(operation.GetOperandsCount());
         ASSERT(meta);
 
+        const auto count = static_cast<u32>(operation.GetOperandsCount());
+        const bool has_array = meta->sampler.IsArray();
+        const bool has_shadow = meta->sampler.IsShadow();
+
         std::string expr = func;
         expr += '(';
         expr += GetSampler(meta->sampler);
         expr += ", ";
 
-        expr += coord_constructors[meta->coords_count - 1];
+        expr += coord_constructors.at(count + (has_array ? 1 : 0) + (has_shadow ? 1 : 0) - 1);
         expr += '(';
         for (u32 i = 0; i < count; ++i) {
-            const bool is_extra = i >= meta->coords_count;
-            const bool is_array = i == meta->array_index;
-
-            std::string operand = [&]() {
-                if (is_extra && is_extra_int) {
-                    if (const auto immediate = std::get_if<ImmediateNode>(operation[i])) {
-                        return std::to_string(static_cast<s32>(immediate->GetValue()));
-                    } else {
-                        return "ftoi(" + Visit(operation[i]) + ')';
-                    }
-                } else {
-                    return Visit(operation[i]);
-                }
-            }();
-            if (is_array) {
-                ASSERT(!is_extra);
-                operand = "float(ftoi(" + operand + "))";
-            }
-
-            expr += operand;
+            expr += Visit(operation[i]);
 
-            if (i + 1 == meta->coords_count) {
-                expr += ')';
-            }
-            if (i + 1 < count) {
+            const u32 next = i + 1;
+            if (next < count || has_array || has_shadow)
+                expr += ", ";
+        }
+        if (has_array) {
+            expr += "float(ftoi(" + Visit(meta->array) + "))";
+        }
+        if (has_shadow) {
+            if (has_array)
                 expr += ", ";
+            expr += Visit(meta->depth_compare);
+        }
+        expr += ')';
+
+        for (const Node extra : meta->extras) {
+            expr += ", ";
+            if (is_extra_int) {
+                if (const auto immediate = std::get_if<ImmediateNode>(extra)) {
+                    // Inline the string as an immediate integer in GLSL (some extra arguments are
+                    // required to be constant)
+                    expr += std::to_string(static_cast<s32>(immediate->GetValue()));
+                } else {
+                    expr += "ftoi(" + Visit(extra) + ')';
+                }
+            } else {
+                expr += Visit(extra);
             }
         }
+
         expr += ')';
         return expr;
     }
@@ -1134,7 +1140,7 @@ private:
                                   Type::HalfFloat);
     }
 
-    std::string F4Texture(Operation operation) {
+    std::string Texture(Operation operation) {
         const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
         ASSERT(meta);
 
@@ -1145,7 +1151,7 @@ private:
         return expr + GetSwizzle(meta->element);
     }
 
-    std::string F4TextureLod(Operation operation) {
+    std::string TextureLod(Operation operation) {
         const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
         ASSERT(meta);
 
@@ -1156,7 +1162,7 @@ private:
         return expr + GetSwizzle(meta->element);
     }
 
-    std::string F4TextureGather(Operation operation) {
+    std::string TextureGather(Operation operation) {
         const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
         ASSERT(meta);
 
@@ -1164,7 +1170,7 @@ private:
                GetSwizzle(meta->element);
     }
 
-    std::string F4TextureQueryDimensions(Operation operation) {
+    std::string TextureQueryDimensions(Operation operation) {
         const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
         ASSERT(meta);
 
@@ -1184,7 +1190,7 @@ private:
         return "0";
     }
 
-    std::string F4TextureQueryLod(Operation operation) {
+    std::string TextureQueryLod(Operation operation) {
         const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
         ASSERT(meta);
 
@@ -1195,29 +1201,32 @@ private:
         return "0";
     }
 
-    std::string F4TexelFetch(Operation operation) {
+    std::string TexelFetch(Operation operation) {
         constexpr std::array<const char*, 4> constructors = {"int", "ivec2", "ivec3", "ivec4"};
         const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
-        const auto count = static_cast<u32>(operation.GetOperandsCount());
         ASSERT(meta);
+        UNIMPLEMENTED_IF(meta->sampler.IsArray());
+        UNIMPLEMENTED_IF(!meta->extras.empty());
+
+        const auto count = static_cast<u32>(operation.GetOperandsCount());
 
         std::string expr = "texelFetch(";
         expr += GetSampler(meta->sampler);
         expr += ", ";
 
-        expr += constructors[meta->coords_count - 1];
+        expr += constructors.at(count - 1);
         expr += '(';
         for (u32 i = 0; i < count; ++i) {
             expr += VisitOperand(operation, i, Type::Int);
 
-            if (i + 1 == meta->coords_count) {
+            const u32 next = i + 1;
+            if (next == count)
                 expr += ')';
-            }
-            if (i + 1 < count) {
+            if (next < count)
                 expr += ", ";
-            }
         }
         expr += ')';
+
         return expr + GetSwizzle(meta->element);
     }
 
@@ -1454,12 +1463,12 @@ private:
         &GLSLDecompiler::Logical2HNotEqual,
         &GLSLDecompiler::Logical2HGreaterEqual,
 
-        &GLSLDecompiler::F4Texture,
-        &GLSLDecompiler::F4TextureLod,
-        &GLSLDecompiler::F4TextureGather,
-        &GLSLDecompiler::F4TextureQueryDimensions,
-        &GLSLDecompiler::F4TextureQueryLod,
-        &GLSLDecompiler::F4TexelFetch,
+        &GLSLDecompiler::Texture,
+        &GLSLDecompiler::TextureLod,
+        &GLSLDecompiler::TextureGather,
+        &GLSLDecompiler::TextureQueryDimensions,
+        &GLSLDecompiler::TextureQueryLod,
+        &GLSLDecompiler::TexelFetch,
 
         &GLSLDecompiler::Branch,
         &GLSLDecompiler::PushFlowStack,
diff --git a/src/video_core/shader/decode/arithmetic_integer.cpp b/src/video_core/shader/decode/arithmetic_integer.cpp
index 38bb692d6..9fd4b273e 100644
--- a/src/video_core/shader/decode/arithmetic_integer.cpp
+++ b/src/video_core/shader/decode/arithmetic_integer.cpp
@@ -41,7 +41,7 @@ u32 ShaderIR::DecodeArithmeticInteger(NodeBlock& bb, u32 pc) {
 
         const Node value = Operation(OperationCode::IAdd, PRECISE, op_a, op_b);
 
-        SetInternalFlagsFromInteger(bb, value, instr.op_32.generates_cc);
+        SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
         SetRegister(bb, instr.gpr0, value);
         break;
     }
@@ -284,4 +284,4 @@ void ShaderIR::WriteLop3Instruction(NodeBlock& bb, Register dest, Node op_a, Nod
     SetRegister(bb, dest, value);
 }
 
-} // namespace VideoCommon::Shader
\ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/conversion.cpp b/src/video_core/shader/decode/conversion.cpp
index a992f73f8..55a6fbbf2 100644
--- a/src/video_core/shader/decode/conversion.cpp
+++ b/src/video_core/shader/decode/conversion.cpp
@@ -118,8 +118,8 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
 
         value = [&]() {
             switch (instr.conversion.f2i.rounding) {
-            case Tegra::Shader::F2iRoundingOp::None:
-                return value;
+            case Tegra::Shader::F2iRoundingOp::RoundEven:
+                return Operation(OperationCode::FRoundEven, PRECISE, value);
             case Tegra::Shader::F2iRoundingOp::Floor:
                 return Operation(OperationCode::FFloor, PRECISE, value);
             case Tegra::Shader::F2iRoundingOp::Ceil:
@@ -146,4 +146,4 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
     return pc;
 }
 
-} // namespace VideoCommon::Shader
\ No newline at end of file
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/decode/memory.cpp b/src/video_core/shader/decode/memory.cpp
index e006f8138..523421794 100644
--- a/src/video_core/shader/decode/memory.cpp
+++ b/src/video_core/shader/decode/memory.cpp
@@ -306,7 +306,6 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
     case OpCode::Id::TLD4S: {
         UNIMPLEMENTED_IF_MSG(instr.tld4s.UsesMiscMode(TextureMiscMode::AOFFI),
                              "AOFFI is not implemented");
-
         if (instr.tld4s.UsesMiscMode(TextureMiscMode::NODEP)) {
             LOG_WARNING(HW_GPU, "TLD4S.NODEP implementation is incomplete");
         }
@@ -315,9 +314,8 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         const Node op_a = GetRegister(instr.gpr8);
         const Node op_b = GetRegister(instr.gpr20);
 
-        std::vector<Node> coords;
-
         // TODO(Subv): Figure out how the sampler type is encoded in the TLD4S instruction.
+        std::vector<Node> coords;
         if (depth_compare) {
             // Note: TLD4S coordinate encoding works just like TEXS's
             const Node op_y = GetRegister(instr.gpr8.Value() + 1);
@@ -328,18 +326,17 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
             coords.push_back(op_a);
             coords.push_back(op_b);
         }
-        const auto num_coords = static_cast<u32>(coords.size());
-        coords.push_back(Immediate(static_cast<u32>(instr.tld4s.component)));
+        std::vector<Node> extras;
+        extras.push_back(Immediate(static_cast<u32>(instr.tld4s.component)));
 
         const auto& sampler =
             GetSampler(instr.sampler, TextureType::Texture2D, false, depth_compare);
 
         Node4 values;
         for (u32 element = 0; element < values.size(); ++element) {
-            auto params = coords;
-            MetaTexture meta{sampler, element, num_coords};
-            values[element] =
-                Operation(OperationCode::F4TextureGather, std::move(meta), std::move(params));
+            auto coords_copy = coords;
+            MetaTexture meta{sampler, {}, {}, extras, element};
+            values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy));
         }
 
         WriteTexsInstructionFloat(bb, instr, values);
@@ -360,12 +357,13 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         switch (instr.txq.query_type) {
         case Tegra::Shader::TextureQueryType::Dimension: {
             for (u32 element = 0; element < 4; ++element) {
-                if (instr.txq.IsComponentEnabled(element)) {
-                    MetaTexture meta{sampler, element};
-                    const Node value = Operation(OperationCode::F4TextureQueryDimensions,
-                                                 std::move(meta), GetRegister(instr.gpr8));
-                    SetTemporal(bb, indexer++, value);
+                if (!instr.txq.IsComponentEnabled(element)) {
+                    continue;
                 }
+                MetaTexture meta{sampler, {}, {}, {}, element};
+                const Node value =
+                    Operation(OperationCode::TextureQueryDimensions, meta, GetRegister(instr.gpr8));
+                SetTemporal(bb, indexer++, value);
             }
             for (u32 i = 0; i < indexer; ++i) {
                 SetRegister(bb, instr.gpr0.Value() + i, GetTemporal(i));
@@ -412,9 +410,8 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
 
         for (u32 element = 0; element < 2; ++element) {
             auto params = coords;
-            MetaTexture meta_texture{sampler, element, static_cast<u32>(coords.size())};
-            const Node value =
-                Operation(OperationCode::F4TextureQueryLod, meta_texture, std::move(params));
+            MetaTexture meta{sampler, {}, {}, {}, element};
+            const Node value = Operation(OperationCode::TextureQueryLod, meta, std::move(params));
             SetTemporal(bb, element, value);
         }
         for (u32 element = 0; element < 2; ++element) {
@@ -535,15 +532,16 @@ void ShaderIR::WriteTexsInstructionHalfFloat(NodeBlock& bb, Instruction instr,
 }
 
 Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
-                               TextureProcessMode process_mode, bool depth_compare, bool is_array,
-                               std::size_t array_offset, std::size_t bias_offset,
-                               std::vector<Node>&& coords) {
-    UNIMPLEMENTED_IF_MSG(
-        (texture_type == TextureType::Texture3D && (is_array || depth_compare)) ||
-            (texture_type == TextureType::TextureCube && is_array && depth_compare),
-        "This method is not supported.");
+                               TextureProcessMode process_mode, std::vector<Node> coords,
+                               Node array, Node depth_compare, u32 bias_offset) {
+    const bool is_array = array;
+    const bool is_shadow = depth_compare;
 
-    const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, depth_compare);
+    UNIMPLEMENTED_IF_MSG((texture_type == TextureType::Texture3D && (is_array || is_shadow)) ||
+                             (texture_type == TextureType::TextureCube && is_array && is_shadow),
+                         "This method is not supported.");
+
+    const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, is_shadow);
 
     const bool lod_needed = process_mode == TextureProcessMode::LZ ||
                             process_mode == TextureProcessMode::LL ||
@@ -552,35 +550,30 @@ Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
     // LOD selection (either via bias or explicit textureLod) not supported in GL for
     // sampler2DArrayShadow and samplerCubeArrayShadow.
     const bool gl_lod_supported =
-        !((texture_type == Tegra::Shader::TextureType::Texture2D && is_array && depth_compare) ||
-          (texture_type == Tegra::Shader::TextureType::TextureCube && is_array && depth_compare));
+        !((texture_type == Tegra::Shader::TextureType::Texture2D && is_array && is_shadow) ||
+          (texture_type == Tegra::Shader::TextureType::TextureCube && is_array && is_shadow));
 
     const OperationCode read_method =
-        lod_needed && gl_lod_supported ? OperationCode::F4TextureLod : OperationCode::F4Texture;
+        lod_needed && gl_lod_supported ? OperationCode::TextureLod : OperationCode::Texture;
 
     UNIMPLEMENTED_IF(process_mode != TextureProcessMode::None && !gl_lod_supported);
 
-    std::optional<u32> array_offset_value;
-    if (is_array)
-        array_offset_value = static_cast<u32>(array_offset);
-
-    const auto coords_count = static_cast<u32>(coords.size());
-
+    std::vector<Node> extras;
     if (process_mode != TextureProcessMode::None && gl_lod_supported) {
         if (process_mode == TextureProcessMode::LZ) {
-            coords.push_back(Immediate(0.0f));
+            extras.push_back(Immediate(0.0f));
         } else {
             // If present, lod or bias are always stored in the register indexed by the gpr20
             // field with an offset depending on the usage of the other registers
-            coords.push_back(GetRegister(instr.gpr20.Value() + bias_offset));
+            extras.push_back(GetRegister(instr.gpr20.Value() + bias_offset));
         }
     }
 
     Node4 values;
     for (u32 element = 0; element < values.size(); ++element) {
-        auto params = coords;
-        MetaTexture meta{sampler, element, coords_count, array_offset_value};
-        values[element] = Operation(read_method, std::move(meta), std::move(params));
+        auto copy_coords = coords;
+        MetaTexture meta{sampler, array, depth_compare, extras, element};
+        values[element] = Operation(read_method, meta, std::move(copy_coords));
     }
 
     return values;
@@ -602,28 +595,22 @@ Node4 ShaderIR::GetTexCode(Instruction instr, TextureType texture_type,
     for (std::size_t i = 0; i < coord_count; ++i) {
         coords.push_back(GetRegister(coord_register + i));
     }
-    // 1D.DC in opengl the 2nd component is ignored.
+    // 1D.DC in OpenGL the 2nd component is ignored.
     if (depth_compare && !is_array && texture_type == TextureType::Texture1D) {
         coords.push_back(Immediate(0.0f));
     }
-    std::size_t array_offset{};
-    if (is_array) {
-        array_offset = coords.size();
-        coords.push_back(GetRegister(array_register));
-    }
+
+    const Node array = is_array ? GetRegister(array_register) : nullptr;
+
+    Node dc{};
     if (depth_compare) {
-        // Depth is always stored in the register signaled by gpr20
-        // or in the next register if lod or bias are used
+        // Depth is always stored in the register signaled by gpr20 or in the next register if lod
+        // or bias are used
         const u64 depth_register = instr.gpr20.Value() + (lod_bias_enabled ? 1 : 0);
-        coords.push_back(GetRegister(depth_register));
-    }
-    // Fill ignored coordinates
-    while (coords.size() < total_coord_count) {
-        coords.push_back(Immediate(0));
+        dc = GetRegister(depth_register);
     }
 
-    return GetTextureCode(instr, texture_type, process_mode, depth_compare, is_array, array_offset,
-                          0, std::move(coords));
+    return GetTextureCode(instr, texture_type, process_mode, coords, array, dc, 0);
 }
 
 Node4 ShaderIR::GetTexsCode(Instruction instr, TextureType texture_type,
@@ -641,6 +628,7 @@ Node4 ShaderIR::GetTexsCode(Instruction instr, TextureType texture_type,
         (is_array || !(lod_bias_enabled || depth_compare) || (coord_count > 2))
             ? static_cast<u64>(instr.gpr20.Value())
             : coord_register + 1;
+    const u32 bias_offset = coord_count > 2 ? 1 : 0;
 
     std::vector<Node> coords;
     for (std::size_t i = 0; i < coord_count; ++i) {
@@ -648,24 +636,17 @@ Node4 ShaderIR::GetTexsCode(Instruction instr, TextureType texture_type,
         coords.push_back(GetRegister(last ? last_coord_register : coord_register + i));
     }
 
-    std::size_t array_offset{};
-    if (is_array) {
-        array_offset = coords.size();
-        coords.push_back(GetRegister(array_register));
-    }
+    const Node array = is_array ? GetRegister(array_register) : nullptr;
+
+    Node dc{};
     if (depth_compare) {
-        // Depth is always stored in the register signaled by gpr20
-        // or in the next register if lod or bias are used
+        // Depth is always stored in the register signaled by gpr20 or in the next register if lod
+        // or bias are used
         const u64 depth_register = instr.gpr20.Value() + (lod_bias_enabled ? 1 : 0);
-        coords.push_back(GetRegister(depth_register));
-    }
-    // Fill ignored coordinates
-    while (coords.size() < total_coord_count) {
-        coords.push_back(Immediate(0));
+        dc = GetRegister(depth_register);
     }
 
-    return GetTextureCode(instr, texture_type, process_mode, depth_compare, is_array, array_offset,
-                          (coord_count > 2 ? 1 : 0), std::move(coords));
+    return GetTextureCode(instr, texture_type, process_mode, coords, array, dc, bias_offset);
 }
 
 Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool depth_compare,
@@ -680,24 +661,16 @@ Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool de
     const u64 coord_register = array_register + (is_array ? 1 : 0);
 
     std::vector<Node> coords;
-
-    for (size_t i = 0; i < coord_count; ++i) {
+    for (size_t i = 0; i < coord_count; ++i)
         coords.push_back(GetRegister(coord_register + i));
-    }
-    std::optional<u32> array_offset;
-    if (is_array) {
-        array_offset = static_cast<u32>(coords.size());
-        coords.push_back(GetRegister(array_register));
-    }
 
     const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, depth_compare);
 
     Node4 values;
     for (u32 element = 0; element < values.size(); ++element) {
-        auto params = coords;
-        MetaTexture meta{sampler, element, static_cast<u32>(coords.size()), array_offset};
-        values[element] =
-            Operation(OperationCode::F4TextureGather, std::move(meta), std::move(params));
+        auto coords_copy = coords;
+        MetaTexture meta{sampler, GetRegister(array_register), {}, {}, element};
+        values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy));
     }
 
     return values;
@@ -705,7 +678,6 @@ Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool de
 
 Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is_array) {
     const std::size_t type_coord_count = GetCoordCount(texture_type);
-    const std::size_t total_coord_count = type_coord_count + (is_array ? 1 : 0);
     const bool lod_enabled = instr.tlds.GetTextureProcessMode() == TextureProcessMode::LL;
 
     // If enabled arrays index is always stored in the gpr8 field
@@ -719,33 +691,22 @@ Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is
             : coord_register + 1;
 
     std::vector<Node> coords;
-
     for (std::size_t i = 0; i < type_coord_count; ++i) {
         const bool last = (i == (type_coord_count - 1)) && (type_coord_count > 1);
         coords.push_back(GetRegister(last ? last_coord_register : coord_register + i));
     }
-    std::optional<u32> array_offset;
-    if (is_array) {
-        array_offset = static_cast<u32>(coords.size());
-        coords.push_back(GetRegister(array_register));
-    }
-    const auto coords_count = static_cast<u32>(coords.size());
 
-    if (lod_enabled) {
-        // When lod is used always is in grp20
-        coords.push_back(GetRegister(instr.gpr20));
-    } else {
-        coords.push_back(Immediate(0));
-    }
+    const Node array = is_array ? GetRegister(array_register) : nullptr;
+    // When lod is used always is in gpr20
+    const Node lod = lod_enabled ? GetRegister(instr.gpr20) : Immediate(0);
 
     const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, false);
 
     Node4 values;
     for (u32 element = 0; element < values.size(); ++element) {
-        auto params = coords;
-        MetaTexture meta{sampler, element, coords_count, array_offset};
-        values[element] =
-            Operation(OperationCode::F4TexelFetch, std::move(meta), std::move(params));
+        auto coords_copy = coords;
+        MetaTexture meta{sampler, array, {}, {lod}, element};
+        values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy));
     }
     return values;
 }
diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h
index 1d4fbef53..52c7f2c4e 100644
--- a/src/video_core/shader/shader_ir.h
+++ b/src/video_core/shader/shader_ir.h
@@ -156,12 +156,12 @@ enum class OperationCode {
     Logical2HNotEqual,     /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
     Logical2HGreaterEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
 
-    F4Texture,                /// (MetaTexture, float[N] coords, float[M] params) -> float4
-    F4TextureLod,             /// (MetaTexture, float[N] coords, float[M] params) -> float4
-    F4TextureGather,          /// (MetaTexture, float[N] coords, float[M] params) -> float4
-    F4TextureQueryDimensions, /// (MetaTexture, float a) -> float4
-    F4TextureQueryLod,        /// (MetaTexture, float[N] coords) -> float4
-    F4TexelFetch,             /// (MetaTexture, int[N], int) -> float4
+    Texture,                /// (MetaTexture, float[N] coords) -> float4
+    TextureLod,             /// (MetaTexture, float[N] coords) -> float4
+    TextureGather,          /// (MetaTexture, float[N] coords) -> float4
+    TextureQueryDimensions, /// (MetaTexture, float a) -> float4
+    TextureQueryLod,        /// (MetaTexture, float[N] coords) -> float4
+    TexelFetch,             /// (MetaTexture, int[N], int) -> float4
 
     Branch,        /// (uint branch_target) -> void
     PushFlowStack, /// (uint branch_target) -> void
@@ -288,9 +288,10 @@ struct MetaHalfArithmetic {
 
 struct MetaTexture {
     const Sampler& sampler;
+    Node array{};
+    Node depth_compare{};
+    std::vector<Node> extras;
     u32 element{};
-    u32 coords_count{};
-    std::optional<u32> array_index;
 };
 
 constexpr MetaArithmetic PRECISE = {true};
@@ -754,9 +755,8 @@ private:
         bool lod_bias_enabled, std::size_t max_coords, std::size_t max_inputs);
 
     Node4 GetTextureCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type,
-                         Tegra::Shader::TextureProcessMode process_mode, bool depth_compare,
-                         bool is_array, std::size_t array_offset, std::size_t bias_offset,
-                         std::vector<Node>&& coords);
+                         Tegra::Shader::TextureProcessMode process_mode, std::vector<Node> coords,
+                         Node array, Node depth_compare, u32 bias_offset);
 
     Node GetVideoOperand(Node op, bool is_chunk, bool is_signed, Tegra::Shader::VideoType type,
                          u64 byte_height);