10 files changed, 121 insertions, 40 deletions

diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp
index ea1555c5d..912e785b9 100644
--- a/src/video_core/engines/fermi_2d.cpp
+++ b/src/video_core/engines/fermi_2d.cpp
@@ -4,11 +4,13 @@
 
 #include "core/memory.h"
 #include "video_core/engines/fermi_2d.h"
+#include "video_core/rasterizer_interface.h"
 #include "video_core/textures/decoders.h"
 
 namespace Tegra::Engines {
 
-Fermi2D::Fermi2D(MemoryManager& memory_manager) : memory_manager(memory_manager) {}
+Fermi2D::Fermi2D(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager)
+    : memory_manager(memory_manager), rasterizer{rasterizer} {}
 
 void Fermi2D::WriteReg(u32 method, u32 value) {
     ASSERT_MSG(method < Regs::NUM_REGS,
@@ -44,27 +46,31 @@ void Fermi2D::HandleSurfaceCopy() {
     u32 src_bytes_per_pixel = RenderTargetBytesPerPixel(regs.src.format);
     u32 dst_bytes_per_pixel = RenderTargetBytesPerPixel(regs.dst.format);
 
-    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, src_bytes_per_pixel,
-                                  dst_bytes_per_pixel, src_buffer, dst_buffer, true,
-                                  regs.src.BlockHeight());
-    } 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, src_bytes_per_pixel,
-                                  dst_bytes_per_pixel, dst_buffer, src_buffer, false,
-                                  regs.dst.BlockHeight());
+    if (!rasterizer.AccelerateSurfaceCopy(regs.src, regs.dst)) {
+        // TODO(bunnei): The below implementation currently will not get hit, as
+        // AccelerateSurfaceCopy tries to always copy and will always return success. This should be
+        // changed once we properly support flushing.
+
+        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, src_bytes_per_pixel,
+                                      dst_bytes_per_pixel, src_buffer, dst_buffer, true,
+                                      regs.src.BlockHeight());
+        } 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, src_bytes_per_pixel,
+                                      dst_bytes_per_pixel, dst_buffer, src_buffer, false,
+                                      regs.dst.BlockHeight());
+        }
     }
 }
 
diff --git a/src/video_core/engines/fermi_2d.h b/src/video_core/engines/fermi_2d.h
index 021b83eaa..81d15c62a 100644
--- a/src/video_core/engines/fermi_2d.h
+++ b/src/video_core/engines/fermi_2d.h
@@ -12,6 +12,10 @@
 #include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
 
+namespace VideoCore {
+class RasterizerInterface;
+}
+
 namespace Tegra::Engines {
 
 #define FERMI2D_REG_INDEX(field_name)                                                              \
@@ -19,7 +23,7 @@ namespace Tegra::Engines {
 
 class Fermi2D final {
 public:
-    explicit Fermi2D(MemoryManager& memory_manager);
+    explicit Fermi2D(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager);
     ~Fermi2D() = default;
 
     /// Write the value to the register identified by method.
@@ -94,6 +98,8 @@ public:
     MemoryManager& memory_manager;
 
 private:
+    VideoCore::RasterizerInterface& rasterizer;
+
     /// Performs the copy from the source surface to the destination surface as configured in the
     /// registers.
     void HandleSurfaceCopy();
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index baa8b63b7..9ba7e3533 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -25,7 +25,7 @@ u32 FramebufferConfig::BytesPerPixel(PixelFormat format) {
 GPU::GPU(VideoCore::RasterizerInterface& rasterizer) {
     memory_manager = std::make_unique<Tegra::MemoryManager>();
     maxwell_3d = std::make_unique<Engines::Maxwell3D>(rasterizer, *memory_manager);
-    fermi_2d = std::make_unique<Engines::Fermi2D>(*memory_manager);
+    fermi_2d = std::make_unique<Engines::Fermi2D>(rasterizer, *memory_manager);
     maxwell_compute = std::make_unique<Engines::MaxwellCompute>();
     maxwell_dma = std::make_unique<Engines::MaxwellDMA>(*memory_manager);
     kepler_memory = std::make_unique<Engines::KeplerMemory>(*memory_manager);
diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h
index cd819d69f..06fc59dbe 100644
--- a/src/video_core/rasterizer_interface.h
+++ b/src/video_core/rasterizer_interface.h
@@ -5,6 +5,7 @@
 #pragma once
 
 #include "common/common_types.h"
+#include "video_core/engines/fermi_2d.h"
 #include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
 
@@ -33,13 +34,9 @@ public:
     /// and invalidated
     virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0;
 
-    /// Attempt to use a faster method to perform a display transfer with is_texture_copy = 0
-    virtual bool AccelerateDisplayTransfer(const void* config) {
-        return false;
-    }
-
-    /// Attempt to use a faster method to perform a display transfer with is_texture_copy = 1
-    virtual bool AccelerateTextureCopy(const void* config) {
+    /// 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) {
         return false;
     }
 
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index edcd5e451..209bdf181 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -617,14 +617,10 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) {
     InvalidateRegion(addr, size);
 }
 
-bool RasterizerOpenGL::AccelerateDisplayTransfer(const void* config) {
+bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
+                                             const Tegra::Engines::Fermi2D::Regs::Surface& dst) {
     MICROPROFILE_SCOPE(OpenGL_Blits);
-    UNREACHABLE();
-    return true;
-}
-
-bool RasterizerOpenGL::AccelerateTextureCopy(const void* config) {
-    UNREACHABLE();
+    res_cache.FermiCopySurface(src, dst);
     return true;
 }
 
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index dc31a2dbc..0dab2018b 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -52,8 +52,8 @@ public:
     void FlushRegion(VAddr addr, u64 size) override;
     void InvalidateRegion(VAddr addr, u64 size) override;
     void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
-    bool AccelerateDisplayTransfer(const void* config) override;
-    bool AccelerateTextureCopy(const void* config) override;
+    bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
+                               const Tegra::Engines::Fermi2D::Regs::Surface& dst) override;
     bool AccelerateFill(const void* config) override;
     bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
                            u32 pixel_stride) override;
diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
index ce967c4d6..56ff83eff 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
@@ -143,6 +143,28 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) {
     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.pixel_format = PixelFormatFromRenderTargetFormat(config.format);
+    params.component_type = ComponentTypeFromRenderTarget(config.format);
+    params.type = GetFormatType(params.pixel_format);
+    params.width = config.width;
+    params.height = config.height;
+    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 = {};
+
+    return params;
+}
+
 static constexpr std::array<FormatTuple, SurfaceParams::MaxPixelFormat> tex_format_tuples = {{
     {GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV, ComponentType::UNorm, false}, // ABGR8U
     {GL_RGBA8, GL_RGBA, GL_BYTE, ComponentType::SNorm, false},                     // ABGR8S
@@ -559,6 +581,18 @@ static bool BlitSurface(const Surface& src_surface, const Surface& dst_surface,
     return true;
 }
 
+static void FastCopySurface(const Surface& src_surface, const Surface& dst_surface) {
+    const auto& src_params{src_surface->GetSurfaceParams()};
+    const auto& dst_params{dst_surface->GetSurfaceParams()};
+
+    const u32 width{std::min(src_params.width, dst_params.width)};
+    const u32 height{std::min(src_params.height, dst_params.height)};
+
+    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);
+}
+
 static void CopySurface(const Surface& src_surface, const Surface& dst_surface,
                         GLuint copy_pbo_handle, GLenum src_attachment = 0,
                         GLenum dst_attachment = 0, std::size_t cubemap_face = 0) {
@@ -1033,6 +1067,26 @@ Surface RasterizerCacheOpenGL::GetUncachedSurface(const SurfaceParams& params) {
     return surface;
 }
 
+void RasterizerCacheOpenGL::FermiCopySurface(
+    const Tegra::Engines::Fermi2D::Regs::Surface& src_config,
+    const Tegra::Engines::Fermi2D::Regs::Surface& dst_config) {
+
+    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));
+}
+
 Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface,
                                                const SurfaceParams& new_params) {
     // Verify surface is compatible for blitting
@@ -1041,6 +1095,15 @@ 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)};
 
+    // 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 &&
+        SurfaceParams::GetFormatBpp(old_params.pixel_format) ==
+            SurfaceParams::GetFormatBpp(new_params.pixel_format)) {
+        FastCopySurface(old_surface, new_surface);
+        return new_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
diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h
index 49025a3fe..0b4940b3c 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h
@@ -13,6 +13,7 @@
 #include "common/common_types.h"
 #include "common/hash.h"
 #include "common/math_util.h"
+#include "video_core/engines/fermi_2d.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/rasterizer_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
@@ -719,6 +720,10 @@ struct SurfaceParams {
                                               Tegra::GPUVAddr zeta_address,
                                               Tegra::DepthFormat format);
 
+    /// Creates SurfaceParams for a Fermi2D surface copy
+    static SurfaceParams CreateForFermiCopySurface(
+        const Tegra::Engines::Fermi2D::Regs::Surface& config);
+
     /// Checks if surfaces are compatible for caching
     bool IsCompatibleSurface(const SurfaceParams& other) const {
         return std::tie(pixel_format, type, width, height, target, depth) ==
@@ -837,6 +842,10 @@ public:
     /// Tries to find a framebuffer using on the provided CPU address
     Surface TryFindFramebufferSurface(VAddr addr) const;
 
+    /// 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);
+
 private:
     void LoadSurface(const Surface& surface);
     Surface GetSurface(const SurfaceParams& params, bool preserve_contents = true);
diff --git a/src/yuzu/main.cpp b/src/yuzu/main.cpp
index ad62a82d0..e11833c5a 100644
--- a/src/yuzu/main.cpp
+++ b/src/yuzu/main.cpp
@@ -485,6 +485,8 @@ QStringList GMainWindow::GetUnsupportedGLExtensions() {
         unsupported_ext.append("ARB_texture_storage");
     if (!GLAD_GL_ARB_multi_bind)
         unsupported_ext.append("ARB_multi_bind");
+    if (!GLAD_GL_ARB_copy_image)
+        unsupported_ext.append("ARB_copy_image");
 
     // Extensions required to support some texture formats.
     if (!GLAD_GL_EXT_texture_compression_s3tc)
diff --git a/src/yuzu_cmd/emu_window/emu_window_sdl2.cpp b/src/yuzu_cmd/emu_window/emu_window_sdl2.cpp
index 0733301b2..155095095 100644
--- a/src/yuzu_cmd/emu_window/emu_window_sdl2.cpp
+++ b/src/yuzu_cmd/emu_window/emu_window_sdl2.cpp
@@ -98,6 +98,8 @@ bool EmuWindow_SDL2::SupportsRequiredGLExtensions() {
         unsupported_ext.push_back("ARB_texture_storage");
     if (!GLAD_GL_ARB_multi_bind)
         unsupported_ext.push_back("ARB_multi_bind");
+    if (!GLAD_GL_ARB_copy_image)
+        unsupported_ext.push_back("ARB_copy_image");
 
     // Extensions required to support some texture formats.
     if (!GLAD_GL_EXT_texture_compression_s3tc)