7 files changed, 188 insertions, 68 deletions

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/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..f585c8884 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
@@ -1021,24 +1021,155 @@ void RasterizerCacheOpenGL::FastLayeredCopySurface(const Surface& src_surface,
     }
 }
 
+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);
 }
 
 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..c742f4a93 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h
@@ -421,7 +421,9 @@ 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);