Merge pull request #2008 from ReinUsesLisp/dirty-framebuffers

gl_rasterizer_cache: Use dirty flags for framebuffers

src/video_core/engines/maxwell_3d.cpp

@@ -135,6 +135,25 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
 
     if (regs.reg_array[method_call.method] != method_call.argument) {
         regs.reg_array[method_call.method] = method_call.argument;
+        // Color buffers
+        constexpr u32 first_rt_reg = MAXWELL3D_REG_INDEX(rt);
+        constexpr u32 registers_per_rt = sizeof(regs.rt[0]) / sizeof(u32);
+        if (method_call.method >= first_rt_reg &&
+            method_call.method < first_rt_reg + registers_per_rt * Regs::NumRenderTargets) {
+            const std::size_t rt_index = (method_call.method - first_rt_reg) / registers_per_rt;
+            dirty_flags.color_buffer |= 1u << static_cast<u32>(rt_index);
+        }
+
+        // Zeta buffer
+        constexpr u32 registers_in_zeta = sizeof(regs.zeta) / sizeof(u32);
+        if (method_call.method == MAXWELL3D_REG_INDEX(zeta_enable) ||
+            method_call.method == MAXWELL3D_REG_INDEX(zeta_width) ||
+            method_call.method == MAXWELL3D_REG_INDEX(zeta_height) ||
+            (method_call.method >= MAXWELL3D_REG_INDEX(zeta) &&
+             method_call.method < MAXWELL3D_REG_INDEX(zeta) + registers_in_zeta)) {
+            dirty_flags.zeta_buffer = true;
+        }
+
         // Shader
         constexpr u32 shader_registers_count =
             sizeof(regs.shader_config[0]) * Regs::MaxShaderProgram / sizeof(u32);

src/video_core/engines/maxwell_3d.h

@@ -1089,12 +1089,17 @@ public:
     MemoryManager& memory_manager;
 
     struct DirtyFlags {
+        u8 color_buffer = 0xFF;
+        bool zeta_buffer = true;
+
         bool shaders = true;
 
         bool vertex_attrib_format = true;
         u32 vertex_array = 0xFFFFFFFF;
 
         void OnMemoryWrite() {
+            color_buffer = 0xFF;
+            zeta_buffer = true;
             shaders = true;
             vertex_array = 0xFFFFFFFF;
         }

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -490,7 +490,19 @@ void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool us
                                              bool using_depth_fb, bool preserve_contents,
                                              std::optional<std::size_t> single_color_target) {
     MICROPROFILE_SCOPE(OpenGL_Framebuffer);
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
+    const auto& regs = gpu.regs;
+
+    const FramebufferConfigState fb_config_state{using_color_fb, using_depth_fb, preserve_contents,
+                                                 single_color_target};
+    if (fb_config_state == current_framebuffer_config_state && gpu.dirty_flags.color_buffer == 0 &&
+        !gpu.dirty_flags.zeta_buffer) {
+        // Only skip if the previous ConfigureFramebuffers call was from the same kind (multiple or
+        // single color targets). This is done because the guest registers may not change but the
+        // host framebuffer may contain different attachments
+        return;
+    }
+    current_framebuffer_config_state = fb_config_state;
 
     Surface depth_surface;
     if (using_depth_fb) {

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -99,6 +99,23 @@ private:
         float max_anisotropic = 1.0f;
     };
 
+    struct FramebufferConfigState {
+        bool using_color_fb{};
+        bool using_depth_fb{};
+        bool preserve_contents{};
+        std::optional<std::size_t> single_color_target;
+
+        bool operator==(const FramebufferConfigState& rhs) const {
+            return std::tie(using_color_fb, using_depth_fb, preserve_contents,
+                            single_color_target) == std::tie(rhs.using_color_fb, rhs.using_depth_fb,
+                                                             rhs.preserve_contents,
+                                                             rhs.single_color_target);
+        }
+        bool operator!=(const FramebufferConfigState& rhs) const {
+            return !operator==(rhs);
+        }
+    };
+
     /**
      * Configures the color and depth framebuffer states.
      * @param use_color_fb If true, configure color framebuffers.
@@ -203,6 +220,7 @@ private:
         vertex_array_cache;
 
     std::map<FramebufferCacheKey, OGLFramebuffer> framebuffer_cache;
+    FramebufferConfigState current_framebuffer_config_state;
 
     std::array<SamplerInfo, Tegra::Engines::Maxwell3D::Regs::NumTextureSamplers> texture_samplers;
 

src/video_core/renderer_opengl/gl_rasterizer_cache.cpp

@@ -919,9 +919,16 @@ Surface RasterizerCacheOpenGL::GetTextureSurface(const Tegra::Texture::FullTextu
 }
 
 Surface RasterizerCacheOpenGL::GetDepthBufferSurface(bool preserve_contents) {
-    const auto& regs{Core::System::GetInstance().GPU().Maxwell3D().regs};
+    auto& gpu{Core::System::GetInstance().GPU().Maxwell3D()};
+    const auto& regs{gpu.regs};
+
+    if (!gpu.dirty_flags.zeta_buffer) {
+        return last_depth_buffer;
+    }
+    gpu.dirty_flags.zeta_buffer = false;
+
     if (!regs.zeta.Address() || !regs.zeta_enable) {
-        return {};
+        return last_depth_buffer = {};
     }
 
     SurfaceParams depth_params{SurfaceParams::CreateForDepthBuffer(
@@ -929,25 +936,31 @@ Surface RasterizerCacheOpenGL::GetDepthBufferSurface(bool preserve_contents) {
         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);
+    return last_depth_buffer = GetSurface(depth_params, preserve_contents);
 }
 
 Surface RasterizerCacheOpenGL::GetColorBufferSurface(std::size_t index, bool preserve_contents) {
-    const auto& regs{Core::System::GetInstance().GPU().Maxwell3D().regs};
+    auto& gpu{Core::System::GetInstance().GPU().Maxwell3D()};
+    const auto& regs{gpu.regs};
+
+    if ((gpu.dirty_flags.color_buffer & (1u << static_cast<u32>(index))) == 0) {
+        return last_color_buffers[index];
+    }
+    gpu.dirty_flags.color_buffer &= ~(1u << static_cast<u32>(index));
 
     ASSERT(index < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets);
 
     if (index >= regs.rt_control.count) {
-        return {};
+        return last_color_buffers[index] = {};
     }
 
     if (regs.rt[index].Address() == 0 || regs.rt[index].format == Tegra::RenderTargetFormat::NONE) {
-        return {};
+        return last_color_buffers[index] = {};
     }
 
     const SurfaceParams color_params{SurfaceParams::CreateForFramebuffer(index)};
 
-    return GetSurface(color_params, preserve_contents);
+    return last_color_buffers[index] = GetSurface(color_params, preserve_contents);
 }
 
 void RasterizerCacheOpenGL::LoadSurface(const Surface& surface) {

src/video_core/renderer_opengl/gl_rasterizer_cache.h

@@ -396,6 +396,9 @@ private:
     /// Use a Pixel Buffer Object to download the previous texture and then upload it to the new one
     /// using the new format.
     OGLBuffer copy_pbo;
+
+    std::array<Surface, Tegra::Engines::Maxwell3D::Regs::NumRenderTargets> last_color_buffers;
+    Surface last_depth_buffer;
 };
 
 } // namespace OpenGL