Implement a helper shader for partial clears

These are not natively supported by Vulkan, so use a helper shader and colorWriteMask for the same behaviour.

app/src/main/cpp/skyline/gpu/interconnect/maxwell_3d/maxwell_3d.cpp

@@ -137,6 +137,10 @@ namespace skyline::gpu::interconnect::maxwell3d {
                 view->range.layerCount != 1 || view->range.baseArrayLayer != 0 || clearSurface.rtArrayIndex != 0;
         }};
 
+        // Always use surfaceClip for render area since it's more likely to match the renderArea of draws and avoid an RP break
+        const auto &surfaceClip{clearEngineRegisters.surfaceClip};
+        vk::Rect2D renderArea{{surfaceClip.horizontal.x, surfaceClip.vertical.y}, {surfaceClip.horizontal.width, surfaceClip.vertical.height}};
+
         auto clearRects{util::MakeFilledArray<vk::ClearRect, 2>(vk::ClearRect{.rect = scissor, .baseArrayLayer = clearSurface.rtArrayIndex, .layerCount = 1})};
         boost::container::small_vector<vk::ClearAttachment, 2> clearAttachments;
 
@@ -147,14 +151,23 @@ namespace skyline::gpu::interconnect::maxwell3d {
             if (auto view{activeState.GetColorRenderTargetForClear(ctx, clearSurface.mrtSelect)}) {
                 ctx.executor.AttachTexture(&*view);
 
-                if (!(clearSurface.rEnable && clearSurface.gEnable && clearSurface.bEnable && clearSurface.aEnable))
+                bool partialClear{!(clearSurface.rEnable && clearSurface.gEnable && clearSurface.bEnable && clearSurface.aEnable)};
-                    Logger::Warn("Partial clears are unimplemented! Performing full clear instead");
+                if (!(view->range.aspectMask & vk::ImageAspectFlagBits::eColor))
-
-                if (!(view->range.aspectMask & vk::ImageAspectFlagBits::eColor)) {
                     Logger::Warn("Colour RT used in clear lacks colour aspect"); // TODO: Drop this check after texman rework
-                }
 
-                if (needsAttachmentClearCmd(view)) {
+
+                if (partialClear) {
+                    ctx.gpu.helperShaders.clearHelperShader.Clear(ctx.gpu, view->range.aspectMask,
+                                                                  (clearSurface.rEnable ? vk::ColorComponentFlagBits::eR : vk::ColorComponentFlags{}) |
+                                                                  (clearSurface.gEnable ? vk::ColorComponentFlagBits::eG : vk::ColorComponentFlags{}) |
+                                                                  (clearSurface.bEnable ? vk::ColorComponentFlagBits::eB : vk::ColorComponentFlags{}) |
+                                                                  (clearSurface.aEnable ? vk::ColorComponentFlagBits::eA : vk::ColorComponentFlags{}),
+                                                                  {clearEngineRegisters.colorClearValue}, &*view, [=](auto &&executionCallback) {
+                        auto dst{view.get()};
+                        ctx.executor.AddSubpass(std::move(executionCallback), renderArea, {}, {}, span<TextureView *>{dst}, nullptr);
+                    });
+                    ctx.executor.NotifyPipelineChange();
+                } else if (needsAttachmentClearCmd(view)) {
                     clearAttachments.push_back({.aspectMask = view->range.aspectMask, .clearValue = {clearEngineRegisters.colorClearValue}});
                     colorView = view;
                 } else {
@@ -190,10 +203,6 @@ namespace skyline::gpu::interconnect::maxwell3d {
         if (clearAttachments.empty())
             return;
 
-        // Always use surfaceClip for render area since it's more likely to match the renderArea of draws and avoid an RP break
-        const auto &surfaceClip{clearEngineRegisters.surfaceClip};
-        vk::Rect2D renderArea{{surfaceClip.horizontal.x, surfaceClip.vertical.y}, {surfaceClip.horizontal.width, surfaceClip.vertical.height}};
-
         std::array<TextureView *, 1> colorAttachments{colorView ? &*colorView : nullptr};
         ctx.executor.AddSubpass([clearAttachments, clearRects](vk::raii::CommandBuffer &commandBuffer, const std::shared_ptr<FenceCycle> &, GPU &, vk::RenderPass, u32) {
             commandBuffer.clearAttachments(clearAttachments, span(clearRects).first(clearAttachments.size()));

app/src/main/cpp/skyline/gpu/shaders/helper_shaders.cpp

@@ -23,7 +23,7 @@ namespace skyline::gpu {
         );
     }
 
-    SimpleColourRTShader::SimpleColourRTShader(GPU &gpu, std::shared_ptr<vfs::Backing> vertexShader, std::shared_ptr<vfs::Backing> fragmentShader)
+    SimpleSingleRtShader::SimpleSingleRtShader(GPU &gpu, std::shared_ptr<vfs::Backing> vertexShader, std::shared_ptr<vfs::Backing> fragmentShader)
         : vertexShaderModule{CreateShaderModule(gpu, *vertexShader)},
           fragmentShaderModule{CreateShaderModule(gpu, *fragmentShader)},
           shaderStages{{
@@ -39,8 +39,10 @@ namespace skyline::gpu {
                            }}
           } {}
 
-    cache::GraphicsPipelineCache::CompiledPipeline SimpleColourRTShader::GetPipeline(GPU &gpu,
+    cache::GraphicsPipelineCache::CompiledPipeline SimpleSingleRtShader::GetPipeline(GPU &gpu,
-                                                                                     TextureView *colorAttachment,
+                                                                                     TextureView *colorAttachment, TextureView *depthStencilAttachment,
+                                                                                     bool depthWrite, bool stencilWrite, u32 stencilValue,
+                                                                                     vk::ColorComponentFlags colorWriteMask,
                                                                                      span<const vk::DescriptorSetLayoutBinding> layoutBindings, span<const vk::PushConstantRange> pushConstantRanges) {
         constexpr static vk::PipelineInputAssemblyStateCreateInfo inputAssemblyState{
             .topology = vk::PrimitiveTopology::eTriangleList,
@@ -73,19 +75,31 @@ namespace skyline::gpu {
             .alphaToOneEnable = false
         };
 
-        constexpr static vk::PipelineDepthStencilStateCreateInfo depthStencilState{
+        vk::PipelineDepthStencilStateCreateInfo depthStencilState{
-            .depthTestEnable = false,
+            .depthTestEnable = depthWrite,
-            .depthWriteEnable = false,
+            .depthWriteEnable = depthWrite,
+            .depthCompareOp = vk::CompareOp::eAlways,
             .depthBoundsTestEnable = false,
-            .stencilTestEnable = false,
+            .stencilTestEnable = stencilWrite,
         };
 
-        constexpr static vk::PipelineColorBlendAttachmentState attachmentState{
+        if (stencilWrite) {
+            depthStencilState.front = depthStencilState.back = {
+                .depthFailOp = vk::StencilOp::eReplace,
+                .passOp = vk::StencilOp::eReplace,
+                .compareOp = vk::CompareOp::eAlways,
+                .compareMask = 0xFF,
+                .writeMask = 0xFF,
+                .reference = stencilValue
+            };
+        }
+
+        vk::PipelineColorBlendAttachmentState attachmentState{
             .blendEnable = false,
-            .colorWriteMask = vk::ColorComponentFlagBits::eR | vk::ColorComponentFlagBits::eG | vk::ColorComponentFlagBits::eB | vk::ColorComponentFlagBits::eA
+            .colorWriteMask = colorWriteMask
         };
 
-        constexpr static vk::PipelineColorBlendStateCreateInfo blendState{
+        vk::PipelineColorBlendStateCreateInfo blendState{
             .logicOpEnable = false,
             .attachmentCount = 1,
             .pAttachments = &attachmentState
@@ -100,11 +114,11 @@ namespace skyline::gpu {
 
         vertexState.unlink<vk::PipelineVertexInputDivisorStateCreateInfoEXT>();
 
-        auto colourAttachmentDimensions{colorAttachment->texture->dimensions};
+        auto attachmentDimensions{colorAttachment ? colorAttachment->texture->dimensions : depthStencilAttachment->texture->dimensions};
 
         vk::Viewport viewport{
-            .height = static_cast<float>(colourAttachmentDimensions.height),
+            .height = static_cast<float>(attachmentDimensions.height),
-            .width = static_cast<float>(colourAttachmentDimensions.width),
+            .width = static_cast<float>(attachmentDimensions.width),
             .x = 0.0f,
             .y = 0.0f,
             .minDepth = 0.0f,
@@ -112,7 +126,7 @@ namespace skyline::gpu {
         };
 
         vk::Rect2D scissor{
-            .extent = colourAttachmentDimensions
+            .extent = attachmentDimensions
         };
 
         vk::PipelineViewportStateCreateInfo viewportState{
@@ -134,7 +148,7 @@ namespace skyline::gpu {
             .colorBlendState = blendState,
             .dynamicState = {},
             .colorAttachments = span<TextureView *>{colorAttachment},
-            .depthStencilAttachment = nullptr,
+            .depthStencilAttachment = depthStencilAttachment,
         }, layoutBindings, pushConstantRanges, true);
     }
 
@@ -142,6 +156,12 @@ namespace skyline::gpu {
         struct Vec2 {
             float x, y;
         };
+
+        struct Vec4 {
+            float x, y, z, w;
+        };
+
+        using Bool = u32;
     }
 
     namespace blit {
@@ -177,7 +197,7 @@ namespace skyline::gpu {
     };
 
     BlitHelperShader::BlitHelperShader(GPU &gpu, std::shared_ptr<vfs::FileSystem> shaderFileSystem)
-        : SimpleColourRTShader{gpu, shaderFileSystem->OpenFile("shaders/blit.vert.spv"), shaderFileSystem->OpenFile("shaders/blit.frag.spv")},
+        : SimpleSingleRtShader{gpu, shaderFileSystem->OpenFile("shaders/blit.vert.spv"), shaderFileSystem->OpenFile("shaders/blit.frag.spv")},
           bilinearSampler{gpu.vkDevice.createSampler(
               vk::SamplerCreateInfo{
                   .addressModeU = vk::SamplerAddressMode::eRepeat,
@@ -231,7 +251,10 @@ namespace skyline::gpu {
                 .dstSrcScaleFactor = {dstSrcScaleFactorX * (srcRect.width / srcImageDimensions.width), dstSrcScaleFactorY * (srcRect.height / srcImageDimensions.height)},
                 .srcHeightRecip = 1.0f / srcImageDimensions.height
             },
-            GetPipeline(gpu, dstImageView, {blit::SamplerLayoutBinding}, blit::PushConstantRanges))
+            GetPipeline(gpu, dstImageView,
+                        nullptr, false, false, 0,
+                        vk::ColorComponentFlagBits::eR | vk::ColorComponentFlagBits::eG | vk::ColorComponentFlagBits::eB | vk::ColorComponentFlagBits::eA,
+                        {blit::SamplerLayoutBinding}, blit::PushConstantRanges))
         };
 
         vk::DescriptorImageInfo imageInfo{
@@ -263,7 +286,63 @@ namespace skyline::gpu {
 
     }
 
+    namespace clear {
+        struct FragmentPushConstantLayout {
+            glsl::Vec4 color;
+            glsl::Bool clearDepth;
+            float depth;
+        };
+
+        constexpr static std::array<vk::PushConstantRange, 1> PushConstantRanges{
+            vk::PushConstantRange{
+                .stageFlags = vk::ShaderStageFlagBits::eFragment,
+                .size = sizeof(FragmentPushConstantLayout),
+                .offset = 0
+            }
+        };
+    }
+
+    ClearHelperShader::ClearHelperShader(GPU &gpu, std::shared_ptr<vfs::FileSystem> shaderFileSystem)
+        : SimpleSingleRtShader{gpu, shaderFileSystem->OpenFile("shaders/clear.vert.spv"), shaderFileSystem->OpenFile("shaders/clear.frag.spv")} {}
+
+    void ClearHelperShader::Clear(GPU &gpu, vk::ImageAspectFlags mask, vk::ColorComponentFlags components, vk::ClearValue value, TextureView *dstImageView,
+                                 std::function<void(std::function<void(vk::raii::CommandBuffer &, const std::shared_ptr<FenceCycle> &, GPU &, vk::RenderPass, u32)> &&)> &&recordCb) {
+        struct DrawState {
+            clear::FragmentPushConstantLayout fragmentPushConstants;
+            cache::GraphicsPipelineCache::CompiledPipeline pipeline;
+
+            DrawState(GPU &gpu,
+                      clear::FragmentPushConstantLayout fragmentPushConstants,
+                      cache::GraphicsPipelineCache::CompiledPipeline pipeline)
+                : fragmentPushConstants{fragmentPushConstants},
+                  pipeline{pipeline} {}
+        };
+
+        bool writeColor{mask & vk::ImageAspectFlagBits::eColor};
+        bool writeDepth{mask & vk::ImageAspectFlagBits::eDepth};
+        bool writeStencil{mask & vk::ImageAspectFlagBits::eStencil};
+
+        auto drawState{std::make_shared<DrawState>(
+            gpu,
+            clear::FragmentPushConstantLayout{
+                .color = {value.color.float32[0], value.color.float32[1], value.color.float32[2], value.color.float32[3]},
+                .clearDepth = (mask & vk::ImageAspectFlagBits::eDepth) != vk::ImageAspectFlags{},
+                .depth = value.depthStencil.depth
+            },
+            GetPipeline(gpu, writeColor ? dstImageView : nullptr, (writeDepth || writeStencil) ? dstImageView : nullptr, writeDepth, writeStencil, value.depthStencil.stencil, components, {}, clear::PushConstantRanges))
+        };
+
+        recordCb([drawState = std::move(drawState)](vk::raii::CommandBuffer &commandBuffer, const std::shared_ptr<FenceCycle> &cycle, GPU &gpu, vk::RenderPass, u32) {
+            cycle->AttachObject(drawState);
+            commandBuffer.bindPipeline(vk::PipelineBindPoint::eGraphics, drawState->pipeline.pipeline);
+            commandBuffer.pushConstants(drawState->pipeline.pipelineLayout, vk::ShaderStageFlagBits::eFragment, 0,
+                                        vk::ArrayProxy<const clear::FragmentPushConstantLayout>{drawState->fragmentPushConstants});
+            commandBuffer.draw(6, 1, 0, 0);
+        });
+    }
+
     HelperShaders::HelperShaders(GPU &gpu, std::shared_ptr<vfs::FileSystem> shaderFileSystem)
-        : blitHelperShader(gpu, std::move(shaderFileSystem)) {}
+        : blitHelperShader(gpu, shaderFileSystem),
+          clearHelperShader(gpu, shaderFileSystem) {}
 
 }

app/src/main/cpp/skyline/gpu/shaders/helper_shaders.h

@@ -16,29 +16,31 @@ namespace skyline::gpu {
     class GPU;
 
     /**
-     * @brief A base class that can be inherited by helper shaders that render to a single color rendertarget to simplify pipeline creation
+     * @brief A base class that can be inherited by helper shaders that render to a single color/depth rendertarget to simplify pipeline creation
      */
-    class SimpleColourRTShader {
+    class SimpleSingleRtShader {
       protected:
         vk::raii::ShaderModule vertexShaderModule;
         vk::raii::ShaderModule fragmentShaderModule;
 
         std::array<vk::PipelineShaderStageCreateInfo, 2> shaderStages; //!< Shader stages for the vertex and fragment shader modules
 
-        SimpleColourRTShader(GPU &gpu, std::shared_ptr<vfs::Backing> vertexShader, std::shared_ptr<vfs::Backing> fragmentShader);
+        SimpleSingleRtShader(GPU &gpu, std::shared_ptr<vfs::Backing> vertexShader, std::shared_ptr<vfs::Backing> fragmentShader);
 
         /**
          * @brief Returns a potentially cached pipeline built according to the supplied input state
          */
         cache::GraphicsPipelineCache::CompiledPipeline GetPipeline(GPU &gpu,
-                                                                   TextureView *colorAttachment,
+                                                                   TextureView *colorAttachment, TextureView *depthStenilAttachment,
+                                                                   bool depthWrite, bool stencilWrite, u32 stencilValue,
+                                                                   vk::ColorComponentFlags colorWriteMask,
                                                                    span<const vk::DescriptorSetLayoutBinding> layoutBindings, span<const vk::PushConstantRange> pushConstantRanges);
     };
 
     /**
      * @brief Simple helper shader for blitting a texture to a rendertarget with subpixel-precision
      */
-    class BlitHelperShader : SimpleColourRTShader {
+    class BlitHelperShader : SimpleSingleRtShader {
       private:
         vk::raii::Sampler bilinearSampler;
         vk::raii::Sampler nearestSampler;
@@ -73,11 +75,30 @@ namespace skyline::gpu {
                       std::function<void(std::function<void(vk::raii::CommandBuffer &, const std::shared_ptr<FenceCycle> &, GPU &, vk::RenderPass, u32)> &&)> &&recordCb);
     };
 
+    /**
+     * @brief Simple helper shader for clearing a texture to a given color
+     */
+    class ClearHelperShader : SimpleSingleRtShader {
+      public:
+        ClearHelperShader(GPU &gpu, std::shared_ptr<vfs::FileSystem> shaderFileSystem);
+
+        /**
+         * @brief Records a sequenced GPU clear operation using a shader
+         * @param mask Mask of which aspects to clear
+         * @param components Mask of which components to clear
+         * @param value The value to clear to
+         * @param recordCb Callback used to record the blit commands for sequenced execution on the GPU
+         */
+        void Clear(GPU &gpu, vk::ImageAspectFlags mask, vk::ColorComponentFlags components, vk::ClearValue value, TextureView *dstImageView,
+                  std::function<void(std::function<void(vk::raii::CommandBuffer &, const std::shared_ptr<FenceCycle> &, GPU &, vk::RenderPass, u32)> &&)> &&recordCb);
+    };
+
     /**
      * @brief Holds all helper shaders to avoid redundantly recreating them on each usage
      */
     struct HelperShaders {
         BlitHelperShader blitHelperShader;
+        ClearHelperShader clearHelperShader;
 
         HelperShaders(GPU &gpu, std::shared_ptr<vfs::FileSystem> shaderFileSystem);
     };

app/src/main/shaders/clear.frag

@@ -0,0 +1,17 @@
+#version 460
+
+layout (location = 0) out vec4 colour;
+
+layout (push_constant) uniform constants {
+    vec4 colour;
+    bool clearDepth;
+    float depth;
+} PC;
+
+void main()
+{
+    if (PC.clearDepth)
+        gl_FragDepth = PC.depth;
+    else
+        colour = PC.colour;
+}

app/src/main/shaders/clear.vert

@@ -0,0 +1,14 @@
+#version 460
+
+void main() {
+	const vec2 lut[6] = vec2[6](
+        vec2(-1, -1),
+        vec2(-1, 1),
+        vec2(1, 1),
+        vec2(1, 1),
+        vec2(1, -1),
+        vec2(-1, -1)
+    );
+
+    gl_Position = vec4(lut[gl_VertexIndex], 0, 1);
+}