Optimize RmlUi render passes

lib/aurora.cpp

@@ -239,13 +239,16 @@ void end_frame() noexcept {
                                                            presentSource.size.width, presentSource.size.height);
 
   wgpu::BindGroup rmlBindGroup;
+  bool rmlOverlay = false;
 #if AURORA_ENABLE_RMLUI
   if (rmlui::is_initialized()) {
-    rmlBindGroup = rmlui::record_frame(viewport);
+    auto rmlFrame = rmlui::record_frame(viewport);
+    rmlBindGroup = std::move(rmlFrame.bindGroup);
+    rmlOverlay = rmlFrame.overlay;
   }
 #endif
 
-  gfx::end_frame([rmlBindGroup = std::move(rmlBindGroup), viewport,
+  gfx::end_frame([rmlBindGroup = std::move(rmlBindGroup), rmlOverlay, viewport,
                   imguiDrawData = std::move(imguiDrawData)](wgpu::CommandEncoder& encoder) {
     wgpu::Texture currentTexture;
     wgpu::TextureView currentView;
@@ -267,7 +270,7 @@ void end_frame() noexcept {
     const bool canPresent = currentTexture && currentView;
     if (canPresent) {
       wgpu::BindGroup presentBindGroup;
-      if (rmlBindGroup) {
+      if (rmlBindGroup && !rmlOverlay) {
         presentBindGroup = rmlBindGroup;
       } else {
         const auto& resampledSource = webgpu::resample_present_source(encoder, viewport);
@@ -294,6 +297,11 @@ void end_frame() noexcept {
         pass.SetViewport(viewport.left, viewport.top, viewport.width, viewport.height, viewport.znear, viewport.zfar);
 
         pass.Draw(3);
+        if (rmlBindGroup && rmlOverlay) {
+          pass.SetPipeline(webgpu::g_CopyPremultipliedAlphaPipeline);
+          pass.SetBindGroup(0, rmlBindGroup, 0, nullptr);
+          pass.Draw(3);
+        }
         pass.End();
       }
       {

lib/rmlui.cpp

@@ -4,6 +4,7 @@
 #include <thread>
 
 #include <RmlUi/Core.h>
+#include <RmlUi/Core/Element.h>
 #include <RmlUi_Backend.h>
 #include <RmlUi_Platform_SDL.h>
 #include <tracy/Tracy.hpp>
@@ -82,6 +83,32 @@ void ensure_render_target(Rml::Vector2i dimensions) noexcept {
   s_renderTargetCopyBindGroup = webgpu::create_copy_bind_group(s_renderTarget);
 }
 
+bool element_has_visible_backdrop_filter(const Rml::Element* element) noexcept {
+  if (element == nullptr || !element->IsVisible(true)) {
+    return false;
+  }
+
+  const auto& computed = element->GetComputedValues();
+  if (computed.opacity() > 0.f && computed.has_backdrop_filter()) {
+    return true;
+  }
+
+  const int childCount = element->GetNumChildren();
+  for (int childIndex = 0; childIndex < childCount; ++childIndex) {
+    if (element_has_visible_backdrop_filter(element->GetChild(childIndex))) {
+      return true;
+    }
+  }
+  return false;
+}
+
+bool context_has_visible_backdrop_filter(Rml::Context* context) noexcept {
+  if (context == nullptr) {
+    return false;
+  }
+  return element_has_visible_backdrop_filter(context->GetRootElement());
+}
+
 struct MappedPoint {
   Rml::Vector2f position;
   bool valid = false;
@@ -408,7 +435,7 @@ void handle_event(SDL_Event& event) noexcept {
   RmlSDL::InputEventHandler(g_context, window::get_sdl_window(), event);
 }
 
-wgpu::BindGroup record_frame(const webgpu::Viewport& presentViewport) noexcept {
+RecordedFrame record_frame(const webgpu::Viewport& presentViewport) noexcept {
   if (g_context == nullptr) {
     return {};
   }
@@ -422,10 +449,12 @@ wgpu::BindGroup record_frame(const webgpu::Viewport& presentViewport) noexcept {
 
   sync_context_metrics(dim);
   g_context->Update();
+  const bool needsBackdrop = context_has_visible_backdrop_filter(g_context);
 
   auto* renderInterface = get_render_interface();
   renderInterface->SetWindowSize(g_context->GetDimensions());
-  renderInterface->BeginFrame(s_renderTarget, webgpu::present_source());
+  renderInterface->BeginFrame(s_renderTarget, webgpu::present_source(),
+                              needsBackdrop ? BaseLayerContent::Scene : BaseLayerContent::Transparent);
 
   Backend::BeginFrame();
   g_context->Render();
@@ -435,7 +464,10 @@ wgpu::BindGroup record_frame(const webgpu::Viewport& presentViewport) noexcept {
     // We didn't render anything
     return {};
   }
-  return s_renderTargetCopyBindGroup;
+  return {
+      .bindGroup = s_renderTargetCopyBindGroup,
+      .overlay = !needsBackdrop,
+  };
 }
 
 void shutdown() noexcept {

lib/rmlui.hpp

@@ -9,9 +9,14 @@
 
 namespace aurora::rmlui {
 
+struct RecordedFrame {
+  wgpu::BindGroup bindGroup;
+  bool overlay = false;
+};
+
 void initialize(const AuroraWindowSize& size) noexcept;
 void handle_event(SDL_Event& event) noexcept;
-wgpu::BindGroup record_frame(const webgpu::Viewport& presentViewport) noexcept;
+RecordedFrame record_frame(const webgpu::Viewport& presentViewport) noexcept;
 void shutdown() noexcept;
 
 } // namespace aurora::rmlui

lib/rmlui/WebGPURenderInterface.cpp

@@ -27,6 +27,7 @@ namespace {
 Module Log("aurora::rmlui::RenderInterface");
 
 constexpr size_t rmlBufferOffsetAlignment = 4;
+constexpr float FilterEpsilon = 0.0001f;
 
 struct Image {
   std::unique_ptr<uint8_t[]> data;
@@ -53,6 +54,23 @@ struct CompiledShaderData {
   GradientUniformBlock gradient;
 };
 
+enum class FilterType {
+  Opacity,
+  Blur,
+  DropShadow,
+  ColorMatrix,
+  MaskImage,
+};
+
+struct CompiledFilter {
+  FilterType type = FilterType::Blur;
+  float opacity = 1.f;
+  float sigma = 0.f;
+  Rml::Vector2f offset;
+  Rml::ColourbPremultiplied color;
+  Rml::Matrix4f colorMatrix;
+};
+
 Image get_image(const Rml::String& source) {
   FileInterface_SDL fileInterface;
   const Rml::FileHandle file = fileInterface.Open(source);
@@ -158,6 +176,74 @@ Rml::ColumnMajorMatrix4f to_shader_matrix(const Rml::Matrix4f& matrix) {
   }
 }
 
+bool is_identity_matrix(const Rml::Matrix4f& matrix) {
+  const Rml::Matrix4f identity = Rml::Matrix4f::Identity();
+  const auto* matrixData = matrix.data();
+  const auto* identityData = identity.data();
+  for (size_t i = 0; i < 16; ++i) {
+    if (std::abs(matrixData[i] - identityData[i]) > FilterEpsilon) {
+      return false;
+    }
+  }
+  return true;
+}
+
+bool is_identity_filter(const CompiledFilter& filter) {
+  switch (filter.type) {
+  case FilterType::Opacity:
+    return std::abs(filter.opacity - 1.f) <= FilterEpsilon;
+  case FilterType::Blur:
+    return filter.sigma < 0.5f;
+  case FilterType::ColorMatrix:
+    return is_identity_matrix(filter.colorMatrix);
+  case FilterType::DropShadow:
+  case FilterType::MaskImage:
+  default:
+    return false;
+  }
+}
+
+std::vector<Rml::CompiledFilterHandle> active_filters(Rml::Span<const Rml::CompiledFilterHandle> filters) {
+  std::vector<Rml::CompiledFilterHandle> activeFilters;
+  activeFilters.reserve(filters.size());
+  for (Rml::CompiledFilterHandle filterHandle : filters) {
+    const auto* filter = reinterpret_cast<const CompiledFilter*>(filterHandle);
+    if (filter != nullptr && !is_identity_filter(*filter)) {
+      activeFilters.push_back(filterHandle);
+    }
+  }
+  return activeFilters;
+}
+
+bool try_fold_simple_filters(Rml::Span<const Rml::CompiledFilterHandle> filters, Rml::Matrix4f& colorMatrix,
+                             float& opacity) {
+  colorMatrix = Rml::Matrix4f::Identity();
+  opacity = 1.f;
+
+  for (Rml::CompiledFilterHandle filterHandle : filters) {
+    const auto* filter = reinterpret_cast<const CompiledFilter*>(filterHandle);
+    if (filter == nullptr) {
+      continue;
+    }
+
+    switch (filter->type) {
+    case FilterType::Opacity:
+      opacity *= filter->opacity;
+      break;
+    case FilterType::ColorMatrix:
+      colorMatrix = filter->colorMatrix * colorMatrix;
+      break;
+    case FilterType::Blur:
+    case FilterType::DropShadow:
+    case FilterType::MaskImage:
+    default:
+      return false;
+    }
+  }
+
+  return true;
+}
+
 Rml::Rectanglei downsample_scissor(Rml::Rectanglei scissor) {
   scissor.p0 = (scissor.p0 + Rml::Vector2i(1)) / 2;
   scissor.p1 = Rml::Math::Max(scissor.p1 / 2, scissor.p0);
@@ -223,6 +309,18 @@ gfx::PipelineRef blit_pipeline(wgpu::TextureFormat colorFormat, uint32_t sampleC
                            blend ? BlendMode::Premultiplied : BlendMode::None));
 }
 
+gfx::PipelineRef simple_filter_pipeline(wgpu::TextureFormat colorFormat, uint32_t sampleCount,
+                                        WebGPURenderInterface::BlitPipelineType type, bool useStencil) {
+  const bool blend = type == WebGPURenderInterface::BlitPipelineType::Blend ||
+                     type == WebGPURenderInterface::BlitPipelineType::BlendMasked;
+  const bool masked = type == WebGPURenderInterface::BlitPipelineType::BlendMasked ||
+                      type == WebGPURenderInterface::BlitPipelineType::ReplaceMasked;
+  return gfx::pipeline_ref(
+      make_pipeline_config(PipelineKind::SimpleFilter, colorFormat, sampleCount, VertexLayoutKind::Fullscreen,
+                           masked ? StencilMode::EqualKeep : (useStencil ? StencilMode::AlwaysKeep : StencilMode::None),
+                           blend ? BlendMode::Premultiplied : BlendMode::None));
+}
+
 gfx::PipelineRef seed_resample_pipeline(wgpu::TextureFormat colorFormat, uint32_t sampleCount, bool useStencil) {
   return gfx::pipeline_ref(
       make_pipeline_config(PipelineKind::SeedResample, colorFormat, sampleCount, VertexLayoutKind::Fullscreen,
@@ -640,6 +738,11 @@ void WebGPURenderInterface::EnsureFrameRenderingStarted() {
   }
 
   m_frameRenderingStarted = true;
+  if (m_baseLayerContent == BaseLayerContent::Transparent) {
+    BeginLayerPass(0, wgpu::LoadOp::Clear, "RmlUi transparent base layer pass", true);
+    return;
+  }
+
   const auto seedBindGroup = texture_bind_group_ref(m_frameSeedView);
   const auto seedUniformRange = gfx::push_uniform(SeedResampleUniformBlock{
       .samplerMode = sampler_mode(),
@@ -832,9 +935,8 @@ void WebGPURenderInterface::RenderBlur(float sigma, const RenderTarget& sourceDe
   }
 }
 
-void WebGPURenderInterface::RenderFilters(Rml::Span<const Rml::CompiledFilterHandle> filters) {
-  constexpr size_t sourceIndex = 0;
-  constexpr size_t shadowIndex = 1;
+size_t WebGPURenderInterface::RenderFilters(Rml::Span<const Rml::CompiledFilterHandle> filters) {
+  size_t sourceIndex = 0;
   constexpr size_t tempIndex = 2;
 
   for (Rml::CompiledFilterHandle filterHandle : filters) {
@@ -842,22 +944,24 @@ void WebGPURenderInterface::RenderFilters(Rml::Span<const Rml::CompiledFilterHan
     if (filter == nullptr) {
       continue;
     }
+    const size_t scratchIndex = sourceIndex == 0 ? 1 : 0;
 
     switch (filter->type) {
     case FilterType::Opacity: {
-      BeginRenderTargetPass(m_postprocessTargets[shadowIndex].view, wgpu::LoadOp::Clear, "RmlUi opacity pass");
+      const SimpleFilterUniformBlock simpleFilterUniform{
+          .matrix = to_shader_matrix(Rml::Matrix4f::Identity()),
+          .opacity = {filter->opacity, 0.f, 0.f, 0.f},
+      };
+      const auto simpleFilterRange = gfx::push_uniform(simpleFilterUniform);
+      BeginRenderTargetPass(m_postprocessTargets[scratchIndex].view, wgpu::LoadOp::Clear, "RmlUi opacity pass");
       DrawFullscreenTexture(texture_bind_group_ref(m_postprocessTargets[sourceIndex].view),
-                            filter_pipeline(PipelineKind::Opacity, m_renderTargetFormat, VertexLayoutKind::Fullscreen,
-                                            BlendMode::Opacity),
-                            0, {}, true, {filter->opacity, filter->opacity, filter->opacity, filter->opacity}, true);
-      CompositeToTarget(texture_bind_group_ref(m_postprocessTargets[shadowIndex].view),
-                        m_postprocessTargets[sourceIndex].view, wgpu::LoadOp::Clear,
-                        blit_pipeline(m_renderTargetFormat, 1, BlitPipelineType::Replace, false),
-                        "RmlUi opacity result copy pass");
+                            filter_pipeline(PipelineKind::SimpleFilter, m_renderTargetFormat), uniform_bind_group_ref(),
+                            simpleFilterRange);
+      sourceIndex = scratchIndex;
       break;
     }
     case FilterType::Blur:
-      RenderBlur(filter->sigma, m_postprocessTargets[sourceIndex], m_postprocessTargets[shadowIndex]);
+      RenderBlur(filter->sigma, m_postprocessTargets[sourceIndex], m_postprocessTargets[tempIndex]);
       break;
     case FilterType::DropShadow: {
       TexCoordLimits texCoordLimits = GetPostprocessTexCoordLimits();
@@ -883,58 +987,52 @@ void WebGPURenderInterface::RenderFilters(Rml::Span<const Rml::CompiledFilterHan
       };
       const auto dropShadowRange = gfx::push_uniform(dropShadowUniform);
       CompositeToTarget(texture_bind_group_ref(m_postprocessTargets[sourceIndex].view),
-                        m_postprocessTargets[shadowIndex].view, wgpu::LoadOp::Clear,
+                        m_postprocessTargets[scratchIndex].view, wgpu::LoadOp::Clear,
                         filter_pipeline(PipelineKind::DropShadow, m_renderTargetFormat), "RmlUi drop shadow pass",
                         uniform_bind_group_ref(), dropShadowRange);
 
-      RenderBlur(filter->sigma, m_postprocessTargets[shadowIndex], m_postprocessTargets[tempIndex]);
+      RenderBlur(filter->sigma, m_postprocessTargets[scratchIndex], m_postprocessTargets[tempIndex]);
 
       CompositeToTarget(texture_bind_group_ref(m_postprocessTargets[sourceIndex].view),
-                        m_postprocessTargets[shadowIndex].view, wgpu::LoadOp::Load,
+                        m_postprocessTargets[scratchIndex].view, wgpu::LoadOp::Load,
                         blit_pipeline(m_renderTargetFormat, 1, BlitPipelineType::Blend, false),
                         "RmlUi drop shadow source composite pass");
-      CompositeToTarget(texture_bind_group_ref(m_postprocessTargets[shadowIndex].view),
-                        m_postprocessTargets[sourceIndex].view, wgpu::LoadOp::Clear,
-                        blit_pipeline(m_renderTargetFormat, 1, BlitPipelineType::Replace, false),
-                        "RmlUi drop shadow result copy pass");
+      sourceIndex = scratchIndex;
       break;
     }
     case FilterType::ColorMatrix: {
-      const ColorMatrixUniformBlock colorMatrixUniform{
+      const SimpleFilterUniformBlock simpleFilterUniform{
           .matrix = to_shader_matrix(filter->colorMatrix),
+          .opacity = {1.f, 0.f, 0.f, 0.f},
       };
-      const auto colorMatrixRange = gfx::push_uniform(colorMatrixUniform);
+      const auto simpleFilterRange = gfx::push_uniform(simpleFilterUniform);
 
       CompositeToTarget(texture_bind_group_ref(m_postprocessTargets[sourceIndex].view),
-                        m_postprocessTargets[shadowIndex].view, wgpu::LoadOp::Clear,
-                        filter_pipeline(PipelineKind::ColorMatrix, m_renderTargetFormat), "RmlUi color matrix pass",
-                        uniform_bind_group_ref(), colorMatrixRange);
-      CompositeToTarget(texture_bind_group_ref(m_postprocessTargets[shadowIndex].view),
-                        m_postprocessTargets[sourceIndex].view, wgpu::LoadOp::Clear,
-                        blit_pipeline(m_renderTargetFormat, 1, BlitPipelineType::Replace, false),
-                        "RmlUi color matrix result copy pass");
+                        m_postprocessTargets[scratchIndex].view, wgpu::LoadOp::Clear,
+                        filter_pipeline(PipelineKind::SimpleFilter, m_renderTargetFormat), "RmlUi color matrix pass",
+                        uniform_bind_group_ref(), simpleFilterRange);
+      sourceIndex = scratchIndex;
       break;
     }
     case FilterType::MaskImage: {
       CompositeToTarget(texture_bind_group_ref(m_postprocessTargets[sourceIndex].view),
-                        m_postprocessTargets[shadowIndex].view, wgpu::LoadOp::Clear,
+                        m_postprocessTargets[scratchIndex].view, wgpu::LoadOp::Clear,
                         filter_pipeline(PipelineKind::MaskImage, m_renderTargetFormat), "RmlUi mask image pass",
                         texture_bind_group_ref(m_blendMaskTarget.view), {}, false);
-      CompositeToTarget(texture_bind_group_ref(m_postprocessTargets[shadowIndex].view),
-                        m_postprocessTargets[sourceIndex].view, wgpu::LoadOp::Clear,
-                        blit_pipeline(m_renderTargetFormat, 1, BlitPipelineType::Replace, false),
-                        "RmlUi mask image result copy pass");
+      sourceIndex = scratchIndex;
       break;
     }
     }
   }
 
   EndActivePass();
+  return sourceIndex;
 }
 
 void WebGPURenderInterface::BeginFrame(const webgpu::TextureWithSampler& target,
-                                       const webgpu::TextureWithSampler& seedTarget) {
-  m_frameSeedView = seedTarget.view;
+                                       const webgpu::TextureWithSampler& sceneTarget,
+                                       BaseLayerContent baseLayerContent) {
+  m_frameSeedView = sceneTarget.view;
   m_frameSize = target.size;
   m_viewport = {
       .left = 0.f,
@@ -950,6 +1048,7 @@ void WebGPURenderInterface::BeginFrame(const webgpu::TextureWithSampler& target,
   m_frameRenderingStarted = false;
   m_frameActive = true;
   m_passActive = false;
+  m_baseLayerContent = baseLayerContent;
 
   NewFrame();
   EnsureFrameTargets(target.size);
@@ -998,6 +1097,7 @@ bool WebGPURenderInterface::EndFrame() {
   m_frameActive = false;
   m_frameSeedView = {};
   m_frameRenderingStarted = false;
+  m_baseLayerContent = BaseLayerContent::Transparent;
   return rendered;
 }
 
@@ -1024,17 +1124,52 @@ void WebGPURenderInterface::CompositeLayers(Rml::LayerHandle source, Rml::LayerH
   }
 
   EnsureFrameRenderingStarted();
-  CompositeToTarget(texture_bind_group_ref(m_layers[source].view), m_postprocessTargets[0].view, wgpu::LoadOp::Clear,
-                    blit_pipeline(m_renderTargetFormat, 1, BlitPipelineType::Replace, false), "RmlUi layer copy pass");
   const Rml::LayerHandle topLayer = m_layerStack.empty() ? 0 : m_layerStack.back();
-  RenderFilters(filters);
+  const std::vector<Rml::CompiledFilterHandle> activeFilterHandles = active_filters(filters);
+  const Rml::Span activeFilters{activeFilterHandles.data(), activeFilterHandles.size()};
 
   const bool replace = blendMode == Rml::BlendMode::Replace;
   const BlitPipelineType pipelineType =
       replace ? (m_clipMaskEnabled ? BlitPipelineType::ReplaceMasked : BlitPipelineType::Replace)
               : (m_clipMaskEnabled ? BlitPipelineType::BlendMasked : BlitPipelineType::Blend);
+  if (activeFilters.empty() && source != destination) {
+    BeginLayerPass(destination, wgpu::LoadOp::Load, "RmlUi layer direct composite pass");
+    DrawFullscreenTexture(texture_bind_group_ref(m_layers[source].view),
+                          blit_pipeline(m_renderTargetFormat, LayerSampleCount, pipelineType, true));
+
+    if (destination != topLayer) {
+      EndActivePass();
+      m_activeLayer = topLayer;
+    }
+    return;
+  }
+
+  Rml::Matrix4f simpleFilterMatrix;
+  float simpleFilterOpacity = 1.f;
+  if (source != destination && try_fold_simple_filters(activeFilters, simpleFilterMatrix, simpleFilterOpacity)) {
+    const SimpleFilterUniformBlock simpleFilterUniform{
+        .matrix = to_shader_matrix(simpleFilterMatrix),
+        .opacity = {simpleFilterOpacity, 0.f, 0.f, 0.f},
+    };
+    const auto simpleFilterRange = gfx::push_uniform(simpleFilterUniform);
+    BeginLayerPass(destination, wgpu::LoadOp::Load, "RmlUi simple filter composite pass");
+    DrawFullscreenTexture(texture_bind_group_ref(m_layers[source].view),
+                          simple_filter_pipeline(m_renderTargetFormat, LayerSampleCount, pipelineType, true),
+                          uniform_bind_group_ref(), simpleFilterRange);
+
+    if (destination != topLayer) {
+      EndActivePass();
+      m_activeLayer = topLayer;
+    }
+    return;
+  }
+
+  CompositeToTarget(texture_bind_group_ref(m_layers[source].view), m_postprocessTargets[0].view, wgpu::LoadOp::Clear,
+                    blit_pipeline(m_renderTargetFormat, 1, BlitPipelineType::Replace, false), "RmlUi layer copy pass");
+  const size_t filteredIndex = RenderFilters(activeFilters);
+
   BeginLayerPass(destination, wgpu::LoadOp::Load, "RmlUi layer composite pass");
-  DrawFullscreenTexture(texture_bind_group_ref(m_postprocessTargets[0].view),
+  DrawFullscreenTexture(texture_bind_group_ref(m_postprocessTargets[filteredIndex].view),
                         blit_pipeline(m_renderTargetFormat, LayerSampleCount, pipelineType, true));
 
   if (destination != topLayer) {

lib/rmlui/WebGPURenderInterface.hpp

@@ -41,8 +41,9 @@ struct DropShadowUniformBlock {
   Rml::Vector2f texCoordMax;
 };
 
-struct ColorMatrixUniformBlock {
+struct SimpleFilterUniformBlock {
   Rml::ColumnMajorMatrix4f matrix;
+  Rml::Vector4f opacity;
 };
 
 struct GradientUniformBlock {
@@ -60,6 +61,11 @@ struct TexCoordLimits {
   Rml::Vector2f max;
 };
 
+enum class BaseLayerContent {
+  Transparent,
+  Scene,
+};
+
 class WebGPURenderInterface : public Rml::RenderInterface {
 public:
   static constexpr wgpu::TextureFormat ClipMaskStencilFormat = wgpu::TextureFormat::Stencil8;
@@ -81,14 +87,6 @@ public:
   };
 
 private:
-  enum class FilterType {
-    Opacity,
-    Blur,
-    DropShadow,
-    ColorMatrix,
-    MaskImage,
-  };
-
   struct RenderTarget {
     wgpu::Texture texture;
     wgpu::TextureView view;
@@ -97,15 +95,6 @@ private:
     wgpu::Extent3D size;
   };
 
-  struct CompiledFilter {
-    FilterType type = FilterType::Blur;
-    float opacity = 1.f;
-    float sigma = 0.f;
-    Rml::Vector2f offset;
-    Rml::ColourbPremultiplied color;
-    Rml::Matrix4f colorMatrix;
-  };
-
   wgpu::TextureView m_frameSeedView;
 
   wgpu::TextureFormat m_renderTargetFormat = wgpu::TextureFormat::Undefined;
@@ -120,10 +109,11 @@ private:
   Rml::Vector2i m_clipResetGeometrySize{};
   std::vector<RenderTarget> m_layers;
   std::array<RenderTarget, 3> m_postprocessTargets{};
-  RenderTarget m_blendMaskTarget;
+  RenderTarget m_blendMaskTarget{};
   std::vector<Rml::LayerHandle> m_layerStack;
   Rml::LayerHandle m_activeLayer = 0;
   Rml::LayerHandle m_nextLayer = 1;
+  BaseLayerContent m_baseLayerContent = BaseLayerContent::Transparent;
 
   Rml::Vector2i m_windowSize{};
   Rml::Matrix4f m_translationMatrix = Rml::Matrix4f::Identity();
@@ -169,7 +159,7 @@ private:
                          gfx::Range extraUniformRange = {}, bool extraBindGroupHasDynamicOffset = true,
                          std::array<float, 4> blendConstant = {}, bool hasBlendConstant = false);
   void RenderBlur(float sigma, const RenderTarget& sourceDestination, const RenderTarget& temp);
-  void RenderFilters(Rml::Span<const Rml::CompiledFilterHandle> filters);
+  size_t RenderFilters(Rml::Span<const Rml::CompiledFilterHandle> filters);
 
 public:
   Rml::CompiledGeometryHandle CompileGeometry(Rml::Span<const Rml::Vertex> vertices,
@@ -199,7 +189,8 @@ public:
                     Rml::TextureHandle texture) override;
   void ReleaseShader(Rml::CompiledShaderHandle shader) override;
 
-  void BeginFrame(const webgpu::TextureWithSampler& target, const webgpu::TextureWithSampler& seed_target);
+  void BeginFrame(const webgpu::TextureWithSampler& target, const webgpu::TextureWithSampler& scene_target,
+                  BaseLayerContent baseLayerContent);
   bool EndFrame();
   void SetWindowSize(const Rml::Vector2i& window_size) { m_windowSize = window_size; }
   void SetRenderTargetFormat(wgpu::TextureFormat render_target_format) { m_renderTargetFormat = render_target_format; }

lib/rmlui/pipeline.cpp

@@ -24,8 +24,8 @@ constexpr uint32_t DynamicGroup2 = 1u << 2u;
 
 constexpr uint64_t CommonUniformBindingSize = AURORA_ALIGN(sizeof(UniformBlock), 16);
 constexpr uint64_t ExtraUniformBindingSize =
-    AURORA_ALIGN(std::max({sizeof(BlurUniformBlock), sizeof(DropShadowUniformBlock), sizeof(ColorMatrixUniformBlock),
-                           sizeof(GradientUniformBlock), sizeof(SeedResampleUniformBlock)}),
+    AURORA_ALIGN(std::max({sizeof(BlurUniformBlock), sizeof(DropShadowUniformBlock), sizeof(SimpleFilterUniformBlock),
+                           sizeof(GradientUniforzxmBlock), sizeof(SeedResampleUniformBlock)}),
                  16);
 
 constexpr std::string_view vertexSource = R"(
@@ -367,20 +367,21 @@ fn main(@builtin(position) position: vec4<f32>, @location(0) uv: vec2<f32>) -> @
 }
 )"sv;
 
-constexpr std::string_view colorMatrixFragmentSource = R"(
-struct ColorMatrixUniforms {
+constexpr std::string_view simpleFilterFragmentSource = R"(
+struct SimpleFilterUniforms {
     matrix: mat4x4<f32>,
+    opacity: vec4<f32>,
 };
 
 @group(0) @binding(1) var s: sampler;
 @group(1) @binding(0) var t: texture_2d<f32>;
-@group(2) @binding(0) var<uniform> color_matrix: ColorMatrixUniforms;
+@group(2) @binding(0) var<uniform> simple_filter: SimpleFilterUniforms;
 
 @fragment
 fn main(@location(0) uv: vec2<f32>) -> @location(0) vec4<f32> {
     let tex_color = textureSample(t, s, uv);
-    let transformed_color = (color_matrix.matrix * tex_color).rgb;
-    return vec4<f32>(transformed_color, tex_color.a);
+    let transformed_color = simple_filter.matrix * tex_color;
+    return vec4<f32>(transformed_color.rgb, tex_color.a) * simple_filter.opacity.x;
 }
 )"sv;
 
@@ -511,21 +512,6 @@ wgpu::BlendState blend_state(BlendMode mode) {
                 .dstFactor = wgpu::BlendFactor::OneMinusSrcAlpha,
             },
     };
-  case BlendMode::Opacity:
-    return {
-        .color =
-            {
-                .operation = wgpu::BlendOperation::Add,
-                .srcFactor = wgpu::BlendFactor::Constant,
-                .dstFactor = wgpu::BlendFactor::Zero,
-            },
-        .alpha =
-            {
-                .operation = wgpu::BlendOperation::Add,
-                .srcFactor = wgpu::BlendFactor::Constant,
-                .dstFactor = wgpu::BlendFactor::Zero,
-            },
-    };
   case BlendMode::None:
   default:
     return {};
@@ -548,7 +534,7 @@ const wgpu::PipelineLayout create_pipeline_layout(PipelineKind kind) {
   case PipelineKind::Blur:
   case PipelineKind::RegionBlit:
   case PipelineKind::DropShadow:
-  case PipelineKind::ColorMatrix:
+  case PipelineKind::SimpleFilter:
   case PipelineKind::SeedResample:
     layouts[layoutCount++] = g_imageBindGroupLayout;
     layouts[layoutCount++] = g_uniformBindGroupLayout;
@@ -556,7 +542,6 @@ const wgpu::PipelineLayout create_pipeline_layout(PipelineKind kind) {
   case PipelineKind::Geometry:
   case PipelineKind::Blit:
   case PipelineKind::OpaqueBlit:
-  case PipelineKind::Opacity:
   default:
     layouts[layoutCount++] = g_imageBindGroupLayout;
     break;
@@ -585,12 +570,11 @@ const std::string_view fragment_source(PipelineKind kind) {
     return regionBlitFragmentSource;
   case PipelineKind::DropShadow:
     return dropShadowFragmentSource;
-  case PipelineKind::ColorMatrix:
-    return colorMatrixFragmentSource;
+  case PipelineKind::SimpleFilter:
+    return simpleFilterFragmentSource;
   case PipelineKind::MaskImage:
     return maskImageFragmentSource;
   case PipelineKind::Blit:
-  case PipelineKind::Opacity:
   default:
     return blitFragmentSource;
   }

lib/rmlui/pipeline.hpp

@@ -10,7 +10,7 @@
 
 namespace aurora::rmlui {
 
-constexpr uint32_t RmlPipelineConfigVersion = 1;
+constexpr uint32_t RmlPipelineConfigVersion = 2;
 
 enum class PipelineKind : uint32_t {
   Geometry,
@@ -18,11 +18,10 @@ enum class PipelineKind : uint32_t {
   Blit,
   OpaqueBlit,
   SeedResample,
-  Opacity,
+  SimpleFilter,
   Blur,
   RegionBlit,
   DropShadow,
-  ColorMatrix,
   MaskImage,
 };
 
@@ -43,7 +42,6 @@ enum class StencilMode : uint32_t {
 enum class BlendMode : uint32_t {
   None,
   Premultiplied,
-  Opacity,
 };
 
 enum class DrawKind : uint32_t {

lib/webgpu/gpu.cpp

@@ -47,6 +47,7 @@ TextureWithSampler g_depthBuffer;
 // EFB -> XFB copy pipeline
 static wgpu::BindGroupLayout g_CopyBindGroupLayout;
 wgpu::RenderPipeline g_CopyPipeline;
+wgpu::RenderPipeline g_CopyPremultipliedAlphaPipeline;
 wgpu::BindGroup g_CopyBindGroup;
 static AuroraSampler g_Resampler = SAMPLER_BILINEAR;
 static wgpu::BindGroupLayout g_ResampleBindGroupLayout;
@@ -411,26 +412,21 @@ fn vs_main(@builtin(vertex_index) vtxIdx: u32) -> VertexOutput {
 }
 
 @fragment
-fn fs_main(in: VertexOutput) -> @location(0) vec4<f32> {
+fn fs_opaque(in: VertexOutput) -> @location(0) vec4<f32> {
     let color = textureSample(efb_texture, efb_sampler, in.uv);
     return vec4(color.rgb, 1.0);
 }
+
+@fragment
+fn fs_premultiplied_alpha(in: VertexOutput) -> @location(0) vec4<f32> {
+    return textureSample(efb_texture, efb_sampler, in.uv);
+}
 )""";
   const wgpu::ShaderModuleDescriptor moduleDescriptor{
       .nextInChain = &sourceDescriptor,
       .label = "XFB Copy Module",
   };
   auto module = g_device.CreateShaderModule(&moduleDescriptor);
-  const std::array colorTargets{wgpu::ColorTargetState{
-      .format = g_graphicsConfig.surfaceConfiguration.format,
-      .writeMask = wgpu::ColorWriteMask::All,
-  }};
-  const wgpu::FragmentState fragmentState{
-      .module = module,
-      .entryPoint = "fs_main",
-      .targetCount = colorTargets.size(),
-      .targets = colorTargets.data(),
-  };
   const std::array bindGroupLayoutEntries{
       wgpu::BindGroupLayoutEntry{
           .binding = 0,
@@ -460,25 +456,58 @@ fn fs_main(in: VertexOutput) -> @location(0) vec4<f32> {
       .bindGroupLayouts = &g_CopyBindGroupLayout,
   };
   auto pipelineLayout = g_device.CreatePipelineLayout(&layoutDescriptor);
-  const wgpu::RenderPipelineDescriptor pipelineDescriptor{
-      .layout = pipelineLayout,
-      .vertex =
-          wgpu::VertexState{
-              .module = module,
-              .entryPoint = "vs_main",
-          },
-      .primitive =
-          wgpu::PrimitiveState{
-              .topology = wgpu::PrimitiveTopology::TriangleList,
-          },
-      .multisample =
-          wgpu::MultisampleState{
-              .count = 1,
-              .mask = UINT32_MAX,
-          },
-      .fragment = &fragmentState,
+
+  const auto make_copy_pipeline = [&](const char* label, const char* fragmentEntryPoint, const wgpu::BlendState* blend) {
+    const std::array colorTargets{wgpu::ColorTargetState{
+        .format = g_graphicsConfig.surfaceConfiguration.format,
+        .blend = blend,
+        .writeMask = wgpu::ColorWriteMask::All,
+    }};
+    const wgpu::FragmentState fragmentState{
+        .module = module,
+        .entryPoint = fragmentEntryPoint,
+        .targetCount = colorTargets.size(),
+        .targets = colorTargets.data(),
+    };
+    const wgpu::RenderPipelineDescriptor pipelineDescriptor{
+        .label = label,
+        .layout = pipelineLayout,
+        .vertex =
+            wgpu::VertexState{
+                .module = module,
+                .entryPoint = "vs_main",
+            },
+        .primitive =
+            wgpu::PrimitiveState{
+                .topology = wgpu::PrimitiveTopology::TriangleList,
+            },
+        .multisample =
+            wgpu::MultisampleState{
+                .count = 1,
+                .mask = UINT32_MAX,
+            },
+        .fragment = &fragmentState,
+    };
+    return g_device.CreateRenderPipeline(&pipelineDescriptor);
   };
-  g_CopyPipeline = g_device.CreateRenderPipeline(&pipelineDescriptor);
+  g_CopyPipeline = make_copy_pipeline("XFB Copy Pipeline", "fs_opaque", nullptr);
+
+  const wgpu::BlendState premultipliedAlphaBlend{
+      .color =
+          {
+              .operation = wgpu::BlendOperation::Add,
+              .srcFactor = wgpu::BlendFactor::One,
+              .dstFactor = wgpu::BlendFactor::OneMinusSrcAlpha,
+          },
+      .alpha =
+          {
+              .operation = wgpu::BlendOperation::Add,
+              .srcFactor = wgpu::BlendFactor::One,
+              .dstFactor = wgpu::BlendFactor::OneMinusSrcAlpha,
+          },
+  };
+  g_CopyPremultipliedAlphaPipeline =
+      make_copy_pipeline("XFB Premultiplied Alpha Copy Pipeline", "fs_premultiplied_alpha", &premultipliedAlphaBlend);
 }
 
 void create_resample_pipeline() {
@@ -983,6 +1012,7 @@ void shutdown() {
   gpu_prof::shutdown();
   g_CopyBindGroupLayout = {};
   g_CopyPipeline = {};
+  g_CopyPremultipliedAlphaPipeline = {};
   g_CopyBindGroup = {};
   g_ResampleBindGroupLayout = {};
   g_ResamplePipeline = {};

lib/webgpu/gpu.hpp

@@ -48,6 +48,7 @@ extern TextureWithSampler g_frameBuffer;
 extern TextureWithSampler g_frameBufferResolved;
 extern TextureWithSampler g_depthBuffer;
 extern wgpu::RenderPipeline g_CopyPipeline;
+extern wgpu::RenderPipeline g_CopyPremultipliedAlphaPipeline;
 extern wgpu::BindGroup g_CopyBindGroup;
 extern wgpu::Instance g_instance;
 extern wgpu::AdapterInfo g_adapterInfo;