Add **Contrast Adaptive Sharpening (CAS)** filter pass

#2423

Source/Engine/Graphics/PostProcessSettings.cpp

@@ -198,6 +198,10 @@ void AntiAliasingSettings::BlendWith(AntiAliasingSettings& other, float weight)
     BLEND_FLOAT(TAA_Sharpness);
     BLEND_FLOAT(TAA_StationaryBlending);
     BLEND_FLOAT(TAA_MotionBlending);
+    BLEND_FLOAT(CAS_SharpeningAmount);
+    BLEND_FLOAT(CAS_EdgeSharpening);
+    BLEND_FLOAT(CAS_MinEdgeThreshold);
+    BLEND_FLOAT(CAS_OverBlurLimit);
 }
 
 void PostFxMaterialsSettings::BlendWith(PostFxMaterialsSettings& other, float weight)

Source/Engine/Graphics/PostProcessSettings.h

@@ -1858,10 +1858,30 @@ API_ENUM(Attributes="Flags") enum class AntiAliasingSettingsOverride : int32
     /// </summary>
     TAA_MotionBlending = 1 << 4,
 
+    /// <summary>
+    /// Overrides <see cref="AntiAliasingSettings.CAS_SharpeningAmount"/> property.
+    /// </summary>
+    CAS_SharpeningAmount = 1 << 5,
+
+    /// <summary>
+    /// Overrides <see cref="AntiAliasingSettings.CAS_EdgeSharpening"/> property.
+    /// </summary>
+    CAS_EdgeSharpening = 1 << 6,
+
+    /// <summary>
+    /// Overrides <see cref="AntiAliasingSettings.CAS_MinEdgeThreshold"/> property.
+    /// </summary>
+    CAS_MinEdgeThreshold = 1 << 7,
+
+    /// <summary>
+    /// Overrides <see cref="AntiAliasingSettings.CAS_OverBlurLimit"/> property.
+    /// </summary>
+    CAS_OverBlurLimit = 1 << 8,
+
     /// <summary>
     /// All properties.
     /// </summary>
-    All = Mode | TAA_JitterSpread | TAA_Sharpness | TAA_StationaryBlending | TAA_MotionBlending,
+    All = Mode | TAA_JitterSpread | TAA_Sharpness | TAA_StationaryBlending | TAA_MotionBlending | CAS_SharpeningAmount | CAS_EdgeSharpening | CAS_MinEdgeThreshold | CAS_OverBlurLimit,
 };
 
 /// <summary>
@@ -1909,6 +1929,30 @@ API_STRUCT() struct FLAXENGINE_API AntiAliasingSettings : ISerializable
     API_FIELD(Attributes="Limit(0, 0.99f, 0.001f), EditorOrder(4), PostProcessSetting((int)AntiAliasingSettingsOverride.TAA_MotionBlending), EditorDisplay(null, \"TAA Motion Blending\"), VisibleIf(nameof(ShowTAASettings))")
     float TAA_MotionBlending = 0.85f;
 
+    /// <summary>
+    /// The sharpening strength for the Contrast Adaptive Sharpening (CAS) pass. Ignored when using TAA that contains own contrast filter.
+    /// </summary>
+    API_FIELD(Attributes = "Limit(0, 10f, 0.001f), EditorOrder(10), PostProcessSetting((int)AntiAliasingSettingsOverride.CAS_SharpeningAmount), EditorDisplay(null, \"CAS Sharpening Amount\"), VisibleIf(nameof(ShowTAASettings), true)")
+    float CAS_SharpeningAmount = 0.0f;
+
+    /// <summary>
+    /// The edge sharpening strength for the Contrast Adaptive Sharpening (CAS) pass. Ignored when using TAA that contains own contrast filter.
+    /// </summary>
+    API_FIELD(Attributes = "Limit(0, 10f, 0.001f), EditorOrder(11), PostProcessSetting((int)AntiAliasingSettingsOverride.CAS_EdgeSharpening), EditorDisplay(null, \"CAS Edge Sharpening\"), VisibleIf(nameof(ShowTAASettings), true)")
+    float CAS_EdgeSharpening = 0.5f;
+
+    /// <summary>
+    /// The minimum edge threshold for the Contrast Adaptive Sharpening (CAS) pass. Ignored when using TAA that contains own contrast filter.
+    /// </summary>
+    API_FIELD(Attributes = "Limit(0, 10f, 0.001f), EditorOrder(12), PostProcessSetting((int)AntiAliasingSettingsOverride.CAS_MinEdgeThreshold), EditorDisplay(null, \"CAS Min Edge Threshold\"), VisibleIf(nameof(ShowTAASettings), true)")
+    float CAS_MinEdgeThreshold = 0.03f;
+
+    /// <summary>
+    /// The over-blur limit for the Contrast Adaptive Sharpening (CAS) pass. Ignored when using TAA that contains own contrast filter.
+    /// </summary>
+    API_FIELD(Attributes = "Limit(0, 100f, 0.001f), EditorOrder(13), PostProcessSetting((int)AntiAliasingSettingsOverride.CAS_OverBlurLimit), EditorDisplay(null, \"CAS Over-blur Limit\"), VisibleIf(nameof(ShowTAASettings), true)")
+    float CAS_OverBlurLimit = 1.0f;
+
 public:
     /// <summary>
     /// Blends the settings using given weight.

Source/Engine/Renderer/AntiAliasing/FXAA.cpp

@@ -65,10 +65,10 @@ void FXAA::Dispose()
 void FXAA::Render(RenderContext& renderContext, GPUTexture* input, GPUTextureView* output)
 {
     auto context = GPUDevice::Instance->GetMainContext();
+    context->SetRenderTarget(output);
     if (checkIfSkipPass())
     {
         // Resources are missing. Do not perform rendering, just copy input frame.
-        context->SetRenderTarget(output);
         context->Draw(input);
         return;
     }
@@ -83,7 +83,6 @@ void FXAA::Render(RenderContext& renderContext, GPUTexture* input, GPUTextureVie
     context->BindSR(0, input);
 
     // Render
-    context->SetRenderTarget(output);
     const auto qualityLevel = Math::Clamp(static_cast<int32>(Graphics::AAQuality), 0, static_cast<int32>(Quality::MAX) - 1);
     context->SetState(_psFXAA.Get(qualityLevel));
     context->DrawFullscreenTriangle();

Source/Engine/Renderer/ContrastAdaptiveSharpeningPass.cpp

@@ -0,0 +1,97 @@
+// Copyright (c) 2012-2024 Wojciech Figat. All rights reserved.
+
+#include "ContrastAdaptiveSharpeningPass.h"
+#include "RenderList.h"
+#include "Engine/Content/Assets/Shader.h"
+#include "Engine/Content/Content.h"
+#include "Engine/Graphics/GPUContext.h"
+#include "Engine/Graphics/GPUDevice.h"
+#include "Engine/Graphics/RenderTargetPool.h"
+#include "Engine/Graphics/RenderTask.h"
+#include "Engine/Graphics/Shaders/GPUShader.h"
+
+GPU_CB_STRUCT(Data {
+    Float2 InputSizeInv;
+    Float2 Padding;
+    float SharpeningAmount;
+    float EdgeSharpening;
+    float MinEdgeThreshold;
+    float OverBlurLimit;
+    });
+
+String ContrastAdaptiveSharpeningPass::ToString() const
+{
+    return TEXT("ContrastAdaptiveSharpening");
+}
+
+void ContrastAdaptiveSharpeningPass::Dispose()
+{
+    RendererPass::Dispose();
+
+    SAFE_DELETE_GPU_RESOURCE(_psCAS);
+    _shader = nullptr;
+}
+
+bool ContrastAdaptiveSharpeningPass::setupResources()
+{
+    // Lazy-load shader
+    if (!_shader)
+    {
+        _shader = Content::LoadAsyncInternal<Shader>(TEXT("Shaders/CAS"));
+        if (!_shader)
+            return false;
+#if COMPILE_WITH_DEV_ENV
+        _shader.Get()->OnReloading.Bind<ContrastAdaptiveSharpeningPass, &ContrastAdaptiveSharpeningPass::OnShaderReloading>(this);
+#endif
+    }
+    if (!_shader->IsLoaded())
+        return true;
+    const auto shader = _shader->GetShader();
+
+    // Validate shader constant buffer size
+    if (shader->GetCB(0)->GetSize() != sizeof(Data))
+    {
+        REPORT_INVALID_SHADER_PASS_CB_SIZE(shader, 0, Data);
+        return true;
+    }
+
+    // Create pipeline stage
+    auto psDesc = GPUPipelineState::Description::DefaultFullscreenTriangle;
+    psDesc.PS = shader->GetPS("PS_CAS");
+    _psCAS = GPUDevice::Instance->CreatePipelineState();
+    if (_psCAS->Init(psDesc))
+        return true;
+
+    return false;
+}
+
+bool ContrastAdaptiveSharpeningPass::CanRender(const RenderContext& renderContext)
+{
+    const AntiAliasingSettings& antiAliasing = renderContext.List->Settings.AntiAliasing;
+    return EnumHasAnyFlags(renderContext.View.Flags, ViewFlags::AntiAliasing) &&
+            antiAliasing.CAS_SharpeningAmount > ZeroTolerance &&
+            !checkIfSkipPass();
+}
+
+void ContrastAdaptiveSharpeningPass::Render(const RenderContext& renderContext, GPUTexture* input, GPUTextureView* output)
+{
+    ASSERT_LOW_LAYER(CanRender(renderContext));
+    PROFILE_GPU_CPU("Contrast Adaptive Sharpening");
+    auto device = GPUDevice::Instance;
+    auto context = device->GetMainContext();
+    const AntiAliasingSettings& antiAliasing = renderContext.List->Settings.AntiAliasing;
+
+    Data data;
+    data.InputSizeInv = Float2::One / input->Size();
+    data.SharpeningAmount = antiAliasing.CAS_SharpeningAmount;
+    data.EdgeSharpening = antiAliasing.CAS_EdgeSharpening;
+    data.MinEdgeThreshold = antiAliasing.CAS_MinEdgeThreshold;
+    data.OverBlurLimit = antiAliasing.CAS_OverBlurLimit;
+    const auto cb = _shader->GetShader()->GetCB(0);
+    context->UpdateCB(cb, &data);
+    context->BindCB(0, cb);
+    context->BindSR(0, input);
+    context->SetState(_psCAS);
+    context->SetRenderTarget(output);
+    context->DrawFullscreenTriangle();
+}

Source/Engine/Renderer/ContrastAdaptiveSharpeningPass.h

@@ -0,0 +1,39 @@
+// Copyright (c) 2012-2024 Wojciech Figat. All rights reserved.
+
+#pragma once
+
+#include "RendererPass.h"
+#include "Engine/Graphics/PostProcessSettings.h"
+
+/// <summary>
+/// Contrast Adaptive Sharpening (CAS) provides a mixed ability to sharpen and optionally scale an image. Based on AMD FidelityFX implementation.
+/// </summary>
+class ContrastAdaptiveSharpeningPass : public RendererPass<ContrastAdaptiveSharpeningPass>
+{
+private:
+    bool _loadShader = true;
+    AssetReference<Shader> _shader;
+    GPUPipelineState* _psCAS = nullptr;
+
+public:
+    bool CanRender(const RenderContext& renderContext);
+    void Render(const RenderContext& renderContext, GPUTexture* input, GPUTextureView* output);
+
+private:
+#if COMPILE_WITH_DEV_ENV
+    void OnShaderReloading(Asset* obj)
+    {
+        _psCAS->ReleaseGPU();
+        invalidateResources();
+    }
+#endif
+
+public:
+    // [RendererPass]
+    String ToString() const override;
+    void Dispose() override;
+
+protected:
+    // [RendererPass]
+    bool setupResources() override;
+};

Source/Engine/Renderer/Renderer.cpp

@@ -22,6 +22,7 @@
 #include "VolumetricFogPass.h"
 #include "HistogramPass.h"
 #include "AtmospherePreCompute.h"
+#include "ContrastAdaptiveSharpeningPass.h"
 #include "GlobalSignDistanceFieldPass.h"
 #include "GI/GlobalSurfaceAtlasPass.h"
 #include "GI/DynamicDiffuseGlobalIllumination.h"
@@ -126,16 +127,41 @@ void RendererService::Dispose()
 void RenderAntiAliasingPass(RenderContext& renderContext, GPUTexture* input, GPUTextureView* output, const Viewport& outputViewport)
 {
     auto context = GPUDevice::Instance->GetMainContext();
-    context->SetViewportAndScissors(outputViewport);
     const auto aaMode = renderContext.List->Settings.AntiAliasing.Mode;
+    if (ContrastAdaptiveSharpeningPass::Instance()->CanRender(renderContext))
+    {
+        if (aaMode == AntialiasingMode::FastApproximateAntialiasing ||
+            aaMode == AntialiasingMode::SubpixelMorphologicalAntialiasing)
+        {
+            // AA -> CAS -> Output
+            auto tmpImage = RenderTargetPool::Get(input->GetDescription());
+            RENDER_TARGET_POOL_SET_NAME(tmpImage, "TmpImage");
+            context->SetViewportAndScissors((float)input->Width(), (float)input->Height());
             if (aaMode == AntialiasingMode::FastApproximateAntialiasing)
+                FXAA::Instance()->Render(renderContext, input, tmpImage->View());
+            else
+                SMAA::Instance()->Render(renderContext, input, tmpImage->View());
+            context->ResetSR();
+            context->ResetRenderTarget();
+            context->SetViewportAndScissors(outputViewport);
+            ContrastAdaptiveSharpeningPass::Instance()->Render(renderContext, tmpImage, output);
+            RenderTargetPool::Release(tmpImage);
+        }
+        else
         {
+            // CAS -> Output
+            context->SetViewportAndScissors(outputViewport);
+            ContrastAdaptiveSharpeningPass::Instance()->Render(renderContext, input, output);
+        }
+    }
+    else
+    {
+        // AA -> Output
+        context->SetViewportAndScissors(outputViewport);
+        if (aaMode == AntialiasingMode::FastApproximateAntialiasing)
             FXAA::Instance()->Render(renderContext, input, output);
-    }
         else if (aaMode == AntialiasingMode::SubpixelMorphologicalAntialiasing)
-    {
             SMAA::Instance()->Render(renderContext, input, output);
-    }
         else
         {
             PROFILE_GPU("Copy frame");
@@ -143,6 +169,7 @@ void RenderAntiAliasingPass(RenderContext& renderContext, GPUTexture* input, GPU
             context->Draw(input);
         }
     }
+}
 
 bool Renderer::IsReady()
 {

Source/Shaders/CAS.shader

@@ -0,0 +1,47 @@
+// Copyright (c) 2012-2024 Wojciech Figat. All rights reserved.
+
+#include "./Flax/Common.hlsl"
+
+META_CB_BEGIN(0, Data)
+float2 InputSizeInv;
+float2 Padding;
+float SharpeningAmount;
+float EdgeSharpening;
+float MinEdgeThreshold;
+float OverBlurLimit;
+META_CB_END
+
+Texture2D Input : register(t0);
+
+// Pixel Shader for Contrast Adaptive Sharpening (CAS) filter. Based on AMD FidelityFX implementation.
+META_PS(true, FEATURE_LEVEL_ES2)
+float4 PS_CAS(Quad_VS2PS input) : SV_Target0
+{
+    // Sample the color texture
+    float4 color = Input.SampleLevel(SamplerLinearClamp, input.TexCoord, 0);
+
+    // Sample neighboring pixels
+    float3 blurred = color.rgb;
+    float3 edges = 0.0;
+    for (int x = -2; x <= 2; x++)
+    {
+        for (int y = -2; y <= 2; y++)
+        {
+            float2 uv = float2(x, y) * InputSizeInv + input.TexCoord;
+            float3 neighbor = Input.SampleLevel(SamplerLinearClamp, uv, 0).rgb;
+            blurred += neighbor;
+            edges += abs(neighbor - color.rgb);
+        }
+    }
+    blurred /= 25.0;
+    edges /= 25.0;
+
+    // Sharpen based on edge detection
+    float edgeAmount = saturate((dot(edges, edges) - MinEdgeThreshold) / (0.001 + dot(edges, edges)));
+    float sharpen = (1.0 - edgeAmount) * SharpeningAmount + edgeAmount * EdgeSharpening;
+    float3 sharpened = color.rgb + (color.rgb - blurred) * sharpen;
+
+    // Limit sharpening to avoid over-blurring
+    sharpened = lerp(color.rgb, sharpened, saturate(OverBlurLimit / (OverBlurLimit + dot(abs(sharpened - color.rgb), float3(1.0, 1.0, 1.0)))));
+    return float4(sharpened, color.a);
+}