ue4系列: shader

发表于 更新于 分类于 ，

简单记录一下ue中global shader的注册、编译和使用。ue的shader里面各种宏乱飞，不得不说，宏是各种引擎里面，最恶心的地方了，尤其是uber shader的大一统写法，不过也算是有舍有得。

本文以PostProcessSubsurface.cpp/PostProcessSubsurface.usf的MainIndirectDispatchCS为例。

  1. global shader中宏的声明

MainIndirectDispatchCS中使用的宏都在这里定义，利用ShaderPermutation.h把每一个宏包装为一个单独的类。 

1
2
3
4
5
6
7
8
9
class FSubsurfacePassFunction : SHADER_PERMUTATION_ENUM_CLASS("SUBSURFACE_PASS", ESubsurfacePass);
class FDimensionQuality : SHADER_PERMUTATION_ENUM_CLASS("SUBSURFACE_QUALITY", EQuality);
class FBilateralFilterKernelFunctionType : SHADER_PERMUTATION_ENUM_CLASS("BILATERAL_FILTER_KERNEL_FUNCTION_TYPE", EBilateralFilterKernelFunctionType);
class FSubsurfaceType : SHADER_PERMUTATION_ENUM_CLASS("SUBSURFACE_TYPE", ESubsurfaceType);
class FDimensionHalfRes : SHADER_PERMUTATION_BOOL("SUBSURFACE_HALF_RES");
class FRunningInSeparable : SHADER_PERMUTATION_BOOL("SUBSURFACE_FORCE_SEPARABLE");
class FDimensionEnableProfileIDCache : SHADER_PERMUTATION_BOOL("ENABLE_PROFILE_ID_CACHE");
using FPermutationDomain = TShaderPermutationDomain<FSubsurfacePassFunction, FDimensionQuality, 
	FBilateralFilterKernelFunctionType, FSubsurfaceType, FDimensionHalfRes, FRunningInSeparable, FDimensionEnableProfileIDCache>;

  1. shader permutation
1
IMPLEMENT_GLOBAL_SHADER(FSubsurfaceIndirectDispatchCS, "/Engine/Private/PostProcessSubsurface.usf", "MainIndirectDispatchCS", SF_Compute);

ue里面各种宏，看的眼晕。IMPLEMENT_GLOBAL_SHADER实例化一个静态对象（通过IMPLEMENT_SHADER_TYPE），渲染模块加载（FEngineLoop::LoadPreInitModules）时候就会执行，在FShaderType::FShaderType的构造（shader.cpp)中调用GlobalListLink.LinkHead把当前类型注册到全局shader链表中，并通过CompileGlobalShaderMap、VerifyGlobalShaders（遍历shader permutation）、FGlobalShaderTypeCompiler::BeginCompileShader编译。

  1. 使用

Engine/Source/Runtime/RenderCore/Public/Shader.h L2105~L2122 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
template<typename ShaderType>
class TShaderMapRef : public TShaderRef<ShaderType>
{
public:
	TShaderMapRef(const typename ShaderType::ShaderMapType* ShaderIndex)
		: TShaderRef<ShaderType>(ShaderIndex->template GetShader<ShaderType>(/* PermutationId = */ 0)) // gcc3 needs the template quantifier so it knows the < is not a less-than
	{
		static_assert(
			TIsSame<typename ShaderType::FPermutationDomain, FShaderPermutationNone>::Value,
			"Missing permutation vector argument for shader that have a permutation domain.");
	}

	TShaderMapRef(
		const typename ShaderType::ShaderMapType* ShaderIndex,
		const typename ShaderType::FPermutationDomain& PermutationVector)
		: TShaderRef<ShaderType>(ShaderIndex->template GetShader<ShaderType>(PermutationVector.ToDimensionValueId())) // gcc3 needs the template quantifier so it knows the < is not a less-than
	{ }
};

1
2
3
4
5
6
7
8
9
10
11
12
   SHADER::FPermutationDomain ComputeShaderPermutationVector;
ComputeShaderPermutationVector.Set<SHADER::FSubsurfacePassFunction>(SubsurfacePassFunction);
ComputeShaderPermutationVector.Set<SHADER::FDimensionQuality>(SHADER::GetQuality());
   ComputeShaderPermutationVector.Set<SHADER::FBilateralFilterKernelFunctionType>(SHADER::GetBilateralFilterKernelFunctionType());
ComputeShaderPermutationVector.Set<SHADER::FSubsurfaceType>(SubsurfaceType);
ComputeShaderPermutationVector.Set<SHADER::FDimensionHalfRes>(bHalfRes);
ComputeShaderPermutationVector.Set<SHADER::FRunningInSeparable>(bForceRunningInSeparable);
ComputeShaderPermutationVector.Set<SHADER::FDimensionEnableProfileIDCache>(bUseProfileIdCache);
TShaderMapRef<SHADER> ComputeShader(View.ShaderMap, ComputeShaderPermutationVector);

   FComputeShaderUtils::AddPass(GraphBuilder, FRDGEventName(PassInfo.Name), ComputeShader, PassParameters, SubsurfaceBufferArgs[SubsurfaceTypeIndex], 0);

TShaderMapRef构造中通过shader permutation id从shadermap中获取对应的shader实例，然后通过RenderGraph添加该pass。