DirectXShaderCompiler/tools/clang/test/CodeGenHLSL/Samples/MiniEngine/UpsampleAndBlurCS.hlsl

// RUN: %dxc -E main -T cs_6_0 %s | FileCheck %s

// CHECK: groupId
// CHECK: threadIdInGroup
// CHECK: threadId
// CHECK: sampleLevel
// CHECK: textureLoad
// CHECK: barrier
// CHECK: textureStore


//
// Copyright (c) Microsoft. All rights reserved.
// This code is licensed under the MIT License (MIT).
// THIS CODE IS PROVIDED *AS IS* WITHOUT WARRANTY OF
// ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING ANY
// IMPLIED WARRANTIES OF FITNESS FOR A PARTICULAR
// PURPOSE, MERCHANTABILITY, OR NON-INFRINGEMENT.
//
// Developed by Minigraph
//
// Author:  James Stanard 
//
// The CS for combining a lower resolution bloom buffer with a higher resolution buffer
// (via bilinear upsampling) and then guassian blurring the resultant buffer.
//
// For the intended bloom blurring algorithm, it is expected that this shader will be
// used repeatedly to upsample and blur successively higher resolutions until the final
// bloom buffer is the destination.
//

#include "PostEffectsRS.hlsli"

Texture2D<float3> HigherResBuf : register( t0 );
Texture2D<float3> LowerResBuf : register( t1 );
SamplerState LinearBorder : register( s1 );
RWTexture2D<float3> Result : register( u0 );

cbuffer cb0 : register(b0)
{
	float2 g_inverseDimensions;
	float g_upsampleBlendFactor;
}

// The guassian blur weights (derived from Pascal's triangle)
static const float Weights5[3] = { 6.0f / 16.0f, 4.0f / 16.0f, 1.0f / 16.0f };
static const float Weights7[4] = { 20.0f / 64.0f, 15.0f / 64.0f, 6.0f / 64.0f, 1.0f / 64.0f };
static const float Weights9[5] = { 70.0f / 256.0f, 56.0f / 256.0f, 28.0f / 256.0f, 8.0f / 256.0f, 1.0f / 256.0f };

float3 Blur5( float3 a, float3 b, float3 c, float3 d, float3 e, float3 f, float3 g, float3 h, float3 i )
{
	return Weights5[0]*e + Weights5[1]*(d+f) + Weights5[2]*(c+g);
}

float3 Blur7( float3 a, float3 b, float3 c, float3 d, float3 e, float3 f, float3 g, float3 h, float3 i )
{
	return Weights7[0]*e + Weights7[1]*(d+f) + Weights7[2]*(c+g) + Weights7[3]*(b+h);
}

float3 Blur9( float3 a, float3 b, float3 c, float3 d, float3 e, float3 f, float3 g, float3 h, float3 i )
{
	return Weights9[0]*e + Weights9[1]*(d+f) + Weights9[2]*(c+g) + Weights9[3]*(b+h) + Weights9[4]*(a+i);
}

#define BlurPixels Blur9

// 16x16 pixels with an 8x8 center that we will be blurring writing out.  Each uint is two color channels packed together
groupshared uint CacheR[128];
groupshared uint CacheG[128];
groupshared uint CacheB[128];

void Store2Pixels( uint index, float3 pixel1, float3 pixel2 )
{
	CacheR[index] = f32tof16(pixel1.r) | f32tof16(pixel2.r) << 16;
	CacheG[index] = f32tof16(pixel1.g) | f32tof16(pixel2.g) << 16;
	CacheB[index] = f32tof16(pixel1.b) | f32tof16(pixel2.b) << 16;
}

void Load2Pixels( uint index, out float3 pixel1, out float3 pixel2 )
{
	uint3 RGB = uint3(CacheR[index], CacheG[index], CacheB[index]);
	pixel1 = f16tof32(RGB);
	pixel2 = f16tof32(RGB >> 16);
}

void Store1Pixel( uint index, float3 pixel )
{
	CacheR[index] = asuint(pixel.r);
	CacheG[index] = asuint(pixel.g);
	CacheB[index] = asuint(pixel.b);
}

void Load1Pixel( uint index, out float3 pixel )
{
	pixel = asfloat( uint3(CacheR[index], CacheG[index], CacheB[index]) );
}

// Blur two pixels horizontally.  This reduces LDS reads and pixel unpacking.
void BlurHorizontally( uint outIndex, uint leftMostIndex )
{
	float3 s0, s1, s2, s3, s4, s5, s6, s7, s8, s9;
	Load2Pixels( leftMostIndex + 0, s0, s1 );
	Load2Pixels( leftMostIndex + 1, s2, s3 );
	Load2Pixels( leftMostIndex + 2, s4, s5 );
	Load2Pixels( leftMostIndex + 3, s6, s7 );
	Load2Pixels( leftMostIndex + 4, s8, s9 );
	
	Store1Pixel(outIndex  , BlurPixels(s0, s1, s2, s3, s4, s5, s6, s7, s8));
	Store1Pixel(outIndex+1, BlurPixels(s1, s2, s3, s4, s5, s6, s7, s8, s9));
}

void BlurVertically( uint2 pixelCoord, uint topMostIndex )
{
	float3 s0, s1, s2, s3, s4, s5, s6, s7, s8;
	Load1Pixel( topMostIndex   , s0 );
	Load1Pixel( topMostIndex+ 8, s1 );
	Load1Pixel( topMostIndex+16, s2 );
	Load1Pixel( topMostIndex+24, s3 );
	Load1Pixel( topMostIndex+32, s4 );
	Load1Pixel( topMostIndex+40, s5 );
	Load1Pixel( topMostIndex+48, s6 );
	Load1Pixel( topMostIndex+56, s7 );
	Load1Pixel( topMostIndex+64, s8 );

	Result[pixelCoord] = BlurPixels(s0, s1, s2, s3, s4, s5, s6, s7, s8);
}

[RootSignature(PostEffects_RootSig)]
[numthreads( 8, 8, 1 )]
void main( uint3 Gid : SV_GroupID, uint3 GTid : SV_GroupThreadID, uint3 DTid : SV_DispatchThreadID )
{
	//
	// Load 4 pixels per thread into LDS
	//
	int2 GroupUL = (Gid.xy << 3) - 4;				// Upper-left pixel coordinate of group read location
	int2 ThreadUL = (GTid.xy << 1) + GroupUL;		// Upper-left pixel coordinate of quad that this thread will read

	//
	// Store 4 blended-but-unblurred pixels in LDS
	//
	float2 uvUL = (float2(ThreadUL) + 0.5) * g_inverseDimensions;
	float2 uvLR = uvUL + g_inverseDimensions;
	float2 uvUR = float2(uvLR.x, uvUL.y);
	float2 uvLL = float2(uvUL.x, uvLR.y);
	int destIdx = GTid.x + (GTid.y << 4);

	float3 pixel1a = lerp(HigherResBuf[ThreadUL + uint2(0, 0)], LowerResBuf.SampleLevel(LinearBorder, uvUL, 0.0f), g_upsampleBlendFactor);
	float3 pixel1b = lerp(HigherResBuf[ThreadUL + uint2(1, 0)], LowerResBuf.SampleLevel(LinearBorder, uvUR, 0.0f), g_upsampleBlendFactor);
	Store2Pixels(destIdx+0, pixel1a, pixel1b);

	float3 pixel2a = lerp(HigherResBuf[ThreadUL + uint2(0, 1)], LowerResBuf.SampleLevel(LinearBorder, uvLL, 0.0f), g_upsampleBlendFactor);
	float3 pixel2b = lerp(HigherResBuf[ThreadUL + uint2(1, 1)], LowerResBuf.SampleLevel(LinearBorder, uvLR, 0.0f), g_upsampleBlendFactor);
	Store2Pixels(destIdx+8, pixel2a, pixel2b);

	GroupMemoryBarrierWithGroupSync();

	//
	// Horizontally blur the pixels in Cache
	//
	uint row = GTid.y << 4;
	BlurHorizontally(row + (GTid.x << 1), row + GTid.x + (GTid.x & 4));

	GroupMemoryBarrierWithGroupSync();

	//
	// Vertically blur the pixels and write the result to memory
	//
	BlurVertically(DTid.xy, (GTid.y << 3) + GTid.x);
}