[Optimization] DXC is missing common factor optimization in some cases

[lizhengxing]

Description
DXC is missing common factor optimization in some cases. Here's an example code:

cbuffer TemporalAAData : register ( b10 )
{
    float2 viewportRelativeSize ;
    float4 outputDimensions ;
}

Texture2D < float4 > HistoryColor : register ( t2 ) ;
SamplerState s_Linear : register ( s1 ) ;
RWTexture1D < float3 > outColorBuffer : register ( u0 ) ;

float4 Test ( in Texture2D < float4 > tex , in SamplerState linearSampler , in float2 uv )
{
    float2 samplePos = uv;
    float2 texPos1 = floor ( samplePos - 0.5f ) + 0.5f ;

    float2 texPos0 = texPos1 - 1 ;
    float2 texPos3 = texPos1 + 2 ;
    float2 texPos12 = texPos1 + samplePos;

    // DXC should recognize (outputDimensions . zw * viewportRelativeSize) is a common factor.
    texPos0 *= outputDimensions . zw * viewportRelativeSize ;
    texPos3 *= outputDimensions . zw * viewportRelativeSize ;
    texPos12 *= outputDimensions . zw * viewportRelativeSize ;

    float4 result = 0.0f ;
    result += tex . SampleLevel ( linearSampler , float2 ( texPos0 . x , texPos0 . y ) , 0.0f );
    result += tex . SampleLevel ( linearSampler , float2 ( texPos12 . x , texPos0 . y ) , 0.0f );
    result += tex . SampleLevel ( linearSampler , float2 ( texPos3 . x , texPos0 . y ) , 0.0f );
    result += tex . SampleLevel ( linearSampler , float2 ( texPos0 . x , texPos12 . y ) , 0.0f );
    result += tex . SampleLevel ( linearSampler , float2 ( texPos12 . x , texPos12 . y ) , 0.0f );
    result += tex . SampleLevel ( linearSampler , float2 ( texPos3 . x , texPos12 . y ) , 0.0f );
    result += tex . SampleLevel ( linearSampler , float2 ( texPos0 . x , texPos3 . y ) , 0.0f );
    result += tex . SampleLevel ( linearSampler , float2 ( texPos12 . x , texPos3 . y ) , 0.0f );
    result += tex . SampleLevel ( linearSampler , float2 ( texPos3 . x , texPos3 . y ) , 0.0f );
    return result ;
}

[ numthreads ( 8 , 8 , 1 ) ] void cs_main ( uint3 GroupID : SV_GroupID , uint GroupIndex : SV_GroupIndex , uint3 GTID : SV_GroupThreadID , uint3 DispatchThreadID : SV_DispatchThreadID )
{
    uint2 pixelCoord = GTID . xy ;

    outColorBuffer [ pixelCoord.x ] = Test ( HistoryColor , s_Linear , outputDimensions . zw ) . rgb;
}

Steps to Reproduce
Run the hlsl code above with the command line below:
dxc.exe -T cs_6_3 -E cs_main dxc-common-factor-optimization.hlsl

Actual Behavior
After compiling the example hlsl, you will see those dxils in the result.

  %24 = fmul fast float %7, %14
  %25 = fmul fast float %24, %22
  %26 = fmul fast float %8, %15
  %27 = fmul fast float %26, %23
  %28 = fmul fast float %7, %16
  %29 = fmul fast float %28, %22
  %30 = fmul fast float %8, %17
  %31 = fmul fast float %30, %23
  %32 = fmul fast float %7, %18
  %33 = fmul fast float %32, %22
  %34 = fmul fast float %8, %20
  %35 = fmul fast float %34, %23

The common factor %7 * %22 is calculated 3 times.

If we rewrite the relevant code like this:

    float2 temp = outputDimensions . zw * viewportRelativeSize;
    texPos0 *= temp ;
    texPos3 *= temp ;
    texPos12 *= temp ;

The generated dxils are:

  %24 = fmul fast float %22, %7
  %25 = fmul fast float %23, %8
  %26 = fmul fast float %24, %14
  %27 = fmul fast float %25, %15
  %28 = fmul fast float %24, %16
  %29 = fmul fast float %25, %17
  %30 = fmul fast float %24, %18
  %31 = fmul fast float %25, %20

The common factor %7 * %22 is only calculated 1 time.

DXC should recognize the common factor and generate the optimized dxils.

Environment

• DXC version: 1.8 - 1.8.0.14576 (main, 381750e)
• Host Operating System: Windows 11

[lizhengxing]

Close it since all the fixes are merged in DXC main branch.