directx11 – Recommendation on methods to finest use if attainable 4 GatherCmpRed to emulate a 4×4 PCF sampling for shadowmap blur

March 5, 2025

7

In part of my stage I’ve shadows near the digital camera the place a superb blur is necessary. Presently I’ve carried out a 4×4 PCF, thus requiring 16 texture samplings.

I am attempting to see if the equal might be achieved with solely 4 gathercmpred calls.

My code is beneath and the consequence within the image for PCF (left) and gathered (proper). If the collect protocol may very well be okay for distant shadows it’s clearly not for shut ones, a minimum of in the best way I did it.

My code:

#outline INV1080 1.0/1080.0 //my shadow map has the dimensions of the display screen
#outline INV1920 1.0/1920.0
#outline INV1920HALF INV1920*0.5 //I feel we want 0.5 offset for gathering
#outline INV1080HALF INV1080*0.5
//precalculated offset for PCF for the dimensions of my shadowmap. Made for as much as 5x5 sampling
static const float2 OffSetPCF[5][5] = {
  {-2*INV1920, -2*INV1080 },{-INV1920, -2*INV1080 },{0, -2*INV1080 },{INV1920, -2*INV1080 },{2*INV1920, -2*INV1080 },
  {-2*INV1920, -INV1080 },{-INV1920, -INV1080 },{0, -INV1080 },{INV1920, -INV1080 },{2*INV1920, -INV1080 },
  {-2*INV1920, 0 },{-INV1920, 0 },{0, 0 },{INV1920, 0},{2*INV1920, 0 },
  {-2*INV1920, INV1080 },{-INV1920, INV1080 },{0, INV1080 },{INV1920, INV1080 },     {2*INV1920, INV1080 },
  {-2*INV1920, 2*INV1080 },{-INV1920, 2*INV1080 },{0, 2*INV1080 },{INV1920, 2*INV1080 },{2*INV1920, 2*INV1080 }
};

//precalculated offset for collect requires the dimensions of my shadowmap
static const float2 OffSetGather[5][5] = {
  {-2*INV1920+INV1920HALF, -2*INV1080+INV1080HALF },{-INV1920+INV1920HALF, -2*INV1080+INV1080HALF },{0+INV1920HALF, -2*INV1080+INV1080HALF },{INV1920+INV1920HALF, -2*INV1080+INV1080HALF },{2*INV1920+INV1920HALF, -2*INV1080+INV1080HALF },
  {-2*INV1920+INV1920HALF,   -INV1080+INV1080HALF },{-INV1920+INV1920HALF,   -INV1080+INV1080HALF },{0+INV1920HALF,   -INV1080+INV1080HALF },{INV1920+INV1920HALF,   -INV1080+INV1080HALF },{2*INV1920+INV1920HALF,   -INV1080+INV1080HALF },
  {-2*INV1920+INV1920HALF,          0+INV1080HALF },{-INV1920+INV1920HALF,          0+INV1080HALF },{0+INV1920HALF,          0+INV1080HALF },{INV1920+INV1920HALF,          0+INV1080HALF },{2*INV1920+INV1920HALF,          0+INV1080HALF },
  {-2*INV1920+INV1920HALF,    INV1080+INV1080HALF },{-INV1920+INV1920HALF,    INV1080+INV1080HALF },{0+INV1920HALF,    INV1080+INV1080HALF },{INV1920+INV1920HALF,    INV1080+INV1080HALF },{2*INV1920+INV1920HALF,    INV1080+INV1080HALF },
  {-2*INV1920+INV1920HALF,  2*INV1080+INV1080HALF },{-INV1920+INV1920HALF,  2*INV1080+INV1080HALF },{0+INV1920HALF, 2*INV1080 +INV1080HALF },{INV1920+INV1920HALF,  2*INV1080+INV1080HALF },{2*INV1920+INV1920HALF,  2*INV1080+INV1080HALF }
};

    float SI = 0;//shadow depth
#ifdef SM_PCF4x4
    for (int x=0;x<4;x++)
    {
        for (int y=0;y<4;y++)
        {
            SI+= txDepthMap.SampleCmp(samCmpLess, txProj.xy + OffSetPCF[x][y], txProj.z - 0.005f);
        }
    }
    SI/=16;
 #endif
 #ifdef SM_GATHER4
    float4 S = float4(0,0,0,0);
    S += txDepthMap.GatherCmpRed(samCmpLess, txProj.xy+ OffSetGather[1][1], txProj.z - 0.005f, int2(-1,-1));
    S += txDepthMap.GatherCmpRed(samCmpLess, txProj.xy+ OffSetGather[0][1], txProj.z - 0.005f, int2(1,-1));
    S += txDepthMap.GatherCmpRed(samCmpLess, txProj.xy+ OffSetGather[1][0], txProj.z - 0.005f, int2(-1,1));
    S += txDepthMap.GatherCmpRed(samCmpLess, txProj.xy+ OffSetGather[1][1], txProj.z - 0.005f, int2(1,1));
    SI = (S.x+S.y+S.z+S.w)/16
  #endif

Can some one give me an recommendation on this feature and methods to enhance it whether it is attainable. You’ll discover that in gathercmpred there’s an offset on the uv and with int2 as nicely. Eradicating the extra int2 offset offers much less blur. I assume we will mix the 2 varieties of offsets in a single one.

Supply hyperlink