GPGPU Programming for Games and Science

Sample Applications 391

In the second (and optimal) attempt, we will use

minmax8, minmax12,and

minmax16 multiple times. Each time, though, we must copy two unprocessed

elements rather than one in order to set up for the next smaller subset. The

compute shader for the median filtering is shown in Listing 7.23.

#include ”MedianShared . h l s l i ”

Texture2D<float> input ;

RWTexture2D<float> output ;

[ numthreads (NUM

X THREADS , NUM Y THREADS , 1 ) ]

void CSMai n ( i nt2 dt : SV

DispatchThreadID)

{

float4 e [7]; // 28 sl o t s , we u s e t h e f i r s t 25

LoadNeighbors( input , dt , e );

minmax16(e ); // D i s c a rd min/max o f v0 . . v1 5 (2n +1=25, n+2<16). ...