What does the for loop means ?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello friends !
In the algorithm of matrixMul of cuda version 3.1.x，

//kernel warmup
matrixMul<<<grid,threads>>>(d_C,d_A,d_B,uiWA,uiWB);
cudaThreadSynchronize();
//create start timer
shrLog(“Run kernels…\n\n”);
unsigned int timer=0;
cutilCheckError(cutCreateTimer(&timer));
cutilCheckError(cutStartTimer(timer));
//execute the kernel
int nInter = 30;
for(int j=0;j<nInter;j++)
{
matrixMul<<<grid,threads>>>(d_C,d_A,d_B,uiWA,uiWB);
}

So I want to know what the for loop means ,and what does the kernel warmup means .Why the height of matrix A is not include int the kernel arguments ?
Expect yout answers ,thank you !

                                                                                                        zcloudz
2

For loop means averaging the runtime over a number of runs.

Warm up means that the first run is not timed - it is not unusual that the first run includes extraneous effects, such as populating cache/TLB, initialization of sorts, etc.

The height of matrix A must have been factored into the grid.

3

Thank you for detailed answer.that’s it…