CUDA: test occupation #1

#include <stdio.h>

const int REPEAT_TEST            = 10;
const int ITERATIONs             = 0xFFF;
const int THREADS_PER_BLOCK_max  = 512;
const int THREADS_PER_BLOCK_step = 32;
const int BLOCKS_max             = 60;


__global__ void Kernel(float *res)
{
    float a = threadIdx.x + blockIdx.x*blockDim.x;
    for(int j = 1; j < ITERATIONs; ++j)
    {
        a += sinf(a*1.01 + 0.01) + cosf(a*0.99 - 0.01);
        //++a;
    }
}


int main(void)
{
    float *res_d;
    float elapsedTime;

    cudaEvent_t start, stop;
    cudaEventCreate(&start);
    cudaEventCreate(&stop);

    FILE *fo = fopen("out.txt", "w");
    fprintf(fo, "%i  %i  %i\n", BLOCKS_max, THREADS_PER_BLOCK_step, THREADS_PER_BLOCK_max);

    cudaMalloc(&res_d, BLOCKS_max*THREADS_PER_BLOCK_max*sizeof(float));
    for(int nb = 1; nb <= BLOCKS_max; ++nb)
    for(int nt = THREADS_PER_BLOCK_step; nt <= THREADS_PER_BLOCK_max; nt += THREADS_PER_BLOCK_step)
    {
        printf("BLOCKS = %3i, THREADS_PER_BLOCK = %3i:\n", nb, nt);

        elapsedTime = 0;
        for(int r = 1; r <= REPEAT_TEST; ++r)
        {
            printf("  r = %i/%i\r", r, REPEAT_TEST);
            fflush(stdout);
            cudaEventRecord(start, 0);
            Kernel <<<nb, nt>>> (res_d);
            cudaEventRecord(stop, 0);
            cudaEventSynchronize(stop);
            cudaEventElapsedTime(&elapsedTime, start, stop);
        }

        printf("  Elapsed Time: %20.14f\n", elapsedTime/REPEAT_TEST);
        fprintf(fo, "%i  %i  %20.14f\n", nb, nt, elapsedTime/REPEAT_TEST);
    }

    cudaFree(res_d);
    fclose(fo);

    return 0;
}