static double bench_norm(int s, long scale)
{
long dims[DIMS] = { 256 * scale, 256 * scale, 1, 16, 1, 1, 1, 1 };
#if 0
complex float* x = md_alloc_gpu(DIMS, dims, CFL_SIZE);
complex float* y = md_alloc_gpu(DIMS, dims, CFL_SIZE);
#else
complex float* x = md_alloc(DIMS, dims, CFL_SIZE);
complex float* y = md_alloc(DIMS, dims, CFL_SIZE);
#endif
md_gaussian_rand(DIMS, dims, x);
md_gaussian_rand(DIMS, dims, y);
double tic = timestamp();
switch (s) {
case 0:
md_zscalar(DIMS, dims, x, y);
break;
case 1:
md_zscalar_real(DIMS, dims, x, y);
break;
case 2:
md_znorm(DIMS, dims, x);
break;
case 3:
md_z1norm(DIMS, dims, x);
break;
}
double toc = timestamp();
md_free(x);
md_free(y);
return toc - tic;
}
int main_sdot(int argc, char* argv[])
{
cmdline(&argc, argv, 2, 2, usage_str, help_str, 0, NULL);
int N = DIMS;
long in1_dims[N];
long in2_dims[N];
complex float* in1_data = load_cfl(argv[1], N, in1_dims);
complex float* in2_data = load_cfl(argv[2], N, in2_dims);
for (int i = 0; i < N; i++)
assert(in1_dims[i] == in2_dims[i]);
// compute scalar product
complex float value = md_zscalar(N, in1_dims, in1_data, in2_data);
printf("%+e%+ei\n", crealf(value), cimagf(value));
unmap_cfl(N, in1_dims, in1_data);
unmap_cfl(N, in2_dims, in2_data);
exit(0);
}
