void sum_apply_adjoint(const void* _data, complex float* dst, const complex float* src)
{
const struct sum_data* data = _data;
md_clear( DIMS, data->imgd_dims, dst, sizeof( complex float ) );
md_zaxpy2( DIMS, data->imgd_dims, data->imgd_strs, dst, 1. / sqrtf( data->levels ) , data->img_strs, src );
}
static double bench_generic_add(long dims[DIMS], unsigned int flags, bool forloop)
{
long dimsX[DIMS];
long dimsY[DIMS];
long dimsC[DIMS];
md_select_dims(DIMS, flags, dimsX, dims);
md_select_dims(DIMS, ~flags, dimsC, dims);
md_select_dims(DIMS, ~0u, dimsY, dims);
long strsX[DIMS];
long strsY[DIMS];
md_calc_strides(DIMS, strsX, dimsX, CFL_SIZE);
md_calc_strides(DIMS, strsY, dimsY, CFL_SIZE);
complex float* x = md_alloc(DIMS, dimsX, CFL_SIZE);
complex float* y = md_alloc(DIMS, dimsY, CFL_SIZE);
md_gaussian_rand(DIMS, dimsX, x);
md_gaussian_rand(DIMS, dimsY, y);
long L = md_calc_size(DIMS, dimsC);
long T = md_calc_size(DIMS, dimsX);
double tic = timestamp();
if (forloop) {
for (long i = 0; i < L; i++) {
for (long j = 0; j < T; j++)
y[i + j * L] += x[j];
}
} else {
md_zaxpy2(DIMS, dims, strsY, y, 1., strsX, x);
}
double toc = timestamp();
md_free(x);
md_free(y);
return toc - tic;
}
