int main()
{
int i;
vsip_randstate *state;
vsip_cmview_d *xy;
vsip_cvview_d *xy_row;
vsip_length stride;
vsip_cvview_d *sum;
vsip_fftm_d *ccfftmip;
vsip_length seed =0, num_procs=1, id=1;
vsip_init((void *)0);
/* Initialize Random Number Generator */
state = vsip_randcreate(seed, num_procs, id, VSIP_PRNG);
/* Row major, 64 (ROWS) of 16 point data vectors */
xy = vsip_cmcreate_d(ROWS, 16, VSIP_ROW, VSIP_MEM_NONE);
/* Bind xy_row view initially to row 0 of xy */
xy_row = vsip_cmrowview_d(xy, 0);
/* Stride between column elements of xy */
stride = vsip_cmgetcolstride_d(xy);
sum = vsip_cvcreate_d(N, VSIP_MEM_NONE);
/* Create an in-place Cmplx->Cmplx Multiple N-pt FFT */
ccfftmip = vsip_ccfftmip_create_d(ROWS, N, 1.0, VSIP_FFT_FWD,
VSIP_ROW, 1, VSIP_ALG_TIME);
/* Initialize xy by rows with complex Gaussian noise N(0,1) */
for (i=0; i<ROWS; i++)
{
vsip_cvputoffset_d(xy_row, i*stride); /* view of row i of xy */
vsip_cvrand_d(state,xy_row); /* Initialize row i of xy */
}
/* Compute an in-place Cmplx->Cmplx Multiple N-pt FFT using the
ccfftmip object*/
vsip_ccfftmip_d(ccfftmip, xy);
/* Coherently sum the rows together (in the Freq domain) */
vsip_cvputoffset_d(xy_row, 0);
vsip_cvcopy_d_d(xy_row, sum); /* sum = row 0 of xy */
for (i=1; i<ROWS; i++)
{
vsip_cvputoffset_d(xy_row, i*stride); /* view of row i of xy */
vsip_cvadd_d(xy_row, sum, sum); /* sum += row i of xy */
}
/* Print it */
printf("\nComplex Output Vector (Real, Imag)\n");
for(i=0; i<N; i++)
printf("%d:\t" SCMPLX "\n", i, ACMPLX(vsip_cvget_d(sum,i)));
printf("\n");
/* Destroy all the objects */
vsip_fftm_destroy_d(ccfftmip);
vsip_cvdestroy(xy_row);
vsip_cvdestroy(sum);
vsip_cmalldestroy_d(xy);
vsip_randdestroy(state);
vsip_finalize((void *)0);
return(0);
}
void
ref_ccorr_d(vsip_bias bias, vsip_support_region sup,
vsip_cvview_d const *ref,
vsip_cvview_d const *in,
vsip_cvview_d const *out)
{
vsip_length M = vsip_cvgetlength_d(ref);
vsip_length N = vsip_cvgetlength_d(in);
vsip_length P = vsip_cvgetlength_d(out);
vsip_length expected_P = ref_corr_output_size(sup, M, N);
vsip_stride shift = ref_expected_shift(sup, M);
assert(expected_P == P);
vsip_cvview_d *sub = vsip_cvcreate_d(M, VSIP_MEM_NONE);
// compute correlation
vsip_index i;
for (i=0; i<P; ++i)
{
vsip_cvfill_d(vsip_cmplx_d(0,0), sub);
vsip_stride pos = (vsip_stride)i + shift;
double scale;
if (pos < 0)
{
vsip_cvview_d *subsub = vsip_cvsubview_d(sub, -pos, M + pos);
vsip_cvview_d *insub = vsip_cvsubview_d(in, 0, M + pos);
vsip_cvcopy_d_d(insub, subsub);
vsip_cvdestroy_d(subsub);
vsip_cvdestroy_d(insub);
scale = M + pos;
}
else if (pos + M > N)
{
vsip_cvview_d *subsub = vsip_cvsubview_d(sub, 0, N - pos);
vsip_cvview_d *insub = vsip_cvsubview_d(in, pos, N - pos);
vsip_cvcopy_d_d(insub, subsub);
vsip_cvdestroy_d(subsub);
vsip_cvdestroy_d(insub);
scale = N - pos;
}
else
{
vsip_cvview_d *insub = vsip_cvsubview_d(in, pos, M);
vsip_cvcopy_d_d(insub, sub);
vsip_cvdestroy_d(insub);
scale = M;
}
#if VSIP_IMPL_CORR_CORRECT_SAME_SUPPORT_SCALING
#else
if (sup == VSIP_SUPPORT_SAME)
{
if (i < (M/2)) scale = i + (M+1)/2; // i + ceil(M/2)
else if (i < N - (M/2)) scale = M; // M
else scale = N - 1 + (M+1)/2 - i; // N-1+ceil(M/2)-i
}
#endif
vsip_cscalar_d val = vsip_cvjdot_d(ref, sub);
if (bias == VSIP_UNBIASED)
{
val.r /= scale;
val.i /= scale;
}
vsip_cvput_d(out, i, val);
}
}
