static void VI_mprodq_d(
vsip_mview_d *C,
vsip_qr_d *qr)
{
vsip_mview_d *H = qr->A;
vsip_scalar_d *beta = qr->beta;
vsip_vview_d *h, hh;
vsip_length j;
vsip_stride k;
vsip_vview_d vv = *(qr->v),
ww = *(qr->w);
vsip_vview_d *v = &vv,
*w = &ww;
h = VI_mcolview_d(H,0,&hh);
w->length = C->col_length;
for(k= 0; k < (vsip_stride)H->row_length; k++){
j = (vsip_length)k;
h->offset = j * H->row_stride +
j * H->col_stride + H->offset;
h->length = H->col_length - j;
v->length = h->length;
v->offset = qr->M - h->length;
VI_vcopy_d_d(h,v);
VI_VPUT_D(v,(vsip_index) 0, (vsip_scalar_d)1.0);
vv = *(qr->v);
v->length = C->col_length;
VI_smvprod_d(-beta[j],C,v,w);
v->length = qr->M;
VI_opu_d(C,w,v);
vv = *(qr->v);
VI_VPUT_D(v,k,0);
}
return;
}
static
void
VI_qprodm_d(
vsip_mview_d *C,
vsip_qr_d *qr)
{
vsip_mview_d *H = qr->A;
vsip_scalar_d *beta = qr->beta;
vsip_vview_d *h, hh;
vsip_length j;
vsip_stride k;
vsip_vview_d vv = *(qr->v),
ww = *(qr->w);
vsip_vview_d *v = &vv, *w = &ww;
VI_vfill_d(0,v);
h = VI_mcolview_d(H,0,&hh);
w->length = C->row_length;
for(k = qr->N -1; k >= 0; k--){
j = (vsip_length)k;
h->offset = j * (H->row_stride + H->col_stride) + H->offset;
h->length = H->col_length -j;
v->length = h->length;
v->offset = qr->M - h->length;
VI_vcopy_d_d(h,v);
VI_VPUT_D(v,(vsip_index)0,(vsip_scalar_d)1);
vv = *(qr->v);
v->length = C->col_length;
VI_svmprod_d(-beta[j],v,C,w);
v->length = qr->M;
VI_opu_d(C,v,w);
}
return;
}
int vsip_firflt_d(
vsip_fir_d *fir,
const vsip_vview_d *xc,
const vsip_vview_d *yc)
{
vsip_length nout,k;
vsip_vview_d xx = *xc,
yy = *yc;
vsip_vview_d H1 = *(fir->h),
H2 = *(fir->h);
vsip_vview_d *x=&xx,*y=&yy;
vsip_vview_d *h1=&H1,*h2=&H2;
vsip_offset oinc;
oinc = (vsip_offset)((vsip_stride)fir->D * x->stride);
/* calculate number of terms in y */
nout = (fir->N - fir->p);
nout = ((nout % fir->D) == 0) ? (nout / fir->D ) : (nout / fir->D + 1);
/* do overlap section */
k = 0;
x->length = fir->p + 1;
h1->length = fir->s->length;
h2->length = x->length;
h2->offset = h1->length;
while(x->length < fir->M){
VI_VPUT_D(y,k++,vsip_vdot_d(h1,fir->s)+vsip_vdot_d(h2,x));
x->length += fir->D;
fir->s->length -= fir->D;
fir->s->offset += fir->D;
h1->length = fir->s->length;
h2->length = x->length;
h2->offset = h1->length;
}
x->offset += (x->length - fir->M) * x->stride;
x->length = fir->M;
while(k < nout){ /* do the rest of the pieces */
VI_VPUT_D(y,k++,vsip_vdot_d(fir->h,x));
x->offset += oinc;
}
{ /* Hacked Mod because In ANSI C it is a machine dependent mod for negative */
vsip_stride temp_p = (fir->p % fir->D) - (fir->N % fir->D);
fir->p = ((temp_p < 0) ? (vsip_length)((vsip_stride)fir->D + temp_p) : (vsip_length)temp_p);
}
fir->s->offset = 0;
fir->s->length = (fir->state == VSIP_STATE_SAVE) ? fir->M - 1 - fir->p : fir->M -1;
x->length = fir->s->length;
/* fix by JMA 31/01/2000, incorrect offset calculation */
/* x->offset = xc->length - fir->s->length; */
x->offset = xc->offset + (xc->length - fir->s->length) * xc->stride;
if((fir->s->length > 0) && (fir->state == VSIP_STATE_SAVE))
VI_vcopy_d_d(x,fir->s);
if(fir->state == VSIP_STATE_NO_SAVE){
VI_vfill_d(0,fir->s);
fir->p = 0;
}
return (int)k;
}
vsip_fir_d *vsip_fir_create_d(
const vsip_vview_d *kernel,
vsip_symmetry symm,
vsip_length N,
vsip_length D,
vsip_obj_state state,
unsigned int ntimes,
vsip_alg_hint hint)
{
vsip_length klength = 0;
vsip_fir_d *fir = (vsip_fir_d*)malloc(sizeof(vsip_fir_d));
if(fir == NULL){ return (vsip_fir_d*) NULL;
} else {
switch(symm){
case 0: klength = kernel->length;
break;
case 1: klength = 2 * kernel->length - 1;
break;
case 2: klength = 2 * kernel->length;
break;
}
fir->h = VI_vcreate_d(klength,VSIP_MEM_NONE);
fir->s = VI_vcreate_d(klength-1,VSIP_MEM_NONE);
}
if((fir->h == NULL) | (fir->s == NULL)){
vsip_fir_destroy_d(fir);
return (vsip_fir_d*) NULL;
} else {
fir->N = N;
fir->M = klength;
fir->D = D;
fir->p = 0;
fir->ntimes = ntimes;
fir->symm = symm;
fir->hint = hint;
switch(symm){
case 0:{
fir->h->offset = fir->h->length - 1;
fir->h->stride = -1;
VI_vcopy_d_d(kernel,fir->h);
}
break;
case 1: {
fir->h->length = kernel->length;
VI_vcopy_d_d(kernel,fir->h);
fir->h->offset = klength - 1;
fir->h->stride = -1;
VI_vcopy_d_d(kernel,fir->h);
fir->h->length = klength;
}
break;
case 2: {
fir->h->length = kernel->length;
VI_vcopy_d_d(kernel,fir->h);
fir->h->offset = klength - 1;
fir->h->stride = -1;
VI_vcopy_d_d(kernel,fir->h);
fir->h->length = klength;
}
break;
}
fir->state = state;
fir->h->offset = 0;
fir->h->stride = 1;
VI_vfill_d(0,fir->s);
return fir;
}
}
