// fills vscale and vshift with values so that algorithm performs
// vdst = vscale * vsrc + vbeta next;
void compute_vscaleshift(const Vmm &vscale, const Vmm &vshift,
const Vmm &vmean, const Vmm &vsqrtvar, size_t offt,
bool need_tail = false) {
if (need_tail) {
uni_vmovups_tail(vmean, mean_ptr(offt));
uni_vmovups_tail(vsqrtvar, var_ptr(offt));
} else {
uni_vmovups(vmean, mean_ptr(offt));
uni_vmovups(vsqrtvar, var_ptr(offt));
}
uni_vaddps(vsqrtvar, vsqrtvar, veps);
uni_vsqrtps(vsqrtvar, vsqrtvar);
if (bdesc_->use_scaleshift()) {
if (need_tail) {
uni_vmovups_tail(vscale, scale_ptr(offt));
uni_vmovups_tail(vshift, shift_ptr(offt));
} else {
uni_vmovups(vscale, scale_ptr(offt));
uni_vmovups(vshift, shift_ptr(offt));
}
vdivps(vscale, vscale, vsqrtvar);
uni_vfnmadd231ps(vshift, vmean, vscale);
} else {
vdivps(vscale, vone, vsqrtvar);
uni_vmulps(vmean, vmean, vscale);
uni_vsubps(vshift, vzero, vmean);
}
};
int main(int argc, char *argv[]) {
int i,j,trials=1, verbose=0,n = 3;
data_t **a,**b,mask=1023,c = 7;
v_vptr scale_ptr;
/* get options */
for (i=1; i<argc; i++)
if (strcmp(argv[i],"-n") == 0) n = atoi(argv[++i]);
else if (strcmp(argv[i],"-t") == 0) trials = atoi(argv[++i]);
else if (strcmp(argv[i],"-m") == 0) mask = atoi(argv[++i]);
else if (strcmp(argv[i],"-c") == 0) c = atoi(argv[++i]);
else if (strcmp(argv[i],"-v") == 0) verbose = 1;
else if (strcmp(argv[i],"-d") == 0) disass = 1;
else if (strcmp(argv[i],"-p") == 0) pixie = 1;
else if (strcmp(argv[i],"-s") == 0) strength_reduce = 0;
/* allocate and initialize */
a = new_matrix(n,n);
b = new_matrix(n,n);
for(i=0;i<n;i++) {
for(j=0;j<n;j++) {
b[i][j] = (random() & mask) + 1;
}
}
if(verbose) {
printf("constant = %u\n",c);
m_print("B", b, n);
}
scale_ptr = scale(n, a, b);
if(verbose) printf("doing constants 1 to %d\n",c);
for(i=0;i<trials;i++) {
int c1;
if(!verbose) startwatch(0);
for(c1 = 0; c1 < c; c1++)
scale_ptr(a, b, c1);
if(!verbose) stopwatch(0);
}
if(verbose) m_print("A", a, n);
return 0;
}
