bool BBox::intersect(const Ray& ray, float *tnear, float *tfar) const {
// you may already have those values hanging around somewhere
const __m128
plus_inf = loadps(ps_cst_plus_inf),
minus_inf = loadps(ps_cst_minus_inf);
// use whatever's apropriate to load.
const __m128
box_min = loadps(&min),
box_max = loadps(&max),
pos = loadps(&ray.o),
inv_dir = loadps(&ray.inv_d);
// use a div if inverted directions aren't available
const __m128 l1 = mulps(subps(box_min, pos), inv_dir);
const __m128 l2 = mulps(subps(box_max, pos), inv_dir);
// the order we use for those min/max is vital to filter out
// NaNs that happens when an inv_dir is +/- inf and
// (box_min - pos) is 0. inf * 0 = NaN
const __m128 filtered_l1a = minps(l1, plus_inf);
const __m128 filtered_l2a = minps(l2, plus_inf);
const __m128 filtered_l1b = maxps(l1, minus_inf);
const __m128 filtered_l2b = maxps(l2, minus_inf);
// now that we're back on our feet, test those slabs.
__m128 lmax = maxps(filtered_l1a, filtered_l2a);
__m128 lmin = minps(filtered_l1b, filtered_l2b);
// unfold back. try to hide the latency of the shufps & co.
const __m128 lmax0 = rotatelps(lmax);
const __m128 lmin0 = rotatelps(lmin);
lmax = minss(lmax, lmax0);
lmin = maxss(lmin, lmin0);
const __m128 lmax1 = muxhps(lmax,lmax);
const __m128 lmin1 = muxhps(lmin,lmin);
lmax = minss(lmax, lmax1);
lmin = maxss(lmin, lmin1);
const bool ret = _mm_comige_ss(lmax, _mm_setzero_ps()) & _mm_comige_ss(lmax,lmin);
storess(lmin, tnear);
storess(lmax, tfar);
return ret;
}
/* min(v, v(Log_p Norm_{F_\p/Q_p}(x))) */
static long
vlognorm(GEN nf, GEN T, GEN x, GEN p, long v)
{
GEN a = nf_to_scalar_or_alg(nf, x);
GEN N = RgXQ_norm(a, T);
if (typ(N) != t_PADIC) N = cvtop(N, p, v);
return minss(v, valp( Qp_log(N) ));
}
inline bool ray_box_intersect(const box_t & b, const ray_t & ray, rayseg_t & rs) {
/* you may already have those values hanging around somewhere */
const __m128
plus_inf = loadps(ps_cst_plus_inf), minus_inf = loadps(ps_cst_minus_inf);
/* use whatever's apropriate to load. */
const __m128
box_min = loadps(&b.min), box_max = loadps(&b.max), pos =
loadps(&ray.pos), inv_dir = loadps(&ray.inv_dir);
/* use a div if inverted directions aren't available */
const __m128 l1 = mulps(subps(box_min, pos), inv_dir);
const __m128 l2 = mulps(subps(box_max, pos), inv_dir);
/* the order we use for those min/max is vital to filter out */
/* NaNs that happens when an inv_dir is +/- inf and */
/* (box_min - pos) is 0. inf * 0 = NaN */
const __m128 filtered_l1a = minps(l1, plus_inf);
const __m128 filtered_l2a = minps(l2, plus_inf);
const __m128 filtered_l1b = maxps(l1, minus_inf);
const __m128 filtered_l2b = maxps(l2, minus_inf);
/* now that we're back on our feet, test those slabs. */
__m128 lmax = maxps(filtered_l1a, filtered_l2a);
__m128 lmin = minps(filtered_l1b, filtered_l2b);
/* unfold back. try to hide the latency of the shufps & co. */
const __m128 lmax0 = rotatelps(lmax);
const __m128 lmin0 = rotatelps(lmin);
lmax = minss(lmax, lmax0);
lmin = maxss(lmin, lmin0);
const __m128 lmax1 = muxhps(lmax, lmax);
const __m128 lmin1 = muxhps(lmin, lmin);
lmax = minss(lmax, lmax1);
lmin = maxss(lmin, lmin1);
const bool ret =
_mm_comige_ss(lmax, _mm_setzero_ps()) & _mm_comige_ss(lmax, lmin);
storess(lmin, &rs.t_near);
storess(lmax, &rs.t_far);
return ret;
}
GEN
shallowmatconcat(GEN v)
{
long i, j, h, l = lg(v), L = 0, H = 0;
GEN M, maxh, maxl;
if (l == 1) return cgetg(1,t_MAT);
switch(typ(v))
{
case t_VEC:
for (i = 1; i < l; i++)
{
GEN c = gel(v,i);
GEN s = _matsize(c);
H = maxss(H, s[1]);
L += s[2];
}
M = zeromatcopy(H, L);
L = 0;
for (i = 1; i < l; i++)
{
GEN c = gel(v,i);
GEN s = _matsize(c);
matfill(M, c, 0, L, 1);
L += s[2];
}
return M;
case t_COL:
for (i = 1; i < l; i++)
{
GEN c = gel(v,i);
GEN s = _matsize(c);
H += s[1];
L = maxss(L, s[2]);
}
M = zeromatcopy(H, L);
H = 0;
for (i = 1; i < l; i++)
{
GEN c = gel(v,i);
GEN s = _matsize(c);
matfill(M, c, H, 0, 1);
H += s[1];
}
return M;
case t_MAT:
h = lgcols(v);
maxh = zero_zv(h-1);
maxl = zero_zv(l-1);
for (j = 1; j < l; j++)
for (i = 1; i < h; i++)
{
GEN c = gcoeff(v,i,j);
GEN s = _matsize(c);
if (s[1] > maxh[i]) maxh[i] = s[1];
if (s[2] > maxl[j]) maxl[j] = s[2];
}
for (i = 1, H = 0; i < h; i++) H += maxh[i];
for (j = 1, L = 0; j < l; j++) L += maxl[j];
M = zeromatcopy(H, L);
for (j = 1, L = 0; j < l; j++)
{
for (i = 1, H = 0; i < h; i++)
{
GEN c = gcoeff(v,i,j);
matfill(M, c, H, L, minss(maxh[i], maxl[j]));
H += maxh[i];
}
L += maxl[j];
}
return M;
default:
pari_err_TYPE("shallowmatconcat", v);
return NULL;
}
}
