__mmask BVH4AOSTriangle1Intersector16Single::occluded(const BVH4AOSTriangle1Intersector16Single* This, Ray16& ray, const __mmask valid_i)
{
/* pointers to node array and triangle array */
const mic_m valid = valid_i;
const BVH4AOS* bvh = This->bvh;
const Node* const nodes = (const Node* ) bvh->nodePtr();
const Triangle1 *const triangles = (const Triangle1*) bvh->triPtr();
mic_i not_occluded = mic_i::minus_one();
long rayIndex = -1;
while((rayIndex = bsf64(rayIndex,valid)) != MIC_NO_BIT_SET_64)
{
// === TODO: precompute SOAtoAOS transformation, load with 4x broadcast
const mic3f ray_rdir = rcp_safe(ray.dir);
const mic_f org_xyz = SOAtoAOS_4f(rayIndex,ray.org.x,ray.org.y,ray.org.z);
const mic_f dir_xyz = SOAtoAOS_4f(rayIndex,ray.dir.x,ray.dir.y,ray.dir.z);
const mic_f rdir_xyz = SOAtoAOS_4f(rayIndex,ray_rdir.x,ray_rdir.y,ray_rdir.z);
const mic_f org_rdir_xyz = org_xyz * rdir_xyz;
const mic_f min_dist_xyz = upconv1f(&ray.tnear[rayIndex]);
const mic_f max_dist_xyz = upconv1f(&ray.tfar[rayIndex]);
if (BVH4AOSTriangle1Intersector1::occluded1(bvh,nodes,triangles,bvh->root,rayIndex,org_xyz,dir_xyz,rdir_xyz,org_rdir_xyz,min_dist_xyz,max_dist_xyz))
not_occluded[rayIndex] = 0;
}
return valid & eq(not_occluded,mic_i::zero());
}
void BVH4AOSTriangle1Intersector16Single::intersect(const BVH4AOSTriangle1Intersector16Single* This, Ray16& ray, const __mmask valid_i)
{
/* pointers to node array and triangle array */
const mic_m valid = valid_i;
const BVH4AOS* bvh = This->bvh;
const Node* const nodes = (const Node* ) bvh->nodePtr();
const Triangle1 *const triangles = (const Triangle1*) bvh->triPtr();
long rayIndex = -1;
while((rayIndex = bsf64(rayIndex,valid)) != MIC_NO_BIT_SET_64)
{
const mic3f ray_rdir = rcp_safe(ray.dir);
const mic_f org_xyz = SOAtoAOS_4f(rayIndex,ray.org.x,ray.org.y,ray.org.z);
const mic_f dir_xyz = SOAtoAOS_4f(rayIndex,ray.dir.x,ray.dir.y,ray.dir.z);
const mic_f rdir_xyz = SOAtoAOS_4f(rayIndex,ray_rdir.x,ray_rdir.y,ray_rdir.z);
const mic_f org_rdir_xyz = org_xyz * rdir_xyz;
const mic_f min_dist_xyz = upconv1f(&ray.tnear[rayIndex]);
const mic_f max_dist_xyz = upconv1f(&ray.tfar[rayIndex]);
BVH4AOSTriangle1Intersector1::intersect1(bvh,nodes,triangles,bvh->root,rayIndex,org_xyz,dir_xyz,rdir_xyz,org_rdir_xyz,min_dist_xyz,max_dist_xyz,ray);
}
}
int mtrropt(u64t wc_addr, u64t wc_len, u32t *memlimit) {
u32t reg;
int ii,
sv4idx = 0;
mtrrentry save4[16];
memset(&save4,0,sizeof(save4));
*memlimit = 0;
if (is_included(wc_addr,wc_len)<0 && is_intersection(wc_addr, wc_len)<0 &&
is_regavail(®))
{
mtrr[reg].start = wc_addr;
mtrr[reg].len = wc_len;
mtrr[reg].cache = MTRRF_WC;
mtrr[reg].on = 1;
return 0;
}
// video memory in not in 4th GB
if (wc_addr<_3GbLL || wc_addr+wc_len>_4GbLL) return OPTERR_VIDMEM3GB;
/* turn off previous write combine on the same memory,
but leave this block to catch low UC border successfully */
ii = is_include(wc_addr, wc_len);
if (ii>=0 && mtrr[ii].cache==MTRRF_WC) mtrr[ii].cache=MTRRF_UC;
// only WB and UC allowed in first 4Gb
for (ii=0; ii<regs; ii++)
if (mtrr[ii].on && mtrr[ii].cache!=MTRRF_UC && mtrr[ii].cache!=MTRRF_WB
&& mtrr[ii].start<_4GbLL) return OPTERR_UNKCT;
// is block intersected with someone?
ii = is_intersection(wc_addr, wc_len);
if (ii>=0) return OPTERR_INTERSECT;
// remove/truncate all above 4Gb (but save it)
for (ii=0; ii<regs; ii++)
if (mtrr[ii].on)
if (mtrr[ii].start<_4GbLL && mtrr[ii].start+mtrr[ii].len>_4GbLL) {
u64t newlen = _4GbLL - mtrr[ii].start,
remain = mtrr[ii].len - newlen;
if (!is_power2(newlen)) return OPTERR_SPLIT4GB;
mtrr[ii].len = newlen;
// save block
if (is_power2(remain)) {
save4[sv4idx].start = _4GbLL;
save4[sv4idx].len = remain;
save4[sv4idx].cache = mtrr[ii].cache;
sv4idx++;
} else
if (is_power2(remain/3)) {
}
} else
if (mtrr[ii].start>=_4GbLL || mtrr[ii].start+mtrr[ii].len>_4GbLL) {
save4[sv4idx].start = mtrr[ii].start;
save4[sv4idx].len = mtrr[ii].len;
save4[sv4idx].cache = mtrr[ii].cache;
sv4idx++;
clearreg(ii);
}
u64t wbend = 0,
ucstart = FFFF64;
// searching for upper WB border
for (ii=0; ii<regs; ii++)
if (mtrr[ii].on)
if (mtrr[ii].cache==MTRRF_WB)
if (mtrr[ii].start+mtrr[ii].len > wbend)
wbend = mtrr[ii].start+mtrr[ii].len;
// searching for lower UC border (but ignore small blocks)
for (ii=0; ii<regs; ii++)
if (mtrr[ii].on)
if (mtrr[ii].cache==MTRRF_UC) {
int pwr = bsf64(mtrr[ii].len);
if (pwr>=27) {
if (ucstart>mtrr[ii].start) ucstart = mtrr[ii].start;
clearreg(ii);
}
}
// pass #2 - removing small blocks above selected border
for (ii=0; ii<regs; ii++)
if (mtrr[ii].on)
if (mtrr[ii].cache==MTRRF_UC && ucstart<=mtrr[ii].start)
clearreg(ii);
// if no UC entries - use the end of WB as border
if (ucstart>wbend) ucstart = wbend;
// this can occur on small video memory size (<128Mb)
if (wc_addr<ucstart) return OPTERR_BELOWUC;
// build new WB list
if (ucstart<wbend) {
if (ucstart<_1GbLL) return OPTERR_LOWUC;
for (ii=0; ii<regs; ii++)
if (mtrr[ii].on)
if (mtrr[ii].cache==MTRRF_WB) clearreg(ii);
int regsfree = regsavail() - sv4idx - 1;
log_it(2, "regs free: %i \n", regsfree);
// force 3 registers (some memory above 4Gb can be lost)
if (regsfree<3) regsfree = 3;
// split memory to list
u64t nextpos = 0;
ii = 0;
for (u64t size=_2GbLL; size>=_64MbLL; size>>=1) {
if (ucstart>=size) {
if (!is_regavail(®)) return OPTERR_NOREG;
mtrr[reg].start = nextpos;
nextpos += (mtrr[reg].len = size);
mtrr[reg].cache = MTRRF_WB;
mtrr[reg].on = 1;
ucstart -= size;
// use only 3 mtrr regs
if (++ii==regsfree) break;
}
}
// save memlimit value
*memlimit = nextpos>>20;
/** and again removing small blocks above selected border...
splitted blocks sum can be smaller than previously selected
UC border and some blocks can be cleared here */
for (ii=0; ii<regs; ii++)
if (mtrr[ii].on)
if (mtrr[ii].cache==MTRRF_UC && nextpos<=mtrr[ii].start)
clearreg(ii);
}
// final check
if (is_included(wc_addr,wc_len)>=0 || is_intersection(wc_addr,wc_len)>=0 ||
is_include(wc_addr,wc_len)>=0) return OPTERR_OPTERR;
// add entry
if (is_regavail(®)) {
mtrr[reg].start = wc_addr;
mtrr[reg].len = wc_len;
mtrr[reg].cache = MTRRF_WC;
mtrr[reg].on = 1;
}
// restore some of above 4Gb memory blocks
if (sv4idx && regsavail()>0) {
for (ii=0; ii<sv4idx; ii++) {
if (!is_regavail(®)) break;
mtrr[reg].start = save4[ii].start;
mtrr[reg].len = save4[ii].len;
mtrr[reg].cache = save4[ii].cache;
mtrr[reg].on = 1;
}
// check lost items for included UC entries
while (ii<sv4idx) {
if (mtrr[ii].cache==MTRRF_UC) {
int idx = is_included(save4[ii].start,save4[ii].len);
if (idx<0) idx = is_intersection(save4[ii].start,save4[ii].len);
// check it multiple times (for intersection)
if (idx>=0) { clearreg(idx); continue; }
}
ii++;
}
}
return 0;
}
/// return 0/1
static int is_power2(u64t length) {
if (length)
if ((u64t)1<<bsf64(length)==length) return 1;
return 0;
}
