-
sparker
thiago
Manta Interactive Ray Tracer Development Mailing List
Text archives Help
- From: bigler@sci.utah.edu
- To: manta@sci.utah.edu
- Subject: [MANTA] r1228 - in trunk: Interface Model/Materials Model/Primitives
- Date: Thu, 26 Oct 2006 16:06:32 -0600 (MDT)
Author: bigler
Date: Thu Oct 26 16:06:31 2006
New Revision: 1228
Modified:
trunk/Interface/RayPacket.h
trunk/Model/Materials/Phong.cc
trunk/Model/Primitives/Sphere.cc
Log:
Interface/RayPacket.h
Added hitFinal, hitWithMask, hitWithoutTminCheck, hit (with no
mask), wereNotHitSSE, and wereHitSSE.
Model/Materials/Phong.cc
Replaced shadow ray hit mask with a call to wereNotHitSSE.
Model/Primitives/Sphere.cc
Replaced hit registration code with calls to RayPacket hit
functions.
Modified: trunk/Interface/RayPacket.h
==============================================================================
--- trunk/Interface/RayPacket.h (original)
+++ trunk/Interface/RayPacket.h Thu Oct 26 16:06:31 2006
@@ -451,10 +451,7 @@
}
bool hit(int which, Real t, const Material* matl, const Primitive* prim,
const TexCoordMapper* tex) {
-
- if(t < T_EPSILON)
- return false;
- if(t < data->minT[which]){
+ if(t > T_EPSILON && t < data->minT[which]){
data->minT[which] = t;
data->hitMatl[which] = matl;
data->hitPrim[which] = prim;
@@ -464,10 +461,161 @@
return false;
}
}
+#ifdef MANTA_SSE
+ private:
+ inline __m128 hitFinal(int start, __m128 mask, __m128 t,
+ const Material* matl, const Primitive* prim,
+ const TexCoordMapper* tex) {
+ mask = _mm_and_ps(mask, _mm_cmplt_ps(t,
_mm_load_ps(&data->minT[start])));
+ int hit = _mm_movemask_ps(mask);
+ if(hit == 15){
+ // All of them hit, so we can store them all
+ _mm_store_ps(&data->minT[start], t);
+#ifdef __x86_64
+ // hitMatl
+ _mm_store_si128((__m128i*)&data->hitMatl[start],
_mm_set1_epi64x((long long)matl));
+ _mm_store_si128((__m128i*)&data->hitMatl[start+2],
_mm_set1_epi64x((long long)matl));
+ // hitPrim
+ _mm_store_si128((__m128i*)&data->hitPrim[start],
_mm_set1_epi64x((long long)prim));
+ _mm_store_si128((__m128i*)&data->hitPrim[start+2],
_mm_set1_epi64x((long long)prim));
+ // hitTex
+ _mm_store_si128((__m128i*)&data->hitTex[start],
_mm_set1_epi64x((long long)tex));
+ _mm_store_si128((__m128i*)&data->hitTex[start+2],
_mm_set1_epi64x((long long)tex));
+#else // __x86_64
+ _mm_store_si128((__m128i*)&data->hitMatl[start],
_mm_set1_epi32((int)matl));
+ _mm_store_si128((__m128i*)&data->hitPrim[start],
_mm_set1_epi32((int)prim));
+ _mm_store_si128((__m128i*)&data->hitTex[start],
_mm_set1_epi32((int)tex));
+#endif // __x86_64
+ } else if(hit != 0){
+ // Some of them hit, some of them do not
+ // Intel has a masked move instruction but it seems to be really slow
+ // It is faster to load the old data and mask it in manually
+ _mm_store_ps(&data->minT[start],
+ _mm_or_ps(_mm_and_ps(mask, t),
+ _mm_andnot_ps(mask,
_mm_load_ps(&data->minT[start]))));
+#ifdef __x86_64
+ __m128i lomask = _mm_castps_si128(_mm_unpacklo_ps(mask, mask));
+ __m128i himask = _mm_castps_si128(_mm_unpackhi_ps(mask, mask));
+ // hitMatl
+ _mm_store_si128((__m128i*)&data->hitMatl[start],
+ _mm_or_si128(_mm_and_si128(lomask,
_mm_set1_epi64x((long long)matl)),
+ _mm_andnot_si128(lomask,
_mm_load_si128((__m128i*)&data->hitMatl[start]))));
+ _mm_store_si128((__m128i*)&data->hitMatl[start+2],
+ _mm_or_si128(_mm_and_si128(himask,
_mm_set1_epi64x((long long)matl)),
+ _mm_andnot_si128(himask,
_mm_load_si128((__m128i*)&data->hitMatl[start+2]))));
+ // hitPrim
+ _mm_store_si128((__m128i*)&data->hitPrim[start],
+ _mm_or_si128(_mm_and_si128(lomask,
_mm_set1_epi64x((long long)prim)),
+ _mm_andnot_si128(lomask,
_mm_load_si128((__m128i*)&data->hitPrim[start]))));
+ _mm_store_si128((__m128i*)&data->hitPrim[start+2],
+ _mm_or_si128(_mm_and_si128(himask,
_mm_set1_epi64x((long long)prim)),
+ _mm_andnot_si128(himask,
_mm_load_si128((__m128i*)&data->hitPrim[start+2]))));
+ // hitTex
+ _mm_store_si128((__m128i*)&data->hitTex[start],
+ _mm_or_si128(_mm_and_si128(lomask,
_mm_set1_epi64x((long long)tex)),
+ _mm_andnot_si128(lomask,
_mm_load_si128((__m128i*)&data->hitTex[start]))));
+ _mm_store_si128((__m128i*)&data->hitTex[start+2],
+ _mm_or_si128(_mm_and_si128(himask,
_mm_set1_epi64x((long long)tex)),
+ _mm_andnot_si128(himask,
_mm_load_si128((__m128i*)&data->hitTex[start+2]))));
+#else // __x86_64
+ __m128i imask = _mm_castps_si128(mask);
+ _mm_store_si128((__m128i*)&data->hitMatl[start],
+ _mm_or_si128(_mm_and_si128(imask,
_mm_set1_epi32((int)matl)),
+ _mm_andnot_si128(imask,
_mm_load_si128((__m128i*)&data->hitMatl[start]))));
+ _mm_store_si128((__m128i*)&data->hitPrim[start],
+ _mm_or_si128(_mm_and_si128(imask,
_mm_set1_epi32((int)prim)),
+ _mm_andnot_si128(imask,
_mm_load_si128((__m128i*)&data->hitPrim[start]))));
+ _mm_store_si128((__m128i*)&data->hitTex[start],
+ _mm_or_si128(_mm_and_si128(imask,
_mm_set1_epi32((int)tex)),
+ _mm_andnot_si128(imask,
_mm_load_si128((__m128i*)&data->hitTex[start]))));
+#endif // __x86_64
+ }
+ return mask;
+ }
+ public:
+ //
+ // These are hit functions for SSE. There are four versions: the
+ // first three have a mask for prior selections (one with a
+ // complemented mask). The fourth has no mask. Both call the
+ // above private function to do the bulk of the work.
+ //
+
+ // Mask is all ones for rays that actually hit, all zeros otherwise
+ inline __m128 hitWithMask(int start, __m128 mask, __m128 t,
+ const Material* matl, const Primitive* prim,
+ const TexCoordMapper* tex) {
+ mask = _mm_and_ps(mask, _mm_cmpgt_ps(t, _mm_set1_ps(T_EPSILON)));
+ return hitFinal(start, mask, t, matl, prim, tex);
+ }
+ // Mask is all zeros for days that actually hit, all ones otherwise
+ inline __m128 hitWithComplementedMask(int start, __m128 mask, __m128 t,
+ const Material* matl,
+ const Primitive* prim,
+ const TexCoordMapper* tex) {
+ mask = _mm_andnot_ps(mask, _mm_cmpgt_ps(t, _mm_set1_ps(T_EPSILON)));
+ return hitFinal(start, mask, t, matl, prim, tex);
+ }
+ // Mask is all ones for rays that actually hit, all zeros
+ // otherwise. It is assumed that mask already contains the results
+ // of checking that t > epsilon.
+ inline __m128 hitWithoutTminCheck(int start, __m128 mask, __m128 t,
+ const Material* matl,
+ const Primitive* prim,
+ const TexCoordMapper* tex) {
+ return hitFinal(start, mask, t, matl, prim, tex);
+ }
+ // No mask
+ inline __m128 hit(int start, __m128 t, const Material* matl,
+ const Primitive* prim, const TexCoordMapper* tex) {
+ __m128 mask = _mm_cmpgt_ps(t, _mm_set1_ps(T_EPSILON));
+ return hitFinal(start, mask, t, matl, prim, tex);
+ }
+
+#endif // #ifdef MANTA_SSE
+
bool wasHit(int which)
{
return data->hitMatl[which] != 0;
}
+
+#ifdef MANTA_SSE
+ ///////////////////////////////////////////////////////////
+ //
+ // Returns a SSE based mask for the rays that have hitMatl == 0
+ //
+ // We are using SSE integer instructions to avoid weird problems
+ // when casting pointers to floats and doing comparisons.
+ // Unfortunately, there aren't any != (cmpneq) integer
+ // instructions. This is why the implementation is for
+ // wereNotHitSSE, and wereHitSSE takes the inverse of
+ // wereNotHitSSE.
+ //
+ // 32 bit: mask = (hitMatl == 0)
+ //
+ // 64 bit: mask = (hitMatl[low_bits] == 0) && (hitMatl[high_bits] == 0)
+ inline __m128 wereNotHitSSE(int start)
+ {
+#ifdef __x86_64
+ __m128 masklo =
_mm_castsi128_ps(_mm_cmpeq_epi32(_mm_load_si128((__m128i*)&data->hitMatl[start]),
+ _mm_setzero_si128()));
+ __m128 maskhi =
_mm_castsi128_ps(_mm_cmpeq_epi32(_mm_load_si128((__m128i*)&data->hitMatl[start+2]),
+ _mm_setzero_si128()));
+ return _mm_and_ps(_mm_shuffle_ps(masklo, maskhi, _MM_SHUFFLE(2, 0, 2,
0)),
+ _mm_shuffle_ps(masklo, maskhi, _MM_SHUFFLE(3, 1, 3,
1)));
+#else // __x86_64
+ return
_mm_castsi128_ps(_mm_cmpeq_epi32(_mm_load_si128((__m128i*)&data->hitMatl[start]),
+ _mm_setzero_si128()));
+
+#endif // __x86_64
+ }
+ ///////////////////////////////////////////////////////////
+ //
+ // Returns a SSE based mask for the rays that have hitMatl != 0
+ //
+ inline __m128 wereHitSSE(int start) {
+ return _mm_cmpeq_ps(wereNotHitSSE(start), _mm_setzero_ps());
+ }
+#endif // #ifdef MANTA_SSE
void setHitMaterial(int which, const Material* matl)
{
Modified: trunk/Model/Materials/Phong.cc
==============================================================================
--- trunk/Model/Materials/Phong.cc (original)
+++ trunk/Model/Materials/Phong.cc Thu Oct 26 16:06:31 2006
@@ -189,86 +189,9 @@
RayPacketData* shadowData = shadowRays.data;
for(;i<e;i+=4){
- // A note about all of this. We are trying to figure out
- // which rays are zero. This is done by checking the hitMatl
- // pointer for being NULL or not. If it is NULL then the
- // shadow ray didn't hit anything and we need to compute the
- // diffuse and specular components.
- //
- // The mask is a funny thing, where (if
- // !shadowRays.hitMatl[i]) meant "add direct illumination",
- // for the mask it is opposite. 0 is where we don't want to
- // add DI. This is done by comparing the hitMatl pointers
- // with zero. If they are 0 then set the mask bits on.
- //
- // The 64 bit stuff is harder. There isn't a good way of
- // doing 64 bit integer comparisons. We tried to cast the
- // pointers to doubles, but this didn't seem to work properly.
- // Probably because when cast to doubles 0xFFFF...FF is NaN.
- //
- // So we do the comparisons on the two halves of the pointer.
- // In these diagrams, FF represents "bit on" and 00 represents
- // "bit off". We need to combine the results of the pairs
- // (done with the SHUFFLEs later and "and" operator).
- //
- // Shadow Pointer Mask (X == 0) Mask Composite Result (and)
- // -------------- ------------- ----------------------
- // FF FF 00 00 00
- // FF 00 00 FF 00
- // 00 FF FF 00 00
- // 00 00 FF FF FF
- //
- // 32 bit
- // FF 00
- // 00 FF
- //
- // So with 64 bit both 32 bit halves need to be 00 for the
- // hitMatl to be shaded.
- //
- // For 32 bit: (mask = !(ShadowPointer == 0) or (mask =
!ShadowPointer)
- //
- // For 64 bit: (mask = !((SP[0] == 0) || (SP[1] == 0)) or
- // (mask = !(SP[0] || SP[1]) or
- // (mask = !SP[0] && !SP[1]) <- which is what we compute
-#ifdef __x86_64
- // This is the golden code
+ // We are interested in the rays that didn't hit anything
+ __m128 mask = shadowRays.wereNotHitSSE(i);
- __m128 masklo =
cast_i2f(_mm_cmpeq_epi32(load44i((__m128i*)&shadowData->hitMatl[i]),
- _mm_setzero_si128()));
- __m128 maskhi =
cast_i2f(_mm_cmpeq_epi32(load44i((__m128i*)&shadowData->hitMatl[i+2]),
- _mm_setzero_si128()));
- __m128 mask = _mm_and_ps(_mm_shuffle_ps(masklo, maskhi,
_MM_SHUFFLE(2, 0, 2, 0)),
- _mm_shuffle_ps(masklo, maskhi,
_MM_SHUFFLE(3, 1, 3, 1)));
-
-#else // __x86_64
-#ifndef SIMULATE_64BIT
- __m128 mask =
cast_i2f(_mm_cmpeq_epi32(load44i((__m128i*)&shadowData->hitMatl[i]),
_mm_setzero_si128()));
-#else // SIMULATE_64BIT
- __m128 mask;
- {
- bool debug = rays.getFlag(RayPacket::DebugPacket);
- if (debug) { cerr << "mask = ";simd_cerr(mask); }
-
- if (debug) { cerr << "mask2 = ";simd_cerr(mask); }
- // MANTA_ALIGN(16) int vals[8] = {0,0,0,0, 0,0,0,0};
- MANTA_ALIGN(16) unsigned long long vals[4];
-
- for(int j = 0; j < 4; ++j)
- vals[j] = (unsigned long long)(shadowData->hitMatl[i+j]);
- __m128 masklo =
cast_i2f(_mm_cmpeq_epi32(load44i((__m128i*)&vals[0]),
- _mm_setzero_si128()));
- __m128 maskhi =
cast_i2f(_mm_cmpeq_epi32(load44i((__m128i*)&vals[2]),
- _mm_setzero_si128()));
- if (debug) { cerr << "masklo = ";simd_cerr(masklo); }
- if (debug) { cerr << "maskhi = ";simd_cerr(maskhi); }
- __m128 mask3 = _mm_and_ps(_mm_shuffle_ps(masklo, maskhi,
_MM_SHUFFLE(2, 0, 2, 0)),
- _mm_shuffle_ps(masklo, maskhi,
_MM_SHUFFLE(3, 1, 3, 1)));
- if (debug) { cerr << "mask3 = ";simd_cerr(mask3); cerr << "\n"; }
- mask = mask3;
- }
-#endif // SIMULATE_64BIT
-#endif // __x86_64
-
if(_mm_movemask_ps(mask) == 0)
continue;
// Not in shadow, so compute the direct and specular contributions.
Modified: trunk/Model/Primitives/Sphere.cc
==============================================================================
--- trunk/Model/Primitives/Sphere.cc (original)
+++ trunk/Model/Primitives/Sphere.cc Thu Oct 26 16:06:31 2006
@@ -119,81 +119,16 @@
// -(r+B) The xor negates the value
sse_t t0 = xor4(add4(r, B), cast_i2f(set4i(0x80000000)));
sse_t hit0 = and4(hit, cmp4_gt(t0, set4(T_EPSILON)));
- if(getmask4(hit0) != 0){
- hit0 = and4(hit, cmp4_lt(t0, load44(&data->minT[i])));
- if(getmask4(hit0) == 15){
- store44(&data->minT[i], t0);
-#ifdef __x86_64
- store44i((sse_int_t*)&data->hitMatl[i], set4l((long
long)getMaterial()));
- store44i((sse_int_t*)&data->hitMatl[i+2], set4l((long
long)getMaterial()));
- store44i((sse_int_t*)&data->hitPrim[i], set4l((long
long)this));
- store44i((sse_int_t*)&data->hitPrim[i+2], set4l((long
long)this));
- store44i((sse_int_t*)&data->hitTex[i], set4l((long
long)getTexCoordMapper()));
- store44i((sse_int_t*)&data->hitTex[i+2], set4l((long
long)getTexCoordMapper()));
-#else
- store44i((sse_int_t*)&data->hitMatl[i],
set4i((int)getMaterial()));
- store44i((sse_int_t*)&data->hitPrim[i], set4i((int)this));
- store44i((sse_int_t*)&data->hitTex[i],
set4i((int)getTexCoordMapper()));
-#endif
- } else {
- maskmove4i(cast_f2i(t0), cast_f2i(hit0),
(char*)&data->minT[i]);
-#ifdef __x86_64
- sse_int_t lohit = cast_f2i(unpacklo(hit0, hit0));
- sse_int_t hihit = cast_f2i(unpackhi(hit0, hit0));
- maskmove4i(set4l((long long)getMaterial()), lohit,
(char*)&data->hitMatl[i]);
- maskmove4i(set4l((long long)getMaterial()), hihit,
(char*)&data->hitMatl[i+2]);
+ rays.hitWithoutTminCheck(i, hit0, t0, getMaterial(), this,
getTexCoordMapper());
- maskmove4i(set4l((long long)this), lohit,
(char*)&data->hitPrim[i]);
- maskmove4i(set4l((long long)this), hihit,
(char*)&data->hitPrim[i+2]);
- maskmove4i(set4l((long long)getTexCoordMapper()), lohit,
(char*)&data->hitTex[i]);
- maskmove4i(set4l((long long)getTexCoordMapper()), hihit,
(char*)&data->hitTex[i+2]);
-#else
- maskmove4i(set4i((int)getMaterial()), cast_f2i(hit0),
(char*)&data->hitMatl[i]);
- maskmove4i(set4i((int)this), cast_f2i(hit0),
(char*)&data->hitPrim[i]);
- maskmove4i(set4i((int)getTexCoordMapper()), cast_f2i(hit0),
(char*)&data->hitTex[i]);
-#endif
- }
- // Mask off rays that successfully hit at t0
- hit = _mm_andnot_ps(hit, hit0);
- if(getmask4(hit) == 0)
- continue;
- }
+ hit = andnot4(hit0, hit);
+ // Rays that successfully hit at t0 are masked off
+ if(getmask4(hit) == 0)
+ continue;
sse_t t1 = sub4(r, B);
sse_t hit1 = and4(hit, cmp4_gt(t1, set4(T_EPSILON)));
- hit1 = and4(hit1, cmp4_lt(t1, load44(&data->minT[i])));
- if(getmask4(hit1) == 15){
- store44(&data->minT[i], t1);
-#ifdef __x86_64
- store44i((sse_int_t*)&data->hitMatl[i], set4l((long
long)getMaterial()));
- store44i((sse_int_t*)&data->hitMatl[i+2], set4l((long
long)getMaterial()));
- store44i((sse_int_t*)&data->hitPrim[i], set4l((long long)this));
- store44i((sse_int_t*)&data->hitPrim[i+2], set4l((long
long)this));
- store44i((sse_int_t*)&data->hitTex[i], set4l((long
long)getTexCoordMapper()));
- store44i((sse_int_t*)&data->hitTex[i+2], set4l((long
long)getTexCoordMapper()));
-#else
- store44i((sse_int_t*)&data->hitMatl[i],
set4i((int)getMaterial()));
- store44i((sse_int_t*)&data->hitPrim[i], set4i((int)this));
- store44i((sse_int_t*)&data->hitTex[i],
set4i((int)getTexCoordMapper()));
-#endif
- } else {
- maskmove4i(cast_f2i(t1), cast_f2i(hit1), (char*)&data->minT[i]);
-#ifdef __x86_64
- sse_int_t lohit = cast_f2i(unpacklo(hit1, hit1));
- sse_int_t hihit = cast_f2i(unpackhi(hit1, hit1));
- maskmove4i(set4l((long long)getMaterial()), lohit,
(char*)&data->hitMatl[i]);
- maskmove4i(set4l((long long)getMaterial()), hihit,
(char*)&data->hitMatl[i+2]);
-
- maskmove4i(set4l((long long)this), lohit,
(char*)&data->hitPrim[i]);
- maskmove4i(set4l((long long)this), hihit,
(char*)&data->hitPrim[i+2]);
- maskmove4i(set4l((long long)getTexCoordMapper()), lohit,
(char*)&data->hitTex[i]);
- maskmove4i(set4l((long long)getTexCoordMapper()), hihit,
(char*)&data->hitTex[i+2]);
-#else
- maskmove4i(set4i((int)getMaterial()), cast_f2i(hit1),
(char*)&data->hitMatl[i]);
- maskmove4i(set4i((int)this), cast_f2i(hit1),
(char*)&data->hitPrim[i]);
- maskmove4i(set4i((int)getTexCoordMapper()), cast_f2i(hit1),
(char*)&data->hitTex[i]);
-#endif
- }
+ rays.hitWithoutTminCheck(i, hit1, t1, getMaterial(), this,
getTexCoordMapper());
}
for(;i<rays.rayEnd;i++){
Vector D(rays.getDirection(i));
@@ -314,81 +249,16 @@
// -(r+B) The xor negates the value
sse_t t0 = xor4(add4(r, B), cast_i2f(set4i(0x80000000)));
sse_t hit0 = and4(hit, cmp4_gt(t0, set4(T_EPSILON)));
- if(getmask4(hit0) != 0){
- hit0 = and4(hit, cmp4_lt(t0, load44(&data->minT[i])));
- if(getmask4(hit0) == 15){
- store44(&data->minT[i], t0);
-#ifdef __x86_64
- store44i((sse_int_t*)&data->hitMatl[i], set4l((long
long)getMaterial()));
- store44i((sse_int_t*)&data->hitMatl[i+2], set4l((long
long)getMaterial()));
- store44i((sse_int_t*)&data->hitPrim[i], set4l((long
long)this));
- store44i((sse_int_t*)&data->hitPrim[i+2], set4l((long
long)this));
- store44i((sse_int_t*)&data->hitTex[i], set4l((long
long)getTexCoordMapper()));
- store44i((sse_int_t*)&data->hitTex[i+2], set4l((long
long)getTexCoordMapper()));
-#else
- store44i((sse_int_t*)&data->hitMatl[i],
set4i((int)getMaterial()));
- store44i((sse_int_t*)&data->hitPrim[i], set4i((int)this));
- store44i((sse_int_t*)&data->hitTex[i],
set4i((int)getTexCoordMapper()));
-#endif
- } else {
- maskmove4i(cast_f2i(t0), cast_f2i(hit0),
(char*)&data->minT[i]);
-#ifdef __x86_64
- sse_int_t lohit = cast_f2i(unpacklo(hit0, hit0));
- sse_int_t hihit = cast_f2i(unpackhi(hit0, hit0));
- maskmove4i(set4l((long long)getMaterial()), lohit,
(char*)&data->hitMatl[i]);
- maskmove4i(set4l((long long)getMaterial()), hihit,
(char*)&data->hitMatl[i+2]);
+ rays.hitWithoutTminCheck(i, hit0, t0, getMaterial(), this,
getTexCoordMapper());
- maskmove4i(set4l((long long)this), lohit,
(char*)&data->hitPrim[i]);
- maskmove4i(set4l((long long)this), hihit,
(char*)&data->hitPrim[i+2]);
- maskmove4i(set4l((long long)getTexCoordMapper()), lohit,
(char*)&data->hitTex[i]);
- maskmove4i(set4l((long long)getTexCoordMapper()), hihit,
(char*)&data->hitTex[i+2]);
-#else
- maskmove4i(set4i((int)getMaterial()), cast_f2i(hit0),
(char*)&data->hitMatl[i]);
- maskmove4i(set4i((int)this), cast_f2i(hit0),
(char*)&data->hitPrim[i]);
- maskmove4i(set4i((int)getTexCoordMapper()), cast_f2i(hit0),
(char*)&data->hitTex[i]);
-#endif
- }
- // Mask off rays that successfully hit at t0
- hit = _mm_andnot_ps(hit, hit0);
- if(getmask4(hit) == 0)
- continue;
- }
+ hit = andnot4(hit0, hit);
+ // Rays that successfully hit at t0 are masked off
+ if(getmask4(hit) == 0)
+ continue;
sse_t t1 = sub4(r, B);
sse_t hit1 = and4(hit, cmp4_gt(t1, set4(T_EPSILON)));
- hit1 = and4(hit1, cmp4_lt(t1, load44(&data->minT[i])));
- if(getmask4(hit1) == 15){
- store44(&data->minT[i], t1);
-#ifdef __x86_64
- store44i((sse_int_t*)&data->hitMatl[i], set4l((long
long)getMaterial()));
- store44i((sse_int_t*)&data->hitMatl[i+2], set4l((long
long)getMaterial()));
- store44i((sse_int_t*)&data->hitPrim[i], set4l((long long)this));
- store44i((sse_int_t*)&data->hitPrim[i+2], set4l((long
long)this));
- store44i((sse_int_t*)&data->hitTex[i], set4l((long
long)getTexCoordMapper()));
- store44i((sse_int_t*)&data->hitTex[i+2], set4l((long
long)getTexCoordMapper()));
-#else
- store44i((sse_int_t*)&data->hitMatl[i],
set4i((int)getMaterial()));
- store44i((sse_int_t*)&data->hitPrim[i], set4i((int)this));
- store44i((sse_int_t*)&data->hitTex[i],
set4i((int)getTexCoordMapper()));
-#endif
- } else {
- maskmove4i(cast_f2i(t1), cast_f2i(hit1), (char*)&data->minT[i]);
-#ifdef __x86_64
- sse_int_t lohit = cast_f2i(unpacklo(hit1, hit1));
- sse_int_t hihit = cast_f2i(unpackhi(hit1, hit1));
- maskmove4i(set4l((long long)getMaterial()), lohit,
(char*)&data->hitMatl[i]);
- maskmove4i(set4l((long long)getMaterial()), hihit,
(char*)&data->hitMatl[i+2]);
-
- maskmove4i(set4l((long long)this), lohit,
(char*)&data->hitPrim[i]);
- maskmove4i(set4l((long long)this), hihit,
(char*)&data->hitPrim[i+2]);
- maskmove4i(set4l((long long)getTexCoordMapper()), lohit,
(char*)&data->hitTex[i]);
- maskmove4i(set4l((long long)getTexCoordMapper()), hihit,
(char*)&data->hitTex[i+2]);
-#else
- maskmove4i(set4i((int)getMaterial()), cast_f2i(hit1),
(char*)&data->hitMatl[i]);
- maskmove4i(set4i((int)this), cast_f2i(hit1),
(char*)&data->hitPrim[i]);
- maskmove4i(set4i((int)getTexCoordMapper()), cast_f2i(hit1),
(char*)&data->hitTex[i]);
-#endif
- }
+ rays.hitWithoutTminCheck(i, hit1, t1, getMaterial(), this,
getTexCoordMapper());
}
for(;i<rays.rayEnd;i++){
Vector O(rays.getOrigin(i)-center);
- [MANTA] r1228 - in trunk: Interface Model/Materials Model/Primitives, bigler, 10/26/2006
Archive powered by MHonArc 2.6.16.
