Bart/bullet3
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Use the add support for double and float triangle vertices. If it has a bad impact on performance, we can hide it behind an #ifdef for double precision vertex support.

Browse Source 
Fix btConvexTriangleMeshShape::calculatePrincipalAxisTransform
Both thanks to Ole.

Use our own union to extract a scalar value from an SSE intrinsic (instead of relying on MSVC-only m128_u32
Thanks to BlindSide
This commit is contained in:
erwin.coumans
2008-10-27 21:36:36 +00:00
parent 876009b053
commit 1ff6420ffb
4 changed files with 71 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
src/BulletCollision/BroadphaseCollision/btDbvt.h
View File

@@ -568,6 +568,8 @@ DBVT_INLINE btScalar Proximity( const btDbvtAabbMm& a,
return(btFabs(d.x())+btFabs(d.y())+btFabs(d.z()));
}
//
DBVT_INLINE int Select( const btDbvtAabbMm& o,
const btDbvtAabbMm& a,
@@ -577,6 +579,14 @@ DBVT_INLINE int Select( const btDbvtAabbMm& o,
static DBVT_ALIGN const unsigned __int32 mask[]={0x7fffffff,0x7fffffff,0x7fffffff,0x7fffffff};
// TODO: the intrinsic version is 11% slower
#if DBVT_USE_INTRINSIC_SSE
union btSSEUnion ///NOTE: if we use more intrinsics, move btSSEUnion into the LinearMath directory
{
__m128 ssereg;
float floats[4];
int ints[4];
};
__m128 omi(_mm_load_ps(o.mi));
omi=_mm_add_ps(omi,_mm_load_ps(o.mx));
__m128 ami(_mm_load_ps(a.mi));
@@ -590,10 +600,14 @@ DBVT_INLINE int Select( const btDbvtAabbMm& o,
__m128 t0(_mm_movehl_ps(ami,ami));
ami=_mm_add_ps(ami,t0);
ami=_mm_add_ss(ami,_mm_shuffle_ps(ami,ami,1));
__m128 t1(_mm_movehl_ps(bmi,bmi));
__m128 t1(_mm_movehl_ps(bmi,bmi));
bmi=_mm_add_ps(bmi,t1);
bmi=_mm_add_ss(bmi,_mm_shuffle_ps(bmi,bmi,1));
return(_mm_cmple_ss(bmi,ami).m128_u32[0]&1);
btSSEUnion tmp;
tmp.ssereg = _mm_cmple_ss(bmi,ami);
return tmp.ints[0]&1;
#else
DBVT_ALIGN __int32 r[1];
__asm