void MapLinearIntensities(FourVectors const &intens,uint32 *p1, uint32 *p2, uint32 *p3, uint32 *p4)
{
// convert four pixels worth of sse-style rgb into argb lwords
// NOTE the _mm_empty macro is voodoo. do not mess with this routine casually - simply throwing
// anything that ends up generating a fpu stack references in here would be bad news.
static fltx4 pixscale={255.0,255.0,255.0,255.0};
fltx4 r,g,b;
r=MinSIMD(pixscale,MulSIMD(pixscale,PowSIMD(intens.x,IGAMMA)));
g=MinSIMD(pixscale,MulSIMD(pixscale,PowSIMD(intens.y,IGAMMA)));
b=MinSIMD(pixscale,MulSIMD(pixscale,PowSIMD(intens.z,IGAMMA)));
// now, convert to integer
r=AndSIMD( AddSIMD( r, Four_MagicNumbers ), PIXMASK );
g=AndSIMD( AddSIMD( g, Four_MagicNumbers ), PIXMASK );
b=AndSIMD( AddSIMD( b, Four_MagicNumbers ), PIXMASK );
*(p1)=(SubInt(r, 0))|(SubInt(g, 0)<<8)|(SubInt(b, 0)<<16);
*(p2)=(SubInt(r, 1))|(SubInt(g, 1)<<8)|(SubInt(b, 1)<<16);
*(p3)=(SubInt(r, 2))|(SubInt(g, 2)<<8)|(SubInt(b, 2)<<16);
*(p4)=(SubInt(r, 3))|(SubInt(g, 3)<<8)|(SubInt(b, 3)<<16);
}
void LightDesc_t::ComputeLightAtPoints( const FourVectors &pos, const FourVectors &normal,
FourVectors &color, bool DoHalfLambert ) const
{
FourVectors delta;
Assert((m_Type==MATERIAL_LIGHT_POINT) || (m_Type==MATERIAL_LIGHT_SPOT) || (m_Type==MATERIAL_LIGHT_DIRECTIONAL));
switch (m_Type)
{
case MATERIAL_LIGHT_POINT:
case MATERIAL_LIGHT_SPOT:
delta.DuplicateVector(m_Position);
delta-=pos;
break;
case MATERIAL_LIGHT_DIRECTIONAL:
ComputeLightAtPointsForDirectional( pos, normal, color, DoHalfLambert );
return;
default:
return;
}
fltx4 dist2 = delta*delta;
dist2=MaxSIMD( Four_Ones, dist2 );
fltx4 falloff;
if( m_Flags & LIGHTTYPE_OPTIMIZATIONFLAGS_HAS_ATTENUATION0 )
{
falloff = ReplicateX4(m_Attenuation0);
}
else
falloff= Four_Epsilons;
if( m_Flags & LIGHTTYPE_OPTIMIZATIONFLAGS_HAS_ATTENUATION1 )
{
falloff=AddSIMD(falloff,MulSIMD(ReplicateX4(m_Attenuation1),SqrtEstSIMD(dist2)));
}
if( m_Flags & LIGHTTYPE_OPTIMIZATIONFLAGS_HAS_ATTENUATION2 )
{
falloff=AddSIMD(falloff,MulSIMD(ReplicateX4(m_Attenuation2),dist2));
}
falloff=ReciprocalEstSIMD(falloff);
// Cull out light beyond this radius
// now, zero out elements for which dist2 was > range^2. !!speed!! lights should store dist^2 in sse format
if (m_Range != 0.f)
{
fltx4 RangeSquared=ReplicateX4(m_RangeSquared); // !!speed!!
falloff=AndSIMD(falloff,CmpLtSIMD(dist2,RangeSquared));
}
delta.VectorNormalizeFast();
fltx4 strength=delta*normal;
if (DoHalfLambert)
{
strength=AddSIMD(MulSIMD(strength,Four_PointFives),Four_PointFives);
}
else
strength=MaxSIMD(Four_Zeros,delta*normal);
switch(m_Type)
{
case MATERIAL_LIGHT_POINT:
// half-lambert
break;
case MATERIAL_LIGHT_SPOT:
{
fltx4 dot2=SubSIMD(Four_Zeros,delta*m_Direction); // dot position with spot light dir for cone falloff
fltx4 cone_falloff_scale=MulSIMD(ReplicateX4(m_OneOverThetaDotMinusPhiDot),
SubSIMD(dot2,ReplicateX4(m_PhiDot)));
cone_falloff_scale=MinSIMD(cone_falloff_scale,Four_Ones);
if ((m_Falloff!=0.0) && (m_Falloff!=1.0))
{
// !!speed!! could compute integer exponent needed by powsimd and store in light
cone_falloff_scale=PowSIMD(cone_falloff_scale,m_Falloff);
}
strength=MulSIMD(cone_falloff_scale,strength);
// now, zero out lighting where dot2<phidot. This will mask out any invalid results
// from pow function, etc
fltx4 OutsideMask=CmpGtSIMD(dot2,ReplicateX4(m_PhiDot)); // outside light cone?
strength=AndSIMD(OutsideMask,strength);
}
break;
default:
break;
}
strength=MulSIMD(strength,falloff);
color.x=AddSIMD(color.x,MulSIMD(strength,ReplicateX4(m_Color.x)));
color.y=AddSIMD(color.y,MulSIMD(strength,ReplicateX4(m_Color.y)));
color.z=AddSIMD(color.z,MulSIMD(strength,ReplicateX4(m_Color.z)));
}
void RayTracingEnvironment::Trace4Rays(const FourRays &rays, fltx4 TMin, fltx4 TMax,
int DirectionSignMask, RayTracingResult *rslt_out,
int32 skip_id, ITransparentTriangleCallback *pCallback)
{
rays.Check();
memset(rslt_out->HitIds,0xff,sizeof(rslt_out->HitIds));
rslt_out->HitDistance=ReplicateX4(1.0e23);
rslt_out->surface_normal.DuplicateVector(Vector(0.,0.,0.));
FourVectors OneOverRayDir=rays.direction;
OneOverRayDir.MakeReciprocalSaturate();
// now, clip rays against bounding box
for(int c=0;c<3;c++)
{
fltx4 isect_min_t=
MulSIMD(SubSIMD(ReplicateX4(m_MinBound[c]),rays.origin[c]),OneOverRayDir[c]);
fltx4 isect_max_t=
MulSIMD(SubSIMD(ReplicateX4(m_MaxBound[c]),rays.origin[c]),OneOverRayDir[c]);
TMin=MaxSIMD(TMin,MinSIMD(isect_min_t,isect_max_t));
TMax=MinSIMD(TMax,MaxSIMD(isect_min_t,isect_max_t));
}
fltx4 active=CmpLeSIMD(TMin,TMax); // mask of which rays are active
if (! IsAnyNegative(active) )
return; // missed bounding box
int32 mailboxids[MAILBOX_HASH_SIZE]; // used to avoid redundant triangle tests
memset(mailboxids,0xff,sizeof(mailboxids)); // !!speed!! keep around?
int front_idx[3],back_idx[3]; // based on ray direction, whether to
// visit left or right node first
if (DirectionSignMask & 1)
{
back_idx[0]=0;
front_idx[0]=1;
}
else
{
back_idx[0]=1;
front_idx[0]=0;
}
if (DirectionSignMask & 2)
{
back_idx[1]=0;
front_idx[1]=1;
}
else
{
back_idx[1]=1;
front_idx[1]=0;
}
if (DirectionSignMask & 4)
{
back_idx[2]=0;
front_idx[2]=1;
}
else
{
back_idx[2]=1;
front_idx[2]=0;
}
NodeToVisit NodeQueue[MAX_NODE_STACK_LEN];
CacheOptimizedKDNode const *CurNode=&(OptimizedKDTree[0]);
NodeToVisit *stack_ptr=&NodeQueue[MAX_NODE_STACK_LEN];
while(1)
{
while (CurNode->NodeType() != KDNODE_STATE_LEAF) // traverse until next leaf
{
int split_plane_number=CurNode->NodeType();
CacheOptimizedKDNode const *FrontChild=&(OptimizedKDTree[CurNode->LeftChild()]);
fltx4 dist_to_sep_plane= // dist=(split-org)/dir
MulSIMD(
SubSIMD(ReplicateX4(CurNode->SplittingPlaneValue),
rays.origin[split_plane_number]),OneOverRayDir[split_plane_number]);
fltx4 active=CmpLeSIMD(TMin,TMax); // mask of which rays are active
// now, decide how to traverse children. can either do front,back, or do front and push
// back.
fltx4 hits_front=AndSIMD(active,CmpGeSIMD(dist_to_sep_plane,TMin));
if (! IsAnyNegative(hits_front))
{
// missed the front. only traverse back
//printf("only visit back %d\n",CurNode->LeftChild()+back_idx[split_plane_number]);
CurNode=FrontChild+back_idx[split_plane_number];
TMin=MaxSIMD(TMin, dist_to_sep_plane);
}
else
{
fltx4 hits_back=AndSIMD(active,CmpLeSIMD(dist_to_sep_plane,TMax));
if (! IsAnyNegative(hits_back) )
{
// missed the back - only need to traverse front node
//printf("only visit front %d\n",CurNode->LeftChild()+front_idx[split_plane_number]);
CurNode=FrontChild+front_idx[split_plane_number];
TMax=MinSIMD(TMax, dist_to_sep_plane);
}
else
{
// at least some rays hit both nodes.
// must push far, traverse near
//printf("visit %d,%d\n",CurNode->LeftChild()+front_idx[split_plane_number],
// CurNode->LeftChild()+back_idx[split_plane_number]);
assert(stack_ptr>NodeQueue);
--stack_ptr;
stack_ptr->node=FrontChild+back_idx[split_plane_number];
stack_ptr->TMin=MaxSIMD(TMin,dist_to_sep_plane);
stack_ptr->TMax=TMax;
CurNode=FrontChild+front_idx[split_plane_number];
TMax=MinSIMD(TMax,dist_to_sep_plane);
}
}
}
// hit a leaf! must do intersection check
int ntris=CurNode->NumberOfTrianglesInLeaf();
if (ntris)
{
int32 const *tlist=&(TriangleIndexList[CurNode->TriangleIndexStart()]);
do
{
int tnum=*(tlist++);
//printf("try tri %d\n",tnum);
// check mailbox
int mbox_slot=tnum & (MAILBOX_HASH_SIZE-1);
TriIntersectData_t const *tri = &( OptimizedTriangleList[tnum].m_Data.m_IntersectData );
if ( ( mailboxids[mbox_slot] != tnum ) && ( tri->m_nTriangleID != skip_id ) )
{
n_intersection_calculations++;
mailboxids[mbox_slot] = tnum;
// compute plane intersection
FourVectors N;
N.x = ReplicateX4( tri->m_flNx );
N.y = ReplicateX4( tri->m_flNy );
N.z = ReplicateX4( tri->m_flNz );
fltx4 DDotN = rays.direction * N;
// mask off zero or near zero (ray parallel to surface)
fltx4 did_hit = OrSIMD( CmpGtSIMD( DDotN,FourEpsilons ),
CmpLtSIMD( DDotN, FourNegativeEpsilons ) );
fltx4 numerator=SubSIMD( ReplicateX4( tri->m_flD ), rays.origin * N );
fltx4 isect_t=DivSIMD( numerator,DDotN );
// now, we have the distance to the plane. lets update our mask
did_hit = AndSIMD( did_hit, CmpGtSIMD( isect_t, FourZeros ) );
//did_hit=AndSIMD(did_hit,CmpLtSIMD(isect_t,TMax));
did_hit = AndSIMD( did_hit, CmpLtSIMD( isect_t, rslt_out->HitDistance ) );
if ( ! IsAnyNegative( did_hit ) )
continue;
// now, check 3 edges
fltx4 hitc1 = AddSIMD( rays.origin[tri->m_nCoordSelect0],
MulSIMD( isect_t, rays.direction[ tri->m_nCoordSelect0] ) );
fltx4 hitc2 = AddSIMD( rays.origin[tri->m_nCoordSelect1],
MulSIMD( isect_t, rays.direction[tri->m_nCoordSelect1] ) );
// do barycentric coordinate check
fltx4 B0 = MulSIMD( ReplicateX4( tri->m_ProjectedEdgeEquations[0] ), hitc1 );
B0 = AddSIMD(
B0,
MulSIMD( ReplicateX4( tri->m_ProjectedEdgeEquations[1] ), hitc2 ) );
B0 = AddSIMD(
B0, ReplicateX4( tri->m_ProjectedEdgeEquations[2] ) );
did_hit = AndSIMD( did_hit, CmpGeSIMD( B0, FourZeros ) );
fltx4 B1 = MulSIMD( ReplicateX4( tri->m_ProjectedEdgeEquations[3] ), hitc1 );
B1 = AddSIMD(
B1,
MulSIMD( ReplicateX4( tri->m_ProjectedEdgeEquations[4]), hitc2 ) );
B1 = AddSIMD(
B1, ReplicateX4( tri->m_ProjectedEdgeEquations[5] ) );
did_hit = AndSIMD( did_hit, CmpGeSIMD( B1, FourZeros ) );
fltx4 B2 = AddSIMD( B1, B0 );
did_hit = AndSIMD( did_hit, CmpLeSIMD( B2, Four_Ones ) );
if ( ! IsAnyNegative( did_hit ) )
continue;
// if the triangle is transparent
if ( tri->m_nFlags & FCACHETRI_TRANSPARENT )
{
if ( pCallback )
{
// assuming a triangle indexed as v0, v1, v2
// the projected edge equations are set up such that the vert opposite the first
// equation is v2, and the vert opposite the second equation is v0
// Therefore we pass them back in 1, 2, 0 order
// Also B2 is currently B1 + B0 and needs to be 1 - (B1+B0) in order to be a real
// barycentric coordinate. Compute that now and pass it to the callback
fltx4 b2 = SubSIMD( Four_Ones, B2 );
if ( pCallback->VisitTriangle_ShouldContinue( *tri, rays, &did_hit, &B1, &b2, &B0, tnum ) )
{
did_hit = Four_Zeros;
}
}
}
// now, set the hit_id and closest_hit fields for any enabled rays
fltx4 replicated_n = ReplicateIX4(tnum);
StoreAlignedSIMD((float *) rslt_out->HitIds,
OrSIMD(AndSIMD(replicated_n,did_hit),
AndNotSIMD(did_hit,LoadAlignedSIMD(
(float *) rslt_out->HitIds))));
rslt_out->HitDistance=OrSIMD(AndSIMD(isect_t,did_hit),
AndNotSIMD(did_hit,rslt_out->HitDistance));
rslt_out->surface_normal.x=OrSIMD(
AndSIMD(N.x,did_hit),
AndNotSIMD(did_hit,rslt_out->surface_normal.x));
rslt_out->surface_normal.y=OrSIMD(
AndSIMD(N.y,did_hit),
AndNotSIMD(did_hit,rslt_out->surface_normal.y));
rslt_out->surface_normal.z=OrSIMD(
AndSIMD(N.z,did_hit),
AndNotSIMD(did_hit,rslt_out->surface_normal.z));
}
} while (--ntris);
// now, check if all rays have terminated
fltx4 raydone=CmpLeSIMD(TMax,rslt_out->HitDistance);
if (! IsAnyNegative(raydone))
{
return;
}
}
if (stack_ptr==&NodeQueue[MAX_NODE_STACK_LEN])
{
return;
}
// pop stack!
CurNode=stack_ptr->node;
TMin=stack_ptr->TMin;
TMax=stack_ptr->TMax;
stack_ptr++;
}
}