GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-14 20:56:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

backporting warning fixes in cache friendly product

Browse Source
This commit is contained in:
Gael Guennebaud 2009-05-19 02:20:04 +02:00
parent 57934b9c30
commit 1c4b4e136b
1 changed files with 29 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

50
Eigen/src/Core/CacheFriendlyProduct.h
View File

@ -361,13 +361,14 @@ static EIGEN_DONT_INLINE void ei_cache_friendly_product_colmajor_times_vector(
#ifdef _EIGEN_ACCUMULATE_PACKETS
#error _EIGEN_ACCUMULATE_PACKETS has already been defined
#endif
#define _EIGEN_ACCUMULATE_PACKETS(A0,A13,A2,OFFSET) \
ei_pstore(&res[j OFFSET], \
ei_padd(ei_pload(&res[j OFFSET]), \
#define _EIGEN_ACCUMULATE_PACKETS(A0,A13,A2) \
ei_pstore(&res[j], \
ei_padd(ei_pload(&res[j]), \
ei_padd( \
ei_padd(ei_pmul(ptmp0,ei_pload ## A0(&lhs0[j OFFSET])),ei_pmul(ptmp1,ei_pload ## A13(&lhs1[j OFFSET]))), \
ei_padd(ei_pmul(ptmp2,ei_pload ## A2(&lhs2[j OFFSET])),ei_pmul(ptmp3,ei_pload ## A13(&lhs3[j OFFSET]))) )))
ei_padd(ei_pmul(ptmp0,EIGEN_CAT(ei_ploa , A0)(&lhs0[j])), \
ei_pmul(ptmp1,EIGEN_CAT(ei_ploa , A13)(&lhs1[j]))), \
ei_padd(ei_pmul(ptmp2,EIGEN_CAT(ei_ploa , A2)(&lhs2[j])), \
ei_pmul(ptmp3,EIGEN_CAT(ei_ploa , A13)(&lhs3[j]))) )))
typedef typename ei_packet_traits<Scalar>::type Packet;
const int PacketSize = sizeof(Packet)/sizeof(Scalar);
@ -433,13 +434,8 @@ static EIGEN_DONT_INLINE void ei_cache_friendly_product_colmajor_times_vector(
{
/* explicit vectorization */
// process initial unaligned coeffs
for (int j=0; j<alignedStart; ++j) {
Scalar s = ei_pfirst(ptmp0)*lhs0[j];
s += ei_pfirst(ptmp1)*lhs1[j];
s += ei_pfirst(ptmp2)*lhs2[j];
s += ei_pfirst(ptmp3)*lhs3[j];
res[j] += s;
}
for (int j=0; j<alignedStart; ++j)
res[j] += ei_pfirst(ptmp0)*lhs0[j] + ei_pfirst(ptmp1)*lhs1[j] + ei_pfirst(ptmp2)*lhs2[j] + ei_pfirst(ptmp3)*lhs3[j];
if (alignedSize>alignedStart)
{
@ -447,11 +443,11 @@ static EIGEN_DONT_INLINE void ei_cache_friendly_product_colmajor_times_vector(
{
case AllAligned:
for (int j = alignedStart; j<alignedSize; j+=PacketSize)
_EIGEN_ACCUMULATE_PACKETS(,,,);
_EIGEN_ACCUMULATE_PACKETS(d,d,d);
break;
case EvenAligned:
for (int j = alignedStart; j<alignedSize; j+=PacketSize)
_EIGEN_ACCUMULATE_PACKETS(,u,,);
_EIGEN_ACCUMULATE_PACKETS(d,du,d);
break;
case FirstAligned:
if(peels>1)
@ -487,11 +483,11 @@ static EIGEN_DONT_INLINE void ei_cache_friendly_product_colmajor_times_vector(
}
}
for (int j = peeledSize; j<alignedSize; j+=PacketSize)
_EIGEN_ACCUMULATE_PACKETS(,u,u,);
_EIGEN_ACCUMULATE_PACKETS(d,du,du);
break;
default:
for (int j = alignedStart; j<alignedSize; j+=PacketSize)
_EIGEN_ACCUMULATE_PACKETS(u,u,u,);
_EIGEN_ACCUMULATE_PACKETS(du,du,du);
break;
}
}
@ -499,7 +495,7 @@ static EIGEN_DONT_INLINE void ei_cache_friendly_product_colmajor_times_vector(
/* process remaining coeffs (or all if there is no explicit vectorization) */
for (int j=alignedSize; j<size; ++j)
res[j] += ei_pfirst(ptmp0)*lhs0[j] + ei_pfirst(ptmp1)*lhs1[j] + ei_pfirst(ptmp2)*lhs2[j] + ei_pfirst(ptmp3)*lhs3[j];
res[j] += ei_pfirst(ptmp0)*lhs0[j] + ei_pfirst(ptmp1)*lhs1[j] + ei_pfirst(ptmp2)*lhs2[j] + ei_pfirst(ptmp3)*lhs3[j];
}
// process remaining first and last columns (at most columnsAtOnce-1)
@ -555,12 +551,12 @@ static EIGEN_DONT_INLINE void ei_cache_friendly_product_rowmajor_times_vector(
#error _EIGEN_ACCUMULATE_PACKETS has already been defined
#endif
#define _EIGEN_ACCUMULATE_PACKETS(A0,A13,A2,OFFSET) {\
#define _EIGEN_ACCUMULATE_PACKETS(A0,A13,A2) {\
Packet b = ei_pload(&rhs[j]); \
ptmp0 = ei_pmadd(b, ei_pload##A0 (&lhs0[j]), ptmp0); \
ptmp1 = ei_pmadd(b, ei_pload##A13(&lhs1[j]), ptmp1); \
ptmp2 = ei_pmadd(b, ei_pload##A2 (&lhs2[j]), ptmp2); \
ptmp3 = ei_pmadd(b, ei_pload##A13(&lhs3[j]), ptmp3); }
ptmp0 = ei_pmadd(b, EIGEN_CAT(ei_ploa,A0) (&lhs0[j]), ptmp0); \
ptmp1 = ei_pmadd(b, EIGEN_CAT(ei_ploa,A13)(&lhs1[j]), ptmp1); \
ptmp2 = ei_pmadd(b, EIGEN_CAT(ei_ploa,A2) (&lhs2[j]), ptmp2); \
ptmp3 = ei_pmadd(b, EIGEN_CAT(ei_ploa,A13)(&lhs3[j]), ptmp3); }
typedef typename ei_packet_traits<Scalar>::type Packet;
const int PacketSize = sizeof(Packet)/sizeof(Scalar);
@ -641,11 +637,11 @@ static EIGEN_DONT_INLINE void ei_cache_friendly_product_rowmajor_times_vector(
{
case AllAligned:
for (int j = alignedStart; j<alignedSize; j+=PacketSize)
_EIGEN_ACCUMULATE_PACKETS(,,,);
_EIGEN_ACCUMULATE_PACKETS(d,d,d);
break;
case EvenAligned:
for (int j = alignedStart; j<alignedSize; j+=PacketSize)
_EIGEN_ACCUMULATE_PACKETS(,u,,);
_EIGEN_ACCUMULATE_PACKETS(d,du,d);
break;
case FirstAligned:
if (peels>1)
@ -684,11 +680,11 @@ static EIGEN_DONT_INLINE void ei_cache_friendly_product_rowmajor_times_vector(
}
}
for (int j = peeledSize; j<alignedSize; j+=PacketSize)
_EIGEN_ACCUMULATE_PACKETS(,u,u,);
_EIGEN_ACCUMULATE_PACKETS(d,du,du);
break;
default:
for (int j = alignedStart; j<alignedSize; j+=PacketSize)
_EIGEN_ACCUMULATE_PACKETS(u,u,u,);
_EIGEN_ACCUMULATE_PACKETS(du,du,du);
break;
}
tmp0 += ei_predux(ptmp0);