mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-21 07:19:46 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix asm comments in 1px1 kernel

Browse Source
This commit is contained in:
Benoit Jacob 2015-03-03 13:44:00 -05:00
parent 5d2fd64a1a
commit 2aa09e6b4e
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
Eigen/src/Core/products/GeneralBlockPanelKernel.h
View File

@ -1481,17 +1481,17 @@ void gebp_kernel<LhsScalar,RhsScalar,Index,DataMapper,mr,nr,ConjugateLhs,Conjuga
for(Index k=0; k<peeled_kc; k+=pk) for(Index k=0; k<peeled_kc; k+=pk)
{ {
EIGEN_ASM_COMMENT("begin gebp micro kernel 2pX1"); EIGEN_ASM_COMMENT("begin gebp micro kernel 1pX1");
RhsPacket B_0; RhsPacket B_0;
#define EIGEN_GEBGP_ONESTEP(K) \ #define EIGEN_GEBGP_ONESTEP(K) \
do { \ do { \
EIGEN_ASM_COMMENT("begin step of gebp micro kernel 2pX1"); \ EIGEN_ASM_COMMENT("begin step of gebp micro kernel 1pX1"); \
EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \ EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
traits.loadLhs(&blA[(0+1*K)*LhsProgress], A0); \ traits.loadLhs(&blA[(0+1*K)*LhsProgress], A0); \
traits.loadRhs(&blB[(0+K)*RhsProgress], B_0); \ traits.loadRhs(&blB[(0+K)*RhsProgress], B_0); \
traits.madd(A0, B_0, C0, B_0); \ traits.madd(A0, B_0, C0, B_0); \
EIGEN_ASM_COMMENT("end step of gebp micro kernel 2pX1"); \ EIGEN_ASM_COMMENT("end step of gebp micro kernel 1pX1"); \
} while(false); } while(false);
EIGEN_GEBGP_ONESTEP(0); EIGEN_GEBGP_ONESTEP(0);
@ -1506,7 +1506,7 @@ void gebp_kernel<LhsScalar,RhsScalar,Index,DataMapper,mr,nr,ConjugateLhs,Conjuga
blB += pk*RhsProgress; blB += pk*RhsProgress;
blA += pk*1*Traits::LhsProgress; blA += pk*1*Traits::LhsProgress;
EIGEN_ASM_COMMENT("end gebp micro kernel 2pX1"); EIGEN_ASM_COMMENT("end gebp micro kernel 1pX1");
} }
// process remaining peeled loop // process remaining peeled loop