GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-12 03:39:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

According to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89101, the previous GCC issue is fixed in GCC trunk (will be gcc 9).

Browse Source
This commit is contained in:
Gael Guennebaud 2019-01-30 11:52:28 +01:00
parent 3775926bba
commit df12fae8b8
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
Eigen/src/Core/products/GeneralBlockPanelKernel.h
View File

@ -1037,7 +1037,7 @@ struct gebp_traits <float, float, false, false,Architecture::NEON>
template<int LaneID> template<int LaneID>
EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacketx4& b, AccPacket& c, RhsPacket& /*tmp*/, const FixedInt<LaneID>&) const EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacketx4& b, AccPacket& c, RhsPacket& /*tmp*/, const FixedInt<LaneID>&) const
{ {
#if EIGEN_COMP_GNUC_STRICT #if EIGEN_COMP_GNUC_STRICT && !(EIGEN_GNUC_AT_LEAST(9,0))
// workaround gcc issue https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89101 // workaround gcc issue https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89101
// vfmaq_laneq_f32 is implemented through a costly dup // vfmaq_laneq_f32 is implemented through a costly dup
if(LaneID==0) asm("fmla %0.4s, %1.4s, %2.s[0]\n" : "+w" (c) : "w" (a), "w" (b) : ); if(LaneID==0) asm("fmla %0.4s, %1.4s, %2.s[0]\n" : "+w" (c) : "w" (a), "w" (b) : );
@ -1093,7 +1093,7 @@ struct gebp_traits <double, double, false, false,Architecture::NEON>
template<int LaneID> template<int LaneID>
EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacketx4& b, AccPacket& c, RhsPacket& /*tmp*/, const FixedInt<LaneID>&) const EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacketx4& b, AccPacket& c, RhsPacket& /*tmp*/, const FixedInt<LaneID>&) const
{ {
#if EIGEN_COMP_GNUC_STRICT #if EIGEN_COMP_GNUC_STRICT && !(EIGEN_GNUC_AT_LEAST(9,0))
// workaround gcc issue https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89101 // workaround gcc issue https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89101
// vfmaq_laneq_f64 is implemented through a costly dup // vfmaq_laneq_f64 is implemented through a costly dup
if(LaneID==0) asm("fmla %0.2d, %1.2d, %2.d[0]\n" : "+w" (c) : "w" (a), "w" (b.B_0) : ); if(LaneID==0) asm("fmla %0.2d, %1.2d, %2.d[0]\n" : "+w" (c) : "w" (a), "w" (b.B_0) : );
@ -1314,7 +1314,7 @@ void gebp_kernel<LhsScalar,RhsScalar,Index,DataMapper,mr,nr,ConjugateLhs,Conjuga
RhsPanel15 rhs_panel; RhsPanel15 rhs_panel;
RhsPacket T0; RhsPacket T0;
LhsPacket A2; LhsPacket A2;
#if EIGEN_COMP_GNUC_STRICT && EIGEN_ARCH_ARM64 && defined(EIGEN_VECTORIZE_NEON) #if EIGEN_COMP_GNUC_STRICT && EIGEN_ARCH_ARM64 && defined(EIGEN_VECTORIZE_NEON) && !(EIGEN_GNUC_AT_LEAST(9,0))
// see http://eigen.tuxfamily.org/bz/show_bug.cgi?id=1633 // see http://eigen.tuxfamily.org/bz/show_bug.cgi?id=1633
// without this workaround A0, A1, and A2 are loaded in the same register, // without this workaround A0, A1, and A2 are loaded in the same register,
// which is not good for pipelining // which is not good for pipelining