GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-06-04 18:54:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix the bug using neon instruction fmla for data type half

Browse Source
This commit is contained in:
Lianhuang Li 2022-12-01 17:28:57 +00:00 committed by Antonio Sánchez
parent 31ab62d347
commit d194167149
1 changed files with 6 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
Eigen/src/Core/arch/NEON/GeneralBlockPanelKernel.h
View File

@ -183,7 +183,11 @@ struct gebp_traits <double,double,false,false,Architecture::NEON>
} }
}; };
#if EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC // The register at operand 3 of fmla for data type half must be v0~v15, the compiler may not
// allocate a required register for the '%2' of inline asm 'fmla %0.8h, %1.8h, %2.h[id]',
// so inline assembly can't be used here to advoid the bug that vfmaq_lane_f16 is implemented
// through a costly dup in gcc compiler.
#if EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC && EIGEN_COMP_CLANG
template<> template<>
struct gebp_traits <half,half,false,false,Architecture::NEON> struct gebp_traits <half,half,false,false,Architecture::NEON>
@ -240,19 +244,10 @@ struct gebp_traits <half,half,false,false,Architecture::NEON>
template<int LaneID> template<int LaneID>
EIGEN_STRONG_INLINE void madd_helper(const LhsPacket& a, const RhsPacketx4& b, AccPacket& c) const EIGEN_STRONG_INLINE void madd_helper(const LhsPacket& a, const RhsPacketx4& b, AccPacket& c) const
{ {
#if EIGEN_COMP_GNUC_STRICT
// 1. vfmaq_lane_f16 is implemented through a costly dup
// 2. workaround the gcc register split problem on arm64-neon
if(LaneID==0) asm("fmla %0.8h, %1.8h, %2.h[0]\n" : "+w" (c) : "w" (a), "w" (b) : );
else if(LaneID==1) asm("fmla %0.8h, %1.8h, %2.h[1]\n" : "+w" (c) : "w" (a), "w" (b) : );
else if(LaneID==2) asm("fmla %0.8h, %1.8h, %2.h[2]\n" : "+w" (c) : "w" (a), "w" (b) : );
else if(LaneID==3) asm("fmla %0.8h, %1.8h, %2.h[3]\n" : "+w" (c) : "w" (a), "w" (b) : );
#else
c = vfmaq_lane_f16(c, a, b, LaneID); c = vfmaq_lane_f16(c, a, b, LaneID);
#endif
} }
}; };
#endif // EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC #endif // EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC && EIGEN_COMP_CLANG
#endif // EIGEN_ARCH_ARM64 #endif // EIGEN_ARCH_ARM64
} // namespace internal } // namespace internal