GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-07-16 18:11:47 +08:00
Code Issues Packages Projects Releases Wiki Activity

Disable FP16 arithmetic for arm32.

Browse Source 
(cherry picked from commit 7465b7651edfb58322557179658853243eb96372)
This commit is contained in:
Antonio Sánchez 2023-06-26 18:39:42 +00:00 committed by Antonio Sanchez
parent ebfdd6bdea
commit e6e921f0e3
2 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
Eigen/src/Core/arch/NEON/PacketMath.h
View File

@ -3941,6 +3941,8 @@ template<> EIGEN_STRONG_INLINE Packet2d prsqrt(const Packet2d& a) {
template<> EIGEN_STRONG_INLINE Packet2d psqrt(const Packet2d& _x){ return vsqrtq_f64(_x); } template<> EIGEN_STRONG_INLINE Packet2d psqrt(const Packet2d& _x){ return vsqrtq_f64(_x); }
#endif // EIGEN_ARCH_ARM64 && !EIGEN_APPLE_DOUBLE_NEON_BUG
// Do we have an fp16 types and supporting Neon intrinsics? // Do we have an fp16 types and supporting Neon intrinsics?
#if EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC #if EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC
typedef float16x4_t Packet4hf; typedef float16x4_t Packet4hf;
@ -4601,8 +4603,6 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet8hf, 8>&
} }
#endif // end EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC #endif // end EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC
#endif // EIGEN_ARCH_ARM64
} // end namespace internal } // end namespace internal
} // end namespace Eigen } // end namespace Eigen

4
Eigen/src/Core/util/Macros.h
View File

@ -545,7 +545,7 @@
/// \internal EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC set to 1 if the architecture /// \internal EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC set to 1 if the architecture
/// supports Neon vector intrinsics for fp16. /// supports Neon vector intrinsics for fp16.
#if EIGEN_ARCH_ARM_OR_ARM64 #if EIGEN_ARCH_ARM64
#ifndef EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC #ifndef EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC
#if defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && !defined(EIGEN_GPU_COMPILE_PHASE) #if defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && !defined(EIGEN_GPU_COMPILE_PHASE)
#define EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC 1 #define EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC 1
@ -557,7 +557,7 @@
/// \internal EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC set to 1 if the architecture /// \internal EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC set to 1 if the architecture
/// supports Neon scalar intrinsics for fp16. /// supports Neon scalar intrinsics for fp16.
#if EIGEN_ARCH_ARM_OR_ARM64 #if EIGEN_ARCH_ARM64
#ifndef EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC #ifndef EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC
#if defined(__ARM_FEATURE_FP16_SCALAR_ARITHMETIC) && !defined(EIGEN_GPU_COMPILE_PHASE) #if defined(__ARM_FEATURE_FP16_SCALAR_ARITHMETIC) && !defined(EIGEN_GPU_COMPILE_PHASE)
#define EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC 1 #define EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC 1