GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-07-05 04:35:46 +08:00
Code Issues Packages Projects Releases Wiki Activity

Ensure EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC is always defined on arm.

Browse Source
This commit is contained in:
Antonio Sánchez 2023-06-26 19:21:54 +00:00 committed by Rasmus Munk Larsen
parent 7465b7651e
commit 31cd2ad371
1 changed files with 8 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
Eigen/src/Core/util/Macros.h
View File

@ -617,9 +617,11 @@
/// \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_ARM64 #if EIGEN_ARCH_ARM_OR_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) // Clang only supports FP16 on aarch64, and not all intrinsics are available
// on A32 anyways even in GCC (e.g. vdiv_f16, vsqrt_f16).
#if EIGEN_ARCH_ARM64 && 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
#else #else
#define EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC 0 #define EIGEN_HAS_ARM64_FP16_VECTOR_ARITHMETIC 0
@ -629,9 +631,11 @@
/// \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_ARM64 #if EIGEN_ARCH_ARM_OR_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) // Clang only supports FP16 on aarch64, and not all intrinsics are available
// on A32 anyways, even in GCC (e.g. vceqh_f16).
#if EIGEN_ARCH_ARM64 && 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
#endif #endif
#endif #endif