diff options
Diffstat (limited to 'Eigen/src/Core/arch/NEON/PacketMath.h')
| -rw-r--r-- | Eigen/src/Core/arch/NEON/PacketMath.h | 15 |
1 files changed, 9 insertions, 6 deletions
diff --git a/Eigen/src/Core/arch/NEON/PacketMath.h b/Eigen/src/Core/arch/NEON/PacketMath.h index 84a56bdcc..836fbc0dd 100644 --- a/Eigen/src/Core/arch/NEON/PacketMath.h +++ b/Eigen/src/Core/arch/NEON/PacketMath.h @@ -51,14 +51,17 @@ typedef uint32x4_t Packet4ui; #define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \ const Packet4i p4i_##NAME = pset1<Packet4i>(X) -// arm64 does have the pld instruction. If available, let's trust the __builtin_prefetch built-in function -// which available on LLVM and GCC (at least) -#if EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC +#if EIGEN_ARCH_ARM64 + // __builtin_prefetch tends to do nothing on ARM64 compilers because the + // prefetch instructions there are too detailed for __builtin_prefetch to map + // meaningfully to them. + #define EIGEN_ARM_PREFETCH(ADDR) __asm__ __volatile__("prfm pldl1keep, [%[addr]]\n" ::[addr] "r"(ADDR) : ); +#elif EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC #define EIGEN_ARM_PREFETCH(ADDR) __builtin_prefetch(ADDR); #elif defined __pld #define EIGEN_ARM_PREFETCH(ADDR) __pld(ADDR) -#elif !EIGEN_ARCH_ARM64 - #define EIGEN_ARM_PREFETCH(ADDR) __asm__ __volatile__ ( " pld [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" ); +#elif EIGEN_ARCH_ARM32 + #define EIGEN_ARM_PREFETCH(ADDR) __asm__ __volatile__ ("pld [%[addr]]\n" :: [addr] "r" (ADDR) : ); #else // by default no explicit prefetching #define EIGEN_ARM_PREFETCH(ADDR) @@ -113,7 +116,7 @@ template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int32_t& from) template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a) { - const float32_t f[] = {0, 1, 2, 3}; + const float f[] = {0, 1, 2, 3}; Packet4f countdown = vld1q_f32(f); return vaddq_f32(pset1<Packet4f>(a), countdown); } |