diff options
Diffstat (limited to 'extern/Eigen3/Eigen/src/Core/arch/NEON/PacketMath.h')
| -rw-r--r-- | extern/Eigen3/Eigen/src/Core/arch/NEON/PacketMath.h | 20 |
1 files changed, 15 insertions, 5 deletions
diff --git a/extern/Eigen3/Eigen/src/Core/arch/NEON/PacketMath.h b/extern/Eigen3/Eigen/src/Core/arch/NEON/PacketMath.h index 163bac215e6..d49670e0410 100644 --- a/extern/Eigen3/Eigen/src/Core/arch/NEON/PacketMath.h +++ b/extern/Eigen3/Eigen/src/Core/arch/NEON/PacketMath.h @@ -48,9 +48,18 @@ typedef uint32x4_t Packet4ui; #define EIGEN_INIT_NEON_PACKET2(X, Y) {X, Y} #define EIGEN_INIT_NEON_PACKET4(X, Y, Z, W) {X, Y, Z, W} #endif - -#ifndef __pld -#define __pld(x) asm volatile ( " pld [%[addr]]\n" :: [addr] "r" (x) : "cc" ); + +// 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) || defined(__GNUC__) + #define EIGEN_ARM_PREFETCH(ADDR) __builtin_prefetch(ADDR); +#elif defined __pld + #define EIGEN_ARM_PREFETCH(ADDR) __pld(ADDR) +#elif !defined(__aarch64__) + #define EIGEN_ARM_PREFETCH(ADDR) __asm__ __volatile__ ( " pld [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" ); +#else + // by default no explicit prefetching + #define EIGEN_ARM_PREFETCH(ADDR) #endif template<> struct packet_traits<float> : default_packet_traits @@ -209,8 +218,8 @@ template<> EIGEN_STRONG_INLINE void pstore<int>(int* to, const Packet4i& f template<> EIGEN_STRONG_INLINE void pstoreu<float>(float* to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_f32(to, from); } template<> EIGEN_STRONG_INLINE void pstoreu<int>(int* to, const Packet4i& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_s32(to, from); } -template<> EIGEN_STRONG_INLINE void prefetch<float>(const float* addr) { __pld(addr); } -template<> EIGEN_STRONG_INLINE void prefetch<int>(const int* addr) { __pld(addr); } +template<> EIGEN_STRONG_INLINE void prefetch<float>(const float* addr) { EIGEN_ARM_PREFETCH(addr); } +template<> EIGEN_STRONG_INLINE void prefetch<int>(const int* addr) { EIGEN_ARM_PREFETCH(addr); } // FIXME only store the 2 first elements ? template<> EIGEN_STRONG_INLINE float pfirst<Packet4f>(const Packet4f& a) { float EIGEN_ALIGN16 x[4]; vst1q_f32(x, a); return x[0]; } @@ -375,6 +384,7 @@ template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a) a_lo = vget_low_s32(a); a_hi = vget_high_s32(a); max = vpmax_s32(a_lo, a_hi); + max = vpmax_s32(max, max); return vget_lane_s32(max, 0); } |