GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-05-11 23:39:03 +08:00
Code Issues Packages Projects Releases Wiki Activity

ARM prefetch fixes: Implement prefetch on ARM64. Do not clobber cc on ARM32.

Browse Source
This commit is contained in:
Benoit Jacob 2017-03-15 06:57:25 -04:00
parent f0f3591118
commit 61160a21d2
1 changed files with 8 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
Eigen/src/Core/arch/NEON/PacketMath.h
View File

@ -51,14 +51,17 @@ typedef uint32x4_t Packet4ui;
#define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \ #define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \
const Packet4i p4i_##NAME = pset1<Packet4i>(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 #if EIGEN_ARCH_ARM64
// which available on LLVM and GCC (at least) // __builtin_prefetch tends to do nothing on ARM64 compilers because the
#if EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC // 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); #define EIGEN_ARM_PREFETCH(ADDR) __builtin_prefetch(ADDR);
#elif defined __pld #elif defined __pld
#define EIGEN_ARM_PREFETCH(ADDR) __pld(ADDR) #define EIGEN_ARM_PREFETCH(ADDR) __pld(ADDR)
#elif !EIGEN_ARCH_ARM64 #elif EIGEN_ARCH_ARM32
#define EIGEN_ARM_PREFETCH(ADDR) __asm__ __volatile__ ( " pld [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" ); #define EIGEN_ARM_PREFETCH(ADDR) __asm__ __volatile__ ("pld [%[addr]]\n" :: [addr] "r" (ADDR) : );
#else #else
// by default no explicit prefetching // by default no explicit prefetching
#define EIGEN_ARM_PREFETCH(ADDR) #define EIGEN_ARM_PREFETCH(ADDR)