devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merged latest updates from the Eigen trunk.

Browse Source
This commit is contained in:
Benoit Steiner
2014-09-15 09:18:16 -07:00
parent efdff15749 9452eb38f8
commit 10a79ca3a3
66 changed files with 882 additions and 1343 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -143,7 +143,7 @@ template<> EIGEN_STRONG_INLINE Packet8f pmadd(const Packet8f& a, const Packet8f&
// so let's enforce it to generate a vfmadd231ps instruction since the most common use case is to accumulate
// the result of the product.
Packet8f res = c;
asm("vfmadd231ps %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));
__asm__("vfmadd231ps %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));
return res;
#else
return _mm256_fmadd_ps(a,b,c);
@@ -153,7 +153,7 @@ template<> EIGEN_STRONG_INLINE Packet4d pmadd(const Packet4d& a, const Packet4d&
#if defined(__clang__) || defined(__GNUC__)
// see above
Packet4d res = c;
asm("vfmadd231pd %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));
__asm__("vfmadd231pd %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));
return res;
#else
return _mm256_fmadd_pd(a,b,c);

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

@@ -52,12 +52,12 @@ typedef uint32x4_t Packet4ui;
// 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 (defined(__has_builtin) && __has_builtin(__builtin_prefetch)) || defined(__GNUC__)
#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" );
#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)