mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-03-07 18:27:40 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix offset argument of ploadu/pstoreu for Altivec

Browse Source 
If no offset is given, them it should be zero.

Also passes full address to vec_vsx_ld/st builtins.

Removes userless _EIGEN_ALIGNED_PTR & _EIGEN_MASK_ALIGNMENT.

Removes unnecessary casts.
This commit is contained in:
João P. L. de Carvalho 2019-08-09 15:59:26 -06:00
parent 66d073c38e
commit 4d29aa0294
1 changed files with 6 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
Eigen/src/Core/arch/AltiVec/PacketMath.h
View File

@ -83,15 +83,6 @@ static Packet4i p4i_COUNTDOWN = { 0, 1, 2, 3 };
static Packet16uc p16uc_REVERSE32 = { 12,13,14,15, 8,9,10,11, 4,5,6,7, 0,1,2,3 };
static Packet16uc p16uc_DUPLICATE32_HI = { 0,1,2,3, 0,1,2,3, 4,5,6,7, 4,5,6,7 };
// Mask alignment
#ifdef __PPC64__
#define _EIGEN_MASK_ALIGNMENT 0xfffffffffffffff0
#else
#define _EIGEN_MASK_ALIGNMENT 0xfffffff0
#endif
#define _EIGEN_ALIGNED_PTR(x) ((std::ptrdiff_t)(x) & _EIGEN_MASK_ALIGNMENT)
// Handle endianness properly while loading constants
// Define global static constants:
#ifdef _BIG_ENDIAN
@ -487,12 +478,12 @@ template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)
template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)
{
EIGEN_DEBUG_UNALIGNED_LOAD
return (Packet4i) vec_vsx_ld((long)from & 15, (const int*) _EIGEN_ALIGNED_PTR(from));
return vec_vsx_ld(0, from);
}
template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from)
{
EIGEN_DEBUG_UNALIGNED_LOAD
return (Packet4f) vec_vsx_ld((long)from & 15, (const float*) _EIGEN_ALIGNED_PTR(from));
return vec_vsx_ld(0, from);
}
#endif
@ -553,12 +544,12 @@ template<> EIGEN_STRONG_INLINE void pstoreu<int>(int* to, const Packet4i& f
template<> EIGEN_STRONG_INLINE void pstoreu<int>(int* to, const Packet4i& from)
{
EIGEN_DEBUG_ALIGNED_STORE
vec_vsx_st(from, (long)to & 15, (int*) _EIGEN_ALIGNED_PTR(to));
vec_vsx_st(from, 0, to);
}
template<> EIGEN_STRONG_INLINE void pstoreu<float>(float* to, const Packet4f& from)
{
EIGEN_DEBUG_ALIGNED_STORE
vec_vsx_st(from, (long)to & 15, (float*) _EIGEN_ALIGNED_PTR(to));
vec_vsx_st(from, 0, to);
}
#endif
@ -1045,7 +1036,7 @@ template<> EIGEN_STRONG_INLINE Packet2d pfloor<Packet2d>(const Packet2d& a) { re
template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from)
{
EIGEN_DEBUG_ALIGNED_LOAD
return (Packet2d) vec_vsx_ld((long)from & 15, (const double*) _EIGEN_ALIGNED_PTR(from));
return vec_vsx_ld(0, from);
}
template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double* from)
@ -1059,7 +1050,7 @@ template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double* from)
template<> EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const Packet2d& from)
{
EIGEN_DEBUG_ALIGNED_STORE
vec_vsx_st((Packet4f)from, (long)to & 15, (float*) _EIGEN_ALIGNED_PTR(to));
vec_vsx_st(from, 0, to);
}
template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { EIGEN_PPC_PREFETCH(addr); }