github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-06-28 16:41:01 +03:00
Code Issues Packages Projects Releases Wiki Activity

blockSigleMask implementations for ARCH_PPC64 added

Browse Source
This commit is contained in:
apostolos 2021-10-14 15:56:13 +03:00
parent 4d2acd59e2
commit d0a41252c8
3 changed files with 30 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/nfa/shufti_simd.hpp
View File

@ -56,6 +56,8 @@ SuperVector<S> blockDoubleMask(SuperVector<S> mask1_lo, SuperVector<S> mask1_hi,
#include "x86/shufti.hpp" #include "x86/shufti.hpp"
#elif defined(ARCH_ARM32) || defined(ARCH_AARCH64) #elif defined(ARCH_ARM32) || defined(ARCH_AARCH64)
#include "arm/shufti.hpp" #include "arm/shufti.hpp"
#elif defined(ARCH_PPC64EL)
#include "ppc64el/shufti.hpp"
#endif #endif
template <uint16_t S> template <uint16_t S>

2
src/nfa/truffle_simd.hpp
View File

@ -49,6 +49,8 @@ const SuperVector<S> blockSingleMask(SuperVector<S> shuf_mask_lo_highclear, Supe
#include "x86/truffle.hpp" #include "x86/truffle.hpp"
#elif defined(ARCH_ARM32) || defined(ARCH_AARCH64) #elif defined(ARCH_ARM32) || defined(ARCH_AARCH64)
#include "arm/truffle.hpp" #include "arm/truffle.hpp"
#elif defined(ARCH_PPC64EL)
#include "ppc64el/truffle.hpp"
#endif #endif
template <uint16_t S> template <uint16_t S>

37
src/util/supervector/arch/ppc64el/impl.cpp
View File

@ -444,7 +444,7 @@ really_inline SuperVector<16> SuperVector<16>::vshl(uint8_t const N) const
} }
template <> template <>
really_inline SuperVector<16> SuperVector<16>::vshr_8 (uint8_t const N) const really_inline SuperVector<16> SuperVector<16>::vshr_8 (uint8_t const UNUSED N) const
{ {
//if (N == 0) return *this; //if (N == 0) return *this;
//if (N == 16) return Zeroes(); //if (N == 16) return Zeroes();
@ -456,7 +456,7 @@ really_inline SuperVector<16> SuperVector<16>::vshr_8 (uint8_t const N) const
} }
template <> template <>
really_inline SuperVector<16> SuperVector<16>::vshr_16 (uint8_t const N) const really_inline SuperVector<16> SuperVector<16>::vshr_16 (uint8_t const UNUSED N) const
{ {
//if (N == 0) return *this; //if (N == 0) return *this;
//if (N == 16) return Zeroes(); //if (N == 16) return Zeroes();
@ -468,7 +468,7 @@ really_inline SuperVector<16> SuperVector<16>::vshr_16 (uint8_t const N) const
} }
template <> template <>
really_inline SuperVector<16> SuperVector<16>::vshr_32 (uint8_t const N) const really_inline SuperVector<16> SuperVector<16>::vshr_32 (uint8_t const UNUSED N) const
{ {
//if (N == 0) return *this; //if (N == 0) return *this;
//if (N == 16) return Zeroes(); //if (N == 16) return Zeroes();
@ -480,7 +480,7 @@ really_inline SuperVector<16> SuperVector<16>::vshr_32 (uint8_t const N) const
} }
template <> template <>
really_inline SuperVector<16> SuperVector<16>::vshr_64 (uint8_t const N) const really_inline SuperVector<16> SuperVector<16>::vshr_64 (uint8_t const UNUSED N) const
{ {
//if (N == 0) return *this; //if (N == 0) return *this;
//if (N == 16) return Zeroes(); //if (N == 16) return Zeroes();
@ -492,7 +492,7 @@ really_inline SuperVector<16> SuperVector<16>::vshr_64 (uint8_t const N) const
} }
template <> template <>
really_inline SuperVector<16> SuperVector<16>::vshr_128(uint8_t const N) const really_inline SuperVector<16> SuperVector<16>::vshr_128(uint8_t const UNUSED N) const
{ {
//if (N == 0) return *this; //if (N == 0) return *this;
//if (N == 16) return Zeroes(); //if (N == 16) return Zeroes();
@ -594,12 +594,6 @@ really_inline SuperVector<16> SuperVector<16>::loadu_maskz(void const *ptr, uint
return mask & v; return mask & v;
} }
template<>
really_inline SuperVector<16> SuperVector<16>::pshufb(SuperVector<16> b)
{
return (m128) vec_permxor((int8x16_t)vec_splat_s8(0), (int8x16_t)u.v128[0], (int8x16_t) b.u.v128[0]);
}
template<> template<>
really_inline SuperVector<16> SuperVector<16>::alignr(SuperVector<16> &other, int8_t offset) really_inline SuperVector<16> SuperVector<16>::alignr(SuperVector<16> &other, int8_t offset)
{ {
@ -626,6 +620,24 @@ really_inline SuperVector<16> SuperVector<16>::alignr(SuperVector<16> &other, in
return *this; return *this;
} }
template<>
template<>
really_inline SuperVector<16> SuperVector<16>::pshufb<false>(SuperVector<16> b)
{
return (m128) vec_permxor((int8x16_t)vec_splat_s8(0), (int8x16_t)u.v128[0], (int8x16_t) b.u.v128[0]);
}
template<>
template<>
really_inline SuperVector<16> SuperVector<16>::pshufb<true>(SuperVector<16> b)
{
/* On Intel, if bit 0x80 is set, then result is zero, otherwise which the lane it is &0xf.
In NEON, if >=16, then the result is zero, otherwise it is that lane.
btranslated is the version that is converted from Intel to NEON. */
SuperVector<16> btranslated = b & SuperVector<16>::dup_s8(0x8f);
return pshufb<false>(btranslated);
}
template<> template<>
really_inline SuperVector<16> SuperVector<16>::pshufb_maskz(SuperVector<16> b, uint8_t const len) really_inline SuperVector<16> SuperVector<16>::pshufb_maskz(SuperVector<16> b, uint8_t const len)
@ -635,6 +647,8 @@ really_inline SuperVector<16> SuperVector<16>::pshufb_maskz(SuperVector<16> b, u
} }
/*
template<> template<>
really_inline SuperVector<16> SuperVector<16>::lshift64(uint8_t const N) really_inline SuperVector<16> SuperVector<16>::lshift64(uint8_t const N)
{ {
@ -661,4 +675,5 @@ really_inline SuperVector<16> SuperVector<16>::rshift128(uint8_t const N)
{ {
return *this >> N; return *this >> N;
} }
*/
#endif #endif