ForkMaster/netgen
1
0
Fork 0
mirror of https://github.com/NGSolve/netgen.git synced 2024-12-25 05:20:34 +05:00
Code Issues Packages Projects Releases Wiki Activity

Fix AVX-Operators for int64_t simd (use generic ones)

Browse Source
This commit is contained in:
Matthias Hochsteger 2020-12-15 09:37:56 +01:00
parent f213a7a5b1
commit 1f3aebcec0
1 changed files with 4 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
libsrc/core/simd_avx.hpp
View File

@ -76,7 +76,9 @@ namespace ngcore
SIMD (std::array<int64_t,4> a) SIMD (std::array<int64_t,4> a)
: data{_mm256_set_epi64x(a[3],a[2],a[1],a[0])} : data{_mm256_set_epi64x(a[3],a[2],a[1],a[0])}
{} {}
// SIMD (SIMD<double,2> v0, SIMD<double,2> v1) : SIMD(v0[0], v0[1], v1[0], v1[1]) { ; } SIMD (SIMD<int64_t,2> v0, SIMD<int64_t,2> v1)
: data(_mm256_set_m128i(v0.Data(),v1.Data()))
{}
SIMD (__m256i _data) { data = _data; } SIMD (__m256i _data) { data = _data; }
NETGEN_INLINE auto operator[] (int i) const { return ((int64_t*)(&data))[i]; } NETGEN_INLINE auto operator[] (int i) const { return ((int64_t*)(&data))[i]; }
@ -94,18 +96,7 @@ namespace ngcore
#ifdef __AVX2__ #ifdef __AVX2__
NETGEN_INLINE SIMD<int64_t,4> operator+ (SIMD<int64_t,4> a, SIMD<int64_t,4> b) { return _mm256_add_epi64(a.Data(),b.Data()); } NETGEN_INLINE SIMD<int64_t,4> operator+ (SIMD<int64_t,4> a, SIMD<int64_t,4> b) { return _mm256_add_epi64(a.Data(),b.Data()); }
NETGEN_INLINE SIMD<int64_t,4> operator- (SIMD<int64_t,4> a, SIMD<int64_t,4> b) { return _mm256_sub_epi64(a.Data(),b.Data()); } NETGEN_INLINE SIMD<int64_t,4> operator- (SIMD<int64_t,4> a, SIMD<int64_t,4> b) { return _mm256_sub_epi64(a.Data(),b.Data()); }
#else #endif // __AVX2__
NETGEN_INLINE SIMD<int64_t,4> operator+ (SIMD<int64_t,4> a, SIMD<int64_t,4> b) {
auto lo_sum = _mm256_extractf128_si256(a.Data(), 0) + _mm256_extractf128_si256(b.Data(), 0);
auto hi_sum = _mm256_extractf128_si256(a.Data(), 1) + _mm256_extractf128_si256(b.Data(), 1);
return _mm256_set_m128i(hi_sum,lo_sum);
}
NETGEN_INLINE SIMD<int64_t,4> operator- (SIMD<int64_t,4> a, SIMD<int64_t,4> b) {
auto lo_sub = _mm256_extractf128_si256(a.Data(), 0) - _mm256_extractf128_si256(b.Data(), 0);
auto hi_sub = _mm256_extractf128_si256(a.Data(), 1) - _mm256_extractf128_si256(b.Data(), 1);
return _mm256_set_m128i(hi_sub,lo_sub);
}
#endif
template<> template<>
class SIMD<double,4> class SIMD<double,4>