ForkMaster/netgen
1
0
Fork 0
mirror of https://github.com/NGSolve/netgen.git synced 2024-12-24 21:10:33 +05:00
Code Issues Packages Projects Releases Wiki Activity

fnma intrinsic for avx512

Browse Source
This commit is contained in:
Joachim Schöberl 2020-12-22 13:06:08 +01:00
parent c1c10174be
commit ea7f6c1e94
3 changed files with 28 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
libsrc/core/simd_avx.hpp
View File

@ -184,6 +184,14 @@ namespace ngcore
{ {
return _mm256_fmadd_pd (_mm256_set1_pd(a), b.Data(), c.Data()); return _mm256_fmadd_pd (_mm256_set1_pd(a), b.Data(), c.Data());
} }
NETGEN_INLINE SIMD<double,4> FNMA (SIMD<double,4> a, SIMD<double,4> b, SIMD<double,4> c)
{
return _mm256_fnmadd_pd (a.Data(), b.Data(), c.Data());
}
NETGEN_INLINE SIMD<double,4> FNMA (const double & a, SIMD<double,4> b, SIMD<double,4> c)
{
return _mm256_fnmadd_pd (_mm256_set1_pd(a), b.Data(), c.Data());
}
#endif #endif
#if defined(__FMA__) && !defined(__AVX512F__) #if defined(__FMA__) && !defined(__AVX512F__)

10
libsrc/core/simd_avx512.hpp
View File

@ -234,6 +234,16 @@ namespace ngcore
{ {
return _mm512_fmadd_pd (_mm512_set1_pd(a), b.Data(), c.Data()); return _mm512_fmadd_pd (_mm512_set1_pd(a), b.Data(), c.Data());
} }
NETGEN_INLINE SIMD<double,8> FNMA (SIMD<double,8> a, SIMD<double,8> b, SIMD<double,8> c)
{
return _mm512_fnmadd_pd (a.Data(), b.Data(), c.Data());
}
NETGEN_INLINE SIMD<double,8> FNMA (const double & a, SIMD<double,8> b, SIMD<double,8> c)
{
return _mm512_fnmadd_pd (_mm512_set1_pd(a), b.Data(), c.Data());
}
} }
#endif // NETGEN_CORE_SIMD_AVX512_HPP #endif // NETGEN_CORE_SIMD_AVX512_HPP

13
libsrc/core/simd_generic.hpp
View File

@ -513,11 +513,17 @@ namespace ngcore
} }
template <typename T1, typename T2, typename T3>
// a*b+c // a*b+c
template <typename T1, typename T2, typename T3>
NETGEN_INLINE auto FMA(T1 a, T2 b, T3 c) NETGEN_INLINE auto FMA(T1 a, T2 b, T3 c)
{ {
return a*b+c; return c+a*b;
}
template <typename T1, typename T2, typename T3>
NETGEN_INLINE auto FNMA(T1 a, T2 b, T3 c)
{
return c-a*b;
} }
// update form of fma // update form of fma
@ -531,7 +537,8 @@ namespace ngcore
template <int N> template <int N>
void FNMAasm (SIMD<double,N> a, SIMD<double,N> b, SIMD<double,N> & sum) void FNMAasm (SIMD<double,N> a, SIMD<double,N> b, SIMD<double,N> & sum)
{ {
sum -= a*b; // sum -= a*b;
sum = FNMA(a,b,sum);
} }