portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 07:17:06 +01:00
Code Releases Activity

Merge branch 'develop' into feature/feynman-rules

Browse Source 
# Conflicts:
#	lib/Threads.h
#	lib/qcd/action/fermion/WilsonFermion.cc
#	lib/qcd/action/fermion/WilsonFermion.h
#	lib/qcd/utils/SUn.h
#	lib/simd/Grid_avx.h
#	lib/simd/Intel512common.h
This commit is contained in:
Antonin Portelli
2016-10-19 18:35:18 +01:00
parent a123dcd7e9 7af9b87318
commit 997fd882ff
84 changed files with 6162 additions and 2851 deletions
Download Patch File Download Diff File Expand all files Collapse all files

62
lib/simd/Grid_avx512.h
View File

@ -370,7 +370,67 @@ namespace Optimization {
//////////////////////////////////////////////
// Some Template specialization
// Hack for CLANG until mm512_reduce_add_ps etc... are implemented in GCC and Clang releases
#undef GNU_CLANG_COMPILER
#ifdef GNU_CLANG_COMPILER
//Complex float Reduce
template<>
inline Grid::ComplexF Reduce<Grid::ComplexF, __m512>::operator()(__m512 in){
__m512 v1,v2;
v1=Optimization::Permute::Permute0(in); // avx 512; quad complex single
v1= _mm512_add_ps(v1,in);
v2=Optimization::Permute::Permute1(v1);
v1 = _mm512_add_ps(v1,v2);
v2=Optimization::Permute::Permute2(v1);
v1 = _mm512_add_ps(v1,v2);
u512f conv; conv.v = v1;
return Grid::ComplexF(conv.f[0],conv.f[1]);
}
//Real float Reduce
template<>
inline Grid::RealF Reduce<Grid::RealF, __m512>::operator()(__m512 in){
__m512 v1,v2;
v1 = Optimization::Permute::Permute0(in); // avx 512; octo-double
v1 = _mm512_add_ps(v1,in);
v2 = Optimization::Permute::Permute1(v1);
v1 = _mm512_add_ps(v1,v2);
v2 = Optimization::Permute::Permute2(v1);
v1 = _mm512_add_ps(v1,v2);
v2 = Optimization::Permute::Permute3(v1);
v1 = _mm512_add_ps(v1,v2);
u512f conv; conv.v=v1;
return conv.f[0];
}
//Complex double Reduce
template<>
inline Grid::ComplexD Reduce<Grid::ComplexD, __m512d>::operator()(__m512d in){
__m512d v1;
v1 = Optimization::Permute::Permute0(in); // sse 128; paired complex single
v1 = _mm512_add_pd(v1,in);
v1 = Optimization::Permute::Permute1(in); // sse 128; paired complex single
v1 = _mm512_add_pd(v1,in);
u512d conv; conv.v = v1;
return Grid::ComplexD(conv.f[0],conv.f[1]);
}
//Real double Reduce
template<>
inline Grid::RealD Reduce<Grid::RealD, __m512d>::operator()(__m512d in){
__m512d v1,v2;
v1 = Optimization::Permute::Permute0(in); // avx 512; quad double
v1 = _mm512_add_pd(v1,in);
v2 = Optimization::Permute::Permute1(v1);
v1 = _mm512_add_pd(v1,v2);
v2 = Optimization::Permute::Permute2(v1);
v1 = _mm512_add_pd(v1,v2);
u512d conv; conv.v = v1;
return conv.f[0];
}
#else
//Complex float Reduce
template<>
inline Grid::ComplexF Reduce<Grid::ComplexF, __m512>::operator()(__m512 in){
@ -382,7 +442,6 @@ namespace Optimization {
return _mm512_reduce_add_ps(in);
}
//Complex double Reduce
template<>
inline Grid::ComplexD Reduce<Grid::ComplexD, __m512d>::operator()(__m512d in){
@ -402,6 +461,7 @@ namespace Optimization {
printf("Reduce : Missing integer implementation -> FIX\n");
assert(0);
}
#endif
}