portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-04 19:25:56 +01:00
Code Releases Activity

Compilation with both single and double precision

Browse Source
This commit is contained in:
Guido Cossu 2016-10-30 12:04:06 +00:00
parent e1042aef77
commit 9b066e94d0
1 changed files with 25 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

55
lib/qcd/action/fermion/WilsonKernelsAsm.cc
View File

@ -10,6 +10,7 @@
Author: Peter Boyle <paboyle@ph.ed.ac.uk> Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk> Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Guido Cossu <guido.cossu@ed.ac.uk>
This program is free software; you can redistribute it and/or modify This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by it under the terms of the GNU General Public License as published by
@ -55,23 +56,24 @@ WilsonKernels<Impl >::DiracOptAsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,
#if defined(AVX512) #if defined(AVX512)
#include <simd/Intel512wilson.h> #include <simd/Intel512wilson.h>
#if defined(GRID_DEFAULT_PRECISION_SINGLE)
/////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////
// If we are AVX512 specialise the single precision routine // If we are AVX512 specialise the single precision routine
/////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////
#include <simd/Intel512single.h> #include <simd/Intel512single.h>
static Vector<vComplexF> signs; static Vector<vComplexF> signsF;
int setupSigns(void ){ template<typename vtype>
Vector<vComplexF> bother(2); int setupSigns(Vector<vtype>& signs ){
Vector<vtype> bother(2);
signs = bother; signs = bother;
vrsign(signs[0]); vrsign(signs[0]);
visign(signs[1]); visign(signs[1]);
return 1; return 1;
} }
static int signInit = setupSigns();
static int signInitF = setupSigns(signsF);
#define label(A) ilabel(A) #define label(A) ilabel(A)
#define ilabel(A) ".globl\n" #A ":\n" #define ilabel(A) ".globl\n" #A ":\n"
@ -80,6 +82,7 @@ static Vector<vComplexF> signs;
#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN(ptr,pf) #define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN(ptr,pf)
#define FX(A) WILSONASM_ ##A #define FX(A) WILSONASM_ ##A
#define COMPLEX_TYPE vComplexF #define COMPLEX_TYPE vComplexF
#define signs signsF
#undef KERNEL_DAG #undef KERNEL_DAG
template<> void template<> void
@ -116,34 +119,22 @@ WilsonKernels<DomainWallVec5dImplF>::DiracOptAsmDhopSiteDag(StencilImpl &st,Lebe
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out) int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef COMPLEX_TYPE #undef COMPLEX_TYPE
#undef signs
#undef VMOVRDUP
#undef MAYBEPERM
#undef MULT_2SPIN
#undef FX
#endif //Single precision ///////////////////////////////////////////////////////////
// If we are AVX512 specialise the double precision routine
#if defined(GRID_DEFAULT_PRECISION_DOUBLE) ///////////////////////////////////////////////////////////
//temporary separating the two sections
//for debug in isolation
//can be unified
///////////////////////////////////////////////////////////
// If we are AVX512 specialise the double precision routine
///////////////////////////////////////////////////////////
#include <simd/Intel512double.h> #include <simd/Intel512double.h>
static Vector<vComplexD> signs; static Vector<vComplexD> signsD;
#define signs signsD
static int signInitD = setupSigns(signsD);
int setupSigns(void ){
Vector<vComplexD> bother(2);
signs = bother;
vrsign(signs[0]);
visign(signs[1]);
return 1;
}
static int signInit = setupSigns();
#define label(A) ilabel(A)
#define ilabel(A) ".globl\n" #A ":\n"
#define MAYBEPERM(A,perm) if (perm) { A ; } #define MAYBEPERM(A,perm) if (perm) { A ; }
#define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN(ptr,pf) #define MULT_2SPIN(ptr,pf) MULT_ADDSUB_2SPIN(ptr,pf)
#define FX(A) WILSONASM_ ##A #define FX(A) WILSONASM_ ##A
@ -185,7 +176,11 @@ WilsonKernels<DomainWallVec5dImplD>::DiracOptAsmDhopSiteDag(StencilImpl &st,Lebe
#include <qcd/action/fermion/WilsonKernelsAsmBody.h> #include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef COMPLEX_TYPE #undef COMPLEX_TYPE
#endif //Double precision #undef signs
#undef VMOVRDUP
#undef MAYBEPERM
#undef MULT_2SPIN
#undef FX
#endif //AVX512 #endif //AVX512