portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-03 21:44:33 +00:00
Code Releases Activity

AVX512 improvements; sigfpe trapping too

Browse Source
This commit is contained in:
paboyle
2016-03-30 08:42:34 +01:00
parent 83b15bfcdd
commit 60d965f79e
2 changed files with 18 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
lib/Init.cc
View File

@@ -318,7 +318,10 @@ void Grid_sa_signal_handler(int sig,siginfo_t *si,void * ptr)
exit(0);
return;
};
#ifdef GRID_FPE
#define _GNU_SOURCE
#include <fenv.h>
#endif
void Grid_debug_handler_init(void)
{
struct sigaction sa,osa;
@@ -327,5 +330,9 @@ void Grid_debug_handler_init(void)
sa.sa_flags = SA_SIGINFO;
sigaction(SIGSEGV,&sa,NULL);
sigaction(SIGTRAP,&sa,NULL);
#ifdef GRID_FPE
feenableexcept( FE_INVALID|FE_OVERFLOW|FE_DIVBYZERO);
sigaction(SIGFPE,&sa,NULL);
#endif
}
}

12
lib/qcd/action/fermion/WilsonKernelsAsm.cc
View File

@@ -43,6 +43,10 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#undef VZERO
#undef VTIMESI
#undef VTIMESMINUSI
#undef VMOVIDUP
#undef VMOVRDUP
#undef VMADDSUB
#undef VSHUF
#define VZERO(A) VZEROf(A)
#define VMOV(A,B) VMOVf(A,B)
@@ -51,8 +55,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define VADD(A,B,C) VADDf(A,B,C)
#define VSUB(A,B,C) VSUBf(A,B,C)
#define VMUL(Uri,Uir,Chi,UChi,Z) VMULf(Uri,Uir,Chi,UChi,Z)
#define VMADD(Uri,Uir,Chi,UChi,Z) VMADDf(Uri,Uir,Chi,UChi,Z)
#define VMUL(Uri,Uir,Chi) VMULf(Uri,Uir,Chi)
#define VMADD(Uri,Uir,Chi) VMADDf(Uri,Uir,Chi)
#define VTIMESI(A,B,C) VTIMESIf(A,B,C)
#define VTIMESMINUSI(A,B,C) VTIMESMINUSIf(A,B,C)
@@ -96,6 +100,10 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define VADDMEM(O,A,B,C) VADDMEMf(O,A,B,C)
#define VSUBMEM(O,A,B,C) VSUBMEMf(O,A,B,C)
#define VMOVIDUP(A,B,C) VMOVIDUPf(A,B,C)
#define VMOVRDUP(A,B,C) VMOVRDUPf(A,B,C)
#define VMADDSUB(A,B,accum) VMADDSUBf(A,B,accum)
#define VSHUF(A,B) VSHUFf(A,B)
#define ZMULMEM2SP(O,P,tmp,B,C,Briir,Biirr,Criir,Ciirr) ZMULMEM2SPf(O,P,tmp,B,C,Briir,Biirr,Criir,Ciirr)
#define ZMADDMEM2SP(O,P,tmp,B,C,Briir,Biirr,Criir,Ciirr) ZMADDMEM2SPf(O,P,tmp,B,C,Briir,Biirr,Criir,Ciirr)