portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-04 05:54:32 +00:00
Code Releases Activity

Merge branch 'develop' into feature/gpu-port

Browse Source 
Conflicts:
	Grid/qcd/action/fermion/WilsonKernelsAsm.cc
	Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermionImplementation.h
	Grid/qcd/action/fermion/implementation/StaggeredKernelsAsm.h
	benchmarks/Benchmark_comms.cc
This commit is contained in:
Peter Boyle
2019-08-14 18:56:54 +01:00
parent 3e49dc8a67 bca36d9bc3
commit 48e6efc7c9
6 changed files with 159 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

127
Grid/qcd/action/fermion/WilsonKernelsAsm.cc Normal file
View File

@@ -0,0 +1,127 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernelsAsm.cc
Copyright (C) 2015
Author: Peter Boyle <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
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/qcd/action/fermion/FermionCore.h>
namespace Grid {
namespace QCD {
///////////////////////////////////////////////////////////
// Default to no assembler implementation
///////////////////////////////////////////////////////////
template<class Impl> void
WilsonKernels<Impl >::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
{
assert(0);
}
template<class Impl> void
WilsonKernels<Impl >::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
{
assert(0);
}
template<class Impl> void
WilsonKernels<Impl >::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
{
assert(0);
}
template<class Impl> void
WilsonKernels<Impl >::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
{
assert(0);
}
template<class Impl> void
WilsonKernels<Impl >::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
{
assert(0);
}
template<class Impl> void
WilsonKernels<Impl >::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
{
assert(0);
}
#include <Grid/qcd/action/fermion/WilsonKernelsAsmAvx512.h>
#include <Grid/qcd/action/fermion/WilsonKernelsAsmQPX.h>
#define INSTANTIATE_ASM(A)\
template void WilsonKernels<A>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,\
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);\
\
template void WilsonKernels<A>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,\
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);\
template void WilsonKernels<A>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,\
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);\
\
template void WilsonKernels<A>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,\
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);\
template void WilsonKernels<A>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,\
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);\
\
template void WilsonKernels<A>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,\
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out);\
INSTANTIATE_ASM(WilsonImplF);
INSTANTIATE_ASM(WilsonImplD);
INSTANTIATE_ASM(ZWilsonImplF);
INSTANTIATE_ASM(ZWilsonImplD);
INSTANTIATE_ASM(GparityWilsonImplF);
INSTANTIATE_ASM(GparityWilsonImplD);
INSTANTIATE_ASM(DomainWallVec5dImplF);
INSTANTIATE_ASM(DomainWallVec5dImplD);
INSTANTIATE_ASM(ZDomainWallVec5dImplF);
INSTANTIATE_ASM(ZDomainWallVec5dImplD);
INSTANTIATE_ASM(WilsonImplFH);
INSTANTIATE_ASM(WilsonImplDF);
INSTANTIATE_ASM(ZWilsonImplFH);
INSTANTIATE_ASM(ZWilsonImplDF);
INSTANTIATE_ASM(GparityWilsonImplFH);
INSTANTIATE_ASM(GparityWilsonImplDF);
INSTANTIATE_ASM(DomainWallVec5dImplFH);
INSTANTIATE_ASM(DomainWallVec5dImplDF);
INSTANTIATE_ASM(ZDomainWallVec5dImplFH);
INSTANTIATE_ASM(ZDomainWallVec5dImplDF);
}}

4
Grid/qcd/action/fermion/implementation/ImprovedStaggeredFermionImplementation.h
View File

@@ -25,8 +25,8 @@ with this program; if not, write to the Free Software Foundation, Inc.,
See the full license in the file "LICENSE" in the top level distribution See the full license in the file "LICENSE" in the top level distribution
directory directory
*************************************************************************************/ *************************************************************************************/
/* END LEGAL */ /* END LEGAL */
#include <Grid.h> #include <Grid/Grid.h>
#pragma once #pragma once

14
Grid/qcd/action/fermion/implementation/StaggeredKernelsAsm.h
View File

@@ -28,11 +28,11 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
/* END LEGAL */ /* END LEGAL */
#pragma once #pragma once
#include <Grid.h> #include <Grid/Grid.h>
#ifdef AVX512 #ifdef AVX512
#include <simd/Intel512common.h> #include <Grid/simd/Intel512common.h>
#include <simd/Intel512avx.h> #include <Grid/simd/Intel512avx.h>
#endif #endif
// Interleave operations from two directions // Interleave operations from two directions
@@ -681,7 +681,7 @@ void StaggeredKernels<Impl>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo,
gauge3 =(uint64_t)&UU[sU]( T ); gauge3 =(uint64_t)&UU[sU]( T );
// This is the single precision 5th direction vectorised kernel // This is the single precision 5th direction vectorised kernel
#include <simd/Intel512single.h> #include <Grid/simd/Intel512single.h>
template <> void StaggeredKernels<StaggeredVec5dImplF>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, template <> void StaggeredKernels<StaggeredVec5dImplF>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo,
DoubledGaugeFieldView &U, DoubledGaugeFieldView &U,
DoubledGaugeFieldView &UUU, DoubledGaugeFieldView &UUU,
@@ -735,7 +735,7 @@ template <> void StaggeredKernels<StaggeredVec5dImplF>::DhopSiteAsm(StencilImpl
} }
#include <simd/Intel512double.h> #include <Grid/simd/Intel512double.h>
template <> void StaggeredKernels<StaggeredVec5dImplD>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, template <> void StaggeredKernels<StaggeredVec5dImplD>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo,
DoubledGaugeFieldView &U, DoubledGaugeFieldView &U,
DoubledGaugeFieldView &UUU, DoubledGaugeFieldView &UUU,
@@ -820,7 +820,7 @@ template <> void StaggeredKernels<StaggeredVec5dImplD>::DhopSiteAsm(StencilImpl
// This is the single precision 5th direction vectorised kernel // This is the single precision 5th direction vectorised kernel
#include <simd/Intel512single.h> #include <Grid/simd/Intel512single.h>
template <> void StaggeredKernels<StaggeredImplF>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, template <> void StaggeredKernels<StaggeredImplF>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo,
DoubledGaugeFieldView &U, DoubledGaugeFieldView &U,
DoubledGaugeFieldView &UUU, DoubledGaugeFieldView &UUU,
@@ -889,7 +889,7 @@ template <> void StaggeredKernels<StaggeredImplF>::DhopSiteAsm(StencilImpl &st,
#endif #endif
} }
#include <simd/Intel512double.h> #include <Grid/simd/Intel512double.h>
template <> void StaggeredKernels<StaggeredImplD>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo, template <> void StaggeredKernels<StaggeredImplD>::DhopSiteAsm(StencilImpl &st, LebesgueOrder &lo,
DoubledGaugeFieldView &U, DoubledGaugeFieldView &U,
DoubledGaugeFieldView &UUU, DoubledGaugeFieldView &UUU,

2
Grid/qcd/action/fermion/implementation/StaggeredKernelsHand.h
View File

@@ -26,7 +26,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
See the full license in the file "LICENSE" in the top level distribution directory See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/ *************************************************************************************/
/* END LEGAL */ /* END LEGAL */
#include <Grid.h> #include <Grid/Grid.h>
#pragma once #pragma once

29
benchmarks/Benchmark_comms.cc
View File

@@ -29,7 +29,6 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
using namespace std; using namespace std;
using namespace Grid; using namespace Grid;
;
struct time_statistics{ struct time_statistics{
double mean; double mean;
@@ -187,9 +186,9 @@ int main (int argc, char ** argv)
rbuf[mu].resize(lat*lat*lat*Ls); rbuf[mu].resize(lat*lat*lat*Ls);
// std::cout << " buffers " << std::hex << (uint64_t)&xbuf[mu][0] <<" " << (uint64_t)&rbuf[mu][0] <<std::endl; // std::cout << " buffers " << std::hex << (uint64_t)&xbuf[mu][0] <<" " << (uint64_t)&rbuf[mu][0] <<std::endl;
} }
uint64_t bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
int ncomm; int ncomm;
int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
for(int i=0;i<Nloop;i++){ for(int i=0;i<Nloop;i++){
double start=usecond(); double start=usecond();
@@ -276,15 +275,22 @@ int main (int argc, char ** argv)
std::vector<HalfSpinColourVectorD *> xbuf(8); std::vector<HalfSpinColourVectorD *> xbuf(8);
std::vector<HalfSpinColourVectorD *> rbuf(8); std::vector<HalfSpinColourVectorD *> rbuf(8);
Grid.ShmBufferFreeAll(); Grid.ShmBufferFreeAll();
uint64_t bytes = lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
for(int d=0;d<8;d++){ for(int d=0;d<8;d++){
<<<<<<< HEAD
xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
// bzero((void *)xbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); // bzero((void *)xbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
// bzero((void *)rbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); // bzero((void *)rbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
=======
xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(bytes);
rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(bytes);
bzero((void *)xbuf[d],bytes);
bzero((void *)rbuf[d],bytes);
>>>>>>> develop
} }
int ncomm; int ncomm;
int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
double dbytes; double dbytes;
for(int i=0;i<Nloop;i++){ for(int i=0;i<Nloop;i++){
@@ -373,15 +379,22 @@ int main (int argc, char ** argv)
std::vector<HalfSpinColourVectorD *> xbuf(8); std::vector<HalfSpinColourVectorD *> xbuf(8);
std::vector<HalfSpinColourVectorD *> rbuf(8); std::vector<HalfSpinColourVectorD *> rbuf(8);
Grid.ShmBufferFreeAll(); Grid.ShmBufferFreeAll();
uint64_t bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
for(int d=0;d<8;d++){ for(int d=0;d<8;d++){
<<<<<<< HEAD
xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
// bzero((void *)xbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); // bzero((void *)xbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
// bzero((void *)rbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); // bzero((void *)rbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
=======
xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(bytes);
rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(bytes);
bzero((void *)xbuf[d],bytes);
bzero((void *)rbuf[d],bytes);
>>>>>>> develop
} }
int ncomm; int ncomm;
int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
double dbytes; double dbytes;
for(int i=0;i<Nloop;i++){ for(int i=0;i<Nloop;i++){
double start=usecond(); double start=usecond();
@@ -471,15 +484,13 @@ int main (int argc, char ** argv)
std::vector<HalfSpinColourVectorD *> xbuf(8); std::vector<HalfSpinColourVectorD *> xbuf(8);
std::vector<HalfSpinColourVectorD *> rbuf(8); std::vector<HalfSpinColourVectorD *> rbuf(8);
Grid.ShmBufferFreeAll(); Grid.ShmBufferFreeAll();
uint64_t bytes = lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
for(int d=0;d<8;d++){ for(int d=0;d<8;d++){
xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(bytes);
rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD)); rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(bytes);
// bzero((void *)xbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
// bzero((void *)rbuf[d],lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
} }
int ncomm; int ncomm;
int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
double dbytes; double dbytes;
for(int i=0;i<Nloop;i++){ for(int i=0;i<Nloop;i++){
double start=usecond(); double start=usecond();

5
documentation/GridXcode/readme.md
View File

@@ -262,7 +262,6 @@ Set HEADER_SEARCH_PATHS to:
$Grid/build$(CONFIGURATION)/Grid $Grid/build$(CONFIGURATION)/Grid
$Grid $Grid
$Grid/Grid
followed by (***the order is important***) the locations reported by `grid-config --cxxflags`, ignoring duplicates, e.g.: followed by (***the order is important***) the locations reported by `grid-config --cxxflags`, ignoring duplicates, e.g.:
@@ -272,7 +271,7 @@ followed by (***the order is important***) the locations reported by `grid-confi
**Note: the easiest way to set this value is to put it all on one line, space separated, and edit the text to the right of `HEADER_SEARCH_PATHS`**, i.e.: **Note: the easiest way to set this value is to put it all on one line, space separated, and edit the text to the right of `HEADER_SEARCH_PATHS`**, i.e.:
$Grid/build$(CONFIGURATION)/Grid $Grid $Grid/Grid $GridPre/openmpi/include $GridPkg/include $GridPre/lime/include $Grid/build$(CONFIGURATION)/Grid $Grid $GridPre/openmpi/include $GridPkg/include $GridPre/lime/include
#### LIBRARY_SEARCH_PATHS #### LIBRARY_SEARCH_PATHS
@@ -298,7 +297,7 @@ The easiest way to link to all required libraries is to obtain a list of all lib
and pasting the output ***with `-lGrid -lHadrons ` prepended*** (including the `-l` switches) directly into `OTHER_LDFLAGS`, e.g.: and pasting the output ***with `-lGrid -lHadrons ` prepended*** (including the `-l` switches) directly into `OTHER_LDFLAGS`, e.g.:
-lGrid -lHadrons -lmpi -lhdf5_cpp -lz -lcrypto -llime -lfftw3f -lfftw3 -lmpfr -lgmp -lstdc++ -lm -lz -lhdf5 -lGrid -lHadrons -lmpi -lhdf5_cpp -lhdf5 -lz -lcrypto -llime -lfftw3f -lfftw3 -lmpfr -lgmp -lm
## Make additional configurations ## Make additional configurations