portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-12 20:27:06 +01:00
Code Releases Activity

5-link first attempt

Browse Source
This commit is contained in:
david clarke
2023-08-11 22:56:30 -06:00
parent 9d263d9a7d
commit 99d879ea7f
3 changed files with 145 additions and 170 deletions
Download Patch File Download Diff File Expand all files Collapse all files

125
tests/smearing/Benchmark_su3mult_vs_lookup.cc
View File

@ -1,125 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./benchmarks/Benchmark_su3mult_vs_lookup.cc
Copyright (C) 2023
Author: D. A. Clarke <clarke.davida@gmail.com>
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
*************************************************************************************/
/*
@file Benchmark_su3mult_vs_lookup.cc
@brief check to see whether su3 multiplication or lookup tables is faster
*/
#include <Grid/Grid.h>
using namespace Grid;
/*! @brief make the logger work like python print */
template<typename... Args>
inline std::string sjoin(Args&&... args) noexcept {
std::ostringstream msg;
(msg << ... << args);
return msg.str();
}
template <typename... Args>
inline void Grid_log(Args&&... args) {
std::string msg = sjoin(std::forward<Args>(args)...);
std::cout << GridLogMessage << msg << std::endl;
}
/*! @brief parameter file to easily adjust Nloop */
struct ConfParameters: Serializable {
GRID_SERIALIZABLE_CLASS_MEMBERS(
ConfParameters,
int, Nloop);
template <class ReaderClass>
ConfParameters(Reader<ReaderClass>& Reader){
read(Reader, "parameters", *this);
}
};
int main (int argc, char** argv) {
// Params for the test.
int Ns = 8;
int Nt = 4;
int threads = GridThread::GetThreads();
std::string conf_in = "nersc.l8t4b3360";
Coordinate latt_size(Nd,0); latt_size[0]=Ns; latt_size[1]=Ns; latt_size[2]=Ns; latt_size[3]=Nt;
Grid_init(&argc,&argv);
Coordinate simd_layout = GridDefaultSimd(Nd,vComplexD::Nsimd());
Coordinate mpi_layout = GridDefaultMpi();
GridCartesian GRID(latt_size,simd_layout,mpi_layout);
Grid_log(" mpi = ",mpi_layout);
Grid_log(" simd = ",simd_layout);
Grid_log(" latt = ",latt_size);
Grid_log("threads = ",threads);
XmlReader Reader("mult_vs_lookup.xml",false, "grid");
ConfParameters param(Reader);
Grid_log(" Nloop = ",param.Nloop);
// Gauge field and accessor
LatticeGaugeField Umu(&GRID);
autoView(U_v, Umu, CpuRead);
// Read the configuration into Umu
FieldMetaData header;
NerscIO::readConfiguration(Umu, header, conf_in);
// Read in lattice sequentially, Nloop times
double lookupTime = 0.;
for(int i=0;i<param.Nloop;i++) {
double start = usecond();
for(int ss=0;ss<U_v.size();ss++)
for(int mu=0;mu<Nd;mu++) {
auto U1 = U_v[ss](mu);
}
double stop = usecond();
lookupTime += stop-start; // microseconds
}
Grid_log("Time to lookup: ",lookupTime,"[ms]");
// Raise a matrix to the power nmat, for each link.
auto U1 = U_v[0](0);
for(int nmat=1;nmat<8;nmat++) {
double multTime = 0.;
for(int i=0;i<param.Nloop;i++) {
double start=usecond();
for(int ss=0;ss<U_v.size();ss++)
for(int mu=0;mu<Nd;mu++) {
auto U2 = U1;
for(int j=1;j<nmat;j++) {
U2 *= U1;
}
}
double stop=usecond();
multTime += stop-start;
}
Grid_log("Time to multiply ",nmat," matrices: ",lookupTime," [ms]");
}
Grid_finalize();
}

65
tests/smearing/Test_fatLinks.cc
View File

@ -51,6 +51,20 @@ inline void Grid_log(Args&&... args) {
std::cout << GridLogMessage << msg << std::endl;
}
/*! @brief parameter file to easily adjust Nloop */
struct ConfParameters: Serializable {
GRID_SERIALIZABLE_CLASS_MEMBERS(
ConfParameters,
int, benchmark,
int, Nloop);
template <class ReaderClass>
ConfParameters(Reader<ReaderClass>& Reader){
read(Reader, "parameters", *this);
}
};
//
// one method: input --> fat
// another : input --> long (naik)
@ -65,16 +79,22 @@ int main (int argc, char** argv) {
Coordinate latt_size(Nd,0); latt_size[0]=Ns; latt_size[1]=Ns; latt_size[2]=Ns; latt_size[3]=Nt;
std::string conf_in = "nersc.l8t4b3360";
std::string conf_out = "nersc.l8t4b3360.3link";
int threads = GridThread::GetThreads();
// Initialize the Grid
Grid_init(&argc,&argv);
Coordinate simd_layout = GridDefaultSimd(Nd,vComplexD::Nsimd());
Coordinate mpi_layout = GridDefaultMpi();
Grid_log(" mpi = ",mpi_layout);
Grid_log("simd = ",simd_layout);
Grid_log("latt = ",latt_size);
Grid_log("mpi = ",mpi_layout);
Grid_log("simd = ",simd_layout);
Grid_log("latt = ",latt_size);
Grid_log("threads = ",threads);
GridCartesian GRID(latt_size,simd_layout,mpi_layout);
XmlReader Reader("fatParams.xml",false,"grid");
ConfParameters param(Reader);
if(param.benchmark) Grid_log(" Nloop = ",param.Nloop);
// Instantiate the LatticeGaugeField objects holding thin (Umu) and fat (U_smr) links
LatticeGaugeField Umu(&GRID);
LatticeGaugeField U_smr(&GRID);
@ -85,6 +105,7 @@ int main (int argc, char** argv) {
// Smear Umu and store result in U_smr
Smear_HISQ_fat<LatticeGaugeField> hisq_fat(&GRID,1/8.,0.,1/16.,1/64.,1/384.,0.);
// Smear_HISQ_fat<LatticeGaugeField> hisq_fat(&GRID,1/8.,0.,1/16.,0.,1/384.,0.);
hisq_fat.smear(U_smr,Umu);
NerscIO::writeConfiguration(U_smr,conf_out,"HISQ");
@ -100,5 +121,43 @@ int main (int argc, char** argv) {
auto absDiff = norm2(diff)/norm2(Umu);
Grid_log(" |Umu-U|/|Umu| = ",absDiff);
if (param.benchmark) {
autoView(U_v, Umu, CpuRead); // Gauge accessor
// Read in lattice sequentially, Nloop times
double lookupTime = 0.;
for(int i=0;i<param.Nloop;i++) {
double start = usecond();
for(int ss=0;ss<U_v.size();ss++)
for(int mu=0;mu<Nd;mu++) {
auto U1 = U_v[ss](mu);
}
double stop = usecond();
lookupTime += stop-start; // microseconds
}
Grid_log("Time to lookup: ",lookupTime,"[ms]");
// Raise a matrix to the power nmat, for each link.
auto U1 = U_v[0](0);
for(int nmat=1;nmat<8;nmat++) {
double multTime = 0.;
for(int i=0;i<param.Nloop;i++) {
double start=usecond();
for(int ss=0;ss<U_v.size();ss++)
for(int mu=0;mu<Nd;mu++) {
auto U2 = U1;
for(int j=1;j<nmat;j++) {
U2 *= U1;
}
}
double stop=usecond();
multTime += stop-start;
}
Grid_log("Time to multiply ",nmat," matrices: ",multTime," [ms]");
}
}
Grid_finalize();
}