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

Optional overlap of comms with compute

Browse Source
This commit is contained in:
paboyle 2016-01-04 14:18:40 +00:00
parent 331768dcff
commit 02452afd36
11 changed files with 97 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
benchmarks/Benchmark_dwf.cc
View File

@ -44,10 +44,16 @@ struct scal {
Gamma::GammaT Gamma::GammaT
}; };
bool overlapComms = false;
int main (int argc, char ** argv) int main (int argc, char ** argv)
{ {
Grid_init(&argc,&argv); Grid_init(&argc,&argv);
if( GridCmdOptionExists(argv,argv+argc,"--asynch") ){
overlapComms = true;
}
int threads = GridThread::GetThreads(); int threads = GridThread::GetThreads();
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl; std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
@ -107,10 +113,15 @@ int main (int argc, char ** argv)
RealD mass=0.1; RealD mass=0.1;
RealD M5 =1.8; RealD M5 =1.8;
DomainWallFermionR Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
typename DomainWallFermionR::ImplParams params;
params.overlapCommsCompute = overlapComms;
DomainWallFermionR Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,params);
std::cout<<GridLogMessage << "Calling Dw"<<std::endl; std::cout<<GridLogMessage << "Calling Dw"<<std::endl;
int ncall=10000; int ncall=1000;
{ {
double t0=usecond(); double t0=usecond();
for(int i=0;i<ncall;i++){ for(int i=0;i<ncall;i++){

11
benchmarks/Benchmark_wilson.cc
View File

@ -44,10 +44,15 @@ struct scal {
Gamma::GammaT Gamma::GammaT
}; };
bool overlapComms = false;
int main (int argc, char ** argv) int main (int argc, char ** argv)
{ {
Grid_init(&argc,&argv); Grid_init(&argc,&argv);
if( GridCmdOptionExists(argv,argv+argc,"--asynch") ){
overlapComms = true;
}
std::vector<int> latt_size = GridDefaultLatt(); std::vector<int> latt_size = GridDefaultLatt();
std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
@ -116,7 +121,11 @@ int main (int argc, char ** argv)
} }
ref = -0.5*ref; ref = -0.5*ref;
RealD mass=0.1; RealD mass=0.1;
WilsonFermionR Dw(Umu,Grid,RBGrid,mass);
typename WilsonFermionR::ImplParams params;
params.overlapCommsCompute = overlapComms;
WilsonFermionR Dw(Umu,Grid,RBGrid,mass,params);
std::cout<<GridLogMessage << "Calling Dw"<<std::endl; std::cout<<GridLogMessage << "Calling Dw"<<std::endl;
int ncall=1000; int ncall=1000;

5
configure vendored
View File

@ -6367,7 +6367,6 @@ echo
echo Checking libraries echo Checking libraries
echo ::::::::::::::::::::::::::::::::::::::::::: echo :::::::::::::::::::::::::::::::::::::::::::
for ac_func in gettimeofday for ac_func in gettimeofday
do : do :
ac_fn_cxx_check_func "$LINENO" "gettimeofday" "ac_cv_func_gettimeofday" ac_fn_cxx_check_func "$LINENO" "gettimeofday" "ac_cv_func_gettimeofday"
@ -8152,8 +8151,10 @@ The following features are enabled:
- graphs and diagrams : `if test "x$enable_dot" = xyes; then echo yes; else echo no; fi` - graphs and diagrams : `if test "x$enable_dot" = xyes; then echo yes; else echo no; fi`
- Supported SIMD flags : $SIMD_FLAGS - Supported SIMD flags : $SIMD_FLAGS
---------------------------------------------------------- ----------------------------------------------------------
- enabled simd support : ${ac_SIMD} (supported: $supported ) - enabled simd support : ${ac_SIMD} (config macro says supported: $supported )
- communications type : ${ac_COMMS} - communications type : ${ac_COMMS}
- default precision : ${ac_PRECISION}
- RNG choice : ${ac_RNG}
" "

5
configure.ac
View File

@ -53,7 +53,6 @@ echo
echo Checking libraries echo Checking libraries
echo ::::::::::::::::::::::::::::::::::::::::::: echo :::::::::::::::::::::::::::::::::::::::::::
AC_CHECK_FUNCS([gettimeofday]) AC_CHECK_FUNCS([gettimeofday])
#AC_CHECK_LIB([gmp],[__gmpf_init],, #AC_CHECK_LIB([gmp],[__gmpf_init],,
@ -262,8 +261,10 @@ The following features are enabled:
- graphs and diagrams : `if test "x$enable_dot" = xyes; then echo yes; else echo no; fi` - graphs and diagrams : `if test "x$enable_dot" = xyes; then echo yes; else echo no; fi`
- Supported SIMD flags : $SIMD_FLAGS - Supported SIMD flags : $SIMD_FLAGS
---------------------------------------------------------- ----------------------------------------------------------
- enabled simd support : ${ac_SIMD} (supported: $supported ) - enabled simd support : ${ac_SIMD} (config macro says supported: $supported )
- communications type : ${ac_COMMS} - communications type : ${ac_COMMS}
- default precision : ${ac_PRECISION}
- RNG choice : ${ac_RNG}
" "

28
lib/Init.cc
View File

@ -238,20 +238,20 @@ void Grid_init(int *argc,char ***argv)
} }
std::cout <<std::endl; std::cout <<std::endl;
std::cout <<Logger::RED << "__|__|__|__|__"<< "|__|__|_"<< "_|__|__|"<< "__|__|__|__|__"<<std::endl; std::cout <<Logger::RED << "__|__|__|__|__"<< "|__|__|_"<<Logger::PURPLE<<"_|__|__|"<< "__|__|__|__|__"<<std::endl;
std::cout <<Logger::RED << "__|__|__|__|__"<< "|__|__|_"<< "_|__|__|"<< "__|__|__|__|__"<<std::endl; std::cout <<Logger::RED << "__|__|__|__|__"<< "|__|__|_"<<Logger::PURPLE<<"_|__|__|"<< "__|__|__|__|__"<<std::endl;
std::cout <<Logger::RED << "__|__|__| | "<< "| | | "<< " | | |"<< " | | | _|__"<<std::endl; std::cout <<Logger::RED << "__|__| | | "<< "| | | "<<Logger::PURPLE<<" | | |"<< " | | | _|__"<<std::endl;
std::cout <<Logger::RED << "__|__ "<< " "<< " "<< " _|__"<<std::endl; std::cout <<Logger::RED << "__|__ "<< " "<<Logger::PURPLE<<" "<< " _|__"<<std::endl;
std::cout <<Logger::GREEN<< "__|_ GGGG "<<Logger::RED<<" RRRR "<<Logger::BLUE<<" III "<<Logger::PURPLE<<"DDDD _|__"<<std::endl; std::cout <<Logger::RED << "__|_ "<<Logger::GREEN<<" GGGG "<<Logger::RED<<" RRRR "<<Logger::BLUE <<" III "<<Logger::PURPLE<<"DDDD "<<Logger::PURPLE<<" _|__"<<std::endl;
std::cout <<Logger::GREEN<< "__|_ G "<<Logger::RED<<" R R "<<Logger::BLUE<<" I "<<Logger::PURPLE<<"D D _|__"<<std::endl; std::cout <<Logger::RED << "__|_ "<<Logger::GREEN<<"G "<<Logger::RED<<" R R "<<Logger::BLUE <<" I "<<Logger::PURPLE<<"D D "<<Logger::PURPLE<<" _|__"<<std::endl;
std::cout <<Logger::GREEN<< "__|_ G "<<Logger::RED<<" R R "<<Logger::BLUE<<" I "<<Logger::PURPLE<<"D D _|__"<<std::endl; std::cout <<Logger::RED << "__|_ "<<Logger::GREEN<<"G "<<Logger::RED<<" R R "<<Logger::BLUE <<" I "<<Logger::PURPLE<<"D D"<<Logger::PURPLE<<" _|__"<<std::endl;
std::cout <<Logger::GREEN<< "__|_ G GG "<<Logger::RED<<" RRRR "<<Logger::BLUE<<" I "<<Logger::PURPLE<<"D D _|__"<<std::endl; std::cout <<Logger::BLUE << "__|_ "<<Logger::GREEN<<"G GG "<<Logger::RED<<" RRRR "<<Logger::BLUE <<" I "<<Logger::PURPLE<<"D D"<<Logger::GREEN <<" _|__"<<std::endl;
std::cout <<Logger::GREEN<< "__|_ G G "<<Logger::RED<<" R R "<<Logger::BLUE<<" I "<<Logger::PURPLE<<"D D _|__"<<std::endl; std::cout <<Logger::BLUE << "__|_ "<<Logger::GREEN<<"G G "<<Logger::RED<<" R R "<<Logger::BLUE <<" I "<<Logger::PURPLE<<"D D "<<Logger::GREEN <<" _|__"<<std::endl;
std::cout <<Logger::GREEN<< "__|_ GGGG "<<Logger::RED<<" R R "<<Logger::BLUE<<" III "<<Logger::PURPLE<<"DDDD _|__"<<std::endl; std::cout <<Logger::BLUE << "__|_ "<<Logger::GREEN<<" GGGG "<<Logger::RED<<" R R "<<Logger::BLUE <<" III "<<Logger::PURPLE<<"DDDD "<<Logger::GREEN <<" _|__"<<std::endl;
std::cout <<Logger::BLUE << "__|__ "<< " "<< " "<< " _|__"<<std::endl; std::cout <<Logger::BLUE << "__|__ "<< " "<<Logger::GREEN <<" "<< " _|__"<<std::endl;
std::cout <<Logger::BLUE << "__|__|__|__|__"<< "|__|____"<< "_|__|__|"<< "__|__|__|__|__"<<std::endl; std::cout <<Logger::BLUE << "__|__|__|__|__"<< "|__|__|_"<<Logger::GREEN <<"_|__|__|"<< "__|__|__|__|__"<<std::endl;
std::cout <<Logger::BLUE << "__|__|__|__|__"<< "|__|__|_"<< "_|__|__|"<< "__|__|__|__|__"<<std::endl; std::cout <<Logger::BLUE << "__|__|__|__|__"<< "|__|__|_"<<Logger::GREEN <<"_|__|__|"<< "__|__|__|__|__"<<std::endl;
std::cout <<Logger::BLUE << " | | | | "<< "| | | "<< " | | |"<< " | | | | "<<std::endl; std::cout <<Logger::BLUE << " | | | | "<< "| | | "<<Logger::GREEN <<" | | |"<< " | | | | "<<std::endl;
std::cout << std::endl; std::cout << std::endl;
std::cout << std::endl; std::cout << std::endl;
std::cout <<Logger::YELLOW<< std::endl; std::cout <<Logger::YELLOW<< std::endl;

8
lib/qcd/action/ActionParams.h
View File

@ -34,12 +34,16 @@ namespace QCD {
// These can move into a params header and be given MacroMagic serialisation // These can move into a params header and be given MacroMagic serialisation
struct GparityWilsonImplParams { struct GparityWilsonImplParams {
bool overlapCommsCompute;
std::vector<int> twists; std::vector<int> twists;
GparityWilsonImplParams () : twists(Nd,0) {}; GparityWilsonImplParams () : twists(Nd,0), overlapCommsCompute(false) {};
}; };
struct WilsonImplParams { }; struct WilsonImplParams {
bool overlapCommsCompute;
WilsonImplParams() : overlapCommsCompute(false) {};
};
struct OneFlavourRationalParams { struct OneFlavourRationalParams {
RealD lo; RealD lo;

7
lib/qcd/action/fermion/FermionOperatorImpl.h
View File

@ -131,9 +131,13 @@ namespace Grid {
typedef WilsonCompressor<SiteHalfSpinor,SiteSpinor> Compressor; typedef WilsonCompressor<SiteHalfSpinor,SiteSpinor> Compressor;
typedef WilsonImplParams ImplParams; typedef WilsonImplParams ImplParams;
typedef CartesianStencil<SiteSpinor,SiteHalfSpinor,Compressor> StencilImpl; typedef CartesianStencil<SiteSpinor,SiteHalfSpinor,Compressor> StencilImpl;
ImplParams Params; ImplParams Params;
WilsonImpl(const ImplParams &p= ImplParams()) : Params(p) {}; WilsonImpl(const ImplParams &p= ImplParams()) : Params(p) {};
bool overlapCommsCompute(void) { return Params.overlapCommsCompute; };
inline void multLink(SiteHalfSpinor &phi,const SiteDoubledGaugeField &U,const SiteHalfSpinor &chi,int mu,StencilEntry *SE,StencilImpl &St){ inline void multLink(SiteHalfSpinor &phi,const SiteDoubledGaugeField &U,const SiteHalfSpinor &chi,int mu,StencilEntry *SE,StencilImpl &St){
mult(&phi(),&U(mu),&chi()); mult(&phi(),&U(mu),&chi());
} }
@ -204,9 +208,12 @@ PARALLEL_FOR_LOOP
typedef CartesianStencil<SiteSpinor,SiteHalfSpinor,Compressor> StencilImpl; typedef CartesianStencil<SiteSpinor,SiteHalfSpinor,Compressor> StencilImpl;
typedef GparityWilsonImplParams ImplParams; typedef GparityWilsonImplParams ImplParams;
ImplParams Params; ImplParams Params;
GparityWilsonImpl(const ImplParams &p= ImplParams()) : Params(p) {}; GparityWilsonImpl(const ImplParams &p= ImplParams()) : Params(p) {};
bool overlapCommsCompute(void) { return Params.overlapCommsCompute; };
// provide the multiply by link that is differentiated between Gparity (with flavour index) and non-Gparity // provide the multiply by link that is differentiated between Gparity (with flavour index) and non-Gparity
inline void multLink(SiteHalfSpinor &phi,const SiteDoubledGaugeField &U,const SiteHalfSpinor &chi,int mu,StencilEntry *SE,StencilImpl &St){ inline void multLink(SiteHalfSpinor &phi,const SiteDoubledGaugeField &U,const SiteHalfSpinor &chi,int mu,StencilEntry *SE,StencilImpl &St){

21
lib/qcd/action/fermion/WilsonFermion.cc
View File

@ -227,7 +227,7 @@ PARALLEL_FOR_LOOP
out.checkerboard = in.checkerboard; out.checkerboard = in.checkerboard;
DhopInternalCommsCompute(Stencil,Umu,in,out,dag); DhopInternal(Stencil,Umu,in,out,dag);
} }
template<class Impl> template<class Impl>
@ -238,7 +238,7 @@ PARALLEL_FOR_LOOP
assert(in.checkerboard==Even); assert(in.checkerboard==Even);
out.checkerboard = Odd; out.checkerboard = Odd;
DhopInternalCommsCompute(StencilEven,UmuOdd,in,out,dag); DhopInternal(StencilEven,UmuOdd,in,out,dag);
} }
template<class Impl> template<class Impl>
@ -249,7 +249,7 @@ PARALLEL_FOR_LOOP
assert(in.checkerboard==Odd); assert(in.checkerboard==Odd);
out.checkerboard = Even; out.checkerboard = Even;
DhopInternalCommsCompute(StencilOdd,UmuEven,in,out,dag); DhopInternal(StencilOdd,UmuEven,in,out,dag);
} }
template<class Impl> template<class Impl>
@ -283,10 +283,19 @@ PARALLEL_FOR_LOOP
}; };
template<class Impl> template<class Impl>
void WilsonFermion<Impl>::DhopInternal(StencilImpl & st,DoubledGaugeField & U, void WilsonFermion<Impl>::DhopInternal(StencilImpl & st,DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag) { const FermionField &in, FermionField &out,int dag)
{
if ( Impl::overlapCommsCompute () ) {
DhopInternalCommsOverlapCompute(st,U,in,out,dag);
} else {
DhopInternalCommsThenCompute(st,U,in,out,dag);
}
}
template<class Impl>
void WilsonFermion<Impl>::DhopInternalCommsThenCompute(StencilImpl & st,DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag) {
assert((dag==DaggerNo) ||(dag==DaggerYes)); assert((dag==DaggerNo) ||(dag==DaggerYes));
@ -322,7 +331,7 @@ PARALLEL_FOR_LOOP
template<class Impl> template<class Impl>
void WilsonFermion<Impl>::DhopInternalCommsCompute(StencilImpl & st,DoubledGaugeField & U, void WilsonFermion<Impl>::DhopInternalCommsOverlapCompute(StencilImpl & st,DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag) { const FermionField &in, FermionField &out,int dag) {
assert((dag==DaggerNo) ||(dag==DaggerYes)); assert((dag==DaggerNo) ||(dag==DaggerYes));

4
lib/qcd/action/fermion/WilsonFermion.h
View File

@ -114,7 +114,9 @@ namespace Grid {
void DhopInternal(StencilImpl & st,DoubledGaugeField & U, void DhopInternal(StencilImpl & st,DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag) ; const FermionField &in, FermionField &out,int dag) ;
void DhopInternalCommsCompute(StencilImpl & st,DoubledGaugeField & U, void DhopInternalCommsThenCompute(StencilImpl & st,DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag) ;
void DhopInternalCommsOverlapCompute(StencilImpl & st,DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag) ; const FermionField &in, FermionField &out,int dag) ;

20
lib/qcd/action/fermion/WilsonFermion5D.cc
View File

@ -272,6 +272,18 @@ template<class Impl>
void WilsonFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOrder &lo, void WilsonFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOrder &lo,
DoubledGaugeField & U, DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag) const FermionField &in, FermionField &out,int dag)
{
if ( Impl::overlapCommsCompute () ) {
DhopInternalCommsOverlapCompute(st,lo,U,in,out,dag);
} else {
DhopInternalCommsThenCompute(st,lo,U,in,out,dag);
}
}
template<class Impl>
void WilsonFermion5D<Impl>::DhopInternalCommsThenCompute(StencilImpl & st, LebesgueOrder &lo,
DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag)
{ {
// assert((dag==DaggerNo) ||(dag==DaggerYes)); // assert((dag==DaggerNo) ||(dag==DaggerYes));
@ -398,7 +410,7 @@ PARALLEL_FOR_LOOP
} }
template<class Impl> template<class Impl>
void WilsonFermion5D<Impl>::DhopInternalCommsCompute(StencilImpl & st, LebesgueOrder &lo, void WilsonFermion5D<Impl>::DhopInternalCommsOverlapCompute(StencilImpl & st, LebesgueOrder &lo,
DoubledGaugeField & U, DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag) const FermionField &in, FermionField &out,int dag)
{ {
@ -536,7 +548,7 @@ void WilsonFermion5D<Impl>::DhopOE(const FermionField &in, FermionField &out,int
assert(in.checkerboard==Even); assert(in.checkerboard==Even);
out.checkerboard = Odd; out.checkerboard = Odd;
DhopInternalCommsCompute(StencilEven,LebesgueEvenOdd,UmuOdd,in,out,dag); DhopInternal(StencilEven,LebesgueEvenOdd,UmuOdd,in,out,dag);
} }
template<class Impl> template<class Impl>
void WilsonFermion5D<Impl>::DhopEO(const FermionField &in, FermionField &out,int dag) void WilsonFermion5D<Impl>::DhopEO(const FermionField &in, FermionField &out,int dag)
@ -547,7 +559,7 @@ void WilsonFermion5D<Impl>::DhopEO(const FermionField &in, FermionField &out,int
assert(in.checkerboard==Odd); assert(in.checkerboard==Odd);
out.checkerboard = Even; out.checkerboard = Even;
DhopInternalCommsCompute(StencilOdd,LebesgueEvenOdd,UmuEven,in,out,dag); DhopInternal(StencilOdd,LebesgueEvenOdd,UmuEven,in,out,dag);
} }
template<class Impl> template<class Impl>
void WilsonFermion5D<Impl>::Dhop(const FermionField &in, FermionField &out,int dag) void WilsonFermion5D<Impl>::Dhop(const FermionField &in, FermionField &out,int dag)
@ -557,7 +569,7 @@ void WilsonFermion5D<Impl>::Dhop(const FermionField &in, FermionField &out,int d
out.checkerboard = in.checkerboard; out.checkerboard = in.checkerboard;
DhopInternalCommsCompute(Stencil,Lebesgue,Umu,in,out,dag); DhopInternal(Stencil,Lebesgue,Umu,in,out,dag);
} }
template<class Impl> template<class Impl>
void WilsonFermion5D<Impl>::DW(const FermionField &in, FermionField &out,int dag) void WilsonFermion5D<Impl>::DW(const FermionField &in, FermionField &out,int dag)

8
lib/qcd/action/fermion/WilsonFermion5D.h
View File

@ -117,7 +117,13 @@ namespace Grid {
FermionField &out, FermionField &out,
int dag); int dag);
void DhopInternalCommsCompute(StencilImpl & st, void DhopInternalCommsThenCompute(StencilImpl & st,
LebesgueOrder &lo,
DoubledGaugeField &U,
const FermionField &in,
FermionField &out,
int dag);
void DhopInternalCommsOverlapCompute(StencilImpl & st,
LebesgueOrder &lo, LebesgueOrder &lo,
DoubledGaugeField &U, DoubledGaugeField &U,
const FermionField &in, const FermionField &in,