portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-12-23 19:35:26 +00:00
Code Releases Activity

Adding persistent communicators

Browse Source
This commit is contained in:
paboyle 2016-07-08 17:16:08 +01:00
parent c667d9fdcc
commit ef97e32152
12 changed files with 347 additions and 95 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

121
benchmarks/Benchmark_comms.cc
View File

@ -196,5 +196,126 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << "= Benchmarking sequential persistent halo exchange in "<<nmu<<" dimensions"<<std::endl;
std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
std::cout<<GridLogMessage << " L "<<"\t\t"<<" Ls "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
for(int lat=4;lat<=32;lat+=2){
for(int Ls=1;Ls<=16;Ls*=2){
std::vector<int> latt_size ({lat,lat,lat,lat});
GridCartesian Grid(latt_size,simd_layout,mpi_layout);
std::vector<std::vector<HalfSpinColourVectorD> > xbuf(8,std::vector<HalfSpinColourVectorD>(lat*lat*lat*Ls));
std::vector<std::vector<HalfSpinColourVectorD> > rbuf(8,std::vector<HalfSpinColourVectorD>(lat*lat*lat*Ls));
int ncomm;
int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
std::vector<CartesianCommunicator::CommsRequest_t> empty;
std::vector<std::vector<CartesianCommunicator::CommsRequest_t> > requests_fwd(Nd,empty);
std::vector<std::vector<CartesianCommunicator::CommsRequest_t> > requests_bwd(Nd,empty);
for(int mu=0;mu<4;mu++){
ncomm=0;
if (mpi_layout[mu]>1 ) {
ncomm++;
int comm_proc;
int xmit_to_rank;
int recv_from_rank;
comm_proc=1;
Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
Grid.SendToRecvFromInit(requests_fwd[mu],
(void *)&xbuf[mu][0],
xmit_to_rank,
(void *)&rbuf[mu][0],
recv_from_rank,
bytes);
comm_proc = mpi_layout[mu]-1;
Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
Grid.SendToRecvFromInit(requests_bwd[mu],
(void *)&xbuf[mu+4][0],
xmit_to_rank,
(void *)&rbuf[mu+4][0],
recv_from_rank,
bytes);
}
}
{
double start=usecond();
for(int i=0;i<Nloop;i++){
for(int mu=0;mu<4;mu++){
if (mpi_layout[mu]>1 ) {
Grid.SendToRecvFromBegin(requests_fwd[mu]);
Grid.SendToRecvFromComplete(requests_fwd[mu]);
Grid.SendToRecvFromBegin(requests_bwd[mu]);
Grid.SendToRecvFromComplete(requests_bwd[mu]);
}
}
Grid.Barrier();
}
double stop=usecond();
double dbytes = bytes;
double xbytes = Nloop*dbytes*2.0*ncomm;
double rbytes = xbytes;
double bidibytes = xbytes+rbytes;
double time = stop-start;
std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
}
{
double start=usecond();
for(int i=0;i<Nloop;i++){
for(int mu=0;mu<4;mu++){
if (mpi_layout[mu]>1 ) {
Grid.SendToRecvFromBegin(requests_fwd[mu]);
Grid.SendToRecvFromBegin(requests_bwd[mu]);
Grid.SendToRecvFromComplete(requests_fwd[mu]);
Grid.SendToRecvFromComplete(requests_bwd[mu]);
}
}
Grid.Barrier();
}
double stop=usecond();
double dbytes = bytes;
double xbytes = Nloop*dbytes*2.0*ncomm;
double rbytes = xbytes;
double bidibytes = xbytes+rbytes;
double time = stop-start;
std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
}
}
}
Grid_finalize();
}

16
lib/cartesian/Cartesian_red_black.h
View File

@ -32,17 +32,11 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
namespace Grid {
static const int CbRed =0;
static const int CbBlack=1;
static const int Even =CbRed;
static const int Odd =CbBlack;
// Perhaps these are misplaced and
// should be in sparse matrix.
// Also should make these a named enum type
static const int DaggerNo=0;
static const int DaggerYes=1;
static const int CbRed =0;
static const int CbBlack=1;
static const int Even =CbRed;
static const int Odd =CbBlack;
// Specialise this for red black grids storing half the data like a chess board.
class GridRedBlackCartesian : public GridBase
{

9
lib/communicator/Communicator_base.h
View File

@ -127,12 +127,21 @@ class CartesianCommunicator {
int recv_from_rank,
int bytes);
void SendToRecvFromInit(std::vector<CommsRequest_t> &list,
void *xmit,
int xmit_to_rank,
void *recv,
int recv_from_rank,
int bytes);
void SendToRecvFromBegin(std::vector<CommsRequest_t> &list,
void *xmit,
int xmit_to_rank,
void *recv,
int recv_from_rank,
int bytes);
void SendToRecvFromBegin(std::vector<CommsRequest_t> &list);
void SendToRecvFromComplete(std::vector<CommsRequest_t> &waitall);
////////////////////////////////////////////////////////////

35
lib/communicator/Communicator_mpi.cc
View File

@ -144,6 +144,28 @@ void CartesianCommunicator::SendRecvPacket(void *xmit,
}
// Basic Halo comms primitive
// Basic Halo comms primitive
void CartesianCommunicator::SendToRecvFromInit(std::vector<CommsRequest_t> &list,
void *xmit,
int dest,
void *recv,
int from,
int bytes)
{
MPI_Request xrq;
MPI_Request rrq;
int rank = _processor;
int ierr;
ierr =MPI_Send_init(xmit, bytes, MPI_CHAR,dest,_processor,communicator,&xrq);
ierr|=MPI_Recv_init(recv, bytes, MPI_CHAR,dest,_processor,communicator,&rrq);
assert(ierr==0);
list.push_back(xrq);
list.push_back(rrq);
}
void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &list)
{
MPI_Startall(list.size(),&list[0]);
}
void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &list,
void *xmit,
int dest,
@ -151,17 +173,8 @@ void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &lis
int from,
int bytes)
{
MPI_Request xrq;
MPI_Request rrq;
int rank = _processor;
int ierr;
ierr =MPI_Isend(xmit, bytes, MPI_CHAR,dest,_processor,communicator,&xrq);
ierr|=MPI_Irecv(recv, bytes, MPI_CHAR,from,from,communicator,&rrq);
assert(ierr==0);
list.push_back(xrq);
list.push_back(rrq);
SendToRecvFromInit(list,xmit,dest,recv,from,bytes);
SendToRecvFromBegin(list);
}
void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
{

13
lib/communicator/Communicator_none.cc
View File

@ -84,6 +84,19 @@ void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &lis
{
assert(0);
}
void CartesianCommunicator::SendToRecvFromInit(std::vector<CommsRequest_t> &list,
void *xmit,
int dest,
void *recv,
int from,
int bytes)
{
assert(0);
}
void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &list)
{
assert(0);
}
void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
{
assert(0);

13
lib/communicator/Communicator_shmem.cc
View File

@ -268,6 +268,10 @@ void CartesianCommunicator::SendRecvPacket(void *xmit,
}
// Basic Halo comms primitive
void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &list)
{
assert(0); //unimplemented
}
void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &list,
void *xmit,
int dest,
@ -280,6 +284,15 @@ void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &lis
// shmem_putmem_nb(recv,xmit,bytes,dest,NULL);
shmem_putmem(recv,xmit,bytes,dest);
}
void CartesianCommunicator::SendToRecvFromInit(std::vector<CommsRequest_t> &list,
void *xmit,
int dest,
void *recv,
int from,
int bytes)
{
assert(0); // Unimplemented
}
void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
{
// shmem_quiet(); // I'm done

1
lib/lattice/Lattice_base.h
View File

@ -101,6 +101,7 @@ public:
int begin(void) { return 0;};
int end(void) { return _odata.size(); }
vobj & operator[](int i) { return _odata[i]; };
const vobj & operator[](int i) const { return _odata[i]; };
public:
typedef typename vobj::scalar_type scalar_type;

6
lib/qcd/QCD.h
View File

@ -60,6 +60,12 @@ namespace QCD {
static const int SpinIndex = 1;
static const int LorentzIndex= 0;
// Also should make these a named enum type
static const int DaggerNo=0;
static const int DaggerYes=1;
static const int InverseNo=0;
static const int InverseYes=1;
// Useful traits is this a spin index
//typename std::enable_if<matchGridTensorIndex<iVector<vtype,Ns>,SpinorIndex>::value,iVector<vtype,Ns> >::type *SFINAE;

214
lib/qcd/action/fermion/CayleyFermion5D.cc
View File

@ -28,7 +28,11 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/Eigen/Dense>
#include <Grid.h>
namespace Grid {
namespace QCD {
@ -95,22 +99,6 @@ namespace QCD {
FermionField Din(psi._grid);
// Assemble Din
/*
for(int s=0;s<Ls;s++){
if ( s==0 ) {
// Din = bs psi[s] + cs[s] psi[s+1}
axpby_ssp_pminus(Din,bs[s],psi,cs[s],psi,s,s+1);
// Din+= -mass*cs[s] psi[s+1}
axpby_ssp_pplus (Din,1.0,Din,-mass*cs[s],psi,s,Ls-1);
} else if ( s==(Ls-1)) {
axpby_ssp_pminus(Din,bs[s],psi,-mass*cs[s],psi,s,0);
axpby_ssp_pplus (Din,1.0,Din,cs[s],psi,s,s-1);
} else {
axpby_ssp_pminus(Din,bs[s],psi,cs[s],psi,s,s+1);
axpby_ssp_pplus(Din,1.0,Din,cs[s],psi,s,s-1);
}
}
*/
Meooe5D(psi,Din);
this->DW(Din,chi,DaggerNo);
@ -152,18 +140,6 @@ namespace QCD {
// Collect the terms in DW
// Chi = bs Din[s] + cs[s] Din[s+1}
// Chi+= -mass*cs[s] psi[s+1}
/*
if ( s==0 ) {
axpby_ssp_pplus (chi,bs[s],Din,cs[s+1],Din,s,s+1);
axpby_ssp_pminus(chi,1.0,chi,-mass*cs[Ls-1],Din,s,Ls-1);
} else if ( s==(Ls-1)) {
axpby_ssp_pplus (chi,bs[s],Din,-mass*cs[0],Din,s,0);
axpby_ssp_pminus(chi,1.0,chi,cs[s-1],Din,s,s-1);
} else {
axpby_ssp_pplus (chi,bs[s],Din,cs[s+1],Din,s,s+1);
axpby_ssp_pminus(chi,1.0,chi,cs[s-1],Din,s,s-1);
}
*/
// FIXME just call MooeeDag??
@ -193,24 +169,6 @@ namespace QCD {
FermionField tmp(psi._grid);
// Assemble the 5d matrix
Meooe5D(psi,tmp);
#if 0
std::cout << "Meooe Test replacement norm2 tmp = " <<norm2(tmp)<<std::endl;
for(int s=0;s<Ls;s++){
if ( s==0 ) {
// tmp = bs psi[s] + cs[s] psi[s+1}
// tmp+= -mass*cs[s] psi[s+1}
axpby_ssp_pminus(tmp,beo[s],psi,-ceo[s],psi ,s, s+1);
axpby_ssp_pplus(tmp,1.0,tmp,mass*ceo[s],psi,s,Ls-1);
} else if ( s==(Ls-1)) {
axpby_ssp_pminus(tmp,beo[s],psi,mass*ceo[s],psi,s,0);
axpby_ssp_pplus(tmp,1.0,tmp,-ceo[s],psi,s,s-1);
} else {
axpby_ssp_pminus(tmp,beo[s],psi,-ceo[s],psi,s,s+1);
axpby_ssp_pplus (tmp,1.0,tmp,-ceo[s],psi,s,s-1);
}
}
std::cout << "Meooe Test replacement norm2 tmp old = " <<norm2(tmp)<<std::endl;
#endif
// Apply 4d dslash
if ( psi.checkerboard == Odd ) {
@ -232,28 +190,10 @@ namespace QCD {
}
MeooeDag5D(tmp,chi);
#if 0
std::cout << "Meooe Test replacement norm2 chi new = " <<norm2(chi)<<std::endl;
// Assemble the 5d matrix
int Ls=this->Ls;
for(int s=0;s<Ls;s++){
if ( s==0 ) {
axpby_ssp_pplus(chi,beo[s],tmp, -ceo[s+1] ,tmp,s,s+1);
axpby_ssp_pminus(chi, 1.0,chi,mass*ceo[Ls-1],tmp,s,Ls-1);
} else if ( s==(Ls-1)) {
axpby_ssp_pplus(chi,beo[s],tmp,mass*ceo[0],tmp,s,0);
axpby_ssp_pminus(chi,1.0,chi,-ceo[s-1],tmp,s,s-1);
} else {
axpby_ssp_pplus(chi,beo[s],tmp,-ceo[s+1],tmp,s,s+1);
axpby_ssp_pminus(chi,1.0 ,chi,-ceo[s-1],tmp,s,s-1);
}
}
std::cout << "Meooe Test replacement norm2 chi old = " <<norm2(chi)<<std::endl;
#endif
}
template<class Impl>
template<class Impl>
void CayleyFermion5D<Impl>::Mooee (const FermionField &psi, FermionField &chi)
{
int Ls=this->Ls;
@ -269,6 +209,10 @@ namespace QCD {
axpby_ssp_pplus (chi,1.0,chi,-cee[s],psi,s,s-1);
}
}
FermionField temp(psi._grid);
this->MooeeDenseInternal(psi,temp,DaggerNo,InverseNo);
temp = temp - chi;
std::cout << "Difference between dense and normal "<< norm2(temp) <<std::endl;
}
template<class Impl>
@ -294,7 +238,7 @@ namespace QCD {
this->DhopDir(tmp,chi,dir,disp);
}
template<class Impl>
template<class Impl>
void CayleyFermion5D<Impl>::MooeeDag (const FermionField &psi, FermionField &chi)
{
int Ls=this->Ls;
@ -311,11 +255,139 @@ namespace QCD {
axpby_ssp_pminus(chi,1.0 ,chi,-cee[s-1],psi,s,s-1);
}
}
FermionField temp(psi._grid);
this->MooeeDenseInternal(psi,temp,DaggerYes,InverseNo);
temp = temp - chi;
std::cout << "Difference between dense and normal "<< norm2(temp) <<std::endl;
}
template<class Impl>
template<class Impl>
void CayleyFermion5D<Impl>::MooeeInv (const FermionField &psi, FermionField &chi)
{
FermionField temp(psi._grid);
this->MooeeLDUInv(psi,chi);
this->MooeeDenseInv(psi,temp);
temp = temp - chi;
std::cout << "Difference between dense and normal "<< norm2(temp) <<std::endl;
}
template<class Impl>
void CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &chi)
{
FermionField temp(psi._grid);
this->MooeeLDUInvDag(psi,chi);
this->MooeeDenseInvDag(psi,temp);
temp = temp - chi;
std::cout << "Difference between dense and normal "<< norm2(temp) <<std::endl;
}
template<class Impl>
void CayleyFermion5D<Impl>::MooeeDenseInvDag (const FermionField &psi, FermionField &chi)
{
this->MooeeDenseInternal(psi,chi,DaggerYes,InverseYes);
}
template<class Impl>
void CayleyFermion5D<Impl>::MooeeDenseInv(const FermionField &psi, FermionField &chi)
{
this->MooeeDenseInternal(psi,chi,DaggerNo,InverseYes);
}
template<class Impl>
void CayleyFermion5D<Impl>::MooeeDenseInternal(const FermionField &psi, FermionField &chi,int dag, int inv)
{
int Ls=this->Ls;
int LLs = psi._grid->_rdimensions[0];
int vol = psi._grid->oSites()/LLs;
chi.checkerboard=psi.checkerboard;
assert(Ls==LLs);
Eigen::MatrixXd Pplus = Eigen::MatrixXd::Zero(Ls,Ls);
Eigen::MatrixXd Pminus = Eigen::MatrixXd::Zero(Ls,Ls);
for(int s=0;s<Ls;s++){
Pplus(s,s) = bee[s];
Pminus(s,s)= bee[s];
}
for(int s=0;s<Ls-1;s++){
Pminus(s,s+1) = -cee[s];
}
for(int s=0;s<Ls-1;s++){
Pplus(s+1,s) = -cee[s];
}
std::cout << " Pplus "<<Pplus<<std::endl;
std::cout << " Pminus "<<Pminus<<std::endl;
Pplus (0,Ls-1) = mass*cee[0];
Pminus(Ls-1,0) = mass*cee[Ls-1];
Eigen::MatrixXd PplusMat ;
Eigen::MatrixXd PminusMat;
if ( inv ) {
PplusMat =Pplus.inverse();
PminusMat=Pminus.inverse();
} else {
PplusMat =Pplus;
PminusMat=Pminus;
}
if(dag){
PplusMat.adjointInPlace();
PminusMat.adjointInPlace();
}
// For the non-vectorised s-direction this is simple
PARALLEL_FOR_LOOP
for(auto site=0;site<vol;site++){
SiteSpinor SiteChi;
SiteHalfSpinor SitePplus;
SiteHalfSpinor SitePminus;
for(int s1=0;s1<Ls;s1++){
SiteChi =zero;
for(int s2=0;s2<Ls;s2++){
int lex2 = s2+Ls*site;
if ( PplusMat(s1,s2) != 0.0 ) {
spProj5p(SitePplus,psi[lex2]);
accumRecon5p(SiteChi,PplusMat (s1,s2)*SitePplus);
}
if ( PminusMat(s1,s2) != 0.0 ) {
spProj5m(SitePminus,psi[lex2]);
accumRecon5m(SiteChi,PminusMat(s1,s2)*SitePminus);
}
}
chi[s1+Ls*site] = SiteChi;
}
}
// For the non-vectorised s-direction we have more work to do in an alternate implementation
//
// ai = Mij bj
//
// where b == {b0} {b1}..
//
// Best to load/bcast b0
//
// ai = Mi0 b0
// for(int i=0;i<Ls;i++){
// ai += Mij bj
// }
//
// For rb5d shamir DWF, the Moo is proportional to the identity
// For rb4d Moo, we need to set a vector of coeffs, and use a stencil type construct, which is cheap.
}
template<class Impl>
void CayleyFermion5D<Impl>::MooeeLDUInv (const FermionField &psi, FermionField &chi)
{
chi.checkerboard=psi.checkerboard;
int Ls=this->Ls;
// Apply (L^{\prime})^{-1}
axpby_ssp (chi,1.0,psi, 0.0,psi,0,0); // chi[0]=psi[0]
@ -340,8 +412,9 @@ namespace QCD {
}
template<class Impl>
void CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &chi)
void CayleyFermion5D<Impl>::MooeeLDUInvDag (const FermionField &psi, FermionField &chi)
{
chi.checkerboard=psi.checkerboard;
int Ls=this->Ls;
// Apply (U^{\prime})^{-dagger}
axpby_ssp (chi,1.0,psi, 0.0,psi,0,0); // chi[0]=psi[0]
@ -364,6 +437,7 @@ namespace QCD {
}
}
// force terms; five routines; default to Dhop on diagonal
template<class Impl>
void CayleyFermion5D<Impl>::MDeriv (GaugeField &mat,const FermionField &U,const FermionField &V,int dag)
@ -416,7 +490,6 @@ namespace QCD {
void CayleyFermion5D<Impl>::SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD b,RealD c)
{
SetCoefficientsZolotarev(1.0,zdata,b,c);
}
//Zolo
template<class Impl>
@ -524,8 +597,11 @@ namespace QCD {
for(int j=0;j<Ls-1;j++) delta_d *= cee[j]/bee[j];
dee[Ls-1] += delta_d;
}
}
FermOpTemplateInstantiate(CayleyFermion5D);
GparityFermOpTemplateInstantiate(CayleyFermion5D);

6
lib/qcd/action/fermion/CayleyFermion5D.h
View File

@ -51,6 +51,11 @@ namespace Grid {
virtual void MooeeDag (const FermionField &in, FermionField &out);
virtual void MooeeInv (const FermionField &in, FermionField &out);
virtual void MooeeInvDag (const FermionField &in, FermionField &out);
virtual void MooeeLDUInv (const FermionField &in, FermionField &out);
virtual void MooeeLDUInvDag (const FermionField &in, FermionField &out);
virtual void MooeeDenseInv (const FermionField &in, FermionField &out);
virtual void MooeeDenseInvDag (const FermionField &in, FermionField &out);
void MooeeDenseInternal(const FermionField &in, FermionField &out,int dag,int inv);
virtual void Instantiatable(void)=0;
// force terms; five routines; default to Dhop on diagonal
@ -94,6 +99,7 @@ namespace Grid {
GridRedBlackCartesian &FourDimRedBlackGrid,
RealD _mass,RealD _M5,const ImplParams &p= ImplParams());
protected:
void SetCoefficientsZolotarev(RealD zolohi,Approx::zolotarev_data *zdata,RealD b,RealD c);
void SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD b,RealD c);

4
prerequisites/Makefile.am
View File

@ -20,10 +20,10 @@ $(top_srcdir)/prerequisites/$(FFTWTAR):
Eigen: $(top_srcdir)/prerequisites/$(EIGENTAR)
tar xvf $(top_srcdir)/prerequisites/$(EIGENTAR)
- rm -rf ../include/Grid/Eigen
- rm -rf $(top_srcdir)/lib/Eigen
mv eigen-eigen*/Eigen .
echo EFILES=`find Eigen -type f -name '*.h' ` > $(top_srcdir)/lib/Eigen.inc
mv Eigen ../include/Grid/
mv Eigen $(top_srcdir)/lib/
touch Eigen
FFTW: $(top_srcdir)/prerequisites/$(FFTWTAR)

4
tests/Makefile.am
View File

@ -1,9 +1,9 @@
# additional include paths necessary to compile the C++ library
SUBDIRS = core
#SUBDIRS = core
# Uncomment to enable complete test suite build
#SUBDIRS = core forces hmc solver debug
SUBDIRS = core forces hmc solver debug
if BUILD_CHROMA_REGRESSION
SUBDIRS+= qdpxx