1
0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-14 17:55:38 +00:00
Grid/lib/communicator/Communicator_mpi3.cc
2016-10-25 06:01:12 +01:00

575 lines
17 KiB
C++

/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/communicator/Communicator_mpi.cc
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.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.h"
#include <mpi.h>
namespace Grid {
///////////////////////////////////////////////////////////////////////////////////////////////////
// Info that is setup once and indept of cartesian layout
///////////////////////////////////////////////////////////////////////////////////////////////////
int CartesianCommunicator::ShmSetup = 0;
MPI_Comm CartesianCommunicator::communicator_world;
MPI_Comm CartesianCommunicator::ShmComm;
MPI_Win CartesianCommunicator::ShmWindow;
std::vector<int> CartesianCommunicator::GroupRanks;
std::vector<int> CartesianCommunicator::MyGroup;
std::vector<void *> CartesianCommunicator::ShmCommBufs;
void *CartesianCommunicator::ShmBufferSelf(void)
{
return ShmCommBufs[ShmRank];
}
void *CartesianCommunicator::ShmBuffer(int rank)
{
int gpeer = GroupRanks[rank];
if (gpeer == MPI_UNDEFINED){
return NULL;
} else {
return ShmCommBufs[gpeer];
}
}
void *CartesianCommunicator::ShmBufferTranslate(int rank,void * local_p)
{
int gpeer = GroupRanks[rank];
if (gpeer == MPI_UNDEFINED){
return NULL;
} else {
uint64_t offset = (uint64_t)local_p - (uint64_t)ShmCommBufs[ShmRank];
uint64_t remote = (uint64_t)ShmCommBufs[gpeer]+offset;
return (void *) remote;
}
}
void CartesianCommunicator::Init(int *argc, char ***argv) {
int flag;
MPI_Initialized(&flag); // needed to coexist with other libs apparently
if ( !flag ) {
MPI_Init(argc,argv);
}
MPI_Comm_dup (MPI_COMM_WORLD,&communicator_world);
MPI_Comm_rank(communicator_world,&WorldRank);
MPI_Comm_size(communicator_world,&WorldSize);
/////////////////////////////////////////////////////////////////////
// Split into groups that can share memory
/////////////////////////////////////////////////////////////////////
MPI_Comm_split_type(communicator_world, MPI_COMM_TYPE_SHARED, 0, MPI_INFO_NULL,&ShmComm);
MPI_Comm_rank(ShmComm ,&ShmRank);
MPI_Comm_size(ShmComm ,&ShmSize);
GroupSize = WorldSize/ShmSize;
/////////////////////////////////////////////////////////////////////
// find world ranks in our SHM group (i.e. which ranks are on our node)
/////////////////////////////////////////////////////////////////////
MPI_Group WorldGroup, ShmGroup;
MPI_Comm_group (communicator_world, &WorldGroup);
MPI_Comm_group (ShmComm, &ShmGroup);
std::vector<int> world_ranks(WorldSize);
GroupRanks.resize(WorldSize);
MyGroup.resize(ShmSize);
for(int r=0;r<WorldSize;r++) world_ranks[r]=r;
MPI_Group_translate_ranks (WorldGroup,WorldSize,&world_ranks[0],ShmGroup, &GroupRanks[0]);
///////////////////////////////////////////////////////////////////
// Identify who is in my group and noninate the leader
///////////////////////////////////////////////////////////////////
int g=0;
for(int rank=0;rank<WorldSize;rank++){
if(GroupRanks[rank]!=MPI_UNDEFINED){
assert(g<ShmSize);
MyGroup[g++] = rank;
}
}
std::sort(MyGroup.begin(),MyGroup.end(),std::less<int>());
int myleader = MyGroup[0];
std::vector<int> leaders_1hot(WorldSize,0);
std::vector<int> leaders_group(GroupSize,0);
leaders_1hot [ myleader ] = 1;
///////////////////////////////////////////////////////////////////
// global sum leaders over comm world
///////////////////////////////////////////////////////////////////
int ierr=MPI_Allreduce(MPI_IN_PLACE,&leaders_1hot[0],WorldSize,MPI_INT,MPI_SUM,communicator_world);
assert(ierr==0);
///////////////////////////////////////////////////////////////////
// find the group leaders world rank
///////////////////////////////////////////////////////////////////
int group=0;
for(int l=0;l<WorldSize;l++){
if(leaders_1hot[l]){
leaders_group[group++] = l;
}
}
///////////////////////////////////////////////////////////////////
// Identify the rank of the group in which I (and my leader) live
///////////////////////////////////////////////////////////////////
GroupRank=-1;
for(int g=0;g<GroupSize;g++){
if (myleader == leaders_group[g]){
GroupRank=g;
}
}
assert(GroupRank!=-1);
//////////////////////////////////////////////////////////////////////////////////////////////////////////
// allocate the shared window for our group
//////////////////////////////////////////////////////////////////////////////////////////////////////////
ShmCommBuf = 0;
ierr = MPI_Win_allocate_shared(MAX_MPI_SHM_BYTES,1,MPI_INFO_NULL,ShmComm,&ShmCommBuf,&ShmWindow);
assert(ierr==0);
// KNL hack -- force to numa-domain 1 in flat
#if 0
//#include <numaif.h>
for(uint64_t page=0;page<MAX_MPI_SHM_BYTES;page+=4096){
void *pages = (void *) ( page + ShmCommBuf );
int status;
int flags=MPOL_MF_MOVE_ALL;
int nodes=1; // numa domain == MCDRAM
unsigned long count=1;
ierr= move_pages(0,count, &pages,&nodes,&status,flags);
if (ierr && (page==0)) perror("numa relocate command failed");
}
#endif
MPI_Win_lock_all (MPI_MODE_NOCHECK, ShmWindow);
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Plan: allocate a fixed SHM region. Scratch that is just used via some scheme during stencil comms, with no allocate free.
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
ShmCommBufs.resize(ShmSize);
for(int r=0;r<ShmSize;r++){
MPI_Aint sz;
int dsp_unit;
MPI_Win_shared_query (ShmWindow, r, &sz, &dsp_unit, &ShmCommBufs[r]);
}
//////////////////////////////////////////////////////////////////////////////////////////////////////////
// Verbose for now
//////////////////////////////////////////////////////////////////////////////////////////////////////////
if (WorldRank == 0){
std::cout<<GridLogMessage<< "Grid MPI-3 configuration: detected ";
std::cout<< WorldSize << " Ranks " ;
std::cout<< GroupSize << " Nodes " ;
std::cout<< ShmSize << " with ranks-per-node "<<std::endl;
std::cout<<GridLogMessage <<"Grid MPI-3 configuration: allocated shared memory region of size ";
std::cout<<std::hex << MAX_MPI_SHM_BYTES <<" ShmCommBuf address = "<<ShmCommBuf << std::dec<<std::endl;
for(int g=0;g<GroupSize;g++){
std::cout<<GridLogMessage<<" Node "<<g<<" led by MPI rank "<<leaders_group[g]<<std::endl;
}
std::cout<<GridLogMessage<<" Boss Node Shm Pointers are {";
for(int g=0;g<ShmSize;g++){
std::cout<<std::hex<<ShmCommBufs[g]<<std::dec;
if(g!=ShmSize-1) std::cout<<",";
else std::cout<<"}"<<std::endl;
}
}
for(int g=0;g<GroupSize;g++){
if ( (ShmRank == 0) && (GroupRank==g) ) std::cout<<GridLogMessage<<"["<<g<<"] Node Group "<<g<<" is ranks {";
for(int r=0;r<ShmSize;r++){
if ( (ShmRank == 0) && (GroupRank==g) ) {
std::cout<<MyGroup[r];
if(r<ShmSize-1) std::cout<<",";
else std::cout<<"}"<<std::endl;
}
MPI_Barrier(communicator_world);
}
}
assert(ShmSetup==0); ShmSetup=1;
}
////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Want to implement some magic ... Group sub-cubes into those on same node
////////////////////////////////////////////////////////////////////////////////////////////////////////////
void CartesianCommunicator::ShiftedRanks(int dim,int shift,int &source,int &dest)
{
std::vector<int> coor = _processor_coor;
assert(std::abs(shift) <_processors[dim]);
coor[dim] = (_processor_coor[dim] + shift + _processors[dim])%_processors[dim];
Lexicographic::IndexFromCoor(coor,source,_processors);
source = LexicographicToWorldRank[source];
coor[dim] = (_processor_coor[dim] - shift + _processors[dim])%_processors[dim];
Lexicographic::IndexFromCoor(coor,dest,_processors);
dest = LexicographicToWorldRank[dest];
}
int CartesianCommunicator::RankFromProcessorCoor(std::vector<int> &coor)
{
int rank;
Lexicographic::IndexFromCoor(coor,rank,_processors);
rank = LexicographicToWorldRank[rank];
return rank;
}
void CartesianCommunicator::ProcessorCoorFromRank(int rank, std::vector<int> &coor)
{
Lexicographic::CoorFromIndex(coor,rank,_processors);
rank = LexicographicToWorldRank[rank];
}
CartesianCommunicator::CartesianCommunicator(const std::vector<int> &processors)
{
int ierr;
communicator=communicator_world;
_ndimension = processors.size();
////////////////////////////////////////////////////////////////
// Assert power of two shm_size.
////////////////////////////////////////////////////////////////
int log2size = -1;
for(int i=0;i<=MAXLOG2RANKSPERNODE;i++){
if ( (0x1<<i) == ShmSize ) {
log2size = i;
break;
}
}
assert(log2size != -1);
////////////////////////////////////////////////////////////////
// Identify subblock of ranks on node spreading across dims
// in a maximally symmetrical way
////////////////////////////////////////////////////////////////
int dim = 0;
std::vector<int> WorldDims = processors;
ShmDims.resize(_ndimension,1);
GroupDims.resize(_ndimension);
ShmCoor.resize(_ndimension);
GroupCoor.resize(_ndimension);
WorldCoor.resize(_ndimension);
for(int l2=0;l2<log2size;l2++){
while ( WorldDims[dim] / ShmDims[dim] <= 1 ) dim=(dim+1)%_ndimension;
ShmDims[dim]*=2;
dim=(dim+1)%_ndimension;
}
////////////////////////////////////////////////////////////////
// Establish torus of processes and nodes with sub-blockings
////////////////////////////////////////////////////////////////
for(int d=0;d<_ndimension;d++){
GroupDims[d] = WorldDims[d]/ShmDims[d];
}
////////////////////////////////////////////////////////////////
// Check processor counts match
////////////////////////////////////////////////////////////////
_Nprocessors=1;
_processors = processors;
_processor_coor.resize(_ndimension);
for(int i=0;i<_ndimension;i++){
_Nprocessors*=_processors[i];
}
assert(WorldSize==_Nprocessors);
////////////////////////////////////////////////////////////////
// Establish mapping between lexico physics coord and WorldRank
//
////////////////////////////////////////////////////////////////
LexicographicToWorldRank.resize(WorldSize,0);
Lexicographic::CoorFromIndex(GroupCoor,GroupRank,GroupDims);
Lexicographic::CoorFromIndex(ShmCoor,ShmRank,ShmDims);
for(int d=0;d<_ndimension;d++){
WorldCoor[d] = GroupCoor[d]*ShmDims[d]+ShmCoor[d];
}
_processor_coor = WorldCoor;
int lexico;
Lexicographic::IndexFromCoor(WorldCoor,lexico,WorldDims);
LexicographicToWorldRank[lexico]=WorldRank;
_processor = lexico;
///////////////////////////////////////////////////////////////////
// global sum Lexico to World mapping
///////////////////////////////////////////////////////////////////
ierr=MPI_Allreduce(MPI_IN_PLACE,&LexicographicToWorldRank[0],WorldSize,MPI_INT,MPI_SUM,communicator);
assert(ierr==0);
};
void CartesianCommunicator::GlobalSum(uint32_t &u){
int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT32_T,MPI_SUM,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalSum(uint64_t &u){
int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT64_T,MPI_SUM,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalSum(float &f){
int ierr=MPI_Allreduce(MPI_IN_PLACE,&f,1,MPI_FLOAT,MPI_SUM,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalSumVector(float *f,int N)
{
int ierr=MPI_Allreduce(MPI_IN_PLACE,f,N,MPI_FLOAT,MPI_SUM,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalSum(double &d)
{
int ierr = MPI_Allreduce(MPI_IN_PLACE,&d,1,MPI_DOUBLE,MPI_SUM,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalSumVector(double *d,int N)
{
int ierr = MPI_Allreduce(MPI_IN_PLACE,d,N,MPI_DOUBLE,MPI_SUM,communicator);
assert(ierr==0);
}
// Basic Halo comms primitive
void CartesianCommunicator::SendToRecvFrom(void *xmit,
int dest,
void *recv,
int from,
int bytes)
{
std::vector<CommsRequest_t> reqs(0);
SendToRecvFromBegin(reqs,xmit,dest,recv,from,bytes);
SendToRecvFromComplete(reqs);
}
void CartesianCommunicator::SendRecvPacket(void *xmit,
void *recv,
int sender,
int receiver,
int bytes)
{
MPI_Status stat;
assert(sender != receiver);
int tag = sender;
if ( _processor == sender ) {
MPI_Send(xmit, bytes, MPI_CHAR,receiver,tag,communicator);
}
if ( _processor == receiver ) {
MPI_Recv(recv, bytes, MPI_CHAR,sender,tag,communicator,&stat);
}
}
// Basic Halo comms primitive
void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &list,
void *xmit,
int dest,
void *recv,
int from,
int bytes)
{
#if 0
this->StencilBarrier();
MPI_Request xrq;
MPI_Request rrq;
static int sequence;
int ierr;
int tag;
int check;
assert(dest != _processor);
assert(from != _processor);
int gdest = GroupRanks[dest];
int gfrom = GroupRanks[from];
int gme = GroupRanks[_processor];
sequence++;
char *from_ptr = (char *)ShmCommBufs[ShmRank];
int small = (bytes<MAX_MPI_SHM_BYTES);
typedef uint64_t T;
int words = bytes/sizeof(T);
assert(((size_t)bytes &(sizeof(T)-1))==0);
assert(gme == ShmRank);
if ( small && (gdest !=MPI_UNDEFINED) ) {
char *to_ptr = (char *)ShmCommBufs[gdest];
assert(gme != gdest);
T *ip = (T *)xmit;
T *op = (T *)to_ptr;
PARALLEL_FOR_LOOP
for(int w=0;w<words;w++) {
op[w]=ip[w];
}
bcopy(&_processor,&to_ptr[bytes],sizeof(_processor));
bcopy(& sequence,&to_ptr[bytes+4],sizeof(sequence));
} else {
ierr =MPI_Isend(xmit, bytes, MPI_CHAR,dest,_processor,communicator,&xrq);
assert(ierr==0);
list.push_back(xrq);
}
this->StencilBarrier();
if (small && (gfrom !=MPI_UNDEFINED) ) {
T *ip = (T *)from_ptr;
T *op = (T *)recv;
PARALLEL_FOR_LOOP
for(int w=0;w<words;w++) {
op[w]=ip[w];
}
bcopy(&from_ptr[bytes] ,&tag ,sizeof(tag));
bcopy(&from_ptr[bytes+4],&check,sizeof(check));
assert(check==sequence);
assert(tag==from);
} else {
ierr=MPI_Irecv(recv, bytes, MPI_CHAR,from,from,communicator,&rrq);
assert(ierr==0);
list.push_back(rrq);
}
this->StencilBarrier();
#else
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);
#endif
}
void CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
void *xmit,
int dest,
void *recv,
int from,
int bytes)
{
MPI_Request xrq;
MPI_Request rrq;
int ierr;
assert(dest != _processor);
assert(from != _processor);
int gdest = GroupRanks[dest];
int gfrom = GroupRanks[from];
int gme = GroupRanks[_processor];
assert(gme == ShmRank);
if ( gdest == MPI_UNDEFINED ) {
ierr =MPI_Isend(xmit, bytes, MPI_CHAR,dest,_processor,communicator,&xrq);
assert(ierr==0);
list.push_back(xrq);
}
if ( gfrom ==MPI_UNDEFINED) {
ierr=MPI_Irecv(recv, bytes, MPI_CHAR,from,from,communicator,&rrq);
assert(ierr==0);
list.push_back(rrq);
}
}
void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &list)
{
SendToRecvFromComplete(list);
}
void CartesianCommunicator::StencilBarrier(void)
{
MPI_Win_sync (ShmWindow);
MPI_Barrier (ShmComm);
MPI_Win_sync (ShmWindow);
}
void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
{
int nreq=list.size();
std::vector<MPI_Status> status(nreq);
int ierr = MPI_Waitall(nreq,&list[0],&status[0]);
assert(ierr==0);
}
void CartesianCommunicator::Barrier(void)
{
int ierr = MPI_Barrier(communicator);
assert(ierr==0);
}
void CartesianCommunicator::Broadcast(int root,void* data, int bytes)
{
int ierr=MPI_Bcast(data,
bytes,
MPI_BYTE,
root,
communicator);
assert(ierr==0);
}
void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes)
{
int ierr= MPI_Bcast(data,
bytes,
MPI_BYTE,
root,
communicator_world);
assert(ierr==0);
}
}