2018-01-13 00:17:16 +00:00
|
|
|
/*************************************************************************************
|
2016-01-02 14:51:32 +00:00
|
|
|
|
|
|
|
Grid physics library, www.github.com/paboyle/Grid
|
|
|
|
|
|
|
|
Source file: ./lib/communicator/Communicator_none.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
|
2018-01-13 00:17:16 +00:00
|
|
|
*************************************************************************************/
|
|
|
|
/* END LEGAL */
|
2017-02-22 18:09:33 +00:00
|
|
|
#include <Grid/GridCore.h>
|
|
|
|
|
2018-01-13 00:16:34 +00:00
|
|
|
NAMESPACE_BEGIN(Grid);
|
2015-03-29 20:35:37 +01:00
|
|
|
|
2016-10-24 17:30:43 +01:00
|
|
|
///////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
// Info that is setup once and indept of cartesian layout
|
|
|
|
///////////////////////////////////////////////////////////////////////////////////////////////////
|
2018-01-08 11:33:47 +00:00
|
|
|
Grid_MPI_Comm CartesianCommunicator::communicator_world;
|
2016-02-14 20:24:38 +00:00
|
|
|
|
2016-10-24 17:30:43 +01:00
|
|
|
void CartesianCommunicator::Init(int *argc, char *** arv)
|
2016-10-22 18:14:27 +01:00
|
|
|
{
|
2018-01-08 11:33:47 +00:00
|
|
|
GlobalSharedMemory::Init(communicator_world);
|
|
|
|
GlobalSharedMemory::SharedMemoryAllocate(
|
2018-01-13 00:17:16 +00:00
|
|
|
GlobalSharedMemory::MAX_MPI_SHM_BYTES,
|
|
|
|
GlobalSharedMemory::Hugepages);
|
2016-10-22 18:14:27 +01:00
|
|
|
}
|
2016-02-14 20:24:38 +00:00
|
|
|
|
2017-10-30 01:14:11 +00:00
|
|
|
CartesianCommunicator::CartesianCommunicator(const std::vector<int> &processors,const CartesianCommunicator &parent,int &srank)
|
2018-01-08 11:33:47 +00:00
|
|
|
: CartesianCommunicator(processors)
|
|
|
|
{
|
|
|
|
srank=0;
|
|
|
|
SetCommunicator(communicator_world);
|
|
|
|
}
|
2017-06-22 08:14:34 +01:00
|
|
|
|
2015-06-02 16:57:12 +01:00
|
|
|
CartesianCommunicator::CartesianCommunicator(const std::vector<int> &processors)
|
2015-03-29 20:35:37 +01:00
|
|
|
{
|
|
|
|
_processors = processors;
|
2015-05-10 15:23:09 +01:00
|
|
|
_ndimension = processors.size();
|
|
|
|
_processor_coor.resize(_ndimension);
|
2015-03-29 20:35:37 +01:00
|
|
|
|
|
|
|
// Require 1^N processor grid for fake
|
2015-05-10 15:23:09 +01:00
|
|
|
_Nprocessors=1;
|
|
|
|
_processor = 0;
|
|
|
|
for(int d=0;d<_ndimension;d++) {
|
|
|
|
assert(_processors[d]==1);
|
|
|
|
_processor_coor[d] = 0;
|
|
|
|
}
|
2018-01-08 11:33:47 +00:00
|
|
|
SetCommunicator(communicator_world);
|
2015-03-29 20:35:37 +01:00
|
|
|
}
|
|
|
|
|
2017-10-25 10:24:14 +01:00
|
|
|
CartesianCommunicator::~CartesianCommunicator(){}
|
|
|
|
|
2015-04-10 04:21:48 +01:00
|
|
|
void CartesianCommunicator::GlobalSum(float &){}
|
|
|
|
void CartesianCommunicator::GlobalSumVector(float *,int N){}
|
|
|
|
void CartesianCommunicator::GlobalSum(double &){}
|
2015-04-22 22:46:48 +01:00
|
|
|
void CartesianCommunicator::GlobalSum(uint32_t &){}
|
2016-03-16 09:27:22 +00:00
|
|
|
void CartesianCommunicator::GlobalSum(uint64_t &){}
|
2015-04-10 04:21:48 +01:00
|
|
|
void CartesianCommunicator::GlobalSumVector(double *,int N){}
|
2017-06-11 23:14:10 +01:00
|
|
|
void CartesianCommunicator::GlobalXOR(uint32_t &){}
|
|
|
|
void CartesianCommunicator::GlobalXOR(uint64_t &){}
|
2015-03-29 20:35:37 +01:00
|
|
|
|
2016-03-03 20:33:28 +00:00
|
|
|
void CartesianCommunicator::SendRecvPacket(void *xmit,
|
|
|
|
void *recv,
|
|
|
|
int xmit_to_rank,
|
|
|
|
int recv_from_rank,
|
|
|
|
int bytes)
|
Binary IO file for generic Grid array parallel I/O.
Number of IO MPI tasks can be varied by selecting which
dimensions use parallel IO and which dimensions use Serial send to boss
I/O.
Thus can neck down from, say 1024 nodes = 4x4x8x8 to {1,8,32,64,128,256,1024} nodes
doing the I/O.
Interpolates nicely between ALL nodes write their data, a single boss per time-plane
in processor space [old UKQCD fortran code did this], and a single node doing all I/O.
Not sure I have the transfer sizes big enough and am not overly convinced fstream
is guaranteed to not give buffer inconsistencies unless I set streambuf size to zero.
Practically it has worked on 8 tasks, 2x1x2x2 writing /cloning NERSC configurations
on my MacOS + OpenMPI and Clang environment.
It is VERY easy to switch to pwrite at a later date, and also easy to send x-strips around from
each node in order to gather bigger chunks at the syscall level.
That would push us up to the circa 8x 18*4*8 == 4KB size write chunk, and by taking, say, x/y non
parallel we get to 16MB contiguous chunks written in multi 4KB transactions
per IOnode in 64^3 lattices for configuration I/O.
I suspect this is fine for system performance.
2015-08-26 13:40:29 +01:00
|
|
|
{
|
|
|
|
assert(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2015-05-02 23:42:30 +01:00
|
|
|
// Basic Halo comms primitive -- should never call in single node
|
2015-03-29 20:35:37 +01:00
|
|
|
void CartesianCommunicator::SendToRecvFrom(void *xmit,
|
2015-04-10 04:21:48 +01:00
|
|
|
int dest,
|
|
|
|
void *recv,
|
|
|
|
int from,
|
|
|
|
int bytes)
|
2015-03-29 20:35:37 +01:00
|
|
|
{
|
2015-05-11 12:43:10 +01:00
|
|
|
assert(0);
|
2015-03-29 20:35:37 +01:00
|
|
|
}
|
2015-05-02 23:42:30 +01:00
|
|
|
void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &list,
|
|
|
|
void *xmit,
|
|
|
|
int dest,
|
|
|
|
void *recv,
|
|
|
|
int from,
|
|
|
|
int bytes)
|
|
|
|
{
|
2015-05-11 12:43:10 +01:00
|
|
|
assert(0);
|
2015-05-02 23:42:30 +01:00
|
|
|
}
|
2016-11-22 13:49:11 +00:00
|
|
|
|
2015-05-02 23:42:30 +01:00
|
|
|
void CartesianCommunicator::SendToRecvFromComplete(std::vector<CommsRequest_t> &list)
|
|
|
|
{
|
2015-05-11 12:43:10 +01:00
|
|
|
assert(0);
|
2015-05-02 23:42:30 +01:00
|
|
|
}
|
2017-10-27 14:20:35 +01:00
|
|
|
void CartesianCommunicator::AllToAll(int dim,void *in,void *out,uint64_t words,uint64_t bytes)
|
2017-10-09 23:16:51 +01:00
|
|
|
{
|
2017-10-27 14:20:35 +01:00
|
|
|
bcopy(in,out,bytes*words);
|
|
|
|
}
|
|
|
|
void CartesianCommunicator::AllToAll(void *in,void *out,uint64_t words,uint64_t bytes)
|
|
|
|
{
|
|
|
|
bcopy(in,out,bytes*words);
|
2017-10-09 23:16:51 +01:00
|
|
|
}
|
2015-03-29 20:35:37 +01:00
|
|
|
|
2016-11-01 11:35:43 +00:00
|
|
|
int CartesianCommunicator::RankWorld(void){return 0;}
|
2016-10-24 17:30:43 +01:00
|
|
|
void CartesianCommunicator::Barrier(void){}
|
|
|
|
void CartesianCommunicator::Broadcast(int root,void* data, int bytes) {}
|
|
|
|
void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes) { }
|
|
|
|
int CartesianCommunicator::RankFromProcessorCoor(std::vector<int> &coor) { return 0;}
|
2016-11-03 16:58:23 +00:00
|
|
|
void CartesianCommunicator::ProcessorCoorFromRank(int rank, std::vector<int> &coor){ coor = _processor_coor; }
|
2015-04-10 04:21:48 +01:00
|
|
|
void CartesianCommunicator::ShiftedRanks(int dim,int shift,int &source,int &dest)
|
|
|
|
{
|
2015-05-10 15:23:09 +01:00
|
|
|
source =0;
|
|
|
|
dest=0;
|
2015-04-10 04:21:48 +01:00
|
|
|
}
|
|
|
|
|
2018-01-08 11:33:47 +00:00
|
|
|
double CartesianCommunicator::StencilSendToRecvFrom( void *xmit,
|
|
|
|
int xmit_to_rank,
|
|
|
|
void *recv,
|
|
|
|
int recv_from_rank,
|
|
|
|
int bytes, int dir)
|
|
|
|
{
|
|
|
|
std::vector<CommsRequest_t> list;
|
|
|
|
// Discard the "dir"
|
|
|
|
SendToRecvFromBegin (list,xmit,xmit_to_rank,recv,recv_from_rank,bytes);
|
|
|
|
SendToRecvFromComplete(list);
|
|
|
|
return 2.0*bytes;
|
|
|
|
}
|
|
|
|
double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
|
|
|
|
void *xmit,
|
|
|
|
int xmit_to_rank,
|
|
|
|
void *recv,
|
|
|
|
int recv_from_rank,
|
|
|
|
int bytes, int dir)
|
|
|
|
{
|
|
|
|
// Discard the "dir"
|
|
|
|
SendToRecvFromBegin(list,xmit,xmit_to_rank,recv,recv_from_rank,bytes);
|
|
|
|
return 2.0*bytes;
|
|
|
|
}
|
|
|
|
void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &waitall,int dir)
|
|
|
|
{
|
|
|
|
SendToRecvFromComplete(waitall);
|
|
|
|
}
|
|
|
|
|
|
|
|
void CartesianCommunicator::StencilBarrier(void){};
|
|
|
|
|
2018-01-13 00:16:34 +00:00
|
|
|
NAMESPACE_END(Grid);
|
2015-04-10 04:21:48 +01:00
|
|
|
|
2015-03-29 20:35:37 +01:00
|
|
|
|