mirror of
https://github.com/paboyle/Grid.git
synced 2024-11-10 07:55:35 +00:00
Significant GPU perf speed up finished
This commit is contained in:
Peter Boyle
parent
5dae6a6dac
commit
86e33c8ab2
@ -1 +0,0 @@
|
|||||||
../CayleyFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../ContinuedFractionFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../DomainWallEOFAFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../MobiusEOFAFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../PartialFractionFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonCloverFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonKernelsInstantiationGparity.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonTMFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
#define IMPLEMENTATION GparityWilsonImplDF
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../CayleyFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../ContinuedFractionFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../DomainWallEOFAFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../MobiusEOFAFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../PartialFractionFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonCloverFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonKernelsInstantiationGparity.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonTMFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
#define IMPLEMENTATION GparityWilsonImplFH
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../CayleyFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../ContinuedFractionFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../DomainWallEOFAFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../MobiusEOFAFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../PartialFractionFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonCloverFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonFermionInstantiation.cc.master
|
|
||||||
@ -1,51 +0,0 @@
|
|||||||
/*************************************************************************************
|
|
||||||
|
|
||||||
Grid physics library, www.github.com/paboyle/Grid
|
|
||||||
|
|
||||||
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
|
|
||||||
|
|
||||||
Copyright (C) 2015, 2020
|
|
||||||
|
|
||||||
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
|
||||||
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
|
|
||||||
Author: paboyle <paboyle@ph.ed.ac.uk>
|
|
||||||
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
|
|
||||||
|
|
||||||
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/qcd/action/fermion/FermionCore.h>
|
|
||||||
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
|
|
||||||
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
|
|
||||||
|
|
||||||
#ifndef AVX512
|
|
||||||
#ifndef QPX
|
|
||||||
#ifndef A64FX
|
|
||||||
#ifndef A64FXFIXEDSIZE
|
|
||||||
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
|
|
||||||
#endif
|
|
||||||
#endif
|
|
||||||
#endif
|
|
||||||
#endif
|
|
||||||
|
|
||||||
NAMESPACE_BEGIN(Grid);
|
|
||||||
|
|
||||||
#include "impl.h"
|
|
||||||
template class WilsonKernels<IMPLEMENTATION>;
|
|
||||||
|
|
||||||
NAMESPACE_END(Grid);
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonTMFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
#define IMPLEMENTATION WilsonImplDF
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../CayleyFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../ContinuedFractionFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../DomainWallEOFAFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../MobiusEOFAFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../PartialFractionFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonCloverFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonFermionInstantiation.cc.master
|
|
||||||
@ -1,51 +0,0 @@
|
|||||||
/*************************************************************************************
|
|
||||||
|
|
||||||
Grid physics library, www.github.com/paboyle/Grid
|
|
||||||
|
|
||||||
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
|
|
||||||
|
|
||||||
Copyright (C) 2015, 2020
|
|
||||||
|
|
||||||
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
|
||||||
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
|
|
||||||
Author: paboyle <paboyle@ph.ed.ac.uk>
|
|
||||||
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
|
|
||||||
|
|
||||||
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/qcd/action/fermion/FermionCore.h>
|
|
||||||
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
|
|
||||||
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
|
|
||||||
|
|
||||||
#ifndef AVX512
|
|
||||||
#ifndef QPX
|
|
||||||
#ifndef A64FX
|
|
||||||
#ifndef A64FXFIXEDSIZE
|
|
||||||
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
|
|
||||||
#endif
|
|
||||||
#endif
|
|
||||||
#endif
|
|
||||||
#endif
|
|
||||||
|
|
||||||
NAMESPACE_BEGIN(Grid);
|
|
||||||
|
|
||||||
#include "impl.h"
|
|
||||||
template class WilsonKernels<IMPLEMENTATION>;
|
|
||||||
|
|
||||||
NAMESPACE_END(Grid);
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonTMFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
#define IMPLEMENTATION WilsonImplFH
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../CayleyFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../ContinuedFractionFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../DomainWallEOFAFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../MobiusEOFAFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../PartialFractionFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonFermion5DInstantiation.cc.master
|
|
||||||
@ -1,51 +0,0 @@
|
|||||||
/*************************************************************************************
|
|
||||||
|
|
||||||
Grid physics library, www.github.com/paboyle/Grid
|
|
||||||
|
|
||||||
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
|
|
||||||
|
|
||||||
Copyright (C) 2015, 2020
|
|
||||||
|
|
||||||
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
|
||||||
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
|
|
||||||
Author: paboyle <paboyle@ph.ed.ac.uk>
|
|
||||||
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
|
|
||||||
|
|
||||||
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/qcd/action/fermion/FermionCore.h>
|
|
||||||
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
|
|
||||||
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
|
|
||||||
|
|
||||||
#ifndef AVX512
|
|
||||||
#ifndef QPX
|
|
||||||
#ifndef A64FX
|
|
||||||
#ifndef A64FXFIXEDSIZE
|
|
||||||
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
|
|
||||||
#endif
|
|
||||||
#endif
|
|
||||||
#endif
|
|
||||||
#endif
|
|
||||||
|
|
||||||
NAMESPACE_BEGIN(Grid);
|
|
||||||
|
|
||||||
#include "impl.h"
|
|
||||||
template class WilsonKernels<IMPLEMENTATION>;
|
|
||||||
|
|
||||||
NAMESPACE_END(Grid);
|
|
||||||
@ -1 +0,0 @@
|
|||||||
#define IMPLEMENTATION ZWilsonImplDF
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../CayleyFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../ContinuedFractionFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../DomainWallEOFAFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../MobiusEOFAFermionInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../PartialFractionFermion5DInstantiation.cc.master
|
|
||||||
@ -1 +0,0 @@
|
|||||||
../WilsonFermion5DInstantiation.cc.master
|
|
||||||
@ -1,51 +0,0 @@
|
|||||||
/*************************************************************************************
|
|
||||||
|
|
||||||
Grid physics library, www.github.com/paboyle/Grid
|
|
||||||
|
|
||||||
Source file: ./lib/qcd/action/fermion/WilsonKernels.cc
|
|
||||||
|
|
||||||
Copyright (C) 2015, 2020
|
|
||||||
|
|
||||||
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
|
||||||
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
|
|
||||||
Author: paboyle <paboyle@ph.ed.ac.uk>
|
|
||||||
Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
|
|
||||||
|
|
||||||
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/qcd/action/fermion/FermionCore.h>
|
|
||||||
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h>
|
|
||||||
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h>
|
|
||||||
|
|
||||||
#ifndef AVX512
|
|
||||||
#ifndef QPX
|
|
||||||
#ifndef A64FX
|
|
||||||
#ifndef A64FXFIXEDSIZE
|
|
||||||
#include <Grid/qcd/action/fermion/implementation/WilsonKernelsAsmImplementation.h>
|
|
||||||
#endif
|
|
||||||
#endif
|
|
||||||
#endif
|
|
||||||
#endif
|
|
||||||
|
|
||||||
NAMESPACE_BEGIN(Grid);
|
|
||||||
|
|
||||||
#include "impl.h"
|
|
||||||
template class WilsonKernels<IMPLEMENTATION>;
|
|
||||||
|
|
||||||
NAMESPACE_END(Grid);
|
|
||||||
@ -1 +0,0 @@
|
|||||||
#define IMPLEMENTATION ZWilsonImplFH
|
|
||||||
@ -57,27 +57,22 @@ NAMESPACE_BEGIN(Grid);
|
|||||||
///////////////////////////////////////////////////////////////////
|
///////////////////////////////////////////////////////////////////
|
||||||
|
|
||||||
void Gather_plane_table_compute (GridBase *grid,int dimension,int plane,int cbmask,
|
void Gather_plane_table_compute (GridBase *grid,int dimension,int plane,int cbmask,
|
||||||
int off,Vector<std::pair<int,int> > & table);
|
int off,std::vector<std::pair<int,int> > & table);
|
||||||
|
|
||||||
template<class vobj,class cobj,class compressor>
|
template<class vobj,class cobj,class compressor>
|
||||||
void Gather_plane_simple_table (Vector<std::pair<int,int> >& table,const Lattice<vobj> &rhs,cobj *buffer,compressor &compress, int off,int so) __attribute__((noinline));
|
void Gather_plane_simple_table (commVector<std::pair<int,int> >& table,const Lattice<vobj> &rhs,cobj *buffer,compressor &compress, int off,int so) __attribute__((noinline));
|
||||||
|
|
||||||
template<class vobj,class cobj,class compressor>
|
template<class vobj,class cobj,class compressor>
|
||||||
void Gather_plane_simple_table (Vector<std::pair<int,int> >& table,const Lattice<vobj> &rhs,cobj *buffer,compressor &compress, int off,int so)
|
void Gather_plane_simple_table (commVector<std::pair<int,int> >& table,const Lattice<vobj> &rhs,cobj *buffer,compressor &compress, int off,int so)
|
||||||
{
|
{
|
||||||
int num=table.size();
|
int num=table.size();
|
||||||
std::pair<int,int> *table_v = & table[0];
|
std::pair<int,int> *table_v = & table[0];
|
||||||
|
|
||||||
auto rhs_v = rhs.View(AcceleratorRead);
|
auto rhs_v = rhs.View(AcceleratorRead);
|
||||||
accelerator_forNB( i,num, vobj::Nsimd(), {
|
accelerator_forNB( i,num, vobj::Nsimd(), {
|
||||||
typedef decltype(coalescedRead(buffer[0])) compressed_t;
|
compress.Compress(buffer[off+table_v[i].first],rhs_v[so+table_v[i].second]);
|
||||||
compressed_t tmp_c;
|
|
||||||
uint64_t o = table_v[i].first;
|
|
||||||
compress.Compress(&tmp_c,0,rhs_v(so+table_v[i].second));
|
|
||||||
coalescedWrite(buffer[off+o],tmp_c);
|
|
||||||
});
|
});
|
||||||
rhs_v.ViewClose();
|
rhs_v.ViewClose();
|
||||||
// Further optimisatoin: i) software prefetch the first element of the next table entry, prefetch the table
|
|
||||||
}
|
}
|
||||||
|
|
||||||
///////////////////////////////////////////////////////////////////
|
///////////////////////////////////////////////////////////////////
|
||||||
@ -85,10 +80,10 @@ void Gather_plane_simple_table (Vector<std::pair<int,int> >& table,const Lattice
|
|||||||
///////////////////////////////////////////////////////////////////
|
///////////////////////////////////////////////////////////////////
|
||||||
template<class cobj,class vobj,class compressor>
|
template<class cobj,class vobj,class compressor>
|
||||||
void Gather_plane_exchange_table(const Lattice<vobj> &rhs,
|
void Gather_plane_exchange_table(const Lattice<vobj> &rhs,
|
||||||
Vector<cobj *> pointers,int dimension,int plane,int cbmask,compressor &compress,int type) __attribute__((noinline));
|
commVector<cobj *> pointers,int dimension,int plane,int cbmask,compressor &compress,int type) __attribute__((noinline));
|
||||||
|
|
||||||
template<class cobj,class vobj,class compressor>
|
template<class cobj,class vobj,class compressor>
|
||||||
void Gather_plane_exchange_table(Vector<std::pair<int,int> >& table,const Lattice<vobj> &rhs,
|
void Gather_plane_exchange_table(commVector<std::pair<int,int> >& table,const Lattice<vobj> &rhs,
|
||||||
Vector<cobj *> pointers,int dimension,int plane,int cbmask,
|
Vector<cobj *> pointers,int dimension,int plane,int cbmask,
|
||||||
compressor &compress,int type)
|
compressor &compress,int type)
|
||||||
{
|
{
|
||||||
@ -100,7 +95,7 @@ void Gather_plane_exchange_table(Vector<std::pair<int,int> >& table,const Lattic
|
|||||||
auto p0=&pointers[0][0];
|
auto p0=&pointers[0][0];
|
||||||
auto p1=&pointers[1][0];
|
auto p1=&pointers[1][0];
|
||||||
auto tp=&table[0];
|
auto tp=&table[0];
|
||||||
accelerator_forNB(j, num, 1, {
|
accelerator_forNB(j, num, vobj::Nsimd(), {
|
||||||
compress.CompressExchange(p0,p1, &rhs_v[0], j,
|
compress.CompressExchange(p0,p1, &rhs_v[0], j,
|
||||||
so+tp[2*j ].second,
|
so+tp[2*j ].second,
|
||||||
so+tp[2*j+1].second,
|
so+tp[2*j+1].second,
|
||||||
@ -266,10 +261,11 @@ public:
|
|||||||
}
|
}
|
||||||
|
|
||||||
int face_table_computed;
|
int face_table_computed;
|
||||||
std::vector<Vector<std::pair<int,int> > > face_table ;
|
std::vector<commVector<std::pair<int,int> > > face_table ;
|
||||||
Vector<int> surface_list;
|
Vector<int> surface_list;
|
||||||
|
|
||||||
stencilVector<StencilEntry> _entries; // Resident in managed memory
|
stencilVector<StencilEntry> _entries; // Resident in managed memory
|
||||||
|
commVector<StencilEntry> _entries_device; // Resident in managed memory
|
||||||
std::vector<Packet> Packets;
|
std::vector<Packet> Packets;
|
||||||
std::vector<Merge> Mergers;
|
std::vector<Merge> Mergers;
|
||||||
std::vector<Merge> MergersSHM;
|
std::vector<Merge> MergersSHM;
|
||||||
@ -342,7 +338,7 @@ public:
|
|||||||
|
|
||||||
void *shm = (void *) _grid->ShmBufferTranslate(recv_from_rank,this->u_recv_buf_p);
|
void *shm = (void *) _grid->ShmBufferTranslate(recv_from_rank,this->u_recv_buf_p);
|
||||||
|
|
||||||
if ( shm==NULL ) return 0;
|
if ( (shm==NULL) || Stencil_force_mpi ) return 0;
|
||||||
|
|
||||||
return 1;
|
return 1;
|
||||||
}
|
}
|
||||||
@ -609,13 +605,14 @@ public:
|
|||||||
template<class decompressor>
|
template<class decompressor>
|
||||||
void CommsMerge(decompressor decompress,std::vector<Merge> &mm,std::vector<Decompress> &dd) {
|
void CommsMerge(decompressor decompress,std::vector<Merge> &mm,std::vector<Decompress> &dd) {
|
||||||
|
|
||||||
|
|
||||||
mergetime-=usecond();
|
mergetime-=usecond();
|
||||||
for(int i=0;i<mm.size();i++){
|
for(int i=0;i<mm.size();i++){
|
||||||
auto mp = &mm[i].mpointer[0];
|
auto mp = &mm[i].mpointer[0];
|
||||||
auto vp0= &mm[i].vpointers[0][0];
|
auto vp0= &mm[i].vpointers[0][0];
|
||||||
auto vp1= &mm[i].vpointers[1][0];
|
auto vp1= &mm[i].vpointers[1][0];
|
||||||
auto type= mm[i].type;
|
auto type= mm[i].type;
|
||||||
accelerator_forNB(o,mm[i].buffer_size/2,1,{
|
accelerator_forNB(o,mm[i].buffer_size/2,vobj::Nsimd(),{
|
||||||
decompress.Exchange(mp,vp0,vp1,type,o);
|
decompress.Exchange(mp,vp0,vp1,type,o);
|
||||||
});
|
});
|
||||||
}
|
}
|
||||||
@ -1039,7 +1036,12 @@ public:
|
|||||||
int so = sx*rhs.Grid()->_ostride[dimension]; // base offset for start of plane
|
int so = sx*rhs.Grid()->_ostride[dimension]; // base offset for start of plane
|
||||||
if ( !face_table_computed ) {
|
if ( !face_table_computed ) {
|
||||||
face_table.resize(face_idx+1);
|
face_table.resize(face_idx+1);
|
||||||
Gather_plane_table_compute ((GridBase *)_grid,dimension,sx,cbmask,u_comm_offset,face_table[face_idx]);
|
std::vector<std::pair<int,int> > face_table_host ;
|
||||||
|
Gather_plane_table_compute ((GridBase *)_grid,dimension,sx,cbmask,u_comm_offset,face_table_host);
|
||||||
|
face_table[face_idx].resize(face_table_host.size());
|
||||||
|
acceleratorCopyToDevice(&face_table_host[0],
|
||||||
|
&face_table[face_idx][0],
|
||||||
|
face_table[face_idx].size()*sizeof(face_table_host[0]));
|
||||||
}
|
}
|
||||||
|
|
||||||
// int rank = _grid->_processor;
|
// int rank = _grid->_processor;
|
||||||
@ -1062,13 +1064,13 @@ public:
|
|||||||
}
|
}
|
||||||
|
|
||||||
send_buf = (cobj *)_grid->ShmBufferTranslate(xmit_to_rank,recv_buf);
|
send_buf = (cobj *)_grid->ShmBufferTranslate(xmit_to_rank,recv_buf);
|
||||||
if ( send_buf==NULL ) {
|
if ( (send_buf==NULL) || Stencil_force_mpi ) {
|
||||||
send_buf = this->u_send_buf_p;
|
send_buf = this->u_send_buf_p;
|
||||||
}
|
}
|
||||||
|
|
||||||
// Find out if we get the direct copy.
|
// Find out if we get the direct copy.
|
||||||
void *success = (void *) _grid->ShmBufferTranslate(recv_from_rank,this->u_send_buf_p);
|
void *success = (void *) _grid->ShmBufferTranslate(recv_from_rank,this->u_send_buf_p);
|
||||||
if (success==NULL) {
|
if ((success==NULL)||Stencil_force_mpi) {
|
||||||
// we found a packet that comes from MPI and contributes to this leg of stencil
|
// we found a packet that comes from MPI and contributes to this leg of stencil
|
||||||
shm_receive_only = 0;
|
shm_receive_only = 0;
|
||||||
}
|
}
|
||||||
@ -1172,11 +1174,18 @@ public:
|
|||||||
|
|
||||||
if ( !face_table_computed ) {
|
if ( !face_table_computed ) {
|
||||||
face_table.resize(face_idx+1);
|
face_table.resize(face_idx+1);
|
||||||
Gather_plane_table_compute ((GridBase *)_grid,dimension,sx,cbmask,u_comm_offset,face_table[face_idx]);
|
std::vector<std::pair<int,int> > face_table_host ;
|
||||||
|
|
||||||
|
Gather_plane_table_compute ((GridBase *)_grid,dimension,sx,cbmask,u_comm_offset,face_table_host);
|
||||||
|
face_table[face_idx].resize(face_table_host.size());
|
||||||
|
acceleratorCopyToDevice(&face_table_host[0],
|
||||||
|
&face_table[face_idx][0],
|
||||||
|
face_table[face_idx].size()*sizeof(face_table_host[0]));
|
||||||
}
|
}
|
||||||
gathermtime-=usecond();
|
gathermtime-=usecond();
|
||||||
|
|
||||||
Gather_plane_exchange_table(face_table[face_idx],rhs,spointers,dimension,sx,cbmask,compress,permute_type); face_idx++;
|
Gather_plane_exchange_table(face_table[face_idx],rhs,spointers,dimension,sx,cbmask,compress,permute_type);
|
||||||
|
face_idx++;
|
||||||
|
|
||||||
gathermtime+=usecond();
|
gathermtime+=usecond();
|
||||||
//spointers[0] -- low
|
//spointers[0] -- low
|
||||||
@ -1208,7 +1217,7 @@ public:
|
|||||||
// shm == receive pointer if offnode
|
// shm == receive pointer if offnode
|
||||||
// shm == Translate[send pointer] if on node -- my view of his send pointer
|
// shm == Translate[send pointer] if on node -- my view of his send pointer
|
||||||
cobj *shm = (cobj *) _grid->ShmBufferTranslate(recv_from_rank,sp);
|
cobj *shm = (cobj *) _grid->ShmBufferTranslate(recv_from_rank,sp);
|
||||||
if (shm==NULL) {
|
if ((shm==NULL)||Stencil_force_mpi) {
|
||||||
shm = rp;
|
shm = rp;
|
||||||
// we found a packet that comes from MPI and contributes to this shift.
|
// we found a packet that comes from MPI and contributes to this shift.
|
||||||
// is_same_node is only used in the WilsonStencil, and gets set for this point in the stencil.
|
// is_same_node is only used in the WilsonStencil, and gets set for this point in the stencil.
|
||||||
|
|||||||
Loading…
Reference in New Issue
Block a user