portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-04 05:54:32 +00:00
Code Releases Activity

Shmem related fixes for shmem compile

Browse Source
This commit is contained in:
Peter Boyle
2016-02-11 07:37:39 -06:00
parent e2f73e3ead
commit 7f927a541c
14 changed files with 3411 additions and 4061 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7309
configure vendored
View File

File diff suppressed because it is too large Load Diff

21
lib/AlignedAllocator.h
View File

@@ -40,6 +40,9 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
#ifdef HAVE_MM_MALLOC_H
#include <mm_malloc.h>
#endif
#ifdef GRID_COMMS_SHMEM
#include <mpp/shmem.h>
#endif
namespace Grid {
@@ -74,19 +77,37 @@ public:
pointer allocate(size_type __n, const void* = 0)
{
#ifdef GRID_COMMS_SHMEM
static void * bcast;
static long psync[_SHMEM_REDUCE_SYNC_SIZE];
shmem_barrier_all();
_Tp *ptr = (_Tp *) shmem_align(128,__n*sizeof(_Tp));
shmem_barrier_all();
bcast = (void *) _Tp;
shmem_broadcast32((void *)&bcast,(void *)&bcast,sizeof(void *)/4,0,0,0,shmem_n_pes(),psync);
assert( bcast == (void *) _Tp);
#else
#ifdef HAVE_MM_MALLOC_H
_Tp * ptr = (_Tp *) _mm_malloc(__n*sizeof(_Tp),128);
#else
_Tp * ptr = (_Tp *) memalign(128,__n*sizeof(_Tp));
#endif
#endif
return ptr;
}
void deallocate(pointer __p, size_type) {
#ifdef GRID_COMMS_SHMEM
shmem_free((void *)__p);
#else
#ifdef HAVE_MM_MALLOC_H
_mm_free((void *)__p);
#else
free((void *)__p);
#endif
#endif
}
void construct(pointer __p, const _Tp& __val) { };

38
lib/Config.h.in
View File

@@ -15,7 +15,7 @@
/* EMPTY_SIMD only for DEBUGGING */
#undef EMPTY_SIMD
/* GRID_COMMS_MPI */
/* GRID_COMMS_SHMEM */
#undef GRID_COMMS_MPI
/* GRID_COMMS_NONE */
@@ -27,15 +27,6 @@
/* GRID_DEFAULT_PRECISION is SINGLE */
#undef GRID_DEFAULT_PRECISION_SINGLE
/* Support Altivec instructions */
#undef HAVE_ALTIVEC
/* Support AVX (Advanced Vector Extensions) instructions */
#undef HAVE_AVX
/* Support AVX2 (Advanced Vector Extensions 2) instructions */
#undef HAVE_AVX2
/* Define to 1 if you have the declaration of `be64toh', and to 0 if you
don't. */
#undef HAVE_DECL_BE64TOH
@@ -50,9 +41,6 @@
/* Define to 1 if you have the <execinfo.h> header file. */
#undef HAVE_EXECINFO_H
/* Support FMA3 (Fused Multiply-Add) instructions */
#undef HAVE_FMA
/* Define to 1 if you have the `gettimeofday' function. */
#undef HAVE_GETTIMEOFDAY
@@ -71,30 +59,9 @@
/* Define to 1 if you have the <memory.h> header file. */
#undef HAVE_MEMORY_H
/* Support mmx instructions */
#undef HAVE_MMX
/* Define to 1 if you have the <mm_malloc.h> header file. */
#undef HAVE_MM_MALLOC_H
/* Support SSE (Streaming SIMD Extensions) instructions */
#undef HAVE_SSE
/* Support SSE2 (Streaming SIMD Extensions 2) instructions */
#undef HAVE_SSE2
/* Support SSE3 (Streaming SIMD Extensions 3) instructions */
#undef HAVE_SSE3
/* Support SSSE4.1 (Streaming SIMD Extensions 4.1) instructions */
#undef HAVE_SSE4_1
/* Support SSSE4.2 (Streaming SIMD Extensions 4.2) instructions */
#undef HAVE_SSE4_2
/* Support SSSE3 (Supplemental Streaming SIMD Extensions 3) instructions */
#undef HAVE_SSSE3
/* Define to 1 if you have the <stdint.h> header file. */
#undef HAVE_STDINT_H
@@ -137,9 +104,6 @@
/* Define to the one symbol short name of this package. */
#undef PACKAGE_TARNAME
/* Define to the home page for this package. */
#undef PACKAGE_URL
/* Define to the version of this package. */
#undef PACKAGE_VERSION

1
lib/Grid.h
View File

@@ -66,6 +66,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#include <AlignedAllocator.h>
#include <Simd.h>
#include <Threads.h>
#include <Lexicographic.h>
#include <Communicator.h>
#include <Cartesian.h>
#include <Tensors.h>

4
lib/Makefile.am
View File

@@ -6,6 +6,10 @@ if BUILD_COMMS_MPI
extra_sources+=communicator/Communicator_mpi.cc
endif
if BUILD_COMMS_SHMEM
extra_sources+=communicator/Communicator_shmem.cc
endif
if BUILD_COMMS_NONE
extra_sources+=communicator/Communicator_none.cc
endif

1
lib/Stencil.h
View File

@@ -150,7 +150,6 @@ namespace Grid {
void CommsMerge(void ) {
//PARALLEL_NESTED_LOOP2
for(int i=0;i<Mergers.size();i++){
spintime-=usecond();
int packet_id = Mergers[i].packet_id;

24
lib/cartesian/Cartesian_base.h
View File

@@ -115,27 +115,11 @@ public:
for(int d=0;d<_ndimension;d++) idx+=_ostride[d]*ocoor[d];
return idx;
}
static inline void CoorFromIndex (std::vector<int>& coor,int index,std::vector<int> &dims){
int nd= dims.size();
coor.resize(nd);
for(int d=0;d<nd;d++){
coor[d] = index % dims[d];
index = index / dims[d];
}
}
inline void oCoorFromOindex (std::vector<int>& coor,int Oindex){
CoorFromIndex(coor,Oindex,_rdimensions);
}
static inline void IndexFromCoor (std::vector<int>& coor,int &index,std::vector<int> &dims){
int nd=dims.size();
int stride=1;
index=0;
for(int d=0;d<nd;d++){
index = index+stride*coor[d];
stride=stride*dims[d];
}
Lexicographic::CoorFromIndex(coor,Oindex,_rdimensions);
}
//////////////////////////////////////////////////////////
// SIMD lane addressing
//////////////////////////////////////////////////////////
@@ -147,7 +131,7 @@ public:
}
inline void iCoorFromIindex(std::vector<int> &coor,int lane)
{
CoorFromIndex(coor,lane,_simd_layout);
Lexicographic::CoorFromIndex(coor,lane,_simd_layout);
}
inline int PermuteDim(int dimension){
return _simd_layout[dimension]>1;
@@ -179,7 +163,7 @@ public:
// Global addressing
////////////////////////////////////////////////////////////////
void GlobalIndexToGlobalCoor(int gidx,std::vector<int> &gcoor){
CoorFromIndex(gcoor,gidx,_gdimensions);
Lexicographic::CoorFromIndex(gcoor,gidx,_gdimensions);
}
void GlobalCoorToGlobalIndex(const std::vector<int> & gcoor,int & gidx){
gidx=0;

34
lib/communicator/Communicator_shmem.cc
View File

@@ -44,12 +44,14 @@ CartesianCommunicator::CartesianCommunicator(const std::vector<int> &processors)
shmem_init_thread(SHMEM_THREAD_FUNNELED);
_processor = shmem_my_pe();
CoorFromIndex(_processor_coor,_processor,_processors);
Lexicographic::CoorFromIndex(_processor_coor,_processor,_processors);
for(int i=0;i<_ndimension;i++){
_Nprocessors*=_processors[i];
}
if ( _processor == 0 ) {
printf("I'm running SHMEM communications %d \n",_processor);
}
int Size = shmem_n_pes();
assert(Size==_Nprocessors);
}
@@ -67,13 +69,13 @@ void CartesianCommunicator::GlobalSum(uint32_t &u){
u = dest;
}
void CartesianCommunicator::GlobalSum(float &f){
static float source = u;
static float source = f;
static float dest = 0 ;
static float llwrk[_SHMEM_REDUCE_MIN_WRKDATA_SIZE];
static long psync[_SHMEM_REDUCE_SYNC_SIZE];
shmem_float_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync);
u = dest;
f = dest;
}
void CartesianCommunicator::GlobalSumVector(float *f,int N)
{
@@ -83,20 +85,20 @@ void CartesianCommunicator::GlobalSumVector(float *f,int N)
static long psync[_SHMEM_REDUCE_SYNC_SIZE];
for(int i=0;i<N;i++){
source = u[i];
source = f[i];
shmem_float_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync);
u[i] = dest;
f[i] = dest;
}
}
void CartesianCommunicator::GlobalSum(double &d)
{
static double source = u;
static double source = d;
static double dest = 0 ;
static double llwrk[_SHMEM_REDUCE_MIN_WRKDATA_SIZE];
static long psync[_SHMEM_REDUCE_SYNC_SIZE];
shmem_double_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync);
u = dest;
d = dest;
}
void CartesianCommunicator::GlobalSumVector(double *d,int N)
{
@@ -106,9 +108,9 @@ void CartesianCommunicator::GlobalSumVector(double *d,int N)
static long psync[_SHMEM_REDUCE_SYNC_SIZE];
for(int i=0;i<N;i++){
source = u[i];
source = d[i];
shmem_double_sum_to_all(&dest,&source,1,0,0,_Nprocessors,llwrk,psync);
u[i] = dest;
d[i] = dest;
}
}
void CartesianCommunicator::ShiftedRanks(int dim,int shift,int &source,int &dest)
@@ -117,21 +119,21 @@ void CartesianCommunicator::ShiftedRanks(int dim,int shift,int &source,int &dest
assert(std::abs(shift) <_processors[dim]);
coor[dim] = (coor[dim] + shift + _processors[dim])%_processors[dim];
IndexFromCoor(coor,source,_processors);
Lexicographic::IndexFromCoor(coor,source,_processors);
coor[dim] = (coor[dim] - shift + _processors[dim])%_processors[dim];
IndexFromCoor(coor,dest,_processors);
Lexicographic::IndexFromCoor(coor,dest,_processors);
}
int CartesianCommunicator::RankFromProcessorCoor(std::vector<int> &coor)
{
int rank;
IndexFromCoor(coor,rank,_processors);
Lexicographic::IndexFromCoor(coor,rank,_processors);
return rank;
}
void CartesianCommunicator::ProcessorCoorFromRank(int rank, std::vector<int> &coor)
{
CoorFromIndex(coor,rank,_processors);
Lexicographic::CoorFromIndex(coor,rank,_processors);
}
// Basic Halo comms primitive
@@ -187,7 +189,9 @@ void CartesianCommunicator::Broadcast(int root,void* data, int bytes)
}
void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes)
{
Broadcast(root,data,bytes);
static long psync[_SHMEM_REDUCE_SYNC_SIZE];
assert( (bytes % 4)==0);
shmem_broadcast32(data,data,bytes/4,root,0,0,shmem_n_pes(),psync);
}
}

2
lib/lattice/Lattice_reduction.h
View File

@@ -178,7 +178,7 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
// sum over reduced dimension planes, breaking out orthog dir
for(int ss=0;ss<grid->oSites();ss++){
GridBase::CoorFromIndex(coor,ss,grid->_rdimensions);
Lexicographic::CoorFromIndex(coor,ss,grid->_rdimensions);
int r = coor[orthogdim];
lvSum[r]=lvSum[r]+Data._odata[ss];
}

16
lib/lattice/Lattice_transfer.h
View File

@@ -115,9 +115,9 @@ inline void blockProject(Lattice<iVector<CComplex,nbasis > > &coarseData,
int sc;
std::vector<int> coor_c(_ndimension);
std::vector<int> coor_f(_ndimension);
GridBase::CoorFromIndex(coor_f,sf,fine->_rdimensions);
Lexicographic::CoorFromIndex(coor_f,sf,fine->_rdimensions);
for(int d=0;d<_ndimension;d++) coor_c[d]=coor_f[d]/block_r[d];
GridBase::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
for(int i=0;i<nbasis;i++) {
@@ -160,9 +160,9 @@ PARALLEL_FOR_LOOP
std::vector<int> coor_c(_ndimension);
std::vector<int> coor_f(_ndimension);
GridBase::CoorFromIndex(coor_f,sf,fine->_rdimensions);
Lexicographic::CoorFromIndex(coor_f,sf,fine->_rdimensions);
for(int d=0;d<_ndimension;d++) coor_c[d]=coor_f[d]/block_r[d];
GridBase::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
// z = A x + y
fineZ._odata[sf]=coarseA._odata[sc]*fineX._odata[sf]+fineY._odata[sf];
@@ -225,9 +225,9 @@ inline void blockSum(Lattice<vobj> &coarseData,const Lattice<vobj> &fineData)
std::vector<int> coor_c(_ndimension);
std::vector<int> coor_f(_ndimension);
GridBase::CoorFromIndex(coor_f,sf,fine->_rdimensions);
Lexicographic::CoorFromIndex(coor_f,sf,fine->_rdimensions);
for(int d=0;d<_ndimension;d++) coor_c[d]=coor_f[d]/block_r[d];
GridBase::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
coarseData._odata[sc]=coarseData._odata[sc]+fineData._odata[sf];
@@ -311,9 +311,9 @@ inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
std::vector<int> coor_c(_ndimension);
std::vector<int> coor_f(_ndimension);
GridBase::CoorFromIndex(coor_f,sf,fine->_rdimensions);
Lexicographic::CoorFromIndex(coor_f,sf,fine->_rdimensions);
for(int d=0;d<_ndimension;d++) coor_c[d]=coor_f[d]/block_r[d];
GridBase::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
Lexicographic::IndexFromCoor(coor_c,sc,coarse->_rdimensions);
for(int i=0;i<nbasis;i++) {
if(i==0) fineData._odata[sf]=coarseData._odata[sc](i) * Basis[i]._odata[sf];

6
lib/parallelIO/BinaryIO.h
View File

@@ -146,7 +146,7 @@ class BinaryIO {
csum = 0;
std::vector<int> lcoor;
for(int l=0;l<grid->lSites();l++){
grid->CoorFromIndex(lcoor,l,grid->_ldimensions);
Lexicographic::CoorFromIndex(lcoor,l,grid->_ldimensions);
peekLocalSite(siteObj,lat,lcoor);
munge(siteObj,fileObj,csum);
}
@@ -451,7 +451,7 @@ class BinaryIO {
std::vector<int> lsite(nd);
std::vector<int> iosite(nd);
grid->CoorFromIndex(tsite,tlex,range);
Lexicographic::CoorFromIndex(tsite,tlex,range);
for(int d=0;d<nd;d++){
lsite[d] = tsite[d]%grid->_ldimensions[d]; // local site
@@ -588,7 +588,7 @@ class BinaryIO {
std::vector<int> lsite(nd);
std::vector<int> iosite(nd);
grid->CoorFromIndex(tsite,tlex,range);
Lexicographic::CoorFromIndex(tsite,tlex,range);
for(int d=0;d<nd;d++){
lsite[d] = tsite[d]%grid->_ldimensions[d]; // local site

2
lib/stencil/Lebesgue.cc
View File

@@ -105,7 +105,7 @@ void LebesgueOrder::IterateI(int ND,
x[d]=xi[d]+xo[d];
}
IndexInteger index;
grid->IndexFromCoor(x,index,grid->_rdimensions);
Lexicographic::IndexFromCoor(x,index,grid->_rdimensions);
_LebesgueReorder.push_back(index);
}
}

4
tests/Test_cshift.cc
View File

@@ -96,13 +96,13 @@ int main (int argc, char ** argv)
std::vector<int> peer(4);
Complex tmp =cm;
Integer index=real(tmp);
Fine.CoorFromIndex(peer,index,latt_size);
Lexicographic::CoorFromIndex(peer,index,latt_size);
if (nrm > 0){
std::cerr<<"FAIL shift "<< shift<<" in dir "<< dir<<" ["<<coor[0]<<","<<coor[1]<<","<<coor[2]<<","<<coor[3]<<"] = "<< cm()()()<<" expect "<<scm<<" "<<nrm<<std::endl;
std::cerr<<"Got "<<index<<" " << peer[0]<<","<<peer[1]<<","<<peer[2]<<","<<peer[3]<<std::endl;
index=real(scm);
Fine.CoorFromIndex(peer,index,latt_size);
Lexicographic::CoorFromIndex(peer,index,latt_size);
std::cerr<<"Expect "<<index<<" " << peer[0]<<","<<peer[1]<<","<<peer[2]<<","<<peer[3]<<std::endl;
}
}}}}

10
tests/Test_cshift_red_black.cc
View File

@@ -132,7 +132,7 @@ int main (int argc, char ** argv)
std::vector<int> peer(4);
Complex ctmp = cm;
Integer index=real(ctmp);
Fine.CoorFromIndex(peer,index,latt_size);
Lexicographic::CoorFromIndex(peer,index,latt_size);
if (nrm > 0){
std::cout<<"FAIL shift "<< shift<<" in dir "<< dir
@@ -140,7 +140,7 @@ int main (int argc, char ** argv)
<< cm()()()<<" expect "<<scm<<" "<<nrm<<std::endl;
std::cout<<"Got "<<index<<" " << peer[0]<<","<<peer[1]<<","<<peer[2]<<","<<peer[3]<<std::endl;
index=real(scm);
Fine.CoorFromIndex(peer,index,latt_size);
Lexicographic::CoorFromIndex(peer,index,latt_size);
std::cout<<"Expect "<<index<<" " << peer[0]<<","<<peer[1]<<","<<peer[2]<<","<<peer[3]<<std::endl;
exit(-1);
}
@@ -180,7 +180,7 @@ int main (int argc, char ** argv)
std::vector<int> peer(4);
Complex ctmp=cmeo;
Integer index=real(ctmp);
Fine.CoorFromIndex(peer,index,latt_size);
Lexicographic::CoorFromIndex(peer,index,latt_size);
double nrm = abs(cmeo()()()-scm);
if (nrm != 0) {
@@ -189,7 +189,7 @@ int main (int argc, char ** argv)
<< cmeo()()()<<" expect "<<scm<<" "<<nrm<<std::endl;
std::cout<<"Got "<<index<<" " << peer[0]<<","<<peer[1]<<","<<peer[2]<<","<<peer[3]<<std::endl;
index=real(scm);
Fine.CoorFromIndex(peer,index,latt_size);
Lexicographic::CoorFromIndex(peer,index,latt_size);
std::cout<<"Expect "<<index<<" " << peer[0]<<","<<peer[1]<<","<<peer[2]<<","<<peer[3]<<std::endl;
exx=1;
@@ -205,7 +205,7 @@ int main (int argc, char ** argv)
<< cm()()()<<" expect "<<scm<<" "<<nrm<<std::endl;
std::cout<<"Got "<<index<<" " << peer[0]<<","<<peer[1]<<","<<peer[2]<<","<<peer[3]<<std::endl;
index=real(scm);
Fine.CoorFromIndex(peer,index,latt_size);
Lexicographic::CoorFromIndex(peer,index,latt_size);
std::cout<<"Expect "<<index<<" " << peer[0]<<","<<peer[1]<<","<<peer[2]<<","<<peer[3]<<std::endl;
exx=1;
} else if (1) {