portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-27 22:25:56 +01:00
Code Releases Activity

Correcting a bug in the IO routines

Browse Source
This commit is contained in:
Guido Cossu 2017-07-27 15:12:50 +01:00
parent c7036f6717
commit 8bd869da37
1 changed files with 86 additions and 47 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

99
lib/parallelIO/BinaryIO.h
View File

@ -99,34 +99,38 @@ class BinaryIO {
NerscChecksum(grid,scalardata,nersc_csum); NerscChecksum(grid,scalardata,nersc_csum);
} }
template<class fobj> static inline void NerscChecksum(GridBase *grid,std::vector<fobj> &fbuf,uint32_t &nersc_csum) template <class fobj>
static inline void NerscChecksum(GridBase *grid, std::vector<fobj> &fbuf, uint32_t &nersc_csum)
{ {
const uint64_t size32 = sizeof(fobj)/sizeof(uint32_t); const uint64_t size32 = sizeof(fobj) / sizeof(uint32_t);
uint64_t lsites = grid->lSites();
uint64_t lsites =grid->lSites(); if (fbuf.size() == 1)
if (fbuf.size()==1) {
lsites=1;
}
#pragma omp parallel
{ {
uint32_t nersc_csum_thr=0; lsites = 1;
}
#pragma omp for #pragma omp parallel
for(uint64_t local_site=0;local_site<lsites;local_site++){ {
uint32_t * site_buf = (uint32_t *)&fbuf[local_site]; uint32_t nersc_csum_thr = 0;
for(uint64_t j=0;j<size32;j++){
nersc_csum_thr=nersc_csum_thr+site_buf[j]; #pragma omp for
for (uint64_t local_site = 0; local_site < lsites; local_site++)
{
uint32_t *site_buf = (uint32_t *)&fbuf[local_site];
for (uint64_t j = 0; j < size32; j++)
{
nersc_csum_thr = nersc_csum_thr + site_buf[j];
} }
} }
#pragma omp critical #pragma omp critical
{ {
nersc_csum += nersc_csum_thr; nersc_csum += nersc_csum_thr;
} }
} }
} }
template<class fobj> static inline void ScidacChecksum(GridBase *grid,std::vector<fobj> &fbuf,uint32_t &scidac_csuma,uint32_t &scidac_csumb) template<class fobj> static inline void ScidacChecksum(GridBase *grid,std::vector<fobj> &fbuf,uint32_t &scidac_csuma,uint32_t &scidac_csumb)
{ {
const uint64_t size32 = sizeof(fobj)/sizeof(uint32_t); const uint64_t size32 = sizeof(fobj)/sizeof(uint32_t);
@ -363,16 +367,20 @@ class BinaryIO {
assert(0); assert(0);
#endif #endif
} else { } else {
std::cout<< GridLogMessage<< "C++ read I/O "<< file<<" : " std::cout << GridLogMessage << "C++ read I/O " << file << " : "
<< iodata.size()*sizeof(fobj)<<" bytes"<<std::endl; << iodata.size() * sizeof(fobj) << " bytes" << std::endl;
std::ifstream fin; std::ifstream fin;
fin.open(file,std::ios::binary|std::ios::in); fin.open(file, std::ios::binary | std::ios::in);
if ( control & BINARYIO_MASTER_APPEND ) { if (control & BINARYIO_MASTER_APPEND)
fin.seekg(-sizeof(fobj),fin.end); {
} else { fin.seekg(-sizeof(fobj), fin.end);
fin.seekg(offset+myrank*lsites*sizeof(fobj));
} }
fin.read((char *)&iodata[0],iodata.size()*sizeof(fobj));assert( fin.fail()==0); else
{
fin.seekg(offset + myrank * lsites * sizeof(fobj));
}
fin.read((char *)&iodata[0], iodata.size() * sizeof(fobj));
assert(fin.fail() == 0);
fin.close(); fin.close();
} }
timer.Stop(); timer.Stop();
@ -416,15 +424,43 @@ class BinaryIO {
assert(0); assert(0);
#endif #endif
} else { } else {
std::ofstream fout; fout.open(file,std::ios::binary|std::ios::out|std::ios::in);
std::cout<< GridLogMessage<< "C++ write I/O "<< file<<" : " std::ofstream fout;
fout.exceptions ( std::fstream::failbit | std::fstream::badbit );
try {
fout.open(file,std::ios::binary|std::ios::out|std::ios::in);
} catch (const std::fstream::failure& exc) {
std::cout << GridLogError << "Error in opening the file " << file << " for output" <<std::endl;
std::cout << GridLogError << "Exception description: " << exc.what() << std::endl;
std::cout << GridLogError << "Probable cause: wrong path, inaccessible location "<< std::endl;
#ifdef USE_MPI_IO
MPI_Abort(MPI_COMM_WORLD,1);
#else
exit(1);
#endif
}
std::cout << GridLogMessage<< "C++ write I/O "<< file<<" : "
<< iodata.size()*sizeof(fobj)<<" bytes"<<std::endl; << iodata.size()*sizeof(fobj)<<" bytes"<<std::endl;
if ( control & BINARYIO_MASTER_APPEND ) { if ( control & BINARYIO_MASTER_APPEND ) {
fout.seekp(0,fout.end); fout.seekp(0,fout.end);
} else { } else {
fout.seekp(offset+myrank*lsites*sizeof(fobj)); fout.seekp(offset+myrank*lsites*sizeof(fobj));
} }
fout.write((char *)&iodata[0],iodata.size()*sizeof(fobj));assert( fout.fail()==0);
try {
fout.write((char *)&iodata[0],iodata.size()*sizeof(fobj));//assert( fout.fail()==0);
}
catch (const std::fstream::failure& exc) {
std::cout << "Exception in writing file " << file << std::endl;
std::cout << GridLogError << "Exception description: "<< exc.what() << std::endl;
#ifdef USE_MPI_IO
MPI_Abort(MPI_COMM_WORLD,1);
#else
exit(1);
#endif
}
fout.close(); fout.close();
} }
timer.Stop(); timer.Stop();
@ -442,12 +478,15 @@ class BinaryIO {
////////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////////
// Safety check // Safety check
////////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////////
// if the data size is 1 we do not want to sum over the MPI ranks
if (iodata.size() != 1){
grid->Barrier(); grid->Barrier();
grid->GlobalSum(nersc_csum); grid->GlobalSum(nersc_csum);
grid->GlobalXOR(scidac_csuma); grid->GlobalXOR(scidac_csuma);
grid->GlobalXOR(scidac_csumb); grid->GlobalXOR(scidac_csumb);
grid->Barrier(); grid->Barrier();
} }
}
///////////////////////////////////////////////////////////////////////////// /////////////////////////////////////////////////////////////////////////////
// Read a Lattice of object // Read a Lattice of object
@ -546,9 +585,9 @@ class BinaryIO {
int gsites = grid->gSites(); int gsites = grid->gSites();
int lsites = grid->lSites(); int lsites = grid->lSites();
uint32_t nersc_csum_tmp; uint32_t nersc_csum_tmp = 0;
uint32_t scidac_csuma_tmp; uint32_t scidac_csuma_tmp = 0;
uint32_t scidac_csumb_tmp; uint32_t scidac_csumb_tmp = 0;
GridStopWatch timer; GridStopWatch timer;