portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-16 06:47:06 +01:00
Code Releases Activity

4GB clean the offsets in parallel IO for multifile records

Browse Source
This commit is contained in:
Peter Boyle
2018-03-16 21:54:56 +00:00
parent 01568b0e62
commit 888838473a
3 changed files with 53 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
lib/parallelIO/BinaryIO.h
View File

@ -91,7 +91,7 @@ class BinaryIO {
typedef typename vobj::scalar_object sobj;
GridBase *grid = lat._grid;
int lsites = grid->lSites();
uint64_t lsites = grid->lSites();
std::vector<sobj> scalardata(lsites);
unvectorizeToLexOrdArray(scalardata,lat);
@ -160,7 +160,9 @@ class BinaryIO {
/*
* Scidac csum is rather more heavyweight
* FIXME -- 128^3 x 256 x 16 will overflow.
*/
int global_site;
Lexicographic::CoorFromIndex(coor,local_site,local_vol);
@ -261,7 +263,7 @@ class BinaryIO {
GridBase *grid,
std::vector<fobj> &iodata,
std::string file,
Integer offset,
uint64_t offset,
const std::string &format, int control,
uint32_t &nersc_csum,
uint32_t &scidac_csuma,
@ -523,7 +525,7 @@ class BinaryIO {
static inline void readLatticeObject(Lattice<vobj> &Umu,
std::string file,
munger munge,
Integer offset,
uint64_t offset,
const std::string &format,
uint32_t &nersc_csum,
uint32_t &scidac_csuma,
@ -533,7 +535,7 @@ class BinaryIO {
typedef typename vobj::Realified::scalar_type word; word w=0;
GridBase *grid = Umu._grid;
int lsites = grid->lSites();
uint64_t lsites = grid->lSites();
std::vector<sobj> scalardata(lsites);
std::vector<fobj> iodata(lsites); // Munge, checksum, byte order in here
@ -544,7 +546,7 @@ class BinaryIO {
GridStopWatch timer;
timer.Start();
parallel_for(int x=0;x<lsites;x++) munge(iodata[x], scalardata[x]);
parallel_for(uint64_t x=0;x<lsites;x++) munge(iodata[x], scalardata[x]);
vectorizeFromLexOrdArray(scalardata,Umu);
grid->Barrier();
@ -560,7 +562,7 @@ class BinaryIO {
static inline void writeLatticeObject(Lattice<vobj> &Umu,
std::string file,
munger munge,
Integer offset,
uint64_t offset,
const std::string &format,
uint32_t &nersc_csum,
uint32_t &scidac_csuma,
@ -569,7 +571,7 @@ class BinaryIO {
typedef typename vobj::scalar_object sobj;
typedef typename vobj::Realified::scalar_type word; word w=0;
GridBase *grid = Umu._grid;
int lsites = grid->lSites();
uint64_t lsites = grid->lSites();
std::vector<sobj> scalardata(lsites);
std::vector<fobj> iodata(lsites); // Munge, checksum, byte order in here
@ -580,7 +582,7 @@ class BinaryIO {
GridStopWatch timer; timer.Start();
unvectorizeToLexOrdArray(scalardata,Umu);
parallel_for(int x=0;x<lsites;x++) munge(scalardata[x],iodata[x]);
parallel_for(uint64_t x=0;x<lsites;x++) munge(scalardata[x],iodata[x]);
grid->Barrier();
timer.Stop();
@ -597,7 +599,7 @@ class BinaryIO {
static inline void readRNG(GridSerialRNG &serial,
GridParallelRNG &parallel,
std::string file,
Integer offset,
uint64_t offset,
uint32_t &nersc_csum,
uint32_t &scidac_csuma,
uint32_t &scidac_csumb)
@ -610,8 +612,8 @@ class BinaryIO {
std::string format = "IEEE32BIG";
GridBase *grid = parallel._grid;
int gsites = grid->gSites();
int lsites = grid->lSites();
uint64_t gsites = grid->gSites();
uint64_t lsites = grid->lSites();
uint32_t nersc_csum_tmp = 0;
uint32_t scidac_csuma_tmp = 0;
@ -626,7 +628,7 @@ class BinaryIO {
nersc_csum,scidac_csuma,scidac_csumb);
timer.Start();
parallel_for(int lidx=0;lidx<lsites;lidx++){
parallel_for(uint64_t lidx=0;lidx<lsites;lidx++){
std::vector<RngStateType> tmp(RngStateCount);
std::copy(iodata[lidx].begin(),iodata[lidx].end(),tmp.begin());
parallel.SetState(tmp,lidx);
@ -659,7 +661,7 @@ class BinaryIO {
static inline void writeRNG(GridSerialRNG &serial,
GridParallelRNG &parallel,
std::string file,
Integer offset,
uint64_t offset,
uint32_t &nersc_csum,
uint32_t &scidac_csuma,
uint32_t &scidac_csumb)
@ -670,8 +672,8 @@ class BinaryIO {
typedef std::array<RngStateType,RngStateCount> RNGstate;
GridBase *grid = parallel._grid;
int gsites = grid->gSites();
int lsites = grid->lSites();
uint64_t gsites = grid->gSites();
uint64_t lsites = grid->lSites();
uint32_t nersc_csum_tmp;
uint32_t scidac_csuma_tmp;
@ -684,7 +686,7 @@ class BinaryIO {
timer.Start();
std::vector<RNGstate> iodata(lsites);
parallel_for(int lidx=0;lidx<lsites;lidx++){
parallel_for(uint64_t lidx=0;lidx<lsites;lidx++){
std::vector<RngStateType> tmp(RngStateCount);
parallel.GetState(tmp,lidx);
std::copy(tmp.begin(),tmp.end(),iodata[lidx].begin());