portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-03 21:44:33 +00:00
Code Releases Activity

Got the NERSC IO working and fixed a bug in cshift.

Browse Source
This commit is contained in:
Peter Boyle
2015-04-22 22:46:48 +01:00
parent 42f167ea37
commit b32c14b433
22 changed files with 925 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
lib/Grid.h
View File

@@ -51,6 +51,7 @@
#include <Grid_where.h>
#include <Grid_stencil.h>
#include <qcd/Grid_QCD.h>
#include <parallelIO/GridNerscIO.h>
namespace Grid {

7
lib/Grid_communicator.h
View File

@@ -35,6 +35,7 @@ class CartesianCommunicator {
// Grid information queries
/////////////////////////////////
int IsBoss(void) { return _processor==0; };
int BossRank(void) { return 0; };
int ThisRank(void) { return _processor; };
const std::vector<int> & ThisProcessorCoor(void) { return _processor_coor; };
const std::vector<int> & ProcessorGrid(void) { return _processors; };
@@ -49,6 +50,8 @@ class CartesianCommunicator {
void GlobalSum(RealD &);
void GlobalSumVector(RealD *,int N);
void GlobalSum(uint32_t &);
void GlobalSum(ComplexF &c)
{
GlobalSumVector((float *)&c,2);
@@ -68,12 +71,10 @@ class CartesianCommunicator {
}
template<class obj> void GlobalSum(obj &o){
typedef typename obj::scalar_type scalar_type;
int words = sizeof(obj)/sizeof(scalar_type);
scalar_type * ptr = (scalar_type *)& o;
GlobalSum(ptr,words);
GlobalSumVector(ptr,words);
}
////////////////////////////////////////////////////////////
// Face exchange

24
lib/Grid_comparison.h
View File

@@ -8,42 +8,42 @@ namespace Grid {
public:
vInteger operator()(const lobj &lhs, const robj &rhs)
{
return lhs == rhs;
return TensorRemove(lhs) == TensorRemove(rhs);
}
};
template<class lobj,class robj> class vne {
public:
vInteger operator()(const lobj &lhs, const robj &rhs)
{
return lhs != rhs;
return TensorRemove(lhs) != TensorRemove(rhs);
}
};
template<class lobj,class robj> class vlt {
public:
vInteger operator()(const lobj &lhs, const robj &rhs)
{
return lhs < rhs;
return TensorRemove(lhs) < TensorRemove(rhs);
}
};
template<class lobj,class robj> class vle {
public:
vInteger operator()(const lobj &lhs, const robj &rhs)
{
return lhs <= rhs;
return TensorRemove(lhs) <= TensorRemove(rhs);
}
};
template<class lobj,class robj> class vgt {
public:
vInteger operator()(const lobj &lhs, const robj &rhs)
{
return lhs > rhs;
return TensorRemove(lhs) > TensorRemove(rhs);
}
};
template<class lobj,class robj> class vge {
public:
vInteger operator()(const lobj &lhs, const robj &rhs)
{
return lhs >= rhs;
return TensorRemove(lhs) >= TensorRemove(rhs);
}
};
@@ -52,42 +52,42 @@ namespace Grid {
public:
Integer operator()(const lobj &lhs, const robj &rhs)
{
return lhs == rhs;
return TensorRemove(lhs) == TensorRemove(rhs);
}
};
template<class lobj,class robj> class sne {
public:
Integer operator()(const lobj &lhs, const robj &rhs)
{
return lhs != rhs;
return TensorRemove(lhs) != TensorRemove(rhs);
}
};
template<class lobj,class robj> class slt {
public:
Integer operator()(const lobj &lhs, const robj &rhs)
{
return lhs < rhs;
return TensorRemove(lhs) < TensorRemove(rhs);
}
};
template<class lobj,class robj> class sle {
public:
Integer operator()(const lobj &lhs, const robj &rhs)
{
return lhs <= rhs;
return TensorRemove(lhs) <= TensorRemove(rhs);
}
};
template<class lobj,class robj> class sgt {
public:
Integer operator()(const lobj &lhs, const robj &rhs)
{
return lhs > rhs;
return TensorRemove(lhs) > TensorRemove(rhs);
}
};
template<class lobj,class robj> class sge {
public:
Integer operator()(const lobj &lhs, const robj &rhs)
{
return lhs >= rhs;
return TensorRemove(lhs) >= TensorRemove(rhs);
}
};

6
lib/Grid_config.h
View File

@@ -16,6 +16,9 @@
/* GRID_COMMS_NONE */
/* #undef GRID_COMMS_NONE */
/* Define to 1 if you have the `be64toh' function. */
/* #undef HAVE_BE64TOH */
/* Define to 1 if you have the `gettimeofday' function. */
#define HAVE_GETTIMEOFDAY 1
@@ -31,6 +34,9 @@
/* Define to 1 if you have the <memory.h> header file. */
#define HAVE_MEMORY_H 1
/* Define to 1 if you have the `ntohll' function. */
/* #undef HAVE_NTOHLL */
/* Define to 1 if you have the <stdint.h> header file. */
#define HAVE_STDINT_H 1

6
lib/Grid_config.h.in
View File

@@ -15,6 +15,9 @@
/* GRID_COMMS_NONE */
#undef GRID_COMMS_NONE
/* Define to 1 if you have the `be64toh' function. */
#undef HAVE_BE64TOH
/* Define to 1 if you have the `gettimeofday' function. */
#undef HAVE_GETTIMEOFDAY
@@ -30,6 +33,9 @@
/* Define to 1 if you have the <memory.h> header file. */
#undef HAVE_MEMORY_H
/* Define to 1 if you have the `ntohll' function. */
#undef HAVE_NTOHLL
/* Define to 1 if you have the <stdint.h> header file. */
#undef HAVE_STDINT_H

12
lib/Grid_lattice.h
View File

@@ -37,7 +37,8 @@ public:
// what about a default grid?
//////////////////////////////////////////////////////////////////
Lattice(GridBase *grid) : _grid(grid) {
_odata.reserve(_grid->oSites());
// _odata.reserve(_grid->oSites());
_odata.resize(_grid->oSites());
assert((((uint64_t)&_odata[0])&0xF) ==0);
checkerboard=0;
}
@@ -49,6 +50,15 @@ public:
}
return *this;
}
template<class robj> inline Lattice<vobj> & operator = (const Lattice<robj> & r){
conformable(*this,r);
std::cout<<"Lattice operator ="<<std::endl;
for(int ss=0;ss<_grid->oSites();ss++){
this->_odata[ss]=r._odata[ss];
}
return *this;
}
// *=,+=,-= operators inherit behvour from correspond */+/- operation
template<class T> inline Lattice<vobj> &operator *=(const T &r) {

41
lib/cartesian/Grid_cartesian_base.h
View File

@@ -6,34 +6,23 @@
namespace Grid{
class GridRNG ; // Forward declaration;
//////////////////////////////////////////////////////////////////////
// Commicator provides information on the processor grid
//////////////////////////////////////////////////////////////////////
// unsigned long _ndimension;
// std::vector<int> _processors; // processor grid
// int _processor; // linear processor rank
// std::vector<int> _processor_coor; // linear processor rank
//////////////////////////////////////////////////////////////////////
class GridBase : public CartesianCommunicator {
public:
// Give Lattice access
template<class object> friend class Lattice;
GridRNG *_rng;
GridBase(std::vector<int> & processor_grid) : CartesianCommunicator(processor_grid) {};
//FIXME
// protected:
// Lattice wide random support. not yet fully implemented. Need seed strategy
// and one generator per site.
// std::default_random_engine generator;
// static std::mt19937 generator( 9 );
//////////////////////////////////////////////////////////////////////
// Commicator provides information on the processor grid
//////////////////////////////////////////////////////////////////////
// unsigned long _ndimension;
// std::vector<int> _processors; // processor grid
// int _processor; // linear processor rank
// std::vector<int> _processor_coor; // linear processor rank
//////////////////////////////////////////////////////////////////////
// Physics Grid information.
std::vector<int> _simd_layout;// Which dimensions get relayed out over simd lanes.
std::vector<int> _fdimensions;// Global dimensions of array prior to cb removal
@@ -101,6 +90,13 @@ public:
for(int d=0;d<_ndimension;d++) idx+=_ostride[d]*ocoor[d];
return idx;
}
inline void CoorFromIndex (std::vector<int>& coor,int index,std::vector<int> &dims){
coor.resize(_ndimension);
for(int d=0;d<_ndimension;d++){
coor[d] = index % dims[d];
index = index / dims[d];
}
}
inline void oCoorFromOindex (std::vector<int>& coor,int Oindex){
coor.resize(_ndimension);
for(int d=0;d<_ndimension;d++){
@@ -146,6 +142,7 @@ public:
inline int lSites(void) { return _isites*_osites; };
inline int gSites(void) { return _isites*_osites*_Nprocessors; };
inline int Nd (void) { return _ndimension;};
inline const std::vector<int> &FullDimensions(void) { return _fdimensions;};
inline const std::vector<int> &GlobalDimensions(void) { return _gdimensions;};
inline const std::vector<int> &LocalDimensions(void) { return _ldimensions;};
@@ -175,10 +172,10 @@ public:
std::vector<int> coor(_ndimension);
ProcessorCoorFromRank(rank,coor);
for(int mu=0;mu<_ndimension;mu++) gcoor[mu] = _ldimensions[mu]&coor[mu];
for(int mu=0;mu<_ndimension;mu++) gcoor[mu] = _ldimensions[mu]*coor[mu];
iCoorFromIindex(coor,i_idx);
for(int mu=0;mu<_ndimension;mu++) gcoor[mu] += _rdimensions[mu]&coor[mu];
for(int mu=0;mu<_ndimension;mu++) gcoor[mu] += _rdimensions[mu]*coor[mu];
oCoorFromOindex (coor,o_idx);
for(int mu=0;mu<_ndimension;mu++) gcoor[mu] += coor[mu];

1
lib/communicator/Grid_communicator_fake.cc
View File

@@ -17,6 +17,7 @@ CartesianCommunicator::CartesianCommunicator(std::vector<int> &processors)
void CartesianCommunicator::GlobalSum(float &){}
void CartesianCommunicator::GlobalSumVector(float *,int N){}
void CartesianCommunicator::GlobalSum(double &){}
void CartesianCommunicator::GlobalSum(uint32_t &){}
void CartesianCommunicator::GlobalSumVector(double *,int N){}
// Basic Halo comms primitive

3
lib/communicator/Grid_communicator_mpi.cc
View File

@@ -28,6 +28,9 @@ CartesianCommunicator::CartesianCommunicator(std::vector<int> &processors)
assert(Size==_Nprocessors);
}
void CartesianCommunicator::GlobalSum(uint32_t &u){
MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT32_T,MPI_SUM,communicator);
}
void CartesianCommunicator::GlobalSum(float &f){
MPI_Allreduce(MPI_IN_PLACE,&f,1,MPI_FLOAT,MPI_SUM,communicator);
}

8
lib/cshift/Grid_cshift_mpi.h
View File

@@ -188,8 +188,8 @@ template<class vobj> void Cshift_comms_simd(Lattice<vobj> &ret,Lattice<vobj> &r
if ( comm_any ) {
for(int i=0;i<Nsimd;i++){
pointers[i] = (scalar_type *)&send_buf_extract[i][0];
for(int i=0;i<Nsimd;i++){ // there is a reversal in extract merge
pointers[i] = (scalar_type *)&send_buf_extract[Nsimd-1-i][0];
}
Gather_plane_extract(rhs,pointers,dimension,sx,cbmask);
@@ -238,9 +238,9 @@ template<class vobj> void Cshift_comms_simd(Lattice<vobj> &ret,Lattice<vobj> &r
for(int i=0;i<Nsimd;i++){
if ( permute_slice ) {
PermuteMap=i^toggle_bit;
pointers[i] = rpointers[PermuteMap];
pointers[Nsimd-1-i] = rpointers[PermuteMap];
} else {
pointers[i] = rpointers[i];
pointers[Nsimd-1-i] = rpointers[i];
}
}

9
lib/lattice/Grid_lattice_coordinate.h
View File

@@ -5,16 +5,17 @@ namespace Grid {
template<class iobj> inline void LatticeCoordinate(Lattice<iobj> &l,int mu)
{
typedef typename iobj::scalar_type scalar_type;
typedef typename iobj::vector_type vector_type;
GridBase *grid = l._grid;
int Nsimd = grid->iSites();
std::vector<int> gcoor;
std::vector<Integer> mergebuf(Nsimd);
std::vector<Integer *> mergeptr(Nsimd);
vInteger vI;
std::vector<scalar_type> mergebuf(Nsimd);
std::vector<scalar_type *> mergeptr(Nsimd);
vector_type vI;
for(int o=0;o<grid->oSites();o++){
for(int i=0;i<grid->iSites();i++){
grid->RankIndexToGlobalCoor(grid->ThisRank(),o,i,gcoor);
// grid->RankIndexToGlobalCoor(0,o,i,gcoor);
mergebuf[i]=gcoor[mu];
mergeptr[i]=&mergebuf[i];
}

84
lib/lattice/Grid_lattice_peekpoke.h
View File

@@ -43,6 +43,7 @@ namespace Grid {
}
return ret;
};
////////////////////////////////////////////////////////////////////////////////////////////////////
// Poke internal indices of a Lattice object
////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -88,26 +89,26 @@ namespace Grid {
assert( sizeof(sobj)*Nsimd == sizeof(vobj));
int rank,odx,idx;
grid->GlobalCoorToRankIndex(rank,odx,idx,site);
// Optional to broadcast from node 0.
grid->Broadcast(0,s);
grid->GlobalCoorToRankIndex(rank,odx,idx,site);
grid->Broadcast(grid->BossRank(),s);
std::vector<sobj> buf(Nsimd);
std::vector<scalar_type *> pointers(Nsimd);
for(int i=0;i<Nsimd;i++) pointers[i] = (scalar_type *)&buf[i];
// extract-modify-merge cycle is easiest way and this is not perf critical
extract(l._odata[odx],pointers);
buf[idx] = s;
for(int i=0;i<Nsimd;i++) pointers[i] = (scalar_type *)&buf[i];
merge(l._odata[odx],pointers);
if ( rank == grid->ThisRank() ) {
for(int i=0;i<Nsimd;i++) pointers[i] = (scalar_type *)&buf[i];
extract(l._odata[odx],pointers);
buf[idx] = s;
for(int i=0;i<Nsimd;i++) pointers[i] = (scalar_type *)&buf[i];
merge(l._odata[odx],pointers);
}
return;
};
//////////////////////////////////////////////////////////
// Peek a scalar object from the SIMD array
//////////////////////////////////////////////////////////
@@ -138,6 +139,69 @@ namespace Grid {
return;
};
//////////////////////////////////////////////////////////
// Peek a scalar object from the SIMD array
//////////////////////////////////////////////////////////
template<class vobj,class sobj>
void peekLocalSite(sobj &s,Lattice<vobj> &l,std::vector<int> &site){
GridBase *grid=l._grid;
typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::vector_type vector_type;
int Nsimd = grid->Nsimd();
assert( l.checkerboard== l._grid->CheckerBoard(site));
assert( sizeof(sobj)*Nsimd == sizeof(vobj));
int odx,idx;
idx= grid->iIndex(site);
odx= grid->oIndex(site);
std::vector<sobj> buf(Nsimd);
std::vector<scalar_type *> pointers(Nsimd);
for(int i=0;i<Nsimd;i++) pointers[i] = (scalar_type *)&buf[i];
extract(l._odata[odx],pointers);
s = buf[idx];
return;
};
template<class vobj,class sobj>
void pokeLocalSite(const sobj &s,Lattice<vobj> &l,std::vector<int> &site){
GridBase *grid=l._grid;
typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::vector_type vector_type;
int Nsimd = grid->Nsimd();
assert( l.checkerboard== l._grid->CheckerBoard(site));
assert( sizeof(sobj)*Nsimd == sizeof(vobj));
int odx,idx;
idx= grid->iIndex(site);
odx= grid->oIndex(site);
std::vector<sobj> buf(Nsimd);
std::vector<scalar_type *> pointers(Nsimd);
for(int i=0;i<Nsimd;i++) pointers[i] = (scalar_type *)&buf[i];
// extract-modify-merge cycle is easiest way and this is not perf critical
extract(l._odata[odx],pointers);
buf[idx] = s;
for(int i=0;i<Nsimd;i++) pointers[i] = (scalar_type *)&buf[i];
merge(l._odata[odx],pointers);
return;
};
}
#endif

30
lib/lattice/Grid_lattice_reduction.h
View File

@@ -40,6 +40,36 @@ namespace Grid {
return nrm;
}
template<class vobj>
inline typename vobj::scalar_object sum(const Lattice<vobj> &arg){
GridBase *grid=arg._grid;
int Nsimd = grid->Nsimd();
typedef typename vobj::scalar_object sobj;
typedef typename vobj::scalar_type scalar_type;
vobj vsum;
sobj ssum;
vsum=zero;
ssum=zero;
for(int ss=0;ss<arg._grid->oSites(); ss++){
vsum = vsum + arg._odata[ss];
}
std::vector<sobj> buf(Nsimd);
std::vector<scalar_type *> pointers(Nsimd);
for(int i=0;i<Nsimd;i++) pointers[i] = (scalar_type *)&buf[i];
extract(vsum,pointers);
for(int i=0;i<Nsimd;i++) ssum = ssum + buf[i];
arg._grid->GlobalSum(ssum);
return ssum;
}
}
#endif

41
lib/math/Grid_math_tensors.h
View File

@@ -13,10 +13,12 @@ template<class vtype> class iScalar
public:
vtype _internal;
typedef typename GridTypeMapper<vtype>::scalar_type scalar_type;
typedef typename GridTypeMapper<vtype>::scalar_type scalar_type;
typedef typename GridTypeMapper<vtype>::vector_type vector_type;
typedef typename GridTypeMapper<vtype>::tensor_reduced tensor_reduced_v;
typedef iScalar<tensor_reduced_v> tensor_reduced;
typedef typename GridTypeMapper<vtype>::scalar_object recurse_scalar_object;
typedef iScalar<recurse_scalar_object> scalar_object;
enum { TensorLevel = GridTypeMapper<vtype>::TensorLevel + 1};
@@ -27,7 +29,7 @@ public:
iScalar(scalar_type s) : _internal(s) {};// recurse down and hit the constructor for vector_type
iScalar(Zero &z){ *this = zero; };
iScalar(const Zero &z){ *this = zero; };
iScalar<vtype> & operator= (const Zero &hero){
zeroit(*this);
@@ -73,11 +75,18 @@ public:
operator ComplexD () const { return(TensorRemove(_internal)); };
operator RealD () const { return(real(TensorRemove(_internal))); }
template<class T,typename std::enable_if<isGridTensor<T>::notvalue, T>::type* = nullptr > inline auto operator = (T arg) -> iScalar<vtype>
{
_internal = arg;
return *this;
}
};
///////////////////////////////////////////////////////////
// Allows to turn scalar<scalar<scalar<double>>>> back to double.
///////////////////////////////////////////////////////////
template<class T> inline typename std::enable_if<isGridTensor<T>::notvalue, T>::type TensorRemove(T arg) { return arg;}
template<class T> inline typename std::enable_if<isGridTensor<T>::notvalue, T>::type TensorRemove(T arg) { return arg;}
template<class vtype> inline auto TensorRemove(iScalar<vtype> arg) -> decltype(TensorRemove(arg._internal))
{
return TensorRemove(arg._internal);
@@ -91,14 +100,17 @@ public:
typedef typename GridTypeMapper<vtype>::scalar_type scalar_type;
typedef typename GridTypeMapper<vtype>::vector_type vector_type;
typedef typename GridTypeMapper<vtype>::tensor_reduced tensor_reduced_v;
typedef typename GridTypeMapper<vtype>::scalar_object recurse_scalar_object;
typedef iScalar<tensor_reduced_v> tensor_reduced;
typedef iVector<recurse_scalar_object,N> scalar_object;
enum { TensorLevel = GridTypeMapper<vtype>::TensorLevel + 1};
typedef iScalar<tensor_reduced_v> tensor_reduced;
iVector(Zero &z){ *this = zero; };
iVector(const Zero &z){ *this = zero; };
iVector() {};// Empty constructure
iVector<vtype,N> & operator= (Zero &hero){
iVector<vtype,N> & operator= (const Zero &hero){
zeroit(*this);
return *this;
}
@@ -153,18 +165,27 @@ public:
typedef typename GridTypeMapper<vtype>::scalar_type scalar_type;
typedef typename GridTypeMapper<vtype>::vector_type vector_type;
typedef typename GridTypeMapper<vtype>::tensor_reduced tensor_reduced_v;
typedef typename GridTypeMapper<vtype>::scalar_object recurse_scalar_object;
typedef iScalar<tensor_reduced_v> tensor_reduced;
typedef iMatrix<recurse_scalar_object,N> scalar_object;
enum { TensorLevel = GridTypeMapper<vtype>::TensorLevel + 1};
typedef iScalar<tensor_reduced_v> tensor_reduced;
iMatrix(Zero &z){ *this = zero; };
iMatrix(const Zero &z){ *this = zero; };
iMatrix() {};
iMatrix<vtype,N> & operator= (Zero &hero){
iMatrix<vtype,N> & operator= (const Zero &hero){
zeroit(*this);
return *this;
}
template<class T,typename std::enable_if<isGridTensor<T>::notvalue, T>::type* = nullptr > inline auto operator = (T arg) -> iMatrix<vtype,N>
{
zeroit(*this);
for(int i=0;i<N;i++)
_internal[i][i] = arg;
return *this;
}
friend void zeroit(iMatrix<vtype,N> &that){
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){

14
lib/math/Grid_math_traits.h
View File

@@ -25,6 +25,7 @@ namespace Grid {
typedef typename T::scalar_type scalar_type;
typedef typename T::vector_type vector_type;
typedef typename T::tensor_reduced tensor_reduced;
typedef typename T::scalar_object scalar_object;
enum { TensorLevel = T::TensorLevel };
};
@@ -36,6 +37,7 @@ namespace Grid {
typedef RealF scalar_type;
typedef RealF vector_type;
typedef RealF tensor_reduced ;
typedef RealF scalar_object;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<RealD> {
@@ -43,6 +45,7 @@ namespace Grid {
typedef RealD scalar_type;
typedef RealD vector_type;
typedef RealD tensor_reduced;
typedef RealD scalar_object;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<ComplexF> {
@@ -50,6 +53,7 @@ namespace Grid {
typedef ComplexF scalar_type;
typedef ComplexF vector_type;
typedef ComplexF tensor_reduced;
typedef ComplexF scalar_object;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<ComplexD> {
@@ -57,6 +61,7 @@ namespace Grid {
typedef ComplexD scalar_type;
typedef ComplexD vector_type;
typedef ComplexD tensor_reduced;
typedef ComplexD scalar_object;
enum { TensorLevel = 0 };
};
@@ -65,6 +70,7 @@ namespace Grid {
typedef RealF scalar_type;
typedef vRealF vector_type;
typedef vRealF tensor_reduced;
typedef RealF scalar_object;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<vRealD> {
@@ -72,6 +78,7 @@ namespace Grid {
typedef RealD scalar_type;
typedef vRealD vector_type;
typedef vRealD tensor_reduced;
typedef RealD scalar_object;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<vComplexF> {
@@ -79,6 +86,7 @@ namespace Grid {
typedef ComplexF scalar_type;
typedef vComplexF vector_type;
typedef vComplexF tensor_reduced;
typedef ComplexF scalar_object;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<vComplexD> {
@@ -86,6 +94,7 @@ namespace Grid {
typedef ComplexD scalar_type;
typedef vComplexD vector_type;
typedef vComplexD tensor_reduced;
typedef ComplexD scalar_object;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<vInteger> {
@@ -93,6 +102,7 @@ namespace Grid {
typedef Integer scalar_type;
typedef vInteger vector_type;
typedef vInteger tensor_reduced;
typedef Integer scalar_object;
enum { TensorLevel = 0 };
};
@@ -102,6 +112,10 @@ namespace Grid {
static const bool notvalue = false;
};
template<> struct isGridTensor<int > {
static const bool value = false;
static const bool notvalue = true;
};
template<> struct isGridTensor<RealD > {
static const bool value = false;
static const bool notvalue = true;

524
lib/parallelIO/GridNerscIO.h Normal file
View File

@@ -0,0 +1,524 @@
#ifndef GRID_NERSC_IO_H
#define GRID_NERSC_IO_H
#include <algorithm>
#include <iostream>
#include <iomanip>
#include <fstream>
#include <arpa/inet.h>
namespace Grid {
using namespace QCD;
////////////////////////////////////////////////////////////////////////////////
// Some data types for intermediate storage
////////////////////////////////////////////////////////////////////////////////
template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, 2>, 4 >;
typedef iLorentzColour2x3<Complex> LorentzColour2x3;
typedef iLorentzColour2x3<ComplexF> LorentzColour2x3F;
typedef iLorentzColour2x3<ComplexD> LorentzColour2x3D;
////////////////////////////////////////////////////////////////////////////////
// header specification/interpretation
////////////////////////////////////////////////////////////////////////////////
class NerscField {
public:
// header strings (not in order)
int dimension[4];
std::string boundary[4];
int data_start;
std::string hdr_version;
std::string storage_format;
// Checks on data
double link_trace;
double plaquette;
uint32_t checksum;
unsigned int sequence_number;
std::string data_type;
std::string ensemble_id ;
std::string ensemble_label ;
std::string creator ;
std::string creator_hardware ;
std::string creation_date ;
std::string archive_date ;
std::string floating_point;
};
////////////////////////////////////////////////////////////////////////////////
// Write and read from fstream; comput header offset for payload
////////////////////////////////////////////////////////////////////////////////
inline unsigned int writeNerscHeader(NerscField &field,std::string file)
{
std::ofstream fout(file,std::ios::out);
fout.seekp(0,std::ios::beg);
fout << "BEGIN_HEADER" << std::endl;
fout << "HDR_VERSION = " << field.hdr_version << std::endl;
fout << "DATATYPE = " << field.data_type << std::endl;
fout << "STORAGE_FORMAT = " << field.storage_format << std::endl;
for(int i=0;i<4;i++){
fout << "DIMENSION_" << i+1 << " = " << field.dimension[i] << std::endl ;
}
// just to keep the space and write it later
fout << "LINK_TRACE = " << std::setprecision(10) << field.link_trace << std::endl;
fout << "PLAQUETTE = " << std::setprecision(10) << field.plaquette << std::endl;
for(int i=0;i<4;i++){
fout << "BOUNDARY_"<<i+1<<" = " << field.boundary[i] << std::endl;
}
fout << "CHECKSUM = "<< std::hex << std::setw(16) << 0 << field.checksum << std::endl;
fout << "ENSEMBLE_ID = " << field.ensemble_id << std::endl;
fout << "ENSEMBLE_LABEL = " << field.ensemble_label << std::endl;
fout << "SEQUENCE_NUMBER = " << field.sequence_number << std::endl;
fout << "CREATOR = " << field.creator << std::endl;
fout << "CREATOR_HARDWARE = "<< field.creator_hardware << std::endl;
fout << "CREATION_DATE = " << field.creation_date << std::endl;
fout << "ARCHIVE_DATE = " << field.archive_date << std::endl;
fout << "FLOATING_POINT = " << field.floating_point << std::endl;
fout << "END_HEADER" << std::endl;
field.data_start = fout.tellp();
return field.data_start;
}
// A little helper
inline void removeWhitespace(std::string &key)
{
key.erase(std::remove_if(key.begin(), key.end(), ::isspace),key.end());
}
// for the header-reader
inline int readNerscHeader(std::string file,GridBase *grid, NerscField &field)
{
int offset=0;
std::map<std::string,std::string> header;
std::string line;
//////////////////////////////////////////////////
// read the header
//////////////////////////////////////////////////
std::ifstream fin(file);
getline(fin,line); // read one line and insist is
removeWhitespace(line);
assert(line==std::string("BEGIN_HEADER"));
do {
getline(fin,line); // read one line
int eq = line.find("=");
if(eq >0) {
std::string key=line.substr(0,eq);
std::string val=line.substr(eq+1);
removeWhitespace(key);
removeWhitespace(val);
header[key] = val;
}
} while( line.find("END_HEADER") == std::string::npos );
field.data_start = fin.tellg();
//////////////////////////////////////////////////
// chomp the values
//////////////////////////////////////////////////
field.hdr_version = header[std::string("HDR_VERSION")];
field.data_type = header[std::string("DATATYPE")];
field.storage_format = header[std::string("STORAGE_FORMAT")];
field.dimension[0] = std::stol(header[std::string("DIMENSION_1")]);
field.dimension[1] = std::stol(header[std::string("DIMENSION_2")]);
field.dimension[2] = std::stol(header[std::string("DIMENSION_3")]);
field.dimension[3] = std::stol(header[std::string("DIMENSION_4")]);
assert(grid->_ndimension == 4);
for(int d=0;d<4;d++){
assert(grid->_fdimensions[d]==field.dimension[d]);
}
field.link_trace = std::stod(header[std::string("LINK_TRACE")]);
field.plaquette = std::stod(header[std::string("PLAQUETTE")]);
field.boundary[0] = header[std::string("BOUNDARY_1")];
field.boundary[1] = header[std::string("BOUNDARY_2")];
field.boundary[2] = header[std::string("BOUNDARY_3")];
field.boundary[3] = header[std::string("BOUNDARY_4")];
field.checksum = std::stoul(header[std::string("CHECKSUM")],0,16);
field.ensemble_id = header[std::string("ENSEMBLE_ID")];
field.ensemble_label = header[std::string("ENSEMBLE_LABEL")];
field.sequence_number = std::stol(header[std::string("SEQUENCE_NUMBER")]);
field.creator = header[std::string("CREATOR")];
field.creator_hardware = header[std::string("CREATOR_HARDWARE")];
field.creation_date = header[std::string("CREATION_DATE")];
field.archive_date = header[std::string("ARCHIVE_DATE")];
field.floating_point = header[std::string("FLOATING_POINT")];
return field.data_start;
}
//////////////////////////////////////////////////////////////////////
// Utilities
//////////////////////////////////////////////////////////////////////
inline void reconstruct3(LorentzColourMatrix & cm)
{
const int x=0;
const int y=1;
const int z=2;
for(int mu=0;mu<4;mu++){
cm(mu)()(2,x) = adj(cm(mu)()(0,y)*cm(mu)()(1,z)-cm(mu)()(0,z)*cm(mu)()(1,y)); //x= yz-zy
cm(mu)()(2,y) = adj(cm(mu)()(0,z)*cm(mu)()(1,x)-cm(mu)()(0,x)*cm(mu)()(1,z)); //y= zx-xz
cm(mu)()(2,z) = adj(cm(mu)()(0,x)*cm(mu)()(1,y)-cm(mu)()(0,y)*cm(mu)()(1,x)); //z= xy-yx
}
}
void inline be32toh(void *file_object,uint32_t bytes)
{
uint32_t * f = (uint32_t *)file_object;
for(int i=0;i*sizeof(uint32_t)<bytes;i++){
f[i] = ntohl(f[i]);
}
}
void inline le32toh(void *file_object,uint32_t bytes)
{
uint32_t *fp = (uint32_t *)file_object;
uint32_t f;
for(int i=0;i*sizeof(uint32_t)<bytes;i++){
f = fp[i];
// got network order and the network to host
f = ((f&0xFF)<<24) | ((f&0xFF00)<<8) | ((f&0xFF0000)>>8) | ((f&0xFF000000UL)>>24) ;
fp[i] = ntohl(f);
}
}
void inline be64toh(void *file_object,uint32_t bytes)
{
uint64_t * f = (uint64_t *)file_object;
for(int i=0;i*sizeof(uint64_t)<bytes;i++){
f[i] = ntohll(f[i]);
}
}
void inline le64toh(void *file_object,uint32_t bytes)
{
uint64_t *fp = (uint64_t *)file_object;
uint64_t f,g;
for(int i=0;i*sizeof(uint64_t)<bytes;i++){
f = fp[i];
// got network order and the network to host
g = ((f&0xFF)<<24) | ((f&0xFF00)<<8) | ((f&0xFF0000)>>8) | ((f&0xFF000000UL)>>24) ;
g = g << 32;
f = f >> 32;
g|= ((f&0xFF)<<24) | ((f&0xFF00)<<8) | ((f&0xFF0000)>>8) | ((f&0xFF000000UL)>>24) ;
fp[i] = ntohl(g);
}
}
inline void NerscChecksum(uint32_t *buf,uint32_t buf_size,uint32_t &csum)
{
for(int i=0;i*sizeof(uint32_t)<buf_size;i++){
csum=csum+buf[i];
}
}
template<class fobj,class sobj>
struct NerscSimpleMunger{
void operator() (fobj &in,sobj &out,uint32_t &csum){
for(int mu=0;mu<4;mu++){
for(int i=0;i<3;i++){
for(int j=0;j<3;j++){
out(mu)()(i,j) = in(mu)()(i,j);
}}}
NerscChecksum((uint32_t *)&in,sizeof(in),csum);
};
};
template<class fobj,class sobj>
struct NerscSimpleUnmunger{
void operator() (sobj &in,fobj &out,uint32_t &csum){
for(int mu=0;mu<4;mu++){
for(int i=0;i<3;i++){
for(int j=0;j<3;j++){
out(mu)()(i,j) = in(mu)()(i,j);
}}}
NerscChecksum((uint32_t *)&out,sizeof(out),csum);
};
};
template<class fobj,class sobj>
struct Nersc3x2munger{
void operator() (fobj &in,sobj &out,uint32_t &csum){
NerscChecksum((uint32_t *)&in,sizeof(in),csum);
for(int mu=0;mu<4;mu++){
for(int i=0;i<2;i++){
for(int j=0;j<3;j++){
out(mu)()(i,j) = in(mu)(i)(j);
}}
}
reconstruct3(out);
}
};
template<class fobj,class sobj>
struct Nersc3x2unmunger{
void operator() (sobj &in,fobj &out,uint32_t &csum){
NerscChecksum((uint32_t *)&out,sizeof(out),csum);
for(int mu=0;mu<4;mu++){
for(int i=0;i<2;i++){
for(int j=0;j<3;j++){
out(mu)(i)(j) = in(mu)()(i,j);
}}
}
}
};
////////////////////////////////////////////////////////////////////////////
// Template wizardry to map types to strings for NERSC in an extensible way
////////////////////////////////////////////////////////////////////////////
template<class vobj> struct NerscDataType {
static void DataType (std::string &str) { str = std::string("4D_BINARY_UNKNOWN"); };
static void FloatingPoint(std::string &str) { str = std::string("IEEE64BIG"); };
};
template<> struct NerscDataType<iColourMatrix<ComplexD> > {
static void DataType (std::string &str) { str = std::string("4D_SU3_GAUGE_3X3"); };
static void FloatingPoint(std::string &str) { str = std::string("IEEE64BIG");};
};
template<> struct NerscDataType<iColourMatrix<ComplexF> > {
static void DataType (std::string &str) { str = std::string("4D_SU3_GAUGE_3X3"); };
static void FloatingPoint(std::string &str) { str = std::string("IEEE32BIG");};
};
//////////////////////////////////////////////////////////////////////
// Bit and Physical Checksumming and QA of data
//////////////////////////////////////////////////////////////////////
/*
template<class vobj> inline uint32_t NerscChecksum(Lattice<vobj> & data)
{
uint32_t sum;
for(int ss=0;ss<data._grid->Osites();ss++){
uint32_t *iptr = (uint32_t *)& data._odata[0] ;
for(int i=0;i<sizeof(vobj);i+=sizeof(uint32_t)){
sum=sum+iptr[i];
}
}
data._grid->globalSum(sum);
return sum;
}
*/
template<class vobj> inline void NerscPhysicalCharacteristics(Lattice<vobj> & data,NerscField &header)
{
header.data_type = NerscDataType<vobj>::DataType;
header.floating_point = NerscDataType<vobj>::FloatingPoint;
return;
}
template<> inline void NerscPhysicalCharacteristics(LatticeGaugeField & data,NerscField &header)
{
NerscDataType<decltype(data._odata[0])>::DataType(header.data_type);
NerscDataType<decltype(data._odata[0])>::FloatingPoint(header.floating_point);
header.link_trace=1.0;
header.plaquette =1.0;
}
template<class vobj> inline void NerscStatisics(Lattice<vobj> & data,NerscField &header)
{
assert(data._grid->_ndimension==4);
for(int d=0;d<4;d++)
header.dimension[d] = data._grid->_fdimensions[d];
// compute checksum and any physical properties contained for this type
// header.checksum = NerscChecksum(data);
NerscPhysicalCharacteristics(data,header);
}
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Now the meat: the object readers
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
template<class vobj,class sobj,class fobj,class munger>
inline void readNerscObject(Lattice<vobj> &Umu,std::string file,munger munge,int offset,std::string &format)
{
GridBase *grid = Umu._grid;
int ieee32big = (format == std::string("IEEE32BIG"));
int ieee32 = (format == std::string("IEEE32"));
int ieee64big = (format == std::string("IEEE64BIG"));
int ieee64 = (format == std::string("IEEE64"));
// Find the location of each site and send to primary node
// for(int site=0; site < Layout::vol(); ++site){
// multi1d<int> coord = crtesn(site, Layout::lattSize());
// for(int dd=0; dd<Nd; dd++){ /* dir */
// cfg_in.readArray(su3_buffer, float_size, mat_size);
//
// Above from Chroma; defines loop order now that NERSC doc no longer
// available (how short sighted is that?)
{
std::ifstream fin(file,std::ios::binary|std::ios::in);
fin.seekg(offset);
Umu = zero;
uint32_t csum=0;
fobj file_object;
sobj munged;
for(int t=0;t<grid->_fdimensions[3];t++){
for(int z=0;z<grid->_fdimensions[2];z++){
for(int y=0;y<grid->_fdimensions[1];y++){
for(int x=0;x<grid->_fdimensions[0];x++){
std::vector<int> site({x,y,z,t});
if ( grid->IsBoss() ) {
fin.read((char *)&file_object,sizeof(file_object));
if(ieee32big) be32toh((void *)&file_object,sizeof(file_object));
if(ieee32) le32toh((void *)&file_object,sizeof(file_object));
if(ieee64big) be64toh((void *)&file_object,sizeof(file_object));
if(ieee64) le64toh((void *)&file_object,sizeof(file_object));
munge(file_object,munged,csum);
}
// The boss who read the file has their value poked
pokeSite(munged,Umu,site);
}}}}
}
}
template<class vobj,class sobj,class fobj,class munger>
inline void writeNerscObject(Lattice<vobj> &Umu,std::string file,munger munge,int offset,
int sequence,double lt,double pl)
{
GridBase *grid = Umu._grid;
NerscField header;
//////////////////////////////////////////////////
// First write the header; this is in wrong place
//////////////////////////////////////////////////
assert(grid->_ndimension == 4);
for(int d=0;d<4;d++){
header.dimension[d]=grid->_fdimensions[d];
header.boundary [d]=std::string("PERIODIC");;
}
header.hdr_version=std::string("WHATDAHECK");
// header.storage_format=storage_format<vobj>::string; // use template specialisation
// header.data_type=data_type<vobj>::string;
header.storage_format=std::string("debug");
header.data_type =std::string("debug");
//FIXME; use template specialisation to fill these out
header.link_trace =lt;
header.plaquette =pl;
header.checksum =0;
//
header.sequence_number =sequence;
header.ensemble_id =std::string("UKQCD");
header.ensemble_label =std::string("UKQCD");
header.creator =std::string("Tadahito");
header.creator_hardware=std::string("BlueGene/Q");
header.creation_date =std::string("AnnoDomini");
header.archive_date =std::string("AnnoDomini");
header.floating_point =std::string("IEEE64BIG");
// header.data_start=;
// unsigned int checksum;
//////////////////////////////////////////////////
// Now write the body
//////////////////////////////////////////////////
{
std::ofstream fout(file,std::ios::binary|std::ios::in);
fout.seekp(offset);
Umu = zero;
uint32_t csum=0;
fobj file_object;
sobj unmunged;
for(int t=0;t<grid->_fdimensions[3];t++){
for(int z=0;z<grid->_fdimensions[2];z++){
for(int y=0;y<grid->_fdimensions[1];y++){
for(int x=0;x<grid->_fdimensions[0];x++){
std::vector<int> site({x,y,z,t});
peekSite(unmunged,Umu,site);
munge(unmunged,file_object,csum);
// broadcast & insert
fout.write((char *)&file_object,sizeof(file_object));
}}}}
}
}
inline void readNerscConfiguration(LatticeGaugeField &Umu,NerscField& header,std::string file)
{
GridBase *grid = Umu._grid;
int offset = readNerscHeader(file,Umu._grid,header);
std::string format(header.floating_point);
int ieee32big = (format == std::string("IEEE32BIG"));
int ieee32 = (format == std::string("IEEE32"));
int ieee64big = (format == std::string("IEEE64BIG"));
int ieee64 = (format == std::string("IEEE64"));
// depending on datatype, set up munger;
// munger is a function of <floating point, Real, data_type>
if ( header.data_type == std::string("4D_SU3_GAUGE") ) {
if ( ieee32 || ieee32big ) {
readNerscObject<vLorentzColourMatrix, LorentzColourMatrix, LorentzColour2x3F>
(Umu,file,
Nersc3x2munger<LorentzColour2x3F,LorentzColourMatrix>(),
offset,format);
}
if ( ieee64 || ieee64big ) {
readNerscObject<vLorentzColourMatrix, LorentzColourMatrix, LorentzColour2x3D>
(Umu,file,
Nersc3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),
offset,format);
}
} else if ( header.data_type == std::string("4D_SU3_GAUGE_3X3") ) {
if ( ieee32 || ieee32big ) {
readNerscObject<vLorentzColourMatrix,LorentzColourMatrix,LorentzColourMatrixF>
(Umu,file,NerscSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format);
}
if ( ieee64 || ieee64big ) {
readNerscObject<vLorentzColourMatrix,LorentzColourMatrix,LorentzColourMatrixD>
(Umu,file,NerscSimpleMunger<LorentzColourMatrixD,LorentzColourMatrix>(),offset,format);
}
} else {
assert(0);
}
}
template<class vobj>
inline void writeNerscConfiguration(Lattice<vobj> &Umu,NerscField &header,std::string file)
{
GridBase &grid = Umu._grid;
NerscStatisics(Umu,header);
int offset = writeNerscHeader(header,file);
writeNerscObject(Umu,NerscSimpleMunger<vobj,vobj>(),offset);
}
}
#endif

4
lib/qcd/Grid_QCD.h
View File

@@ -42,6 +42,8 @@ namespace QCD {
typedef iColourMatrix<Complex > ColourMatrix;
typedef iSpinColourMatrix<Complex > SpinColourMatrix;
typedef iLorentzColourMatrix<Complex > LorentzColourMatrix;
typedef iLorentzColourMatrix<ComplexF > LorentzColourMatrixF;
typedef iLorentzColourMatrix<ComplexD > LorentzColourMatrixD;
typedef iSpinVector<Complex > SpinVector;
typedef iColourVector<Complex > ColourVector;
@@ -66,7 +68,7 @@ namespace QCD {
typedef Lattice<vTReal> LatticeReal;
typedef Lattice<vTComplex> LatticeComplex;
typedef Lattice<vInteger> LatticeInteger; // Predicates for "where"
typedef Lattice<vTInteger> LatticeInteger; // Predicates for "where"
typedef Lattice<vColourMatrix> LatticeColourMatrix;
typedef Lattice<vSpinMatrix> LatticeSpinMatrix;