portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-12 20:27:06 +01:00
Code Releases Activity

Merge branch 'master' of https://github.com/paboyle/Grid

Browse Source
This commit is contained in:
Azusa Yamaguchi
2015-12-09 12:48:44 +00:00
parent 967be91692 26161addd0
commit a32a59fc43
24 changed files with 406 additions and 221 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
lib/Config.h.in
View File

@ -137,9 +137,6 @@
/* 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/Init.cc
View File

@ -17,7 +17,6 @@
#define __X86_64
#ifdef HAVE_EXECINFO_H
#include <execinfo.h>
#endif

36
lib/PerfCount.h
View File

@ -5,22 +5,29 @@
#include <ctime>
#include <chrono>
#include <string.h>
#include <sys/ioctl.h>
#include <sys/syscall.h>
#include <linux/perf_event.h>
#include <sys/ioctl.h>
#ifdef __linux__
#include <syscall.h>
#include <linux/perf_event.h>
#else
#include <sys/syscall.h>
#endif
namespace Grid {
#ifdef __linux__
static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid,
int cpu, int group_fd, unsigned long flags)
{
int ret;
int ret=0;
ret = syscall(__NR_perf_event_open, hw_event, pid, cpu,
group_fd, flags);
return ret;
}
#endif
class PerformanceCounter {
@ -63,7 +70,6 @@ public:
int PCT;
struct perf_event_attr pe;
long long count;
int fd;
uint64_t elapsed;
@ -74,15 +80,19 @@ public:
}
PerformanceCounter(int _pct) {
#ifdef __linux__
assert(_pct>=0);
assert(_pct<PERFORMANCE_COUNTER_NUM_TYPES);
fd=-1;
count=0;
PCT =_pct;
Open();
#endif
}
void Open(void)
{
#ifdef __linux__
struct perf_event_attr pe;
memset(&pe, 0, sizeof(struct perf_event_attr));
pe.size = sizeof(struct perf_event_attr);
@ -99,32 +109,48 @@ public:
fprintf(stderr, "Error opening leader %llx for event %s\n", pe.config,name);
perror("Error is");
}
#endif
}
void Start(void)
{
#ifdef __linux__
if ( fd!= -1) {
ioctl(fd, PERF_EVENT_IOC_RESET, 0);
ioctl(fd, PERF_EVENT_IOC_ENABLE, 0);
}
begin =__rdtsc();
#else
begin = 0;
#endif
}
void Stop(void) {
count=0;
#ifdef __linux__
if ( fd!= -1) {
ioctl(fd, PERF_EVENT_IOC_DISABLE, 0);
::read(fd, &count, sizeof(long long));
}
elapsed = __rdtsc() - begin;
#else
elapsed = 0;
#endif
}
void Report(void) {
#ifdef __linux__
printf("%llu cycles %s = %20llu\n", elapsed , PerformanceCounterConfigs[PCT].name, count);
#else
printf("%llu cycles \n", elapsed );
#endif
}
~PerformanceCounter()
{
#ifdef __linux__
close(fd);
#endif
}
};

2
lib/Threads.h
View File

@ -44,7 +44,7 @@ class GridThread {
};
static void SetMaxThreads(void) {
#ifdef GRID_OMP
setenv("KMP_AFFINITY","balanced",1);
// setenv("KMP_AFFINITY","balanced",1);
_threads = omp_get_max_threads();
omp_set_num_threads(_threads);
#else

3
lib/algorithms/CoarsenedMatrix.h
View File

@ -264,6 +264,9 @@ PARALLEL_FOR_LOOP
for(int i=0;i<nbasis;i++){
phi=Subspace.subspace[i];
std::cout<<GridLogMessage<<"("<<i<<").."<<std::endl;
for(int p=0;p<geom.npoint;p++){
int dir = geom.directions[p];

1
lib/algorithms/approx/Chebyshev.h
View File

@ -166,7 +166,6 @@ namespace Grid {
Field *Tn = &T1;
Field *Tnp = &T2;
std::cout<<GridLogMessage << "Chebyshev ["<<lo<<","<<hi<<"]"<< " order "<<order <<std::endl;
// Tn=T1 = (xscale M + mscale)in
RealD xscale = 2.0/(hi-lo);
RealD mscale = -(hi+lo)/(hi-lo);

3
lib/algorithms/iterative/DenseMatrix.h
View File

@ -25,6 +25,9 @@ template<class T> void SizeSquare(DenseMatrix<T> & mat, int &N)
assert(N==M);
}
template<class T> void Resize(DenseVector<T > & mat, int N) {
mat.resize(N);
}
template<class T> void Resize(DenseMatrix<T > & mat, int N, int M) {
mat.resize(N);
for(int i=0;i<N;i++){

75
lib/algorithms/iterative/ImplicitlyRestartedLanczos.h
View File

@ -1,4 +1,3 @@
#if 0
#ifndef GRID_IRL_H
#define GRID_IRL_H
@ -18,8 +17,9 @@ template<class Field>
const RealD small = 1.0e-16;
public:
int lock;
int converged;
int get;
int Niter;
int converged;
int Nk; // Number of converged sought
int Np; // Np -- Number of spare vecs in kryloc space
@ -59,6 +59,7 @@ public:
// Sanity checked this routine (step) against Saad.
/////////////////////////
void RitzMatrix(DenseVector<Field>& evec,int k){
if(1) return;
GridBase *grid = evec[0]._grid;
@ -451,8 +452,9 @@ until convergence
std::cout << " -- Nconv = "<< Nconv << "\n";
}
/////////////////////////////////////////////////
// Adapted from Rudy's lanczos factor routine
/////////////////////////////////////////////////
int Lanczos_Factor(int start, int end, int cont,
DenseVector<Field> & bq,
Field &bf,
@ -546,10 +548,16 @@ until convergence
std::cout << "alpha = " << alpha << " fnorm = " << fnorm << '\n';
///Iterative refinement of orthogonality V = [ bq[0] bq[1] ... bq[M] ]
#if 0
int re = 0;
// FIXME undefined params; how set in Rudy's code
int ref =0;
Real rho = 1.0e-8;
while( re == ref || (sqbt < rho * bck && re < 5) ){
Field tmp2(grid);
Field tmp1(grid);
//bex = V^dag bf
DenseVector<ComplexD> bex(j+1);
for(int k=0;k<j+1;k++){
@ -566,14 +574,14 @@ until convergence
//bf = bf - V V^dag bf. Subtracting off any component in span { V[j] }
RealD btc = axpy_norm(bf,-1.0,tmp2,bf);
alpha = alpha + bex[j]; sqbt = sqrt(real(btc));
alpha = alpha + real(bex[j]); sqbt = sqrt(real(btc));
// FIXME is alpha real in RUDY's code?
RealD nmbex = 0;for(int k=0;k<j+1;k++){nmbex = nmbex + real( conjugate(bex[k])*bex[k] );}
bck = sqrt( nmbex );
re++;
}
std::cout << "Iteratively refined orthogonality, changes alpha\n";
if(re > 1) std::cout << "orthagonality refined " << re << " times" <<std::endl;
#endif
H[j][j]=alpha;
}
@ -641,7 +649,7 @@ until convergence
int M=Nm;
DenseMatrix<RealD> H; Resize(H,Nm,Nm);
Resize(evals,Nm,Nm);
Resize(evals,Nm);
Resize(evecs,Nm);
int ff = Lanczos_Factor(0, M, cont, bq,bf,H); // 0--M to begin with
@ -702,7 +710,6 @@ until convergence
RealD beta;
Householder_vector<RealD>(ck, 0, 2, v, beta);
Householder_mult<RealD>(H,v,beta,0,lock_num+0,lock_num+2,0);
Householder_mult<RealD>(H,v,beta,0,lock_num+0,lock_num+2,1);
///Accumulate eigenvector
@ -758,11 +765,11 @@ until convergence
RealD resid_nrm= norm2(bf);
if(!lock) converged = 0;
#if 0
for(int i = SS - lock_num - 1; i >= SS - Nk && i >= 0; --i){
RealD diff = 0;
diff = abs(tevecs[i][Nm - 1 - lock_num]) * resid_nrm;
diff = abs( tevecs[i][Nm - 1 - lock_num] ) * resid_nrm;
std::cout << "residual estimate " << SS-1-i << " " << diff << " of (" << tevals[i] << ")" << std::endl;
@ -785,53 +792,29 @@ until convergence
break;
}
}
#endif
std::cout << "Got " << converged << " so far " <<std::endl;
}
#if 0
///Check
void Check(void) {
DenseVector<RealD> goodval(get);
///Check
void Check(DenseVector<RealD> &evals,
DenseVector<DenseVector<RealD> > &evecs) {
DenseVector<RealD> goodval(this->get);
EigenSort(evals,evecs);
int NM = Nm;
int Nget = this->get;
S **V;
V = new S* [NM];
RealD *QZ;
QZ = new RealD [NM*NM];
DenseVector< DenseVector<RealD> > V; Size(V,NM);
DenseVector<RealD> QZ(NM*NM);
for(int i = 0; i < NM; i++){
for(int j = 0; j < NM; j++){
QZ[i*NM+j] = this->evecs[i][j];
int f_size_cb = 24*dop.cbLs*dop.node_cbvol;
for(int cb = this->prec; cb < 2; cb++){
for(int i = 0; i < NM; i++){
V[i] = (S*)(this->bq[i][cb]);
const int m0 = 4 * 4; // this is new code
assert(m0 % 16 == 0); // see the reason in VtimesQ.C
const int row_per_thread = f_size_cb / (bfmarg::threads);
{
{
DenseVector<RealD> vrow_tmp0(m0*NM);
DenseVector<RealD> vrow_tmp1(m0*NM);
RealD *row_tmp0 = vrow_tmp0.data();
RealD *row_tmp1 = vrow_tmp1.data();
VtimesQ(QZ, NM, V, row_tmp0, row_tmp1, id * row_per_thread, m0, (id + 1) * row_per_thread);
}
}
}
}
// evecs[i][j];
}
}
}
#endif
/**
@ -1020,4 +1003,4 @@ static void Lock(DenseMatrix<T> &H, ///Hess mtx
}
#endif
#endif

11
lib/lattice/Lattice_unary.h
View File

@ -24,6 +24,17 @@ PARALLEL_FOR_LOOP
return ret;
}
template<class obj> Lattice<obj> div(const Lattice<obj> &rhs,Integer y){
Lattice<obj> ret(rhs._grid);
ret.checkerboard = rhs.checkerboard;
conformable(ret,rhs);
PARALLEL_FOR_LOOP
for(int ss=0;ss<rhs._grid->oSites();ss++){
ret._odata[ss]=div(rhs._odata[ss],y);
}
return ret;
}
template<class obj> Lattice<obj> expMat(const Lattice<obj> &rhs, ComplexD alpha, Integer Nexp = DEFAULT_MAT_EXP){
Lattice<obj> ret(rhs._grid);
ret.checkerboard = rhs.checkerboard;

51
lib/qcd/action/fermion/WilsonFermion5D.cc
View File

@ -266,11 +266,8 @@ void WilsonFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOrder &lo,
if( this->HandOptDslash ) {
#pragma omp parallel for schedule(static)
for(int ss=0;ss<U._grid->oSites();ss++){
int sU=ss;
for(int s=0;s<Ls;s++){
int sU=ss;
if ( LebesgueOrder::UseLebesgueOrder ) {
sU=lo.Reorder(ss);
}
int sF = s+Ls*sU;
Kernels::DiracOptHandDhopSiteDag(st,U,comm_buf,sF,sU,in,out);
}
@ -323,52 +320,42 @@ PARALLEL_FOR_LOOP
// Counter.Report();
// }
} else if( this->HandOptDslash ) {
/*
#pragma omp parallel for
#pragma omp parallel for schedule(static)
for(int t=0;t<threads;t++){
int hyperthread = t%HT;
int core = t/HT;
int sswork, swork,soff, sU,sF;
sswork = (nwork + cores-1)/cores;
int sswork, swork,soff,ssoff, sU,sF;
GridThread::GetWork(nwork,core,sswork,ssoff,cores);
GridThread::GetWork(Ls , hyperthread, swork, soff,HT);
for(int ss=0;ss<sswork;ss++){
sU=ss+core*sswork; // max locality within an L2 slice
if ( LebesgueOrder::UseLebesgueOrder ) {
sU = lo.Reorder(sU);
sU=ss+ ssoff;
for(int s=soff;s<soff+swork;s++){
sF = s+Ls*sU;
Kernels::DiracOptHandDhopSite(st,U,comm_buf,sF,sU,in,out);
}
if ( sU < nwork ) {
for(int s=soff;s<soff+swork;s++){
sF = s+Ls*sU;
Kernels::DiracOptHandDhopSite(st,U,comm_buf,sF,sU,in,out);
}
}
}
}
/*
#pragma omp parallel for schedule(static)
for(int ss=0;ss<U._grid->oSites();ss++){
for(int s=0;s<Ls;s++){
int sU=ss;
if ( LebesgueOrder::UseLebesgueOrder ) {
sU=lo.Reorder(ss);
}
int sF = s+Ls*sU;
Kernels::DiracOptHandDhopSite(st,U,comm_buf,sF,sU,in,out);
}
}
*/
#pragma omp parallel for schedule(static)
for(int ss=0;ss<U._grid->oSites();ss++){
int sU=ss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
Kernels::DiracOptHandDhopSite(st,U,comm_buf,sF,sU,in,out);
}
}
} else {
PARALLEL_FOR_LOOP
for(int ss=0;ss<U._grid->oSites();ss++){
int sU=ss;
for(int s=0;s<Ls;s++){
// int sU=lo.Reorder(ss);
int sU=ss;
int sF = s+Ls*sU;
Kernels::DiracOptDhopSite(st,U,comm_buf,sF,sU,in,out);
}

2
lib/qcd/action/fermion/WilsonKernels.h
View File

@ -29,7 +29,7 @@ namespace Grid {
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out,int dirdisp,int gamma);
#if defined(AVX512) || defined(IMCI)
void DiracOptAsmDhopSite(CartesianStencil &st,DoubledGaugeField &U,
void DiracOptAsmDhopSite(StencilImpl &st,DoubledGaugeField &U,
std::vector<SiteHalfSpinor,alignedAllocator<SiteHalfSpinor> > &buf,
int sF,int sU,const FermionField &in, FermionField &out,uint64_t *);
#else

16
lib/serialisation/BinaryIO.cc
View File

@ -1,15 +1,14 @@
#include <Grid.h>
using namespace Grid;
using namespace std;
namespace Grid {
// Writer implementation ///////////////////////////////////////////////////////
BinaryWriter::BinaryWriter(const string &fileName)
: file_(fileName, ios::binary|ios::out)
BinaryWriter::BinaryWriter(const std::string &fileName)
: file_(fileName, std::ios::binary|std::ios::out)
{}
template <>
void BinaryWriter::writeDefault(const string &s, const string &output)
void BinaryWriter::writeDefault(const std::string &s, const std::string &output)
{
uint64_t sz = output.size();
@ -21,12 +20,12 @@ void BinaryWriter::writeDefault(const string &s, const string &output)
}
// Reader implementation ///////////////////////////////////////////////////////
BinaryReader::BinaryReader(const string &fileName)
: file_(fileName, ios::binary|ios::in)
BinaryReader::BinaryReader(const std::string &fileName)
: file_(fileName, std::ios::binary|std::ios::in)
{}
template <>
void BinaryReader::readDefault(const string &s, string &output)
void BinaryReader::readDefault(const std::string &s, std::string &output)
{
uint64_t sz;
@ -34,3 +33,4 @@ void BinaryReader::readDefault(const string &s, string &output)
output.reserve(sz);
file_.read((char *)output.data(), sz);
}
}

25
lib/serialisation/TextIO.cc
View File

@ -1,14 +1,12 @@
#include <Grid.h>
using namespace Grid;
using namespace std;
namespace Grid {
// Writer implementation ///////////////////////////////////////////////////////
TextWriter::TextWriter(const string &fileName)
: file_(fileName, ios::out)
TextWriter::TextWriter(const std::string &fileName)
: file_(fileName, std::ios::out)
{}
void TextWriter::push(const string &s)
void TextWriter::push(const std::string &s)
{
level_++;
};
@ -27,11 +25,11 @@ void TextWriter::indent(void)
};
// Reader implementation ///////////////////////////////////////////////////////
TextReader::TextReader(const string &fileName)
: file_(fileName, ios::in)
TextReader::TextReader(const std::string &fileName)
: file_(fileName, std::ios::in)
{}
void TextReader::push(const string &s)
void TextReader::push(const std::string &s)
{
level_++;
};
@ -50,17 +48,18 @@ void TextReader::checkIndent(void)
file_.get(c);
if (c != '\t')
{
cerr << "mismatch on tab " << c << " level " << level_;
cerr << " i "<< i <<endl;
abort();
std::cerr << "mismatch on tab " << c << " level " << level_;
std::cerr << " i "<< i <<std::endl;
std::abort();
}
}
}
template <>
void TextReader::readDefault(const string &s, string &output)
void TextReader::readDefault(const std::string &s, std::string &output)
{
checkIndent();
output.clear();
getline(file_, output);
}
}

21
lib/serialisation/XmlIO.cc
View File

@ -1,10 +1,8 @@
#include <Grid.h>
using namespace Grid;
using namespace std;
namespace Grid {
// Writer implementation ///////////////////////////////////////////////////////
XmlWriter::XmlWriter(const string &fileName)
XmlWriter::XmlWriter(const std::string &fileName)
: fileName_(fileName)
{
node_ = doc_.append_child();
@ -16,7 +14,7 @@ XmlWriter::~XmlWriter(void)
doc_.save_file(fileName_.c_str(), " ");
}
void XmlWriter::push(const string &s)
void XmlWriter::push(const std::string &s)
{
node_ = node_.append_child(s.c_str());
}
@ -27,22 +25,22 @@ void XmlWriter::pop(void)
}
// Reader implementation ///////////////////////////////////////////////////////
XmlReader::XmlReader(const string &fileName)
XmlReader::XmlReader(const std::string &fileName)
: fileName_(fileName)
{
pugi::xml_parse_result result = doc_.load_file(fileName_.c_str());
if ( !result )
{
cerr << "XML error description: " << result.description() << "\n";
cerr << "XML error offset : " << result.offset << "\n";
abort();
std::cerr << "XML error description: " << result.description() << "\n";
std::cerr << "XML error offset : " << result.offset << "\n";
std::abort();
}
node_ = doc_.child("grid");
}
void XmlReader::push(const string &s)
void XmlReader::push(const std::string &s)
{
node_ = node_.child(s.c_str());
}
@ -53,7 +51,8 @@ void XmlReader::pop(void)
}
template <>
void XmlReader::readDefault(const string &s, string &output)
void XmlReader::readDefault(const std::string &s, std::string &output)
{
output = node_.child(s.c_str()).first_child().value();
}
}

1
lib/serialisation/XmlIO.h
View File

@ -96,6 +96,7 @@ namespace Grid
node_.child("elem").set_name("elem-done");
i++;
}
// assert( is.tellg()==-1);
pop();
}

12
lib/simd/Grid_vector_unops.h
View File

@ -67,6 +67,14 @@ namespace Grid {
}
};
template<class scalar> struct DivIntFunctor {
Integer y;
DivIntFunctor(Integer _y) : y(_y) {};
scalar operator()(const scalar &a) const {
return Integer(a)/y;
}
};
template<class scalar> struct RealFunctor {
scalar operator()(const scalar &a) const {
return real(a);
@ -131,6 +139,10 @@ namespace Grid {
inline Grid_simd<S,V> mod(const Grid_simd<S,V> &r,Integer y) {
return SimdApply(ModIntFunctor<S>(y),r);
}
template < class S, class V >
inline Grid_simd<S,V> div(const Grid_simd<S,V> &r,Integer y) {
return SimdApply(DivIntFunctor<S>(y),r);
}
////////////////////////////////////////////////////////////////////////////
// Allows us to assign into **conformable** real vectors from complex
////////////////////////////////////////////////////////////////////////////

2
lib/tensors/Tensor_unary.h
View File

@ -111,7 +111,7 @@ template<class obj,int N> inline auto toComplex(const iMatrix<obj,N> &z) -> type
return ret;
}
BINARY_RSCALAR(div,Integer);
BINARY_RSCALAR(mod,Integer);
BINARY_RSCALAR(pow,RealD);