jiri/LA_library
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

tensor: added some timing

Browse Source
This commit is contained in:
Jiri Pittner
2025-11-19 16:19:45 +01:00
parent e0f5485b94
commit 37973a7161
2 changed files with 29 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
t.cc
View File

@@ -4394,7 +4394,7 @@ cout <<"Error = "<<(z-zzz).norm()<<endl;
if(1)
{
//for profiling
//for profiling and timing
int nn;
cin>>nn;
int r=4;

27
tensor.cc
View File

@@ -25,6 +25,7 @@
#include "bitvector.h"
#include <string.h>
#include <bits/stdc++.h>
#include <time.h>
namespace LA {
@@ -1138,13 +1139,21 @@ for(int i=0; i<nn; ++i) for(int j=0; j<mm; ++j)
template<>
void auxmatmult<double>(int nn, int mm, int kk, double *r, const double *a, const double *b, double alpha, double beta, bool conjugate)
{
clock_t t0=clock();
cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasTrans, nn, mm, kk, alpha, a, kk, b, kk, beta, r, mm);
#ifdef TIMING
std::cout<<"cblas_dgemm CPU time " << (clock()-t0)/(double)CLOCKS_PER_SEC<<std::endl;
#endif
}
template<>
void auxmatmult<std::complex<double> >(int nn, int mm, int kk, std::complex<double> *r, const std::complex<double> *a, const std::complex<double> *b, std::complex<double> alpha, std::complex<double> beta, bool conjugate)
{
clock_t t0=clock();
cblas_zgemm(CblasRowMajor, CblasNoTrans, (conjugate?CblasConjTrans:CblasTrans), nn, mm, kk, &alpha, a, kk, b, kk, &beta, r, mm);
#ifdef TIMING
std::cout<<"cblas_zgemm CPU time " << (clock()-t0)/(double)CLOCKS_PER_SEC<<std::endl;
#endif
}
@@ -1306,8 +1315,12 @@ for(int i=0; i<il1.size(); ++i)
for(int j=0; j<i; ++j) if(il1[i]==il1[j]||il2[i]==il2[j]) laerror("repeated index in the list");
}
clock_t t0=clock();
const Tensor<T> u = conjugate1? (rhs1.unwind_indices(il1)).conjugateme() : rhs1.unwind_indices(il1);
const Tensor<T> rhsu = rhs2.unwind_indices(il2);
#ifdef TIMING
std::cout <<"unwinding CPU time = "<<(clock()-t0)/(double)CLOCKS_PER_SEC<<std::endl;
#endif
NRVec<INDEXGROUP> newshape(u.shape.size()+rhsu.shape.size()-2*il1.size());
@@ -1493,10 +1506,13 @@ if(rhs.is_named())
{
NRVec<INDEXNAME> namperm = rhs.names.permuted(pa[0].perm,!inverse);
if(is_named())
{
if(names!=namperm)
{
std::cout <<"LHS names = "<<names<<std::endl;
std::cout <<"permuted RHS names = "<<namperm<<std::endl;
if(names!=namperm) laerror("inconsistent index names in apply_permutation_algebra");
laerror("inconsistent index names in apply_permutation_algebra");
}
}
else
names=namperm;
@@ -2166,10 +2182,19 @@ if(pb.is_identity())
}
//create an intermediate contracted tensor and the permute it
clock_t t0=clock();
Tensor<T> tmp=rhs1.contractions(il1,rhs2,il2,alpha,conjugate1,conjugate2);
#ifdef TIMING
std::cout<<"contractions CPU time = "<<(1.*clock()-t0)/CLOCKS_PER_SEC<<std::endl;
#endif
if(rank()!=tmp.rank()) laerror("rank mismatch in add_permuted_contractions");
if(tmp.names!=tmpnames) laerror("internal error in add_permuted_contractions");
clock_t t1=clock();
apply_permutation_algebra(tmp,pb,true,(T)1,beta);
#ifdef TIMING
std::cout<<"apply_permutation_algebra CPU time = "<<(1.*clock()-t1)/CLOCKS_PER_SEC<<std::endl;
#endif
}