GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-12 11:49:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix bench_gemm

Browse Source
This commit is contained in:
Gael Guennebaud 2011-09-09 10:36:20 +02:00
parent d52d8e4a53
commit 3e7aaadb1d
1 changed files with 7 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
bench/bench_gemm.cpp
View File

@ -161,10 +161,11 @@ int main(int argc, char ** argv)
A a(m,p); a.setRandom(); A a(m,p); a.setRandom();
B b(p,n); b.setRandom(); B b(p,n); b.setRandom();
C c(m,n); c.setOnes(); C c(m,n); c.setOnes();
C rc = c;
std::cout << "Matrix sizes = " << m << "x" << p << " * " << p << "x" << n << "\n"; std::cout << "Matrix sizes = " << m << "x" << p << " * " << p << "x" << n << "\n";
std::ptrdiff_t mc(m), nc(n), kc(p); std::ptrdiff_t mc(m), nc(n), kc(p);
computeProductBlockingSizes<Scalar,Scalar>(kc, mc, nc); internal::computeProductBlockingSizes<Scalar,Scalar>(kc, mc, nc);
std::cout << "blocking size (mc x kc) = " << mc << " x " << kc << "\n"; std::cout << "blocking size (mc x kc) = " << mc << " x " << kc << "\n";
C r = c; C r = c;
@ -188,23 +189,22 @@ int main(int argc, char ** argv)
blas_gemm(a,b,r); blas_gemm(a,b,r);
c.noalias() += a * b; c.noalias() += a * b;
if(!r.isApprox(c)) std::cerr << "Warning, your product is crap!\n\n"; if(!r.isApprox(c)) std::cerr << "Warning, your product is crap!\n\n";
// std::cerr << r << "\n\n" << c << "\n\n";
#else #else
gemm(a,b,c); gemm(a,b,c);
r.noalias() += a.cast<Scalar>() * b.cast<Scalar>(); r.noalias() += a.cast<Scalar>() * b.cast<Scalar>();
if(!r.isApprox(c)) std::cerr << "Warning, your product is crap!\n\n"; if(!r.isApprox(c)) std::cerr << "Warning, your product is crap!\n\n";
// std::cerr << c << "\n\n";
// std::cerr << r << "\n\n";
#endif #endif
#ifdef HAVE_BLAS #ifdef HAVE_BLAS
BenchTimer tblas; BenchTimer tblas;
c = rc;
BENCH(tblas, tries, rep, blas_gemm(a,b,c)); BENCH(tblas, tries, rep, blas_gemm(a,b,c));
std::cout << "blas cpu " << tblas.best(CPU_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tblas.best(CPU_TIMER))*1e-9 << " GFLOPS \t(" << tblas.total(CPU_TIMER) << "s)\n"; std::cout << "blas cpu " << tblas.best(CPU_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tblas.best(CPU_TIMER))*1e-9 << " GFLOPS \t(" << tblas.total(CPU_TIMER) << "s)\n";
std::cout << "blas real " << tblas.best(REAL_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tblas.best(REAL_TIMER))*1e-9 << " GFLOPS \t(" << tblas.total(REAL_TIMER) << "s)\n"; std::cout << "blas real " << tblas.best(REAL_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tblas.best(REAL_TIMER))*1e-9 << " GFLOPS \t(" << tblas.total(REAL_TIMER) << "s)\n";
#endif #endif
BenchTimer tmt; BenchTimer tmt;
c = rc;
BENCH(tmt, tries, rep, gemm(a,b,c)); BENCH(tmt, tries, rep, gemm(a,b,c));
std::cout << "eigen cpu " << tmt.best(CPU_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tmt.best(CPU_TIMER))*1e-9 << " GFLOPS \t(" << tmt.total(CPU_TIMER) << "s)\n"; std::cout << "eigen cpu " << tmt.best(CPU_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tmt.best(CPU_TIMER))*1e-9 << " GFLOPS \t(" << tmt.total(CPU_TIMER) << "s)\n";
std::cout << "eigen real " << tmt.best(REAL_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tmt.best(REAL_TIMER))*1e-9 << " GFLOPS \t(" << tmt.total(REAL_TIMER) << "s)\n"; std::cout << "eigen real " << tmt.best(REAL_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tmt.best(REAL_TIMER))*1e-9 << " GFLOPS \t(" << tmt.total(REAL_TIMER) << "s)\n";
@ -213,8 +213,9 @@ int main(int argc, char ** argv)
if(procs>1) if(procs>1)
{ {
BenchTimer tmono; BenchTimer tmono;
//omp_set_num_threads(1); omp_set_num_threads(1);
Eigen::setNbThreads(1); Eigen::internal::setNbThreads(1);
c = rc;
BENCH(tmono, tries, rep, gemm(a,b,c)); BENCH(tmono, tries, rep, gemm(a,b,c));
std::cout << "eigen mono cpu " << tmono.best(CPU_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tmono.best(CPU_TIMER))*1e-9 << " GFLOPS \t(" << tmono.total(CPU_TIMER) << "s)\n"; std::cout << "eigen mono cpu " << tmono.best(CPU_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tmono.best(CPU_TIMER))*1e-9 << " GFLOPS \t(" << tmono.total(CPU_TIMER) << "s)\n";
std::cout << "eigen mono real " << tmono.best(REAL_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tmono.best(REAL_TIMER))*1e-9 << " GFLOPS \t(" << tmono.total(REAL_TIMER) << "s)\n"; std::cout << "eigen mono real " << tmono.best(REAL_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tmono.best(REAL_TIMER))*1e-9 << " GFLOPS \t(" << tmono.total(REAL_TIMER) << "s)\n";