diff options
| author | Pjotr Prins | 2017-12-19 08:56:17 +0000 |
|---|---|---|
| committer | Pjotr Prins | 2017-12-19 08:56:17 +0000 |
| commit | c760aa09c2aa91ca6270b5f898c27e9aed376a73 (patch) | |
| tree | 8b792cbb40abe9ee1656ec7332f69d432a86399a /INSTALL.md | |
| parent | fe883b9730925c75ac2bd09e7060c042929a942e (diff) | |
| download | pangemma-c760aa09c2aa91ca6270b5f898c27e9aed376a73.tar.gz | |
Optimizing OpenBlas
Diffstat (limited to 'INSTALL.md')
| -rw-r--r-- | INSTALL.md | 16 |
1 files changed, 8 insertions, 8 deletions
@@ -130,14 +130,14 @@ and/or play with the switches (listed in OpenBlas Makefile.rule) and you should see something like - OpenBLAS build complete. (BLAS CBLAS) + OpenBLAS build complete. (BLAS CBLAS LAPACK LAPACKE) - OS ... Linux - Architecture ... x86_64 - BINARY ... 64bit - C compiler ... GCC (command line : gcc) - Fortran compiler ... GFORTRAN (command line : gfortran) - Library Name ... libopenblas_haswellp-r0.3.0.dev.a (Multi threaded; Max num-threads is 64) + OS ... Linux + Architecture ... x86_64 + BINARY ... 64bit + C compiler ... GCC (command line : gcc) + Fortran compiler ... GFORTRAN (command line : gfortran) + Library Name ... libopenblas_haswellp-r0.3.0.dev.a (Multi threaded; Max num-threads is 64) Note that OpenBlas by default uses a 32-bit integer API which can overflow with large matrix sizes. We don't include LAPACK - the @@ -146,7 +146,7 @@ OpenBlas version gives problems around eigenvalues for some reason. We now have a static library which you can link using the full path with using the GEMMA Makefile: - time env OPENBLAS_NUM_THREADS=4 make EIGEN_INCLUDE_PATH=~/.guix-profile/include/eigen3 LIBS="~/tmp/OpenBLAS/libopenblas_haswellp-r0.3.0.dev.a -lgsl -lgslcblas -pthread -lz -llapack" -j 4 unittests + time env OPENBLAS_NUM_THREADS=4 make EIGEN_INCLUDE_PATH=~/.guix-profile/include/eigen3 LIBS="~/tmp/OpenBLAS/libopenblas_haswellp-r0.3.0.dev.a -lgsl -pthread -lz" -j 4 unittests Latest (INT64, no gslcblas): |