CODEWITHSUNDEEP
pythonnumpyeigen
varantir
varantir

Reputation: 6854

Build Numpy with Eigen instead of ATLAS or OpenBlas?

since Eigen looks very promising (Benchmark) I wonder if it was possible to compile numpy with the Eigen library, as it is possible to build numpy with ATLAS or with OpenBlas (or with intel-mkl). I was trying to play a little bit in the config but I guess this is beyond my reach. Benchmark from different linear algebra cores, Higher = better

Upvotes: 2

Views: 1267

Answers (1)

ggael
ggael

Reputation: 29205

From the eigen sources, you can build a BLAS interface library and then link Numpy to it. For the first part:

make build-eigen
cd build-eigen
cmake -DCMAKE_BUILD_TYPE=Release path/to/eigen
make blas lapack

This gives you a BLAS and LAPACK libraries with both static and shared versions. Then simply link Numpy to, e.g., libeigen_blas_static.a instead of ATLAS.

Also, note that this benchmark is quite old, and done on a single SSE processor. The figure might be different with, e.g., AVX.

Upvotes: 2

Related Questions