[SciPy-User] Build Questions, Intel Compilers/MKL (16), CentOS 7.2

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view

[SciPy-User] Build Questions, Intel Compilers/MKL (16), CentOS 7.2

Novosielski, Ryan
Hi there,

1) I’m trying to get an answer to something that isn’t explained in any of the documentation I’ve found about building NumPy/SciPy with the Intel compilers. What I’ve been looking at is both here:

…and here:

…which agree with one another. If you look at these, both say to build NumPy, do this:

$ python setup.py config --compiler=intelem build_clib --compiler=intelem build_ext --compiler=intelem install

For SciPy, they say to do this:

$ python setup.py config --compiler=intelem --fcompiler=intelem build_clib --compiler=intelem --fcompiler=intelem build_ext --compiler=intelem --fcompiler=intelem install

…but neither one says why we’re not using the Intel Fortran compiler for NumPy. I assumed at first it might be because NumPy doesn’t have Fortran code or something, but I see during the build process, gfortran is definitely used. Can anyone shed some light on this one? It builds fine if I use the same command to build NumPy as the recommended one for SciPy.

Somewhat related, the Intel link above recommends a set of compiler flags:
self.cc_exe = 'icc -O3 -g -fPIC -fp-model strict -fomit-frame-pointer -openmp -xhost’

…which does work to build NumPy. However, when I build SciPy, I’m seeing a lot of -xSSE4.2 on the Fortran compiles:

compiling Fortran sources
Fortran f77 compiler: /opt/intel/compilers_and_libraries_2016.1.150/linux/bin/intel64/ifort -FI -fPIC -openmp -fp-model strict -xSSE4.2
Fortran f90 compiler: /opt/intel/compilers_and_libraries_2016.1.150/linux/bin/intel64/ifort -FR -fPIC -openmp -fp-model strict -xSSE4.2
Fortran fix compiler: /opt/intel/compilers_and_libraries_2016.1.150/linux/bin/intel64/ifort -FI -fPIC -openmp -fp-model strict -xSSE4.2

2) Ideally I’d want to use -xAVX if it were possible (which is what -xhost is doing since I’m building on an AVX-capable machine). There doesn’t seem to be anyplace to set this in SciPy, and there is some indication from reading about SciPy that it uses NumPy’s build setup somehow. Any idea why this is happening?


____ *Note: UMDNJ is now Rutgers-Biomedical and Health Sciences*
|| \\UTGERS      |---------------------*O*---------------------
||_// Biomedical | Ryan Novosielski - Senior Technologist
|| \\ and Health | [hidden email] - 973/972.0922 (2x0922)
||  \\  Sciences | OIRT/High Perf & Res Comp - MSB C630, Newark

SciPy-User mailing list
[hidden email]

signature.asc (211 bytes) Download Attachment