You are not logged in.

#1 2018-04-10 15:48:46

kgizdov
Package Maintainer (PM)
From: Edinburgh, UK
Registered: 2015-12-08
Posts: 113

TensorFlow CPU feature support and python2

Hi,

Just looking for some general comments suggestions here.

Firstly, I'm using the community package for TensorFlow. So far so good, but one thing that gets me is when I use it I get this message:

2018-04-10 17:36:12.318201: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA

I'm guessing this is because package is compiled with generic flags. But I wonder if this can be corrected in the shipped package, given that it is capable of checking CPU support at runtime.

Another thing, some of the projects I work on require python2 support. I tried to recompile for python2 using asp to get the PKGBUILD, but things didn't go so well. I tried to look on the AUR, Wiki and forums, but so far no luck. Does anyone have/know of a python2 PKGBUILD? Compilation time is so long that it will take me a huge amount of time to check everything and get it properly working.

Thanks.

Offline

#2 2018-04-10 16:23:00

Potomac
Member
Registered: 2011-12-25
Posts: 556

Re: TensorFlow CPU feature support and python2

Hello,

edit the official PKGBUILD:
https://git.archlinux.org/svntogit/comm … tensorflow

and check the line "export CC_OPT_FLAGS="-march=x86-64", replace this line by this :

export CC_OPT_FLAGS="-march=native"

Last edited by Potomac (2018-04-10 16:23:32)

Offline

Board footer

Powered by FluxBB