Does PyPy generate AVX2?  I’ve seen some codes running well with AVX2 on KNL.

Jeff

From: <intel-nda-bounces@lists.jlse.anl.gov> on behalf of "William \"Wild Bill\" Scullin" <wscullin@alcf.anl.gov>
Reply-To: "William \"Wild Bill\" Scullin" <wscullin@alcf.anl.gov>
Date: Thursday, September 22, 2016 at 8:09 AM
To: John Taylor Childers III <jchilders@anl.gov>
Cc: "intel-nda@lists.jlse.anl.gov" <intel-nda@lists.jlse.anl.gov>
Subject: Re: [JLSE-Intel-NDA] pypy?

I do, but it's not quite the miracle it is on Xeon as it doesn't generate AVX512 instructions. On the other hand, it is still better than CPython. I did opt to use the standard NumPy with it as opposed to PyPy NumPy. I'm not sure how much of a difference that would make on KNL as the standard NumPy built with MKL performed better for my needs. I do need to repeat my tests with a larger application and a more recent PyPy, when I do, I'll try and find a way to leave everything in a public space.

- William


On Sep 22, 2016 8:40 AM, "Childers, Taylor" <jchilders@anl.gov> wrote:
Hi Anyone have Pypy installed on KNL?
Taylor


J. Taylor Childers, Assistant Physicist
Argonne National Laboratory
HEP Division Bldg. 360, Rm. C-217
Office: (630) 252-4212 Mobile: (331) 302-4647
jchilders@anl.gov


_______________________________________________
Intel-nda mailing list
Intel-nda@lists.jlse.anl.gov
https://lists.jlse.anl.gov/mailman/listinfo/intel-nda