Compiling m4ri: msse/ mavx flags

[Aditya Shrotri]

Hello,

I have some code that heavily uses M4RI (primarily mzd_combine_even_in_place; no matrix multiplication). I have compiled M4RI 'as is' without any modifications to the makefile or configure script. During compilation, I noticed that only the msse, msse2 and msse3 flags are enabled for gcc. I was wondering if it would help to enable mavx or any other gcc flags, since I am compiling for relatively new intel processors (6th gen i7) which I believe have these features. If so, how should I go about it? Is there a reason why these flags don't come enabled out of the box? Will using march=native help?

Disclaimer: I have next to no knowledge of SIMD instruction sets, and what the relevant flags really do. Pardon my ignorance if the answer is trivial.

Thanks,
Aditya.

[Martin R. Albrecht]

Hi there,

enabling those extensions might indeed help, either by using
"-march=native" or explicitly "-mavx". If you notice speed
difference, please report back :) We explicitly deal with SSE2
because we have some explicit code making use of it.

I know that some people are working towards contributing AVX2 code
along the same lines.

Cheers,
Martin

--

_pgp: https://keybase.io/martinralbrecht
_www: https://martinralbrecht.wordpress.com
_jab: martinr...@jabber.ccc.de
_otr: 47F43D1A 5D68C36F 468BAEBA 640E8856 D7951CCF