[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 6 May 2013 01:12:42 +0200
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: SSE - PARA
On 4 May, 2013, at 23:29 , jfoug@....net wrote:
> We may want to audit our SSE-PARA values. I just checked a few on my 64 bit VM, building with icc, and both 32 and 64 bit were non-optimal.
>
> Current icc: SHA1_SSE_PARA set to 2 for both 32 and 64 bit.
>
> Timing using dyna_26
>
> para-2 64 bit, 12.5k 32 bit 10.2k
> para-1 64 bit, 13.5k 32 bit 12.8k
>
> I have not looked much deeper. I am working on some SSE porting, and was simply doing some testing at different sse-para settings. Right now, I have bugs to work through, if PARA is > 1 for my pbkdf2-hmac-sha1 logic, but will get that fixed up 'soon'. I just wanted to report these findings while it was fresh in my memory.
Yes I never checked this thoroughly for the new ICC version (some quick checks indicated no change was needed but I may have neglected SHA-1).
You could set CC=icc in top of Makefile and try "make clean testpara". There's also "testpara32".
Preferably we should do that on a recent Intel and on a recent AMD and compare the results.
magnum
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
