Date: Wed, 30 Oct 2013 16:05:11 +0400 From: Solar Designer <solar@...nwall.com> To: john-dev@...ts.openwall.com Subject: Re: bcrypt-parallella on 64-core (was: Katja's weekly report #13) Yaniv, Katja - On Wed, Oct 30, 2013 at 07:51:38AM -0400, Yaniv Sapir wrote: > The problem is not with certain registers, but with certain cores. Cores on > the 2nd row should not *load *or *fetch *from external memory. DMA reads > are OK. Oh. I was referring to: http://www.openwall.com/lists/john-dev/2013/09/18/3 "(A very long shot:) In the stuck cores, do you have external loads into registers R53, R0, R2, R38, respectively?" so it sounded like the problem had something to do with certain registers (as well as certain cores). Further in that thread, Katja confirmed that "there exists external load intto r38 in compiler generated code": http://www.openwall.com/lists/john-dev/2013/09/19/6 and asked "Is it possible to force e-gcc not to use r38 and r53?": http://www.openwall.com/lists/john-dev/2013/10/01/8 Maybe this can in fact be done by placing some local variables into those problematic registers (using the syntax I just posted about), and either only using those variables for other than external loads or keeping them unused. Like I said, I am unsure if this syntax prevents gcc from making other use of the same registers, though. Yaniv, it'd be helpful if you post a full set of relevant Epiphany-IV errata (specifically for E64G401?) in one message. For example, I am confused whether the problem with certain registers occurs only on certain cores (that is, when both the core and the register number are problematic) or whether it's independent (that is, a problem occurs when either register or core number is problematic). Thanks, Alexander
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.