[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20150824223740.GA26955@openwall.com>
Date: Tue, 25 Aug 2015 01:37:40 +0300
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: PHC: Argon2 on GPU
On Tue, Aug 25, 2015 at 12:24:49AM +0200, Agnieszka Bielec wrote:
> 2015-08-23 9:21 GMT+02:00 Solar Designer <solar@...nwall.com>:
> > Also, right now opencl_blake2-round-no-msg.h uses rotate() with negative
> > shift counts. We should change this to use the corresponding positive
> > shift counts instead.
>
> why? I just changed -63 to 1 in one case but the rest have values
> bigger than -32
Is it stated anywhere that negative or/and over-large shift counts have
a defined behavior? In C, that's UB. Even if for OpenCL's rotate()
this is defined in a specific way(*), it's better not to get into that
bad habit (it will then bite you in C or the like).
(*) For amd_bitalign(), it is in fact said that the shift count is taken
"& 31" (which I found inconvenient since 32 would have been useful as a
way to fully shift in the high 32 bits when using a runtime computed
shift count). However, I didn't see anything like that for rotate().
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.
