Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 28 Jan 2013 20:14:09 +0100
From: magnum <>
Subject: Re: Shared GWS tuning function

On 28 Jan, 2013, at 19:24 , Claudio André <> wrote:
> Em 28-01-2013 15:54, magnum escreveu:
>> On 28 Jan, 2013, at 18:32 , Claudio André <> wrote:
>>> Em 27-01-2013 22:58, magnum escreveu:
>>>> In bleeding, Claudio has added a shared function for tuning GWS. I haven't had time to try it out yet.
>>> It is not hard to use, as you can see attached.
>> Yes, I'll do all my formats in one batch when I get some time.
>> BTW in this example patch, I see you changed the buffer sizes macros back to variables:
>> -#define insize (sizeof(pwsafe_pass) * global_work_size)
>> -#define outsize (sizeof(pwsafe_hash) * global_work_size)
>> -#define saltsize (sizeof(pwsafe_salt))
>> +static int insize;
>> +static int outsize;
>> +static int saltsize;
>> I changed them to macros when I made it honour count argument, just for not having to adjust them whenever GWS changes (including in crypt_all()). Doesn't matter when running, but using the macros you don't have to remember that.
> During crypt_all_benchmark you can't change global_work_size (i mean, the user might selected GWS=0).
> It is ok to revert to macros, but you have to save global_work_size inside init (or control if you have to call find_best_gws using something else, not global_work_size).
> A solution to this (not really tested is attached).
>> Anyway, that patch is ready-to-go for bleeding, right? We might just as well apply it.
>> magnum
> It worked here and on Bull.

I tried it but GWS is still pegged at the default. Oh, I see now, I need to force it with GWS=0. I'll change that.


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.