Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 18 Dec 2011 23:22:46 +0400
From: Solar Designer <>
Subject: Re: 1.7.9-jumbo

On Sun, Dec 18, 2011 at 07:41:07PM +0100, magnum wrote:
> Memory comes back to me now... for the non-utf8 versions of set_key(), 
> the 1.7.8 code was buggy. It truncated at 28 instead of 27 (but this was 
> only a problem when running external mode). I have a patch somewhere 
> that I can post soon, that re-introduces the tests as if external.c was 
> not truncating, but that does it in a more effective way - and properly 
> at max length. And if 1.7.8 would still be a tad faster after that, I 
> will just say that's because it was buggy.

Are you saying that 1.7.8-jumbo's benchmark results for NT were
inflated because of a bug (I don't see how this can be the case...),
and that this has been corrected?

> For set_key_utf8(), the code present in jumbo-5 has got to be faster 
> than 1.7.8, and now properly truncates at 27. In UTF-8 mode we must 
> always take care of truncation since the "reported" max length is 3x27. 
> Try running "-test -enc:utf8" comparing 178J8 and 179J5, that's a quick 
> test only including Unicode formats.

$ ./john -te -fo=nt -enc=utf8
Benchmarking: NT MD4 [128/128 X2 SSE2-16] in UTF-8 mode... DONE
Raw:    21421K c/s real, 21421K c/s virtual

$ ./john -te -fo=nt -enc=utf8
Benchmarking: NT MD4 [128/128 X2 SSE2-16] in UTF-8 mode... DONE
Raw:    20427K c/s real, 20427K c/s virtual

(I repeated both runs twice, as usual - got similar results.)

So 1.7.8-jumbo-8 reports slightly faster speed even for utf8.


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.