Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [day] [month] [year] [list]
Date: Tue, 12 May 2015 00:23:20 +0200
From: magnum <>
Subject: sunmd5 assumes a SIMD width of 4

Solar, Lei,

I just saw this in sunmd5:

  * 576 divides evenly by 4, 8, 12, 16 (para = 1, 2, 3, 4) Each of these
  * will get an even number of para buckets, so hopefully we get the best
  * average fill we possibly can get. 960 would also allow us to pick up 
20 (para 5)
  * NOTE, made 'smaller' fit. 240 would get all 4,8,12,16,20, but now it 
is multiple
  * defines. this allows self tests to run much faster, BUT the speed of 
the format
  * is still about the same.
//#define MAX_KEYS_PER_CRYPT		576
//#define MAX_KEYS_PER_CRYPT		240
#if !defined(MD5_SSE_PARA) || MD5_SSE_PARA==1
#elif MD5_SSE_PARA==2
#elif MD5_SSE_PARA==3 || MD5_SSE_PARA==4
#elif MD5_SSE_PARA==5
#define MAX_KEYS_PER_CRYPT 100

Unfortunately Jim has just assumed a SIMD width of 4 (again, doh). The 
above will fail for certain combinations of width and interlaving 
factors. For example, AVX2 at para 5 is 8x4 so 100 is not a very good 
number to use.

I'm not sure why we're not just using eg.


Perhaps with some fixed multipler if we really want to (not sure why we 
would but I haven't looked much at the code).


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.