Re: [Monotone-devel] Re: sha1/botan stuff

[Nathaniel Smith]

On Fri, Aug 11, 2006 at 08:15:41AM +0200, Florian Weimer wrote:
> * Nathaniel Smith:
> 
> >   mozilla                                  :  36.9 MiB/s
> >   openssl ("no-asm debian-i386")           :  56.5 MiB/s
> >   openssl ("no-asm debian-i386-i686/cmov") :  56.5 MiB/s
> >   botan                                    :  64.1 MiB/s
> >   beecrypt (from debian)                   :  87.0 MiB/s
> >   nettle (lsh's library, from debian)      : 147.1 MiB/s
> >   openssl (with asm)                       : 263.2 MiB/s
> 
> Have you tried libgcrypt?  I suppose it's in the Mozilla range,
> though.

Huh.  I tried a few more, but gcrypt is indeed the surprise:

gcrypt                                         :  82.3 MiB/s
dean gaudet's SIMD impl[1]                     : 115.1 MiB/s

[1] http://www.arctic.org/~dean/crypto/sha1.html ; not using the same
harness as the others, but just the timing script included with the
code.  Since the code doesn't even a full sha1 implementation, I
wasn't all that interested in spending time on it :-).

The "mozilla" line above refers to the C code that git extracted from
mozilla; mozilla itself appears to have two sha1 implementations, one
called "sha1" and one called "sha1_fast".  I tried installing libnss
(the mozilla crypto library) and benchmarking it, but I totally failed
to get it to actually work, so no new benchmark for that.


At a cursory glance, gcrypt's implementation looks just the same as
all the others; I don't know what makes it so fast.

-- Nathaniel

-- 
"Lull'd in the countless chambers of the brain,
Our thoughts are link'd by many a hidden chain:
Awake but one, and lo! what myriads rise!
Each stamps its image as the other flies"
  -- Ann Ward Radcliffe, The Mysteries of Udolpho