SHA1

[John Tapsell]

Hi all,

  There is a long conversation going on on the git mailing list over
SHA1 implementations, and in particular measuring them in 'real world'
situations.  Since xorg also relies on a fast SHA1 implementation for
comparing pixmaps, I think you'll find the results interesting.

  I have summarised the results below for your convenience.

  I was particularly amazed at just how much of a difference there is
between implementations.

These results are from running "git fsck -a" on the linux kernel and
so represent "real" world situations.

These are the results that Linus Torvalds got on x86-64:

- Mozilla SHA1 portable C-code (sucky sucky): MOZILLA_SHA1=1

       real    0m38.194s
       user    0m37.838s
       sys     0m0.356s

 - The SHA1 implementation used by git  ("half-portable C code"): BLK_SHA1=1

       real    0m28.120s
       user    0m27.930s
       sys     0m0.192s

 - OpenSSL assembler code:

       real    0m26.327s
       user    0m26.194s
       sys     0m0.136s


And on 32 bit we see GCC clearly struggling with register starvation:

 - Mozilla SHA:

       real    0m47.063s
       user    0m46.815s
       sys     0m0.252s

 - BLK_SHA1=1

       real    0m34.705s
       user    0m34.394s
       sys     0m0.312s

 - OPENSSL:

       real    0m29.754s
       user    0m29.446s
       sys     0m0.288s




Encouraged by the fact that the git implementation is almost as good
as the ASM, Linus then started to tweak the C code with the most evil
and foul code that I've seen.  The end result was (on 64-bit)


 - OpenSSL:

       real    0m26.363s
       user    0m26.174s
       sys     0m0.188s

 - This C implementation:

       real    0m26.594s
       user    0m26.310s
       sys     0m0.256s


And on a slow 32bit atom (with the evil-and-foul changes they made):


 - BLK_SHA1:

       real    2m27.160s
       user    2m23.651s
       sys     0m2.392s

 - OpenSSL:

       real    2m12.580s
       user    2m9.998s
       sys     0m1.811s



John