[PATCH 1/1] Improved ripemd160 perfromance

Jussi Kivilinna jussi.kivilinna at iki.fi
Fri Aug 22 09:56:08 CEST 2014

On 2014-08-22 01:34, Meta Schima wrote:

> Hello, I am new to the mailing list. I have a patch here to improve ripemd160 performance by interleaving left and right lanes. On my system before:


Nice improvement...

When submitting benchmarking results, I personally am interested to know
what kind of CPU these tests were run (Intel, AMD, model). There can be
quite big differences in results even between x86-64 CPUs.

> Hash:
>                 |  nanosecs/byte   mebibytes/sec   cycles/byte
>  RIPEMD160      |      3.31 ns/B     288.1 MiB/s         - c/B
> after:
> Hash:
>                 |  nanosecs/byte   mebibytes/sec   cycles/byte
>  RIPEMD160      |      2.08 ns/B     458.3 MiB/s         - c/B
> Note that I removed one temporary u32 variable, and this reduces the stack burn from 108 to 104, hope this is right.
> ~ Andrei

You need to check doc/HACKING about how to contribute to libgcrypt. You
need to check the part about DCO & Signed-off-by and how to start commit


More information about the Gcrypt-devel mailing list