On Sun, Jan 25, 2004 at 07:07:17PM +0100, Christian Biere wrote:
> Hi,
> 
> can anyone confirm that OpenSSL's md5 uses only about half as much CPU/user
> time as NetBSD's own md5?

It actually really depends which architecture you're on, whether OpenSSL
has asm for that architecture, etc. etc.

Unfortunately, SSLeay/OpenSSL used a *very similar* but still divergent
API for the message-digest functions, compared to everyone else in the
world.  So we have them named one way in libc, with one API, and named
another way in libcrypto, with a different API.  Merging them would be
nontrivial, but I did set out a few months ago to do so anyway.

Unfortunately, what I discovered during some initial testing was that
the OpenSSL implementations were *not* generally faster than the ones
in libc.  In at least one case, the OpenSSL asm implementation was
slightly slower than our libc C implementation.  Eventually I threw
up my hands and gave up.

Thor