Re: CRC was: Re: beta testing version

On Fri, Dec 08, 2000 at 10:36:39AM -0800, Ian Lance Taylor wrote:
>    Incidentally, I benchmarked the previously mentioned 64-bit fingerprint,
>    the standard 32-bit CRC, MD5 and SHA, and the fastest algorithm on my
>    Celeron and on a PIII was MD5.  The 64-bit fingerprint was only a hair
>    slower, the CRC was (quite surprisingly) about 40% slower, and the
>    implementation of SHA that I had available was a real dog.  Taking an
>    arbitrary 32 bits of a MD5 would likely be less collision prone than
>    using a 32-bit CRC, and it appears faster as well.
>
> I just want to confirm that you used something like the fast 32-bit
> CRC algorithm, appended.  The one posted earlier was accurate but
> slow.

Yes.  I just rebuilt the framework using this exact code, and it
performed identically to the previous CRC code (which didn't have an
unrolled inner loop).  These were compiled with -O6 with egcs 1.1.2.
--
Bruce Guenter <bruceg@em.ca>                       http://em.ca/~bruceg/