RE: New Clients Available

andrew meggs (insect@antennahead.com)
Fri, 9 May 1997 05:02:40 -0400


>Justin, can the person(s) that did the Mac client please recompile for 603e?
>

I am rewriting the PowerPC clients in assembly. They are still being
debugged, maybe by Monday. There will be only one version of the code,
which will be tuned for all CPU models (except maybe prototype Arthur
systems since I'm not on the A-list for seeding), because I never
encountered any situations where the optimal code for one CPU was mutually
exclusive with the optimal code for another.

>My 3400/240 performs close to my S900/250 on other clients,

To me, this indicates that those other clients are very poorly tuned for
the 604e, not that they're tuned well for the 603e. An S900/250 should be
smoking a 3400/240 unless there are some serious scheduling or misalignment
problems in the code that keep it from taking advantage of 2 extra integer
pipelines, larger L1 and L2 caches, much more intelligent branch
prediction, and a better system bus.

> but the 601/604
>DESCHALL client running on the 603e based 3400 sux; only doing 2^28
>repetitions, rather than 2^30.

The new code shows a larger improvement on 603's than on any other CPU.

____________________________________________________________________________
Andrew Meggs, content provider Antennahead Industries, Inc.
<mailto:insect@antennahead.com> <http://www.antennahead.com>
_________asserting_my_individuality_through_really_really_bad_taste_________