RE: kernel 2.4.9-34 ia64 bugfix list?

David,

>>>>>> On Thu, 13 Jun 2002 13:49:35 +0100, Daniel Kidger
><[email protected]> said:
>  Daniel> yes - but note that you get "floating point assist faults"
>  Daniel> when for example your application code has the odd
>  Daniel> underflow.  A side-effect is that messages get written to
>  Daniel> the console and even /var/log/messages (caused just by
>  Daniel> normal round-off in a scientific application!).

>  Daniel> The result is the code goes slowly - in one case of ours the
>  Daniel> code run 13 times slower than it should.

>The code doesn't run slowly _because_ of the console messages.  The
>console messages warn you that the code will run slowly because the
>faults take some time to emulate.  (Note that the console logging is
>rate-limited; you'll never see more than one message per second on
>average.)

True - it is not the logging message that cause the slowdown, but it is an
effect of the same cause.

The problem I found with the floating point assist being the default on IA64
was two-fold:
  1. As an application programmer,  by default codes often ran very slowly.

  2. As a system administrator, my console and /var/log/messages were
grossly pollutaed by thousands of line of these warnings, just caused by a
roundoff in one of our users applications. My view is that since underflows
are 'normal' in scientific codes, they shouldn't get 'escalated' to the
system logs.




Yours,
Daniel.

--------------------------------------------------------------
Dr. Dan Kidger, Quadrics Ltd.      [email protected]
One Bridewell St., Bristol, BS1 2AA, UK         0117 915 5505
----------------------- www.quadrics.com --------------------