Subject: Re: IEEE math on Alpha
To: None <port-alpha@netbsd.org>
From: Paul Mather <paul@gromit.dlib.vt.edu>
List: port-alpha
Date: 01/01/2001 11:28:23
On Sun, 31 Dec 2000, Thomas T. Thai wrote:

=> again i'm working to get PostgreSQL to pass on all regression tests on the
=> Alpha. however, i'm running into this problem with NetBSD failing on 'inf'
=> and 'nan' with SIGFPE (kernel reports user traps). if compiled with -mieee
=> the attached code will cause the user trap.
=>
=> i've taken the -mieee flag out of the Makefile in pgsql, and
=> recompiled, but the program keeps dumping user trap errors and it's
=> always something related to NAN. any suggestion is appreciated.

IEEE FP has been broken for a long while on alpha.  I ran into this a
long while ago when trying to get R from pkgsrc to run.  That package
uses fpsetmask() to control IEEE FP errors.  When I looked a little in
the alpha sources, fpsetmask() seems (from memory) to be implemented via
a system call on the alpha, but the code to handle it is just a stub
with a comment to the effect of "someone needs to write this..."

There is a related PR on file (#6805), but I don't think it has
received any attention.

Cheers,

Paul.

e-mail: paul@gromit.dlib.vt.edu

"Without music to decorate it, time is just a bunch of boring production
 deadlines or dates by which bills must be paid."
        --- Frank Vincent Zappa