[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: netbsd build update

From: Ben Pfaff
Subject: Re: netbsd build update
Date: Wed, 17 May 2006 20:08:42 -0700
User-agent: Gnus/5.110004 (No Gnus v0.4) Emacs/21.4 (gnu/linux)

Jason Stover <address@hidden> writes:

> Short answer: The alpha chip does not comply entirely with ieee 754.
> The -mieee option to gcc fixes the problem. Maybe this option should be
> added conditionally in (or wherever such things go).

The problem is a floating point exception (FPE), right?  (Before,
I thought it was a segmentation fault, but I guess I
misunderstood.)  I'd expect that increased IEEE compliance would
only prevent an FPE if it prevented, say, underflow by instead
resulting in a denormal.  If that's the case, then an underflow
from a denormal to zero would still, presumably, cause an FPE and
we'd only be papering over the problem, not curing.

I do think that -mieee is a good idea, but I think we should look
for a secondary problem too.  Here is one idea: PSPP tries to
turn off FPEs entirely, in fpu_init() in main.c:

    static void
    fpu_init (void) 
      fenv_t foo;
      feholdexcept (&foo);
    #elif HAVE___SETFPUCW && defined(_FPU_IEEE)
      __setfpucw (_FPU_IEEE);

Does NetBSD support either of the approaches we try there?  That
defined when that function is compiled?  If not, then we should
find out how we turn off FPEs on NetBSD and add that to the
possibilities.  (We should possibly warn about this at configure
time, too.)

"...I've forgotten where I was going with this,
 but you can bet it was scathing."

reply via email to

[Prev in Thread] Current Thread [Next in Thread]