Subject: Re: Setting FP precision: fpgetprec() and fpsetprec()
To: Jason Thorpe <email@example.com>
From: Krister Walfridsson <firstname.lastname@example.org>
Date: 06/30/2003 12:13:09
On Sun, 29 Jun 2003, Jason Thorpe wrote:
> Unfortunately, applying the change to use intel_96_round_53 on
> i386-netbsd* seems to have caused regressions. This puzzles me
> I guess there are two interesting things here:
> 1. ieee/rbug.c succeeds without the patch at -O3. I don't
> recall this being the case before. I need to go look
> back through older results.
> 2. If it succeeds at -O3, why does it fail at < -O3?
> 3. The patch causes the i387 pi constant to not be recognized
> properly. Why didn't this happen for FreeBSD, too?
The rbug.c has previously randomly succeeded/failed depending on if the
opimization steps saw that operands were constants and combined them, or
if it let the hardware do it...
I think Loren's mail mentioned that there were a couple of failures that
he thought might be "bugs" in the testsuite (in that it needlessly assume
double precision), but he had not analyzed them yet.
I'm travelling right now, but I can look into it next week (if you don't
want to do it before...)