Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 15 Oct 2002 12:39:26 +1000
From:      Tim Robbins <tjr@FreeBSD.ORG>
To:        Kris Kennaway <kris@obsecurity.org>
Cc:        Andrew Gallatin <gallatin@cs.duke.edu>, Nate Lawson <nate@root.org>, current@FreeBSD.ORG
Subject:   Re: X problems & 5.0... -RELEASE?
Message-ID:  <20021015123926.A76421@dilbert.robbins.dropbear.id.au>
In-Reply-To: <20021014200045.GA51207@xor.obsecurity.org>; from kris@obsecurity.org on Mon, Oct 14, 2002 at 01:00:46PM -0700
References:  <15786.50928.826403.959175@grasshopper.cs.duke.edu> <Pine.BSF.4.21.0210141053301.33371-100000@root.org> <15787.1474.824422.286474@grasshopper.cs.duke.edu> <20021014200045.GA51207@xor.obsecurity.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, Oct 14, 2002 at 01:00:46PM -0700, Kris Kennaway wrote:

[...]
> Did anyone test -current with the various FP test suites people posted
> about last week?

Yes. I ran paranoia from http://cm.bell-labs.com/netlib/paranoia/ and
found that FP arithmethic is satisfactory when -O is not used, and no
-march or -mcpu options are used.

However, compiling with -O causes a lot of failures.

Here are the messages:

Seeking Underflow thresholds UfThold and E0.
DEFECT:  Difference underflows at a higher threshold than products.
...
Can `Z = -Y' overflow?
Trying it on Y = -inf .
finds a FLAW:  -(-Y) differs from Y.
...
FAILURE:  Comparisons involving +--inf, +-inf
and +-4.94066e-324 are confused by Overflow.
...
DEFECT:  Badly unbalanced range; UfThold * V = -inf
        is too far from 1.

SERIOUS DEFECT:    X / X differs from 1 when X = -inf
  instead, X / X - 1/2 - 1/2 = nan .


The summary message:

The number of  FAILUREs  encountered =       1.
The number of  SERIOUS DEFECTs  discovered = 1.
The number of  DEFECTs  discovered =         2.
The number of  FLAWs  discovered =           1.

The arithmetic diagnosed has unacceptable Serious Defects.
Potentially fatal FAILURE may have spoiled this program's subsequent diagnoses.


It's worth noting that 4.7-RELEASE (w/ gcc 2.95.4) fails *more* test cases
than -CURRENT when -O is used to compile paranoia.

f77 -O seems to also generate bad code for dpara.f, the FORTRAN version of
paranoia.


Tim

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20021015123926.A76421>