Subject: Re: accuracy of "long double"
To: Matthias Drochner <>
From: Krister Walfridsson <>
List: port-amd64
Date: 08/23/2007 00:05:07
On Mon, 20 Aug 2007, Matthias Drochner wrote:
> said:
>> I was planning to submit the TARGET_96_ROUND_53_LONG_DOUBLE to
>> gcc-current some weeks ago, but testing showed that this had some
>> issues
> I've enabled it on a -current box and found that it works
> as expected. It makes some problems more obvious which we
> have anyway, eg that DBL_MAX overflows, but this is not
> gcc's fault.
> Didn't try the gcc testsuite.
> Which problems do you see?

The problems I saw was some random failures in the gcc test suite.  I 
have not analyzed the failures in details, but I do not expect them to be 
serious; it is probably just some special cases where gcc forget to 
honor TARGET_96_ROUND_53_LONG_DOUBLE while optimizing (and a couple of the 
failures are due to incorrect test cases that assume that "everything
work as Linux"...)

I plan to fix this before gcc 4.3 is released...