This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: MIPS: test-float fma math tests trigger underflow exception
- From: Andrew Pinski <pinskia at gmail dot com>
- To: Joseph Myers <joseph at codesourcery dot com>
- Cc: Aurelien Jarno <aurelien at aurel32 dot net>, GNU C Library <libc-alpha at sourceware dot org>
- Date: Wed, 29 Jun 2016 14:38:55 -0700
- Subject: Re: MIPS: test-float fma math tests trigger underflow exception
- Authentication-results: sourceware.org; auth=none
- References: <20160629212012 dot GB3688 at aurel32 dot net> <alpine dot DEB dot 2 dot 20 dot 1606292132290 dot 18822 at digraph dot polyomino dot org dot uk>
On Wed, Jun 29, 2016 at 2:36 PM, Joseph Myers <joseph@codesourcery.com> wrote:
> On Wed, 29 Jun 2016, Aurelien Jarno wrote:
>
>> When building the GNU libc on MIPS (at least o32 and n64), one get the
>> following testsuite errors for the fma float tests:
>
> These are all processor bugs. MIPS processors are specified to detect
> whether floating-point results are tiny, and so determine whether to raise
> the underflow exception, on an after-rounding basis. Some MIPS processors
> appear to have bugs involving using before-rounding tininess detection in
> some cases. These are tests where the expectations depend on whether the
> architecture is before-rounding or after-rounding (which is required by
> IEEE 754 to be consistent for all operations with binary floating-point
> results).
>
>> Does someone has an idea how this can be fixed?
>
> Report the bug to the processor providers.
Well Octeon is not going to be fixed. There is no roadmap for an MIPS
based Octeon any more.
So the only thing we can do is issue an errata :).
Thanks,
Andrew Pinski
>
> --
> Joseph S. Myers
> joseph@codesourcery.com