This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: Add another fma test
- From: Andreas Jaeger <aj at suse dot com>
- To: "Joseph S. Myers" <joseph at codesourcery dot com>
- Cc: libc-alpha at sourceware dot org
- Date: Sat, 15 Jun 2013 21:27:58 +0200
- Subject: Re: Add another fma test
- References: <Pine dot LNX dot 4 dot 64 dot 1306142039000 dot 4414 at digraph dot polyomino dot org dot uk>
On 06/14/2013 10:39 PM, Joseph S. Myers wrote:
When I reviewed some MIPS test results after my math-tests.h addition,
I was suspicious that in one respect they looked too good: the
math/test-ldouble results showed no fma failures, despite the existing
implementations of fma requiring FE_TOWARDZERO and FE_INEXACT support
to work for the relevant types (and math-tests.h only being about
tests using particular features directly, not about what's internally
used in the implementation of a function).
This patch adds an fma test for ldbl-128 to libm-test.inc that fails
if the code needing FE_TOWARDZERO gets run in round-to-nearest instead
(such tests already are present for float and double, and I'm not
aware of any ldbl-96 systems for which such lack of exception /
rounding mode support is an issue). Tested on MIPS64 that this does
show the desired failure in test-ldouble.out.
(Bug 13304 is the bug for defective fma functions in the absence of
FE_TOWARDZERO / FE_INEXACT support for the relevant types, and for
ldbl-128ibm.)
ok, thanks,
Andreas
--
Andreas Jaeger aj@{suse.com,opensuse.org} Twitter/Identica: jaegerandi
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn,Jennifer Guild,Felix Imendörffer,HRB16746 (AG Nürnberg)
GPG fingerprint = 93A3 365E CE47 B889 DF7F FED1 389A 563C C272 A126