This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Speed up math/test-tgmath2.c
- From: Steve Ellcey <sellcey at caviumnetworks dot com>
- To: Joseph Myers <joseph at codesourcery dot com>
- Cc: <libc-alpha at sourceware dot org>
- Date: Fri, 4 Nov 2016 11:06:30 -0700
- Subject: Re: [PATCH] Speed up math/test-tgmath2.c
- Authentication-results: sourceware.org; auth=none
- Authentication-results: spf=none (sender IP is ) smtp.mailfrom=Steve dot Ellcey at cavium dot com;
- References: <1477354555.8523.78.camel@caviumnetworks.com> <1478280156.4630.18.camel@caviumnetworks.com> <alpine.DEB.2.20.1611041724170.25235@digraph.polyomino.org.uk>
- Spamdiagnosticmetadata: NSPM
- Spamdiagnosticoutput: 1:99
On Fri, 2016-11-04 at 17:25 +0000, Joseph Myers wrote:
> On Fri, 4 Nov 2016, Steve Ellcey wrote:
>
> >
> > >
> > > + return test_cos (vint1, vllong1)
> > > + + test_fabs (vint1, vllong1)
> > > + + test_conj (vint1, vllong1)
> > > + + test_expm1 (vint1, vllong1)
> > > + + test_lrint (vint1, vllong1)
> > > + + test_ldexp (vint1, vllong1)
> > > + + test_atan2 (vint1, vllong1)
> > > + + test_remquo (vint1, vllong1)
> > > + + test_pow (vint1, vllong1)
> > > + + test_fma_1 (vint1, vllong1)
> > > + + test_fma_2 (vint1, vllong1)
> > > + + test_fma_3 (vint1, vllong1)
> > > + + test_fma_4 (vint1, vllong1);
> You don't want to add up test results like that; you should OR them
> together instead, to avoid the potential for test results adding up
> to 77 and wrongly becoming UNSUPPORTED. (In this case of course
> there aren't enough tests to get to 77, but the principle applies
> that you should avoid the test return being a count of something.)
I'll fix that and resubmit. Is there documentation on what do_test
should return? I looked at test-skeleton.c but there was nothing in
there about what do_test/TEST_FUNCTION should return for different
results (PASS/FAIL/UNSUPPORTED). I didn't see anything in the Wiki
about return values from do_test either.
Steve Ellcey
sellcey@caviumnetworks.com