[PATCH] rs6000: Fix 2 for PR92661, Do not define builtins that overload disabled builtins

Segher Boessenkool segher@kernel.crashing.org
Wed Dec 4 20:51:00 GMT 2019


On Wed, Dec 04, 2019 at 01:56:45PM -0600, Peter Bergner wrote:
> On 12/4/19 1:16 PM, Segher Boessenkool wrote:
> > Why do you need/want the check_effective_target_dfp test?  If for example
> > this is to prevent ICEs, that is no good, that is *hiding* problems.
> > 
> > I suspect it is to stop the testsuite from complaining if you configure
> > with --disable-decimal-float.  What is different there then, compared to
> > targets that do actually not support decimal float?
> 
> Well, yes.  I saw those tests being run for my --disable-decimal-float
> runs, which resulted in FAILs for all of those tests.  They had ICE's
> on unpatched trunk and FAILed gracefully using my patch, but they all
> still FAILed, since these are DFP tests and DFP is disabled.
> There's no sense in running these tests when DFP is disabled, either
> manually due to --disable-decimal-float or implicitly because of the
> specific target.

It would be nice to keep *some* dfp test(s) to make sure we don't ICE.
If we disabled all such tests already, like with this patch, we wouldn't
have ICEd or seen this problem.  That can be a separate test of course
(and could be outside gcc.target/powerpc/dfp/).

OTOH, if you add this check, we can lose the

/* { dg-require-effective-target powerpc_p9vector_ok } */
/* { dg-skip-if "" { powerpc*-*-aix* } } */

from all the dtstsfi-* tests, etc.  (Well, no, need to keep the p9 part).

Making such changes to the testsuite needs testing on *all* subtargets :-(

> Why isn't just testing check_effective_target_dfp enough to disable the
> tests on Darwin, --disable-decimal-float, etc.?  That would seem to imply
> that one of those targets we're testing against enables DFP, but somehow
> we don't want to run the tests or they all still FAIL for some reason???

It should be enough.  Currently we just directly skip all tests on OSes
that do not support DFP, but that is not as nice as the effective target.


Segher



More information about the Gcc-patches mailing list