Re: [PATCH 1/5] testsuite: attr-alloc_size-11.c (PR79356)

[Jeff Law <law at redhat dot com>]

On 03/10/2017 09:20 AM, Martin Sebor wrote:
> On 03/10/2017 05:57 AM, Rainer Orth wrote:
> > Hi Segher,
> >
> > > On Fri, Feb 10, 2017 at 11:56:39AM +0100, Rainer Orth wrote:
> > > > Segher Boessenkool <segher@kernel.crashing.org> writes:
> > > >
> > > > > As stated in the PR, this test now passes on aarch64, ia64, powerpc,
> > > > > and s390x.  This patch disables the xfails for those targets.
> > > >
> > > > As I'd mentioned in PR tree-optimization/78775, the test XPASSes on
> > > > SPARC, too.
> > > >
> > > > > Tested on powerpc64-linux {-m32,-m64}.  Is this okay for trunk?
> > > > [...]
> > > > > 2017-02-09  Segher Boessenkool  <segher@kernel.crashing.org>
> > > > >
> > > > > gcc/testsuite/
> > > > >     PR testsuite/79356
> > > > >     * gcc.dg/attr-alloc_size-11.c: Don't xfail on aarch64, ia64,
> > > > > powerpc,
> > > > >     or s390x.
> > > >
> > > > TBH, I'd strongly prefer to have a proper analysis instead of just
> > > > un-xfail-ing the test on an ever growing apparently random list of
> > > > targets.
> > >
> > > Yeah, I agree.  I thought it was just another test that stopped failing,
> > > but it seems to fail in two ways now, making the testcase succeed?
> > > Lovely.  Please consider this patch withdrawn.
> >
> > I just noticed that nothing has happened at all in a month, so anything
> > is better than the tests XPASSing on a number of targets.
> >
> > So the patch is ok for mainline with sparc*-*-* added to the target
> > lists and a reference to PR testsuite/79356 in the comment.
> >
> > I'd still be very grateful if Martin could have a look what's really
> > going on here, though.
>
> Sorry, I haven't had a chance to look more deeply into why these
> assertions pass on some targets and fail on others.  There is at
> least one bug that tracks a VRP problem that manifests only on
> some targets and not others (79054).  I don't know if this is
> a symptom of the same bug or something different.  I'll see if
> I can find some time to isolate it.
It could well be a BRANCH_COST effect.  BRANCH_COST is probably the most 
annoying target property that bleeds into the tree/gimple world.  From 
looking at the gimple in the BZ that could well be the case.

See logical_op_short_circuit for how this is often dealt with in the 
testsuite.

jeff