> On Jul 15, 2016, at 2:24 AM, Richard Biener <richard.guent...@gmail.com> > wrote: > > On Thu, Jul 14, 2016 at 6:10 PM, Martin Liška <mli...@suse.cz> wrote: >> On 07/14/2016 01:21 PM, Richard Biener wrote: >>> On Thu, Jul 14, 2016 at 1:06 PM, Martin Liška <mli...@suse.cz> wrote: >>>> On 07/13/2016 07:21 PM, Jeff Law wrote: >>>>> Isn't that a code quality regression? So instead shouldn't we be keeping >>>>> the same expectation, but xfailing the test? >>>>> >>>>> jeff >>>> >>>> Hello. >>>> >>>> Disabling a pass before slsr makes the test to catch both opportunities. >>>> Is the patch fine? >>> >>> No, this is still a code quality regression. What happens is that for >>> some reason we fail to sink for GCC 6. >> >> So should I just mark the test-case as a xfail? > > Leave it FAIL and open a bug. We need to fix SLSR to handle the situation.
Please CC me on the bug (wschm...@gcc.gnu.org). Thanks, Bill > > You can try going back to the point where the testcase was added and look at > the > IL that it was supposed to test, on the GCC 6 branch we sink into one > arm but not > the other, on trunk we sink into both. Iff the original IL was > without any sinking > then adding a testcase variant with sinking turned off might be good as well. > > I'll also note that if we'd do these kind of tests as unit-tests we'd > never notice > that in real-world the testcase would have started failing due to > previous passes > messing up the IL. > > Richard. > >> M. >> >>> >>> Richard. >>> >>>> Thanks, >>>> Martin >> >
