https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93017
--- Comment #3 from rguenther at suse dot de <rguenther at suse dot de> --- On Wed, 8 Jan 2020, dave.anglin at bell dot net wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93017 > > --- Comment #2 from dave.anglin at bell dot net --- > On 2020-01-08 9:38 a.m., rguenth at gcc dot gnu.org wrote: > > What ISL version are you using? > > dave@mx3210:~$ dpkg -l '*isl*' > Desired=Unknown/Install/Remove/Purge/Hold > | Status=Not/Inst/Conf-files/Unpacked/halF-conf/Half-inst/trig-aWait/Trig-pend > |/ Err?=(none)/Reinst-required (Status,Err: uppercase=bad) > ||/ Name Version Architecture Description > +++-==================-============-============-==============================> > ii libcloog-isl4:hppa 0.18.4-2 hppa Chunky Loop Generator > (runtime> > ii libisl-dev:hppa 0.22-2 hppa manipulating sets and > relation> > ii libisl15:hppa 0.18-4 hppa manipulating sets and > relation> > ii libisl19:hppa 0.20-2 hppa manipulating sets and > relation> > ii libisl22:hppa 0.22-2 hppa manipulating sets and > relation> > > Debian 0.22-2 should be the one being used. The older libraries are linked > into various gcc versions. OK, 0.22 has some issues which is why I didn't try it yet. Most of the testcases really are "non-testcases", we do not do any transform there and the dump scans are simply spurious... The testcase in the summary already says it: /*FIXME: Between isl 0.12 and isl 0.15 the schedule optimizer needs to print something canonical so that it can be checked in the test. The final code generated by both are same in this case but the messaged printed are not consistent. */ and since there's a reduction involved in the testcase I _know_ we're not interchanging the loops. On x86_64 we actually dump "not tiled" (as I said, scanning for "tiled" is useless).
