gcc.dg/tree-ssa/loop-4.c scan-tree-dump-times iter 0 has started failing on mainline between 20050321 and 20030323, on hppa64-hpux and ia64-hpux (-mlp64 only) but not on hppa2.0w-hpux or ia64-hpux -lilp32. Results on gcc-testresults also show it failing on ia64-linux.
That should of course be 20050323 (not 20030323).
This regression affects ia64-linux which is a secondary platform.
Binary search on ia64-hpux shows the failure appeared between 2005-03-22 22:04 UTC and 2005-03-22 22:30 UTC. The only plausibly responsible patch in that range is 2005-03-22 Zdenek Dvorak <dvorakz@suse.cz> * tree-ssa-loop-ivopts.c (determine_iv_cost): Do not try to preserve artificial original candidates.
Hmm, maybe we are just selecting the orignal IV as the IV which we should use, this is just a missed optimization.
Or it might be the testcase which is matching it wrong, someone already asked about this testcase before on IRC (IIRC) and the regex was just matching wrongly.
Subject: Re: [4.1 Regression] gcc.dg/tree-ssa/loop-4.c scan-tree-dump-times iter 0 fails On Mon, 16 May 2005, pinskia at gcc dot gnu dot org wrote: > Or it might be the testcase which is matching it wrong, someone already asked about this testcase > before on IRC (IIRC) and the regex was just matching wrongly. No, that's bug 21341 (if indeed the problem was the regex, the submitter of the patch which changed the regex didn't mention the bug and its prior change from PASS to FAIL). There are multiple test assertions within the same test file, and when any one changes from PASS to FAIL that is in principle, and in practice in this case, a separate bug. http://gcc.gnu.org/ml/gcc-patches/2005-05/msg01057.html
Appears fixed on ia64-hpux and hppa64-hpux 20050901.
Fixed.