On Fri, Jan 22, 2016 at 11:25 AM, Bernd Schmidt <bernds_...@t-online.de> wrote: > On 01/22/2016 08:03 PM, Ian Lance Taylor wrote: >> >> Updated patch. >> >> I've verified that I'm changing the only relevant place in >> tree-ssa-loop-ivopts.c that creates a POINTER_PLUS_EXPR, so I do think >> that this is the only changed to fix the problem for ivopts. > > > I don't think so. One of the problems with ivopts is that it likes to cast > everything to unsigned int, so looking for POINTER_PLUS_EXPR wouldn't find > all affected spots. At least this used to happen, I didn't check recently. > Also, a lot of the generated expressions are built by tree-affine.c rather > than in ivopts directly.
Thanks for the tip. I moved the check to add_candidate_1 instead. This is before the point where it converts to an integer type. This approach is better anyhow, as it permits a pointer loop to use an integer induction variable for the offset. My tests still pass, as does bootstrap/testsuite on x86_64-pc-linux-gnu. Does this look OK for mainline? Ian gcc/ChangeLog: 2016-01-22 Ian Lance Taylor <i...@google.com> * common.opt (fkeep-gc-roots-live): New option. * tree-ssa-loop-ivopts.c (add_candidate_1): If -fkeep-gc-roots-live, skip pointers. (add_iv_candidate_for_biv): Handle add_candidate_1 returning NULL. * doc/invoke.texi (Optimize Options): Document -fkeep-gc-roots-live. gcc/testsuite/ChangeLog: 2016-01-22 Ian Lance Taylor <i...@google.com> * gcc.dg/tree-ssa/ivopt_5.c: New test.
