On 10/12/18 10:38 AM, Peter Bergner wrote: > On 10/11/18 3:51 PM, Vladimir Makarov wrote: >> I think it has a sense because even if LRA has the same problem, it will be >> hard to fix it in reload and LRA. Nobody worked on reload pass for a long >> time and it is not worth to fix it because we are moving from reload. > [snip] >> In any case, the patch is ok for me. > > Ok, I committed the patch as revision 265113 with a slightly longer comment > explaining why we're disabling it for reload. Thanks! > > > >> I suspect that LRA might be immune to these failures because it generates >> new reload pseudos if it is necessary for insn constraints. Plus there is >> some primitive value numbering in LRA which can avoid the problem. > > Maybe. We still have some LRA targets with issues, but we haven't gotten > to the point of identifying what the problem is. It could well be target > constraints, etc. that is the problem. > > I built a newer binutils on our s390x box and I have now recreated the > selftest ICE in stage3 and I still have the largish aarch64 failure I'm > looking into. ACK. In that case I'll look into the one arm issue and see if I can conclusively rule in or out the IRA/LRA work.
jeff
