> On Wed, Jan 23, 2013 at 11:18:31AM +0100, Jan Hubicka wrote: > > Bootstrapped/regtested x86_64-linux, compensated the 3 testcases in tesuite > > testing > > the feature to both test the feature with parameter increased and thest > > that the > > optimization happens anyway at IPA level by default. > > Not objecting to the patch, just wondering, if the fix is only about > changing a default of a param, won't the PR55797 testcase still ICE > if (to the -O -fno-guess-branch-probability -fno-tree-forwprop > --param=early-inlining-insns=176 > options) one also adds --param=max-early-inliner-iterations=2 ? > With --param=max-early-inliner-iterations=1 it indeed no longer ICEs.
It is not meant to be complette fix for PR55797, only the code quality issue Richi observed while debugging it. I am looking into the ICE now, too. Honza
