On Mon, Jul 16, 2012 at 9:35 PM, Jakub Jelinek <ja...@redhat.com> wrote: > On Sun, Jul 08, 2012 at 02:44:15PM +0200, Steven Bosscher wrote: >> IMHO, no RTL pass implementation file should have to include tree.h. >> So ira-conflicts.c shouldn't need tree.h. The reason it needed it, was >> because there is code to avoid putting user variables in callee >> clobbered registers, and ira-conflict.c looked at DECL_ARTIFICIAL >> (REG_EXPR (reg)). But it can just look at REG_USERVAR_P(reg) instead, >> which is simper and cleaner. > > But it regresses PR53948.
And your proposed solution in the PR is to revert the patch? I don't see how that helps, and frankly I don't think it's a constructive attitude. The variable that is optimized out is a user variable. Instead of reverting my patch, we should find out why REG_USERVAR_P isn't set for it. Ciao! Steven
