https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125456
--- Comment #2 from Hu Lin <lin1.hu at intel dot com> --- (In reply to Drea Pinski from comment #1) > I wonder if we move the unroller and make it better to the gimple level? > Before ivopts? Do you mean moving the whole pass_rtl_unroll_loops over to before pass_iv_optimize? That feels like a big change to take on at once. I'm not familiar with the GIMPLE side. Would doing this optimization at the GIMPLE level actually be easier and better than at RTL? If so, I could try adding a new GIMPLE pass (doing the same thing as pass_rtl_unroll_loops) before ivopts.
