madhur13490 wrote: @antoniofrighetto You're right — that case is PRE-able (both successors reach backedge), and I was bailing over-conservativily and regressing it. I now only split when a successor leaves the loop; when all successors stay in the loop the reload just goes in the blocker. `test_load_on_cold_path_multi_inloop_succ` now shows it PRE'd, and I added `test_load_on_exiting_cold_path_multi_inloop_succ` (two in-loop successors + a loop-exit successor) for the case we genuinely must bail on.
https://github.com/llvm/llvm-project/pull/206694 _______________________________________________ llvm-branch-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
