nikic wrote: I think this is the right canonicalization at the IR level, as select allows better analysis than zext(icmp). However, this seems to be universally worse for the backend: https://llvm.godbolt.org/z/Yh7brfc8b So I think we would need an SDAG undo transform. Interestingly, this already happens for riscv64, so maybe the undo transform already exists but is not enabled for all target.
@goldsteinn What do you think? https://github.com/llvm/llvm-project/pull/66740 _______________________________________________ lldb-commits mailing list lldb-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits
