shiltian wrote: I still don't think we should enable LTO by default for both compilation modes this way.
I also noticed that `--no-lto` is not equivalent to having the compiler emit an object file directly. It looks like the middle-end pass pipeline is different. I still think we need to carefully evaluate the potential performance impact before switching to the LTO pipeline by default. I agree that the PR being reverted here is a hack, but I think the new driver should match the old driver's behavior until we've evaluated the performance impact and confirmed there won't be any downstream regressions. Also, the old driver's behavior was affected by the recent change to enable LTO by default. That needs to be fixed as well. Otherwise, we completely lose the ability to compile code the way we did before. https://github.com/llvm/llvm-project/pull/206961 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
