mysterymath wrote:

> In order to fix LTO+llvm-libc, is it only the ELF backend of LLD that needs 
> special handling? Or only ELF is considered as a prototype for now? I 
> experimented with LTO+overlay llvm-libc+-fbuiltin on Windows/COFF and 
> everything seems to work fine, even without these changes.

I'll admit; I haven't spent much time looking at the other file formats. I 
suspect that all of them would have this issue, but I'd need to spend some time 
on them, and I didn't want that to block getting this out for review. This 
change should still partially fix the semantics, since effect (2) in the 
description doesn't need any linker support.

It's worth noting that LTO of libc *often* works without encountering this 
issue. Most libcall->libcall transforms occur in compilcation before LTO; these 
are fine. The trouble only occurs when libcall->libcall transforms occur during 
LTO; and even then only a subset of these transforms.

https://github.com/llvm/llvm-project/pull/164916
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to