mikolaj-pirog wrote: > As I'm not a Windows developer, I defer to other reviewers' expertise on > MSVC's PGO/LTO feature. > > However, to be honest, I'm unsure about the value of porting the strings > given the large feature differences between Clang and MSVC on PGO and LTO. > `clang -flto -c` output is LLVM bitcode files, which are distinguishable on > their own. With LLVM LTO, you can have 1 bitcode file and 99 object files, or > 99 bitcode files and 1 object file. I am not sure identifying that LTO > happens is very useful. > > There are many PGO flavors. If we add this, there could be some inconsistency > everytime someone adds a new flavor of PGO and does not port this piece of > code.
I agree that it's a little awkward to port this behavior to clang, since msvc does pgo/lto differently; the lto is enabled by default; it's impossible to do pgo without lto. The value of this patch, as I have seen when creating it, is the feature parity with msvc, accepting the awkwardness. https://github.com/llvm/llvm-project/pull/114260 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
