pcc wrote: > I think the idea behind this PR is that, from an architectural perspective, > it shouldn't be necessary to parse inline assembly to fetch the symbol table > of an object in the first place.
The idea behind the bitcode symbol table was to avoid needing to parse the inline assembly (and read other parts of the IR from bitcode) in most cases when we read a bitcode file. In cases such as version mismatches we do need to read the bitcode to get the symbol table, but that's fairly expensive anyway and the rarely used module level inline asm feature should usually be a small part of the cost. The mapping from inline assembly to a list of symbols is, at least theoretically, a pure function, so I don't see an architectural problem with doing so in the rare cases where we need to do so. https://github.com/llvm/llvm-project/pull/174995 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
