rnk added a comment. I would really prefer to avoid adding a new variant of AttrBuilder. What is the main blocker to making AttrBuilder more efficient? It just needs an `LLVMContext`, right? Would that be feasible instead? Most AttrBuilders are constructed from existing AttributeLists, which have a context.
> Internally it maintains two SmallVector of sorted Attributes, which turns out > to > be more efficient than temporary hashmap and bitfields as used by AttrBuilder. Surely the bitset is more efficient than a vector of attributes. I think it's the `std::map<SmallString, SmallString>` that's slow. I think replacing that with a sorted vector of hashed string attributes would be just as efficient. CHANGES SINCE LAST ACTION https://reviews.llvm.org/D115798/new/ https://reviews.llvm.org/D115798 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
