https://github.com/mordante commented: Thanks for doing this. Just curious did you do some benchmarks to see how much this performance this saves in our implementation? In general looks good, some minor remarks. https://github.com/llvm/llvm-project/pull/132408