erichkeane wrote: > Cannot we just use `mlir::SymbolTAbleAnalysis` attached to top-level module > op, and pick symboltablecallection from it? one does not need to polute API > with passing tables everywhere I guess, and hopefully we can reuse analysis > accross passes not to compute SymbolTAble for each pass again? > > See: https://mlir.llvm.org/doxygen/classmlir_1_1SymbolTableAnalysis.html for > reference. Also might be interresting to measure if we create one global > analysis whether the numbers get even lower.
I was going to suggest we stop threading this through. It is like it is, as this is how you suggested it on my initial patch, but the more I work wiht this, the more I suspect we should have SOME sort of member. it isn't clear to me the value of `SymbolTableAnalysis` vs just storing a collection though. https://github.com/llvm/llvm-project/pull/195919 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
