hi @Wheest, I’m sorry for the confusion. I mean that the name would be added to the /existing/ attribute dictionary that every operation carries.
I’m not so sure about the multi-threading problem with an implement like the LLVM one: since we’re talking about SSA value and the associated scope of an isolated region it seems to me that it matches perfectly the granularity of MLIR multi-threading.
I’m think you and I are imagining your suggestion as different implementation approaches. Can you explain it in more detail?
The approach out outline above has /zero cost/ when it isn’t used, except in the parser/printer. I don’t see any downside to it at all.
-Chris