[flang-commits] [flang] [flang] Option to drop iter_arg for do-concurrent nested loops (PR #207816)

[Scott Manley via flang-commits]

rscottmanley wrote:

> My main question is why not always lower the DO LOOP in such way if the normal representation is cumbersome to work with?

I agree with the question and actually would prefer that the IV iter_arg shadow is removed completely. My understanding from @vzakhari is that this was done for performance reasons. I may not be recalling correctly, but I think it was for vectorization when the IV is used ... e.g. `a(i) = i;`? In which case, I sort of understand that it would be a bigger fix to deal with that in vectorization than it would be here, but I feel the IV shadow is cumbersome in the MLIR stages

https://github.com/llvm/llvm-project/pull/207816