[llvm] [Uniformity] Fixed control-div early stop (PR #139667)

Junjie Gu via llvm-commits llvm-commits at lists.llvm.org
Mon May 19 10:06:25 PDT 2025


jgu222 wrote:

> Continuing along the previous comment, now I am wondering if we even need the ModifiedPO ordering. As far as I can see, for a node N with an IPDom P, if there is a common cycle C containing N and P, then we need to propagate labels through all the nodes in C until a fixed point is reached, else if there is no C, then propagate labels up to P. In either case, a simple BFS is enough and ModifiedPO gives no advantage.

It seems correct.  But It seems to me that we may choose to stay the current course without using IPD ?

https://github.com/llvm/llvm-project/pull/139667


More information about the llvm-commits mailing list