[PATCH] D74642: [CodeGenPrepare] Speed up placeDbgValues, NFC

[Vedant Kumar via Phabricator via llvm-commits]

vsk added a comment.

In D74642#1881690 <https://reviews.llvm.org/D74642#1881690>, @rnk wrote:

> I think for the case of reordering two existing instructions, the ordering will be invalidated. So, after D51664 <https://reviews.llvm.org/D51664>, this is still O(n^2) in the case where every other instruction is a dbg.value that needs sinking. However, that's a pretty pathological input, and I think we could address that use case in a general way by leaving holes in the numbering, rather than by keeping this patch around.


Agreed (point "b" in my last comment was my hand-waving about this being a fairly hard-to-hit case). I don't expect it to turn up, but if it does I'll look at poking holes in the instruction ordering.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D74642/new/

https://reviews.llvm.org/D74642