[llvm] [AArch64] Take cmn into account when adjusting compare constants (PR #98634)

[David Sherwood via llvm-commits]

david-arm wrote:

> The PR in question alone doesn't change anything without this. It's a different PR because that PR does additional cmn folds, whereas this just has more accurate profitability checks that reflect the changes to be made.

Perhaps it's worth explaining a bit more in the commit message that while this change doesn't directly reduce the number of instructions, it will enable further optimisations to be made in a subsequent patch.

With some of the tests in this patch it seems at first glance to increase register dependency between instructions, but I can believe those tests are a bit artificial and not representative of the cases that you actually want to optimise.

https://github.com/llvm/llvm-project/pull/98634