[llvm] [AArch64] Disable Xor merging for certain neoverse cores (PR #165393)

[David Green via llvm-commits]

https://github.com/davemgreen requested changes to this pull request.

Do you have any (realistic) benchmark results for this change? Having less instructions can sometimes be a benefit in itself, and if you are not limited in V0 vector pipeline bandwidth, the eor3 will be quicker. In practice you need to look at the surrounding instructions to check if eor3 will be less profitable that 2 eors.

https://github.com/llvm/llvm-project/pull/165393