[llvm] [ARM][MVE] Invalid tail predication in LowOverheadLoop pass (PR #163217)

[David Green via llvm-commits]

https://github.com/davemgreen commented:

Hello. I am not around at the moment but the intent of a MQPRCopy is to allow tail predication in a loop with register spills / copies inside it. The MQPRCopy should be lowered to either a MVE_ORR if it can be predicated, or a pair of FMOVD's if it cannot, so that it still performs the full vector write. The 2 instructions it needs to generate is worse, but the tail predications otherwise is quite a lot better IIRC.

It sounds like here we might need to look through the tied operand of the VADD to check that the source is copy that need to be expanded or something like that?

https://github.com/llvm/llvm-project/pull/163217