[llvm] [AArch64] Ensure we transferImpOps on BSP pseudo expansions. (PR #149456)

[David Green via llvm-commits]

davemgreen wrote:

I think it is fine, if a bit inefficient. They come from register allocation of the large vectors (something that subregliveness helps with). The values are not coming from the store, the BSP will implicit-def $q16_q17_q18_q19 and insert q16 into it. The others are overridden later before they are used, I believe (or unused, but it looks like they updated by other BSPs). q19 is inserted in the last instruction, the others I think get inserted into q13 and q14 after the moves.

https://github.com/llvm/llvm-project/pull/149456