[llvm] [AMDGPU] Prevent VOP3 from receiving 64b relocations as immediates (PR #221932)

[Carl Ritson via llvm-commits]

perlfu wrote:

> Is this addressing the same issue as #219141 (which is in progress too - and needs further work).

Closing in favour of the existing PR.
Although I spent a while unraveling operand types for vop3, so might follow up with a separate PR for that.

https://github.com/llvm/llvm-project/pull/221932