[llvm] [BOLT] Optimize the codegen of createLoadImmediate for AArch64. (PR #137413)

[Maksim Panchenko via llvm-commits]

maksfb wrote:

Hi Rodrigo, thanks for improving the codegen. While looking at LLVM AArch64 backend, I've noticed that there's `AArch64_IMM::expandMOVImm()` that handles more complex cases of immediate generation. It might be an overkill at this point and there's a level of post-processing required for converting `SmallVectorImpl<ImmInsnModel>` into a sequence of `MCInst`, but perhaps it's not that hard and worth the effort in the long run. The conversion into `MachineInstr` sequence is implemented in `AArch64ExpandPseudo::expandMOVImm()`. 

https://github.com/llvm/llvm-project/pull/137413