[llvm] [GlobalISel][AArch64][AMDGPU] Expand FPOWI into series of multiplication (PR #95217)

[via llvm-commits]

isuckatcs wrote:

Sorry, I made a typo in my previous comment. I meant that I haven't seen many checks for size optimization in GlobalISel. I added one regardless. 

Also it turned out I accidentally expanded G_FPOWI into a lot of multiplications instead of using the more optimal binary exponentiation algorithm, which is also used in the DAG and results in fewer multiplications.

https://github.com/llvm/llvm-project/pull/95217