[llvm] [RegAlloc] Account for use availability when applying rematerializable weight discount (PR #159180)

[Luke Lau via llvm-commits]

lukel97 wrote:

> In principle it looks correct, but it is really hard to reason about massive test changes.

The sheer size of the test diff mostly comes from those amdgcn.bitcast tests. amgdcn.bitcast.1024.bit.ll is 237k lines alone. Do people read the FileCheck output of it or is it mostly to check that it doesn't crash?

It's worth noting though that most of the test diffs here are just noise. In practice on AArch64/RISC-V/X86 there are no changes in llvm-test-suite/SPEC CPU 2017. 

https://github.com/llvm/llvm-project/pull/159180