[llvm] [AMDGPU] Save and restore SCC using only 32-bit SGPR. (PR #68367)

[Pierre van Houtryve via llvm-commits]

================
@@ -0,0 +1,87 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
+; RUN:  llc -march=amdgcn -mcpu=gfx906 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX906 %s
----------------
Pierre-vh wrote:

I'm not sure I like a IR test alone for this kind of fix. I would rather see a MIR test with `-run-pass=si-fix-sgpr-copies". Is that doable?
A MIR test can probably be smaller as well, and test more edge cases.

The IR test can stay too - It's just that I would like a more focused MIR test.
For the IR test I would also check more than one arch. Not sure which ones are relevant in this case but I think at least GFX8/10/11 would be nice to see. Perhaps gfx90a too. 

In any case I also think at least one wave32/wave64 test is needed (for both the MIR and IR tests)


https://github.com/llvm/llvm-project/pull/68367