[llvm] [AMDGPU] Move architected SGPR implementation into isel (PR #79120)

[Jay Foad via llvm-commits]

================
@@ -0,0 +1,147 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=amdgcn-amd-hsa -mcpu=gfx900 -mattr=+architected-sgprs --verify-machineinstrs < %s | FileCheck -check-prefix=GFX9-SDAG %s
----------------
jayfoad wrote:

> I don't think we have standalone tests that cover workgroup IDs so far.

What about `test/CodeGen/AMDGPU/llvm.amdgcn.workgroup.id.ll`?

> Also, why no RUN line for gfx1200?

We have not enabled FeatureArchitectedSGPRs for GFX12 yet (I was going to do that in a later patch) so it would not test anything interesting yet.

https://github.com/llvm/llvm-project/pull/79120