[llvm] [AMDGPU][NFC] New tests for uncovered cases in SIInstrInfo.cpp (PR #200414)

[via llvm-commits]

================
@@ -1,9 +1,38 @@
-; RUN: llc < %s -mtriple=r600 -mcpu=redwood | FileCheck %s
-
-; CHECK: {{^}}fcmp_sext:
-; CHECK: SETE_DX10  T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc < %s -mtriple=r600 -mcpu=redwood | FileCheck -check-prefix=CHECK-REDWOOD %s
----------------
ambergorzynski wrote:

OK - we looked at quite a few existing test files. [This one](https://github.com/ambergorzynski/llvm-project/blob/amdgpu-siinstrinfo-new-tests/llvm/test/CodeGen/AMDGPU/scalar-float-sopc.ll) was added with the relevant lines of code, so it could be the most appropriate place to add them. But now the test is checking two different -mcpu (see [this commit](https://github.com/ambergorzynski/llvm-project/commit/2d6d723a85c2d007b0359c206d66cd2e5a9f00e1#diff-ebc738e793f7b06a7498b553bfe59a662bc17d7a4b6f6bf6e09736f9526e5908) where the second was added). Our tests produce different assembly for these mcpus, so the file checks break. To add our tests here, we would need to:
- Change the prefixes in the existing run lines to be mcpu-specific 
- Change the name of the test, since it refers to scalars but we use vectors

Should we do that, or is there a reason why the assembly should be checked to be the same across `gfx1150` and `gfx1200` here?

https://github.com/llvm/llvm-project/pull/200414