[llvm] AMDGPU: Add baseline tests for gep flag handling (PR #110814)

[Mikael Holmén via llvm-commits]

mikaelholmen wrote:

> Looks like these are failing: http://45.33.8.238/linux/149769/step_11.txt
> 
> Please take a look and revert for now if it takes a while to fix.

It looks like the operand order in several instructions is not deterministic, but depends on the compiler used to compile llc. I've seen it pass when using gcc and fail with clang 18 and the differences when failing are like
```
-; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v2, v0
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
```

https://github.com/llvm/llvm-project/pull/110814