[llvm] [AMDGPU] Change default loop alignment for GFX9 and higher targets (PR #153065)

[Matt Arsenault via llvm-commits]

================
@@ -1431,6 +1431,14 @@ def FeatureDisable : SubtargetFeature<"",
   "Dummy feature to disable assembler instructions"
 >;
 
+// GFX-9 & higher targets have a 16-dword Instruction Buffer and per-SQ
----------------
arsenm wrote:

```suggestion
// GFX9 & higher targets have a 16-dword Instruction Buffer and per-SQ
```

https://github.com/llvm/llvm-project/pull/153065