[llvm] [AMDGPU] Use subtarget feature for flat offset bit width instead of arch checks (PR #183742)

[Mariusz Sikora via llvm-commits]

mariusz-sikora-at-amd wrote:

Merge conflict with other (`FeatureDX10ClampAndIEEEMode`) change:

```diff
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td
remerge CONFLICT (content): Merge conflict in llvm/lib/Target/AMDGPU/AMDGPU.td
index dddcfdf33bd5..21daf613cfa5 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.td
@@ -1416,13 +1416,7 @@ def FeatureGFX10 : GCNSubtargetFeatureGeneration<"GFX10",
    FeatureVmemWriteVgprInOrder, FeatureVMemToLDSLoad, FeatureCubeInsts,
    FeatureLerpInst, FeatureSadInsts, FeatureQsadInsts,
    FeatureCvtNormInsts, FeatureCvtPkNormVOP2Insts,
-<<<<<<< dac78e98f526 (Merge remote-tracking branch 'external-upstream/main' into users/mariusz-sikora-at-amd/add-flat-offset-bits-feature)
-   FeatureCvtPkNormVOP3Insts, FeatureFlatOffsetBits12
-||||||| de5e081a8339
-   FeatureCvtPkNormVOP3Insts
-=======
-   FeatureCvtPkNormVOP3Insts, FeatureDX10ClampAndIEEEMode
->>>>>>> 0af2d43e0641 ([Clang] Warn if both of `dllexport`/`dllimport` and `exclude_from_explicit_instantiation` are specified (#183515))
+   FeatureCvtPkNormVOP3Insts, FeatureDX10ClampAndIEEEMode, FeatureFlatOffsetBits12
   ]
 >;
```

https://github.com/llvm/llvm-project/pull/183742