[llvm] AMDGPU: Fix incorrectly selecting fp8/bf8 conversion intrinsics (PR #107291)

[Stanislav Mekhanoshin via llvm-commits]

================
@@ -1657,7 +1664,7 @@ def FeatureISAVersion12 : FeatureSet<
    FeatureFlatAtomicFaddF32Inst,
    FeatureImageInsts,
    FeatureExtendedImageInsts,
-   FeatureFP8ConversionInsts,
+   FeatureFP8ConversionInsts, FeatureCvtFP8VOP1Bug,
----------------
rampitec wrote:

Does not apply to gfx12, it has no SDWA.

https://github.com/llvm/llvm-project/pull/107291