[llvm] [AMDGPU] Add custom lowering of llvm.convert.from.arbitrary.fp for FP8 (PR #194144)

[Dmitry Sidorov via llvm-commits]

================
@@ -811,6 +820,17 @@ let OtherPredicates = [HasFP8ConversionInsts] in {
   }
   let SubtargetPredicate = isGFX11Plus in
     def : Cvt_F_F8_Pat_ByteSel<int_amdgcn_cvt_f32_bf8, V_CVT_F32_BF8_OP_SEL_e64>;
+
+  // llvm.convert.from.arbitrary.fp i8 -> f32 OCP scalar patterns.
+  let SubtargetPredicate = isGFX11PlusNot12_50 in
----------------
MrSidims wrote:

Should be better now, though isGFX11Plus, isNotGFX1250Plus are still explicitly mentioned. 

https://github.com/llvm/llvm-project/pull/194144