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

[Dmitry Sidorov via llvm-commits]

================
@@ -785,11 +795,11 @@ def V_CVT_F16_F8_Fake16_Profile : VOP3_Profile_Fake16<V_CVT_F16_F8_Profile>;
 let SubtargetPredicate = isGFX11Plus, OtherPredicates = [HasFP8ConversionInsts],
     mayRaiseFPException = 0, SchedRW = [WriteFloatCvt] in {
   let SubtargetPredicate = isGFX11PlusNot12_50 in
----------------
MrSidims wrote:

Should be better now. 

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