[clang] [compiler-rt] [llvm] [SystemZ] Add support for half (fp16) (PR #109164)

[Ulrich Weigand via cfe-commits]

================
@@ -541,11 +543,26 @@ SystemZTargetLowering::SystemZTargetLowering(const TargetMachine &TM,
   }
 
   // Handle floating-point types.
+  // Promote all f16 operations to float, with some exceptions below.
+  for (unsigned Opc = 0; Opc < ISD::BUILTIN_OP_END; ++Opc)
+    setOperationAction(Opc, MVT::f16, Promote);
+  setOperationAction(ISD::ConstantFP, MVT::f16, Expand);
+  for (MVT VT : {MVT::f32, MVT::f64, MVT::f128}) {
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::f16, Expand);
+    setTruncStoreAction(VT, MVT::f16, Expand);
+  }
+  for (auto Op : {ISD::LOAD, ISD::ATOMIC_LOAD, ISD::STORE, ISD::ATOMIC_STORE})
+    setOperationAction(Op, MVT::f16, Subtarget.hasVector() ? Legal : Custom);
+  setOperationAction(ISD::FP_ROUND, MVT::f16, LibCall);
+  setOperationAction(ISD::STRICT_FP_ROUND, MVT::f16, LibCall);
+  if (Subtarget.hasVector()) // f16 <-> i16 bitcasts.
+    setOperationAction(ISD::BITCAST, MVT::i16, Custom);
----------------
uweigand wrote:

Not sure about the whole `BITCAST` story here.  `f16` should now be legal always, whether we have vector support or not (either in VR16 or FP16).  That seems to imply that we'll have to have a `BITCAST` implementation in both cases too - does it not?

https://github.com/llvm/llvm-project/pull/109164