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

[Jonas Paulsson via llvm-commits]

================
@@ -6769,11 +6898,20 @@ SystemZTargetLowering::LowerOperationWrapper(SDNode *N,
     break;
   }
   case ISD::BITCAST: {
+    SDLoc DL(N);
     SDValue Src = N->getOperand(0);
-    if (N->getValueType(0) == MVT::i128 && Src.getValueType() == MVT::f128 &&
-        !useSoftFloat()) {
-      SDLoc DL(N);
+    EVT SrcVT = Src.getValueType();
+    EVT ResVT = N->getValueType(0);
+    if (ResVT == MVT::i128 && SrcVT == MVT::f128 && !useSoftFloat())
       Results.push_back(expandBitCastF128ToI128(DAG, Src, DL));
+    else if (SrcVT == MVT::i16 && ResVT == MVT::f16) {
----------------
JonPsson1 wrote:

Wrapping all setXXXAction() calls for f16 with a guard against soft-float, which should make sure that no custom handlings that emit vector / fp instructions are reached.

Soft-float fails with calls to non-existing library functions it seems, including calls to the conversion functions we are now using, except with args in GPRs instead.

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