[AArch64][CodeGen] Fix crash when fptrunc returns fp16 with +nofp attr

[Nashe Mncube via llvm-commits]

When performing lowering of the fptrunc opcode returning fp16 with the +nofp flag enabled we could trigger a compiler crash. This is because we had no custom lowering implemented. This patch implements a custom lowering for the case in which we need to promote an fp16 return type for fptrunc when the +nofp attr is enabled.


diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 28005880eceb..d88b560ac479 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -25046,6 +25046,15 @@ void AArch64TargetLowering::ReplaceNodeResults(
       Results.push_back(
           LowerToPredicatedOp(SDValue(N, 0), DAG, AArch64ISD::MULHU_PRED));
     return;
+  case ISD::FP_ROUND:{
+    if (N->getValueType(0) == MVT::f16 && !Subtarget->hasFullFP16()) {
+        // Promote fp16 result to legal type
+        SDLoc DL(N);
+        EVT NVT = getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
+        Results.push_back(DAG.getNode(ISD::FP16_TO_FP, DL, NVT, N->getOperand(0)));
+    }
+    return;
+  }
   case ISD::FP_TO_UINT:
   case ISD::FP_TO_SINT:
   case ISD::STRICT_FP_TO_SINT:
diff --git a/llvm/test/CodeGen/AArch64/float16-promotion-with-nofp.ll b/llvm/test/CodeGen/AArch64/float16-promotion-with-nofp.ll
new file mode 100644
index 000000000000..03426579131a
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/float16-promotion-with-nofp.ll
@@ -0,0 +1,21 @@
+; RUN: llc -mcpu=cortex-r82 -O1 -o - %s | FileCheck %s
+
+; Source used:
+; __fp16 f2h(float a) { return a; }
+; Compiled with: clang --target=aarch64-arm-none-eabi -march=armv8-r+nofp
+
+define hidden noundef nofpclass(nan inf) half @f2h(float noundef nofpclass(nan inf) %a) local_unnamed_addr #0 {
+;CHECK:      f2h:                                    // @f2h
+;CHECK-NEXT: // %bb.0:                               // %entry
+;CHECK-NEXT:     str x30, [sp, #-16]!                // 8-byte Folded Spill
+;CHECK-NEXT:     bl  __gnu_h2f_ieee
+;CHECK-NEXT:     ldr x30, [sp], #16                  // 8-byte Folded Reload
+;CHECK-NEXT:     ret
+entry:
+  %0 = fptrunc float %a to half
+  ret half %0
+}
+
+
+attributes #0 = { mustprogress nofree norecurse nosync nounwind willreturn memory(none) "denormal-fp-math"="preserve-sign,preserve-sign" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+crc,+lse,+pauth,+ras,+rcpc,+sb,+ssbs,+v8r,-complxnum,-dotprod,-fmv,-fp-armv8,-fp16fml,-fullfp16,-jsconv,-neon,-rdm" }
+

https://github.com/llvm/llvm-project/pull/81724
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20240214/eeac4a52/attachment.html>