Merging r339535:

------------------------------------------------------------------------
r339535 | ctopper | 2018-08-13 08:53:47 +0200 (Mon, 13 Aug 2018) | 5 lines

[SelectionDAG] In PromoteIntRes_BITCAST, when the input is TypePromoteFloat, make sure the output type is scalar. For vectors, use a store and load of temporary.

Previously if the result type was a vector, we emitted a FP_TO_FP16 with a vector result type which isn't valid.

This is basically the opposite case of the root cause of PR38533.
------------------------------------------------------------------------

2 files changed, 21 insertions, 2 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
index 63a1ea13a5f..133831fa76f 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -269,8 +269,8 @@ SDValue DAGTypeLegalizer::PromoteIntRes_BITCAST(SDNode *N) {
     return DAG.getNode(ISD::ANY_EXTEND, dl, NOutVT, GetSoftenedFloat(InOp));
   case TargetLowering::TypePromoteFloat: {
     // Convert the promoted float by hand.
-    SDValue PromotedOp = GetPromotedFloat(InOp);
-    return DAG.getNode(ISD::FP_TO_FP16, dl, NOutVT, PromotedOp);
+    if (!NOutVT.isVector())
+      return DAG.getNode(ISD::FP_TO_FP16, dl, NOutVT, GetPromotedFloat(InOp));
     break;
   }
   case TargetLowering::TypeExpandInteger:
diff --git a/llvm/test/CodeGen/X86/pr38533.ll b/llvm/test/CodeGen/X86/pr38533.ll
index 4c57d84fe65..d6429b7a1fa 100644
--- a/llvm/test/CodeGen/X86/pr38533.ll
+++ b/llvm/test/CodeGen/X86/pr38533.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown | FileCheck %s
 
+; This test makes sure that a vector that needs to be promoted that is bitcasted to fp16 is legalized correctly without causing a width mismatch.
 define void @constant_fold_vector_to_half() {
 ; CHECK-LABEL: constant_fold_vector_to_half:
 ; CHECK:       # %bb.0:
@@ -9,3 +10,21 @@ define void @constant_fold_vector_to_half() {
   store volatile half bitcast (<4 x i4> <i4 0, i4 0, i4 0, i4 4> to half), half* undef
   ret void
 }
+
+; Similarly this makes sure that the opposite bitcast of the above is also legalized without crashing.
+define void @pr38533_2(half %x) {
+; CHECK-LABEL: pr38533_2:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    pushq %rax
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    callq __gnu_f2h_ieee
+; CHECK-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
+; CHECK-NEXT:    movzwl {{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movw %ax, (%rax)
+; CHECK-NEXT:    popq %rax
+; CHECK-NEXT:    .cfi_def_cfa_offset 8
+; CHECK-NEXT:    retq
+  %a = bitcast half %x to <4 x i4>
+  store volatile <4 x i4> %a, <4 x i4>* undef
+  ret void
+}