[LegalizeTypes] Properly handle the case when UpdateNodeOperands in PromoteIntOp_MLOA...

The caller can't handle the node having multiple results like a
masked load does. So we need to detect the case and do our own
result replacement.

Fixes PR46532.

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
index 873812d..54dff1c 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -1723,7 +1723,14 @@ SDValue DAGTypeLegalizer::PromoteIntOp_MLOAD(MaskedLoadSDNode *N,
   SDValue Mask = PromoteTargetBoolean(N->getOperand(OpNo), DataVT);
   SmallVector<SDValue, 4> NewOps(N->op_begin(), N->op_end());
   NewOps[OpNo] = Mask;
-  return SDValue(DAG.UpdateNodeOperands(N, NewOps), 0);
+  SDNode *Res = DAG.UpdateNodeOperands(N, NewOps);
+  if (Res == N)
+    return SDValue(Res, 0);
+
+  // Update triggered CSE, do our own replacement since caller can't.
+  ReplaceValueWith(SDValue(N, 0), SDValue(Res, 0));
+  ReplaceValueWith(SDValue(N, 1), SDValue(Res, 1));
+  return SDValue();
 }
 
 SDValue DAGTypeLegalizer::PromoteIntOp_MGATHER(MaskedGatherSDNode *N,

diff --git a/llvm/test/CodeGen/X86/pr46532.ll b/llvm/test/CodeGen/X86/pr46532.ll
new file mode 100644 (file)
index 0000000..3e0638b
--- /dev/null
+++ b/llvm/test/CodeGen/X86/pr46532.ll
@@ -0,0 +1,33 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-linux-gnu -mcpu=haswell | FileCheck %s
+
+define void @WhileWithLoopInvariantOperation.21() {
+; CHECK-LABEL: WhileWithLoopInvariantOperation.21:
+; CHECK:       # %bb.0: # %while.1.body.preheader
+; CHECK-NEXT:    movq (%rax), %rax
+; CHECK-NEXT:    vxorps %xmm0, %xmm0, %xmm0
+; CHECK-NEXT:    vmovaps %xmm0, 32(%rax)
+; CHECK-NEXT:    vmovaps {{.*#+}} ymm0 = [4294967295,4294967295,0,0,0,0,0,0]
+; CHECK-NEXT:    vmaskmovps %ymm0, %ymm0, (%rax)
+while.1.body.preheader:
+  %0 = load i8*, i8** undef, align 8, !invariant.load !0, !dereferenceable !1, !align !2
+  %1 = getelementptr inbounds i8, i8* %0, i64 32
+  tail call void @llvm.memset.p0i8.i64(i8* nonnull align 16 dereferenceable(16) %1, i8 0, i64 16, i1 false)
+  %2 = tail call <8 x float> @llvm.masked.load.v8f32.p0v8f32(<8 x float>* undef, i32 4, <8 x i1> <i1 true, i1 true, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false>, <8 x float> <float undef, float undef, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00>)
+  %3 = tail call <16 x float> @llvm.masked.load.v16f32.p0v16f32(<16 x float>* undef, i32 4, <16 x i1> <i1 true, i1 true, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false>, <16 x float> <float undef, float undef, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00>)
+  tail call void @llvm.masked.store.v16f32.p0v16f32(<16 x float> undef, <16 x float>* nonnull undef, i32 4, <16 x i1> <i1 true, i1 true, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false>)
+  unreachable
+}
+
+declare void @llvm.memset.p0i8.i64(i8* nocapture writeonly, i8, i64, i1 immarg)
+
+declare <8 x float> @llvm.masked.load.v8f32.p0v8f32(<8 x float>*, i32 immarg, <8 x i1>, <8 x float>)
+
+declare <16 x float> @llvm.masked.load.v16f32.p0v16f32(<16 x float>*, i32 immarg, <16 x i1>, <16 x float>)
+
+declare void @llvm.masked.store.v16f32.p0v16f32(<16 x float>, <16 x float>*, i32 immarg, <16 x i1>)
+
+!0 = !{}
+!1 = !{i64 65}
+!2 = !{i64 16}
+