Index: llvm/lib/Target/ARM/ARMParallelDSP.cpp
===================================================================
--- llvm/lib/Target/ARM/ARMParallelDSP.cpp
+++ llvm/lib/Target/ARM/ARMParallelDSP.cpp
@@ -459,6 +459,10 @@
     if (ValidLHS && ValidRHS)
       return true;
 
+    // Ensure we don't add the root as the incoming accumulator.
+    if (R.getRoot() == I)
+      return false;
+
     return R.InsertAcc(I);
   }
   case Instruction::Mul: {
@@ -535,6 +539,7 @@
       InsertParallelMACs(R);
       Changed = true;
       AllAdds.insert(R.getAdds().begin(), R.getAdds().end());
+      LLVM_DEBUG(dbgs() << "BB after inserting parallel MACs:\n" << BB);
     }
   }
 
Index: llvm/test/CodeGen/ARM/ParallelDSP/self-ref-bug.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/ARM/ParallelDSP/self-ref-bug.ll
@@ -0,0 +1,35 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -mtriple=thumbv7-unknown-linux-gnueabihf < %s -arm-parallel-dsp -verify -dce -S | FileCheck %s
+
+target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
+
+define i32 @test(ptr %b) {
+; CHECK-LABEL: @test(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[B:%.*]], align 2
+; CHECK-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 3
+; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[ARRAYIDX1]], align 2
+; CHECK-NEXT:    [[TMP2:%.*]] = call i32 @llvm.arm.smlad(i32 [[TMP1]], i32 [[TMP0]], i32 0)
+; CHECK-NEXT:    [[MUL13:%.*]] = mul nsw i32 0, 0
+; CHECK-NEXT:    [[ADD15:%.*]] = add nsw i32 [[MUL13]], [[TMP2]]
+; CHECK-NEXT:    ret i32 [[ADD15]]
+;
+entry:
+  %0 = load i16, ptr %b, align 2
+  %conv = sext i16 %0 to i32
+  %arrayidx1 = getelementptr inbounds i16, ptr %b, i32 3
+  %1 = load i16, ptr %arrayidx1, align 2
+  %conv2 = sext i16 %1 to i32
+  %mul = mul nsw i32 %conv2, %conv
+  %incdec.ptr = getelementptr inbounds i16, ptr %b, i32 1
+  %2 = load i16, ptr %incdec.ptr, align 2
+  %conv4 = sext i16 %2 to i32
+  %arrayidx5 = getelementptr inbounds i16, ptr %b, i32 4
+  %3 = load i16, ptr %arrayidx5, align 2
+  %conv6 = sext i16 %3 to i32
+  %mul7 = mul nsw i32 %conv6, %conv4
+  %add9 = add nsw i32 %mul7, %mul
+  %mul13 = mul nsw i32 0, 0
+  %add15 = add nsw i32 %mul13, %add9
+  ret i32 %add15
+}