-declare <8 x i16> @llvm.arm.neon.vshiftls.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
-declare <4 x i32> @llvm.arm.neon.vshiftls.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
-declare <2 x i64> @llvm.arm.neon.vshiftls.v2i64(<2 x i32>, <2 x i32>) nounwind readnone
+; And these have a shift just out of range so separate vmovl and vshl
+; instructions are needed.
+define <8 x i16> @vshllu8_bad(<8 x i8>* %A) nounwind {
+; CHECK-LABEL: vshllu8_bad:
+; CHECK: vmovl.u8
+; CHECK: vshl.i16
+ %tmp1 = load <8 x i8>* %A
+ %zext = zext <8 x i8> %tmp1 to <8 x i16>
+ %shift = shl <8 x i16> %zext, <i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9>
+ ret <8 x i16> %shift
+}
+
+define <4 x i32> @vshlls16_bad(<4 x i16>* %A) nounwind {
+; CHECK-LABEL: vshlls16_bad:
+; CHECK: vmovl.s16
+; CHECK: vshl.i32
+ %tmp1 = load <4 x i16>* %A
+ %sext = sext <4 x i16> %tmp1 to <4 x i32>
+ %shift = shl <4 x i32> %sext, <i32 17, i32 17, i32 17, i32 17>
+ ret <4 x i32> %shift
+}