Add support for ARM's Advanced SIMD (NEON) instruction set.

This is still a work in progress but most of the NEON instruction set
is supported.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73919 91177308-0d34-0410-b5e6-96231b3b80d8

88 files changed, 6904 insertions, 0 deletions

diff --git a/test/CodeGen/ARM/2009-06-02-ISelCrash.ll b/test/CodeGen/ARM/2009-06-02-ISelCrash.ll
new file mode 100644
index 0000000000..7cd35b9557
--- /dev/null
+++ b/test/CodeGen/ARM/2009-06-02-ISelCrash.ll
@@ -0,0 +1,62 @@
+; RUN: llvm-as < %s | llc -mtriple=arm-apple-darwin -relocation-model=pic -mattr=+v6,+vfp2
+
+@"\01LC" = external constant [15 x i8]		; <[15 x i8]*> [#uses=1]
+
+declare i32 @printf(i8* nocapture, ...) nounwind
+
+define i32 @main() nounwind {
+entry:
+	br label %bb.i1.i
+
+bb.i1.i:		; preds = %Cos.exit.i.i, %entry
+	br label %bb.i.i.i
+
+bb.i.i.i:		; preds = %bb.i.i.i, %bb.i1.i
+	br i1 undef, label %Cos.exit.i.i, label %bb.i.i.i
+
+Cos.exit.i.i:		; preds = %bb.i.i.i
+	br i1 undef, label %bb2.i.i, label %bb.i1.i
+
+bb2.i.i:		; preds = %Cos.exit.i.i
+	br label %bb3.i.i
+
+bb3.i.i:		; preds = %bb5.i.i, %bb2.i.i
+	br label %bb4.i.i
+
+bb4.i.i:		; preds = %bb4.i.i, %bb3.i.i
+	br i1 undef, label %bb5.i.i, label %bb4.i.i
+
+bb5.i.i:		; preds = %bb4.i.i
+	br i1 undef, label %bb.i, label %bb3.i.i
+
+bb.i:		; preds = %bb.i, %bb5.i.i
+	br i1 undef, label %bb1.outer2.i.i.outer, label %bb.i
+
+bb1.outer2.i.i.outer:		; preds = %Fft.exit.i, %bb5.i12.i, %bb.i
+	br label %bb1.outer2.i.i
+
+bb1.outer2.i.i:		; preds = %bb2.i9.i, %bb1.outer2.i.i.outer
+	br label %bb1.i.i
+
+bb1.i.i:		; preds = %bb1.i.i, %bb1.outer2.i.i
+	br i1 undef, label %bb2.i9.i, label %bb1.i.i
+
+bb2.i9.i:		; preds = %bb1.i.i
+	br i1 undef, label %bb4.i11.i, label %bb1.outer2.i.i
+
+bb4.i11.i:		; preds = %bb4.i11.i, %bb2.i9.i
+	br i1 undef, label %bb5.i12.i, label %bb4.i11.i
+
+bb5.i12.i:		; preds = %bb4.i11.i
+	br i1 undef, label %bb7.i.i, label %bb1.outer2.i.i.outer
+
+bb7.i.i:		; preds = %bb7.i.i, %bb5.i12.i
+	br i1 undef, label %Fft.exit.i, label %bb7.i.i
+
+Fft.exit.i:		; preds = %bb7.i.i
+	br i1 undef, label %bb5.i, label %bb1.outer2.i.i.outer
+
+bb5.i:		; preds = %Fft.exit.i
+	%0 = tail call i32 (i8*, ...)* @printf(i8* getelementptr ([15 x i8]* @"\01LC", i32 0, i32 0), double undef, double undef) nounwind		; <i32> [#uses=0]
+	unreachable
+}
diff --git a/test/CodeGen/ARM/neon_arith1.ll b/test/CodeGen/ARM/neon_arith1.ll
new file mode 100644
index 0000000000..18b516fc1a
--- /dev/null
+++ b/test/CodeGen/ARM/neon_arith1.ll
@@ -0,0 +1,7 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon | grep vadd
+
+define <8 x i8> @t_i8x8(<8 x i8> %a, <8 x i8> %b) nounwind {
+entry:
+	%0 = add <8 x i8> %a, %b
+	ret <8 x i8> %0
+}
diff --git a/test/CodeGen/ARM/neon_ld1.ll b/test/CodeGen/ARM/neon_ld1.ll
new file mode 100644
index 0000000000..8901ba177d
--- /dev/null
+++ b/test/CodeGen/ARM/neon_ld1.ll
@@ -0,0 +1,22 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon | grep fldd | count 4
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon | grep fstd
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon | grep fmrrd
+
+define void @t1(<2 x i32>* %r, <4 x i16>* %a, <4 x i16>* %b) nounwind {
+entry:
+	%0 = load <4 x i16>* %a, align 8		; <<4 x i16>> [#uses=1]
+	%1 = load <4 x i16>* %b, align 8		; <<4 x i16>> [#uses=1]
+	%2 = add <4 x i16> %0, %1		; <<4 x i16>> [#uses=1]
+	%3 = bitcast <4 x i16> %2 to <2 x i32>		; <<2 x i32>> [#uses=1]
+	store <2 x i32> %3, <2 x i32>* %r, align 8
+	ret void
+}
+
+define <2 x i32> @t2(<4 x i16>* %a, <4 x i16>* %b) nounwind readonly {
+entry:
+	%0 = load <4 x i16>* %a, align 8		; <<4 x i16>> [#uses=1]
+	%1 = load <4 x i16>* %b, align 8		; <<4 x i16>> [#uses=1]
+	%2 = sub <4 x i16> %0, %1		; <<4 x i16>> [#uses=1]
+	%3 = bitcast <4 x i16> %2 to <2 x i32>		; <<2 x i32>> [#uses=1]
+	ret <2 x i32> %3
+}
diff --git a/test/CodeGen/ARM/neon_ld2.ll b/test/CodeGen/ARM/neon_ld2.ll
new file mode 100644
index 0000000000..a26904afca
--- /dev/null
+++ b/test/CodeGen/ARM/neon_ld2.ll
@@ -0,0 +1,23 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon | grep vldmia | count 4
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon | grep vstmia | count 1
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon | grep fmrrd  | count 2
+
+define void @t1(<4 x i32>* %r, <2 x i64>* %a, <2 x i64>* %b) nounwind {
+entry:
+	%0 = load <2 x i64>* %a, align 16		; <<2 x i64>> [#uses=1]
+	%1 = load <2 x i64>* %b, align 16		; <<2 x i64>> [#uses=1]
+	%2 = add <2 x i64> %0, %1		; <<2 x i64>> [#uses=1]
+	%3 = bitcast <2 x i64> %2 to <4 x i32>		; <<4 x i32>> [#uses=1]
+	store <4 x i32> %3, <4 x i32>* %r, align 16
+	ret void
+}
+
+define <4 x i32> @t2(<2 x i64>* %a, <2 x i64>* %b) nounwind readonly {
+entry:
+	%0 = load <2 x i64>* %a, align 16		; <<2 x i64>> [#uses=1]
+	%1 = load <2 x i64>* %b, align 16		; <<2 x i64>> [#uses=1]
+	%2 = sub <2 x i64> %0, %1		; <<2 x i64>> [#uses=1]
+	%3 = bitcast <2 x i64> %2 to <4 x i32>		; <<4 x i32>> [#uses=1]
+	ret <4 x i32> %3
+}
+
diff --git a/test/CodeGen/ARM/vaba.ll b/test/CodeGen/ARM/vaba.ll
new file mode 100644
index 0000000000..98ee1e155b
--- /dev/null
+++ b/test/CodeGen/ARM/vaba.ll
@@ -0,0 +1,119 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vaba\\.s8} %t | count 2
+; RUN: grep {vaba\\.s16} %t | count 2
+; RUN: grep {vaba\\.s32} %t | count 2
+; RUN: grep {vaba\\.u8} %t | count 2
+; RUN: grep {vaba\\.u16} %t | count 2
+; RUN: grep {vaba\\.u32} %t | count 2
+
+define <8 x i8> @vabas8(<8 x i8>* %A, <8 x i8>* %B, <8 x i8>* %C) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = load <8 x i8>* %C
+	%tmp4 = call <8 x i8> @llvm.arm.neon.vabas.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2, <8 x i8> %tmp3)
+	ret <8 x i8> %tmp4
+}
+
+define <4 x i16> @vabas16(<4 x i16>* %A, <4 x i16>* %B, <4 x i16>* %C) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = load <4 x i16>* %C
+	%tmp4 = call <4 x i16> @llvm.arm.neon.vabas.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2, <4 x i16> %tmp3)
+	ret <4 x i16> %tmp4
+}
+
+define <2 x i32> @vabas32(<2 x i32>* %A, <2 x i32>* %B, <2 x i32>* %C) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = load <2 x i32>* %C
+	%tmp4 = call <2 x i32> @llvm.arm.neon.vabas.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2, <2 x i32> %tmp3)
+	ret <2 x i32> %tmp4
+}
+
+define <8 x i8> @vabau8(<8 x i8>* %A, <8 x i8>* %B, <8 x i8>* %C) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = load <8 x i8>* %C
+	%tmp4 = call <8 x i8> @llvm.arm.neon.vabau.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2, <8 x i8> %tmp3)
+	ret <8 x i8> %tmp4
+}
+
+define <4 x i16> @vabau16(<4 x i16>* %A, <4 x i16>* %B, <4 x i16>* %C) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = load <4 x i16>* %C
+	%tmp4 = call <4 x i16> @llvm.arm.neon.vabau.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2, <4 x i16> %tmp3)
+	ret <4 x i16> %tmp4
+}
+
+define <2 x i32> @vabau32(<2 x i32>* %A, <2 x i32>* %B, <2 x i32>* %C) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = load <2 x i32>* %C
+	%tmp4 = call <2 x i32> @llvm.arm.neon.vabau.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2, <2 x i32> %tmp3)
+	ret <2 x i32> %tmp4
+}
+
+define <16 x i8> @vabaQs8(<16 x i8>* %A, <16 x i8>* %B, <16 x i8>* %C) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = load <16 x i8>* %C
+	%tmp4 = call <16 x i8> @llvm.arm.neon.vabas.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2, <16 x i8> %tmp3)
+	ret <16 x i8> %tmp4
+}
+
+define <8 x i16> @vabaQs16(<8 x i16>* %A, <8 x i16>* %B, <8 x i16>* %C) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = load <8 x i16>* %C
+	%tmp4 = call <8 x i16> @llvm.arm.neon.vabas.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2, <8 x i16> %tmp3)
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vabaQs32(<4 x i32>* %A, <4 x i32>* %B, <4 x i32>* %C) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = load <4 x i32>* %C
+	%tmp4 = call <4 x i32> @llvm.arm.neon.vabas.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2, <4 x i32> %tmp3)
+	ret <4 x i32> %tmp4
+}
+
+define <16 x i8> @vabaQu8(<16 x i8>* %A, <16 x i8>* %B, <16 x i8>* %C) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = load <16 x i8>* %C
+	%tmp4 = call <16 x i8> @llvm.arm.neon.vabau.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2, <16 x i8> %tmp3)
+	ret <16 x i8> %tmp4
+}
+
+define <8 x i16> @vabaQu16(<8 x i16>* %A, <8 x i16>* %B, <8 x i16>* %C) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = load <8 x i16>* %C
+	%tmp4 = call <8 x i16> @llvm.arm.neon.vabau.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2, <8 x i16> %tmp3)
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vabaQu32(<4 x i32>* %A, <4 x i32>* %B, <4 x i32>* %C) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = load <4 x i32>* %C
+	%tmp4 = call <4 x i32> @llvm.arm.neon.vabau.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2, <4 x i32> %tmp3)
+	ret <4 x i32> %tmp4
+}
+
+declare <8 x i8>  @llvm.arm.neon.vabas.v8i8(<8 x i8>, <8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vabas.v4i16(<4 x i16>, <4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vabas.v2i32(<2 x i32>, <2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vabau.v8i8(<8 x i8>, <8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vabau.v4i16(<4 x i16>, <4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vabau.v2i32(<2 x i32>, <2 x i32>, <2 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vabas.v16i8(<16 x i8>, <16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vabas.v8i16(<8 x i16>, <8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vabas.v4i32(<4 x i32>, <4 x i32>, <4 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vabau.v16i8(<16 x i8>, <16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vabau.v8i16(<8 x i16>, <8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vabau.v4i32(<4 x i32>, <4 x i32>, <4 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vabal.ll b/test/CodeGen/ARM/vabal.ll
new file mode 100644
index 0000000000..ed9460345b
--- /dev/null
+++ b/test/CodeGen/ARM/vabal.ll
@@ -0,0 +1,63 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vabal\\.s8} %t | count 1
+; RUN: grep {vabal\\.s16} %t | count 1
+; RUN: grep {vabal\\.s32} %t | count 1
+; RUN: grep {vabal\\.u8} %t | count 1
+; RUN: grep {vabal\\.u16} %t | count 1
+; RUN: grep {vabal\\.u32} %t | count 1
+
+define <8 x i16> @vabals8(<8 x i16>* %A, <8 x i8>* %B, <8 x i8>* %C) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = load <8 x i8>* %C
+	%tmp4 = call <8 x i16> @llvm.arm.neon.vabals.v8i16(<8 x i16> %tmp1, <8 x i8> %tmp2, <8 x i8> %tmp3)
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vabals16(<4 x i32>* %A, <4 x i16>* %B, <4 x i16>* %C) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = load <4 x i16>* %C
+	%tmp4 = call <4 x i32> @llvm.arm.neon.vabals.v4i32(<4 x i32> %tmp1, <4 x i16> %tmp2, <4 x i16> %tmp3)
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @vabals32(<2 x i64>* %A, <2 x i32>* %B, <2 x i32>* %C) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = load <2 x i32>* %C
+	%tmp4 = call <2 x i64> @llvm.arm.neon.vabals.v2i64(<2 x i64> %tmp1, <2 x i32> %tmp2, <2 x i32> %tmp3)
+	ret <2 x i64> %tmp4
+}
+
+define <8 x i16> @vabalu8(<8 x i16>* %A, <8 x i8>* %B, <8 x i8>* %C) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = load <8 x i8>* %C
+	%tmp4 = call <8 x i16> @llvm.arm.neon.vabalu.v8i16(<8 x i16> %tmp1, <8 x i8> %tmp2, <8 x i8> %tmp3)
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vabalu16(<4 x i32>* %A, <4 x i16>* %B, <4 x i16>* %C) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = load <4 x i16>* %C
+	%tmp4 = call <4 x i32> @llvm.arm.neon.vabalu.v4i32(<4 x i32> %tmp1, <4 x i16> %tmp2, <4 x i16> %tmp3)
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @vabalu32(<2 x i64>* %A, <2 x i32>* %B, <2 x i32>* %C) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = load <2 x i32>* %C
+	%tmp4 = call <2 x i64> @llvm.arm.neon.vabalu.v2i64(<2 x i64> %tmp1, <2 x i32> %tmp2, <2 x i32> %tmp3)
+	ret <2 x i64> %tmp4
+}
+
+declare <8 x i16> @llvm.arm.neon.vabals.v8i16(<8 x i16>, <8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vabals.v4i32(<4 x i32>, <4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vabals.v2i64(<2 x i64>, <2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vabalu.v8i16(<8 x i16>, <8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vabalu.v4i32(<4 x i32>, <4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vabalu.v2i64(<2 x i64>, <2 x i32>, <2 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vabd.ll b/test/CodeGen/ARM/vabd.ll
new file mode 100644
index 0000000000..0fe5ddb94b
--- /dev/null
+++ b/test/CodeGen/ARM/vabd.ll
@@ -0,0 +1,126 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vabd\\.s8} %t | count 2
+; RUN: grep {vabd\\.s16} %t | count 2
+; RUN: grep {vabd\\.s32} %t | count 2
+; RUN: grep {vabd\\.u8} %t | count 2
+; RUN: grep {vabd\\.u16} %t | count 2
+; RUN: grep {vabd\\.u32} %t | count 2
+; RUN: grep {vabd\\.f32} %t | count 2
+
+define <8 x i8> @vabds8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vabds.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vabds16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vabds.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vabds32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vabds.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <8 x i8> @vabdu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vabdu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vabdu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vabdu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vabdu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vabdu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <2 x float> @vabdf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = call <2 x float> @llvm.arm.neon.vabdf.v2f32(<2 x float> %tmp1, <2 x float> %tmp2)
+	ret <2 x float> %tmp3
+}
+
+define <16 x i8> @vabdQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vabds.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vabdQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vabds.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vabdQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vabds.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <16 x i8> @vabdQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vabdu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vabdQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vabdu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vabdQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vabdu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <4 x float> @vabdQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = call <4 x float> @llvm.arm.neon.vabdf.v4f32(<4 x float> %tmp1, <4 x float> %tmp2)
+	ret <4 x float> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vabds.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vabds.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vabds.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vabdu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vabdu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vabdu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <2 x float> @llvm.arm.neon.vabdf.v2f32(<2 x float>, <2 x float>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vabds.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vabds.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vabds.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vabdu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vabdu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vabdu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+
+declare <4 x float> @llvm.arm.neon.vabdf.v4f32(<4 x float>, <4 x float>) nounwind readnone
diff --git a/test/CodeGen/ARM/vabdl.ll b/test/CodeGen/ARM/vabdl.ll
new file mode 100644
index 0000000000..b351d0f8d3
--- /dev/null
+++ b/test/CodeGen/ARM/vabdl.ll
@@ -0,0 +1,57 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vabdl\\.s8} %t | count 1
+; RUN: grep {vabdl\\.s16} %t | count 1
+; RUN: grep {vabdl\\.s32} %t | count 1
+; RUN: grep {vabdl\\.u8} %t | count 1
+; RUN: grep {vabdl\\.u16} %t | count 1
+; RUN: grep {vabdl\\.u32} %t | count 1
+
+define <8 x i16> @vabdls8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vabdls.v8i16(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vabdls16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vabdls.v4i32(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vabdls32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vabdls.v2i64(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <8 x i16> @vabdlu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vabdlu.v8i16(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vabdlu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vabdlu.v4i32(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vabdlu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vabdlu.v2i64(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+declare <8 x i16> @llvm.arm.neon.vabdls.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vabdls.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vabdls.v2i64(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vabdlu.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vabdlu.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vabdlu.v2i64(<2 x i32>, <2 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vabs.ll b/test/CodeGen/ARM/vabs.ll
new file mode 100644
index 0000000000..629baa762a
--- /dev/null
+++ b/test/CodeGen/ARM/vabs.ll
@@ -0,0 +1,64 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vabs\\.s8} %t | count 2
+; RUN: grep {vabs\\.s16} %t | count 2
+; RUN: grep {vabs\\.s32} %t | count 2
+; RUN: grep {vabs\\.f32} %t | count 2
+
+define <8 x i8> @vabss8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vabs.v8i8(<8 x i8> %tmp1)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vabss16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vabs.v4i16(<4 x i16> %tmp1)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vabss32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vabs.v2i32(<2 x i32> %tmp1)
+	ret <2 x i32> %tmp2
+}
+
+define <2 x float> @vabsf32(<2 x float>* %A) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = call <2 x float> @llvm.arm.neon.vabsf.v2f32(<2 x float> %tmp1)
+	ret <2 x float> %tmp2
+}
+
+define <16 x i8> @vabsQs8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vabs.v16i8(<16 x i8> %tmp1)
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vabsQs16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vabs.v8i16(<8 x i16> %tmp1)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vabsQs32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vabs.v4i32(<4 x i32> %tmp1)
+	ret <4 x i32> %tmp2
+}
+
+define <4 x float> @vabsQf32(<4 x float>* %A) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = call <4 x float> @llvm.arm.neon.vabsf.v4f32(<4 x float> %tmp1)
+	ret <4 x float> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vabs.v8i8(<8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vabs.v4i16(<4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vabs.v2i32(<2 x i32>) nounwind readnone
+declare <2 x float> @llvm.arm.neon.vabsf.v2f32(<2 x float>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vabs.v16i8(<16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vabs.v8i16(<8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vabs.v4i32(<4 x i32>) nounwind readnone
+declare <4 x float> @llvm.arm.neon.vabsf.v4f32(<4 x float>) nounwind readnone
+
diff --git a/test/CodeGen/ARM/vacge.ll b/test/CodeGen/ARM/vacge.ll
new file mode 100644
index 0000000000..5703dd15f3
--- /dev/null
+++ b/test/CodeGen/ARM/vacge.ll
@@ -0,0 +1,19 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vacge\\.f32} %t | count 2
+
+define <2 x i32> @vacgef32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vacged(<2 x float> %tmp1, <2 x float> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <4 x i32> @vacgeQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vacgeq(<4 x float> %tmp1, <4 x float> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+declare <2 x i32> @llvm.arm.neon.vacged(<2 x float>, <2 x float>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vacgeq(<4 x float>, <4 x float>) nounwind readnone
diff --git a/test/CodeGen/ARM/vacgt.ll b/test/CodeGen/ARM/vacgt.ll
new file mode 100644
index 0000000000..ebf590e98a
--- /dev/null
+++ b/test/CodeGen/ARM/vacgt.ll
@@ -0,0 +1,19 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vacgt\\.f32} %t | count 2
+
+define <2 x i32> @vacgtf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vacgtd(<2 x float> %tmp1, <2 x float> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <4 x i32> @vacgtQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vacgtq(<4 x float> %tmp1, <4 x float> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+declare <2 x i32> @llvm.arm.neon.vacgtd(<2 x float>, <2 x float>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vacgtq(<4 x float>, <4 x float>) nounwind readnone
diff --git a/test/CodeGen/ARM/vadd.ll b/test/CodeGen/ARM/vadd.ll
new file mode 100644
index 0000000000..b2b0e2397c
--- /dev/null
+++ b/test/CodeGen/ARM/vadd.ll
@@ -0,0 +1,76 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vadd\\.i8} %t | count 2
+; RUN: grep {vadd\\.i16} %t | count 2
+; RUN: grep {vadd\\.i32} %t | count 2
+; RUN: grep {vadd\\.i64} %t | count 2
+; RUN: grep {vadd\\.f32} %t | count 2
+
+define <8 x i8> @vaddi8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = add <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vaddi16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = add <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vaddi32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = add <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vaddi64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = add <1 x i64> %tmp1, %tmp2
+	ret <1 x i64> %tmp3
+}
+
+define <2 x float> @vaddf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = add <2 x float> %tmp1, %tmp2
+	ret <2 x float> %tmp3
+}
+
+define <16 x i8> @vaddQi8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = add <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vaddQi16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = add <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vaddQi32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = add <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vaddQi64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = add <2 x i64> %tmp1, %tmp2
+	ret <2 x i64> %tmp3
+}
+
+define <4 x float> @vaddQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = add <4 x float> %tmp1, %tmp2
+	ret <4 x float> %tmp3
+}
diff --git a/test/CodeGen/ARM/vaddhn.ll b/test/CodeGen/ARM/vaddhn.ll
new file mode 100644
index 0000000000..b2d1b142f3
--- /dev/null
+++ b/test/CodeGen/ARM/vaddhn.ll
@@ -0,0 +1,29 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vaddhn\\.i16} %t | count 1
+; RUN: grep {vaddhn\\.i32} %t | count 1
+; RUN: grep {vaddhn\\.i64} %t | count 1
+
+define <8 x i8> @vaddhni16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vaddhn.v8i8(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vaddhni32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vaddhn.v4i16(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vaddhni64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vaddhn.v2i32(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vaddhn.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vaddhn.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vaddhn.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vaddl.ll b/test/CodeGen/ARM/vaddl.ll
new file mode 100644
index 0000000000..26ab17bbc0
--- /dev/null
+++ b/test/CodeGen/ARM/vaddl.ll
@@ -0,0 +1,57 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vaddl\\.s8} %t | count 1
+; RUN: grep {vaddl\\.s16} %t | count 1
+; RUN: grep {vaddl\\.s32} %t | count 1
+; RUN: grep {vaddl\\.u8} %t | count 1
+; RUN: grep {vaddl\\.u16} %t | count 1
+; RUN: grep {vaddl\\.u32} %t | count 1
+
+define <8 x i16> @vaddls8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vaddls.v8i16(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vaddls16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vaddls.v4i32(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vaddls32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vaddls.v2i64(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <8 x i16> @vaddlu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vaddlu.v8i16(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vaddlu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vaddlu.v4i32(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vaddlu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vaddlu.v2i64(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+declare <8 x i16> @llvm.arm.neon.vaddls.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vaddls.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vaddls.v2i64(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vaddlu.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vaddlu.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vaddlu.v2i64(<2 x i32>, <2 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vaddw.ll b/test/CodeGen/ARM/vaddw.ll
new file mode 100644
index 0000000000..e06f94a8df
--- /dev/null
+++ b/test/CodeGen/ARM/vaddw.ll
@@ -0,0 +1,57 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vaddw\\.s8} %t | count 1
+; RUN: grep {vaddw\\.s16} %t | count 1
+; RUN: grep {vaddw\\.s32} %t | count 1
+; RUN: grep {vaddw\\.u8} %t | count 1
+; RUN: grep {vaddw\\.u16} %t | count 1
+; RUN: grep {vaddw\\.u32} %t | count 1
+
+define <8 x i16> @vaddws8(<8 x i16>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vaddws.v8i16(<8 x i16> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vaddws16(<4 x i32>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vaddws.v4i32(<4 x i32> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vaddws32(<2 x i64>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vaddws.v2i64(<2 x i64> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <8 x i16> @vaddwu8(<8 x i16>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vaddwu.v8i16(<8 x i16> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vaddwu16(<4 x i32>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vaddwu.v4i32(<4 x i32> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vaddwu32(<2 x i64>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vaddwu.v2i64(<2 x i64> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+declare <8 x i16> @llvm.arm.neon.vaddws.v8i16(<8 x i16>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vaddws.v4i32(<4 x i32>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vaddws.v2i64(<2 x i64>, <2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vaddwu.v8i16(<8 x i16>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vaddwu.v4i32(<4 x i32>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vaddwu.v2i64(<2 x i64>, <2 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vand.ll b/test/CodeGen/ARM/vand.ll
new file mode 100644
index 0000000000..b3eaf1c5dc
--- /dev/null
+++ b/test/CodeGen/ARM/vand.ll
@@ -0,0 +1,59 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep vand %t | count 8
+; Note: function names do not include "vand" to allow simple grep for opcodes
+
+define <8 x i8> @v_andi8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = and <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @v_andi16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = and <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @v_andi32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = and <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @v_andi64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = and <1 x i64> %tmp1, %tmp2
+	ret <1 x i64> %tmp3
+}
+
+define <16 x i8> @v_andQi8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = and <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @v_andQi16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = and <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @v_andQi32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = and <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @v_andQi64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = and <2 x i64> %tmp1, %tmp2
+	ret <2 x i64> %tmp3
+}
diff --git a/test/CodeGen/ARM/vbic.ll b/test/CodeGen/ARM/vbic.ll
new file mode 100644
index 0000000000..dbc11ea58c
--- /dev/null
+++ b/test/CodeGen/ARM/vbic.ll
@@ -0,0 +1,67 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep vbic %t | count 8
+; Note: function names do not include "vbic" to allow simple grep for opcodes
+
+define <8 x i8> @v_bici8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = xor <8 x i8> %tmp2, < i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1 >
+	%tmp4 = and <8 x i8> %tmp1, %tmp3
+	ret <8 x i8> %tmp4
+}
+
+define <4 x i16> @v_bici16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = xor <4 x i16> %tmp2, < i16 -1, i16 -1, i16 -1, i16 -1 >
+	%tmp4 = and <4 x i16> %tmp1, %tmp3
+	ret <4 x i16> %tmp4
+}
+
+define <2 x i32> @v_bici32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = xor <2 x i32> %tmp2, < i32 -1, i32 -1 >
+	%tmp4 = and <2 x i32> %tmp1, %tmp3
+	ret <2 x i32> %tmp4
+}
+
+define <1 x i64> @v_bici64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = xor <1 x i64> %tmp2, < i64 -1 >
+	%tmp4 = and <1 x i64> %tmp1, %tmp3
+	ret <1 x i64> %tmp4
+}
+
+define <16 x i8> @v_bicQi8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = xor <16 x i8> %tmp2, < i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1 >
+	%tmp4 = and <16 x i8> %tmp1, %tmp3
+	ret <16 x i8> %tmp4
+}
+
+define <8 x i16> @v_bicQi16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = xor <8 x i16> %tmp2, < i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1 >
+	%tmp4 = and <8 x i16> %tmp1, %tmp3
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @v_bicQi32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = xor <4 x i32> %tmp2, < i32 -1, i32 -1, i32 -1, i32 -1 >
+	%tmp4 = and <4 x i32> %tmp1, %tmp3
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @v_bicQi64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = xor <2 x i64> %tmp2, < i64 -1, i64 -1 >
+	%tmp4 = and <2 x i64> %tmp1, %tmp3
+	ret <2 x i64> %tmp4
+}
diff --git a/test/CodeGen/ARM/vbsl.ll b/test/CodeGen/ARM/vbsl.ll
new file mode 100644
index 0000000000..37ddf4de6d
--- /dev/null
+++ b/test/CodeGen/ARM/vbsl.ll
@@ -0,0 +1,91 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep vbsl %t | count 8
+; Note: function names do not include "vbsl" to allow simple grep for opcodes
+
+define <8 x i8> @v_bsli8(<8 x i8>* %A, <8 x i8>* %B, <8 x i8>* %C) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = load <8 x i8>* %C
+	%tmp4 = and <8 x i8> %tmp1, %tmp2
+	%tmp5 = xor <8 x i8> %tmp1, < i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1 >
+	%tmp6 = and <8 x i8> %tmp5, %tmp3
+	%tmp7 = or <8 x i8> %tmp4, %tmp6
+	ret <8 x i8> %tmp7
+}
+
+define <4 x i16> @v_bsli16(<4 x i16>* %A, <4 x i16>* %B, <4 x i16>* %C) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = load <4 x i16>* %C
+	%tmp4 = and <4 x i16> %tmp1, %tmp2
+	%tmp5 = xor <4 x i16> %tmp1, < i16 -1, i16 -1, i16 -1, i16 -1 >
+	%tmp6 = and <4 x i16> %tmp5, %tmp3
+	%tmp7 = or <4 x i16> %tmp4, %tmp6
+	ret <4 x i16> %tmp7
+}
+
+define <2 x i32> @v_bsli32(<2 x i32>* %A, <2 x i32>* %B, <2 x i32>* %C) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = load <2 x i32>* %C
+	%tmp4 = and <2 x i32> %tmp1, %tmp2
+	%tmp5 = xor <2 x i32> %tmp1, < i32 -1, i32 -1 >
+	%tmp6 = and <2 x i32> %tmp5, %tmp3
+	%tmp7 = or <2 x i32> %tmp4, %tmp6
+	ret <2 x i32> %tmp7
+}
+
+define <1 x i64> @v_bsli64(<1 x i64>* %A, <1 x i64>* %B, <1 x i64>* %C) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = load <1 x i64>* %C
+	%tmp4 = and <1 x i64> %tmp1, %tmp2
+	%tmp5 = xor <1 x i64> %tmp1, < i64 -1 >
+	%tmp6 = and <1 x i64> %tmp5, %tmp3
+	%tmp7 = or <1 x i64> %tmp4, %tmp6
+	ret <1 x i64> %tmp7
+}
+
+define <16 x i8> @v_bslQi8(<16 x i8>* %A, <16 x i8>* %B, <16 x i8>* %C) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = load <16 x i8>* %C
+	%tmp4 = and <16 x i8> %tmp1, %tmp2
+	%tmp5 = xor <16 x i8> %tmp1, < i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1 >
+	%tmp6 = and <16 x i8> %tmp5, %tmp3
+	%tmp7 = or <16 x i8> %tmp4, %tmp6
+	ret <16 x i8> %tmp7
+}
+
+define <8 x i16> @v_bslQi16(<8 x i16>* %A, <8 x i16>* %B, <8 x i16>* %C) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = load <8 x i16>* %C
+	%tmp4 = and <8 x i16> %tmp1, %tmp2
+	%tmp5 = xor <8 x i16> %tmp1, < i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1 >
+	%tmp6 = and <8 x i16> %tmp5, %tmp3
+	%tmp7 = or <8 x i16> %tmp4, %tmp6
+	ret <8 x i16> %tmp7
+}
+
+define <4 x i32> @v_bslQi32(<4 x i32>* %A, <4 x i32>* %B, <4 x i32>* %C) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = load <4 x i32>* %C
+	%tmp4 = and <4 x i32> %tmp1, %tmp2
+	%tmp5 = xor <4 x i32> %tmp1, < i32 -1, i32 -1, i32 -1, i32 -1 >
+	%tmp6 = and <4 x i32> %tmp5, %tmp3
+	%tmp7 = or <4 x i32> %tmp4, %tmp6
+	ret <4 x i32> %tmp7
+}
+
+define <2 x i64> @v_bslQi64(<2 x i64>* %A, <2 x i64>* %B, <2 x i64>* %C) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = load <2 x i64>* %C
+	%tmp4 = and <2 x i64> %tmp1, %tmp2
+	%tmp5 = xor <2 x i64> %tmp1, < i64 -1, i64 -1 >
+	%tmp6 = and <2 x i64> %tmp5, %tmp3
+	%tmp7 = or <2 x i64> %tmp4, %tmp6
+	ret <2 x i64> %tmp7
+}
diff --git a/test/CodeGen/ARM/vceq.ll b/test/CodeGen/ARM/vceq.ll
new file mode 100644
index 0000000000..77f1890d08
--- /dev/null
+++ b/test/CodeGen/ARM/vceq.ll
@@ -0,0 +1,61 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vceq\\.i8} %t | count 2
+; RUN: grep {vceq\\.i16} %t | count 2
+; RUN: grep {vceq\\.i32} %t | count 2
+; RUN: grep {vceq\\.f32} %t | count 2
+
+define <8 x i8> @vceqi8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = vicmp eq <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vceqi16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = vicmp eq <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vceqi32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = vicmp eq <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <2 x i32> @vceqf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp oeq <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <16 x i8> @vceqQi8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = vicmp eq <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vceqQi16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = vicmp eq <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vceqQi32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = vicmp eq <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <4 x i32> @vceqQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = vfcmp oeq <4 x float> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
diff --git a/test/CodeGen/ARM/vcge.ll b/test/CodeGen/ARM/vcge.ll
new file mode 100644
index 0000000000..14c623ea08
--- /dev/null
+++ b/test/CodeGen/ARM/vcge.ll
@@ -0,0 +1,106 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vcge\\.s8} %t | count 2
+; RUN: grep {vcge\\.s16} %t | count 2
+; RUN: grep {vcge\\.s32} %t | count 2
+; RUN: grep {vcge\\.u8} %t | count 2
+; RUN: grep {vcge\\.u16} %t | count 2
+; RUN: grep {vcge\\.u32} %t | count 2
+; RUN: grep {vcge\\.f32} %t | count 2
+
+define <8 x i8> @vcges8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = vicmp sge <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vcges16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = vicmp sge <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vcges32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = vicmp sge <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <8 x i8> @vcgeu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = vicmp uge <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vcgeu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = vicmp uge <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vcgeu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = vicmp uge <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <2 x i32> @vcgef32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp oge <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <16 x i8> @vcgeQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = vicmp sge <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vcgeQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = vicmp sge <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vcgeQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = vicmp sge <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <16 x i8> @vcgeQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = vicmp uge <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vcgeQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = vicmp uge <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vcgeQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = vicmp uge <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <4 x i32> @vcgeQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = vfcmp oge <4 x float> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
diff --git a/test/CodeGen/ARM/vcgt.ll b/test/CodeGen/ARM/vcgt.ll
new file mode 100644
index 0000000000..3f7e550787
--- /dev/null
+++ b/test/CodeGen/ARM/vcgt.ll
@@ -0,0 +1,106 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vcgt\\.s8} %t | count 2
+; RUN: grep {vcgt\\.s16} %t | count 2
+; RUN: grep {vcgt\\.s32} %t | count 2
+; RUN: grep {vcgt\\.u8} %t | count 2
+; RUN: grep {vcgt\\.u16} %t | count 2
+; RUN: grep {vcgt\\.u32} %t | count 2
+; RUN: grep {vcgt\\.f32} %t | count 2
+
+define <8 x i8> @vcgts8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = vicmp sgt <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vcgts16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = vicmp sgt <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vcgts32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = vicmp sgt <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <8 x i8> @vcgtu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = vicmp ugt <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vcgtu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = vicmp ugt <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vcgtu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = vicmp ugt <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <2 x i32> @vcgtf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp ogt <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <16 x i8> @vcgtQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = vicmp sgt <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vcgtQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = vicmp sgt <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vcgtQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = vicmp sgt <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <16 x i8> @vcgtQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = vicmp ugt <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vcgtQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = vicmp ugt <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vcgtQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = vicmp ugt <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <4 x i32> @vcgtQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = vfcmp ogt <4 x float> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
diff --git a/test/CodeGen/ARM/vcls.ll b/test/CodeGen/ARM/vcls.ll
new file mode 100644
index 0000000000..69e4bdbe16
--- /dev/null
+++ b/test/CodeGen/ARM/vcls.ll
@@ -0,0 +1,48 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vcls\\.s8} %t | count 2
+; RUN: grep {vcls\\.s16} %t | count 2
+; RUN: grep {vcls\\.s32} %t | count 2
+
+define <8 x i8> @vclss8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vcls.v8i8(<8 x i8> %tmp1)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vclss16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vcls.v4i16(<4 x i16> %tmp1)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vclss32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vcls.v2i32(<2 x i32> %tmp1)
+	ret <2 x i32> %tmp2
+}
+
+define <16 x i8> @vclsQs8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vcls.v16i8(<16 x i8> %tmp1)
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vclsQs16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vcls.v8i16(<8 x i16> %tmp1)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vclsQs32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vcls.v4i32(<4 x i32> %tmp1)
+	ret <4 x i32> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vcls.v8i8(<8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vcls.v4i16(<4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vcls.v2i32(<2 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vcls.v16i8(<16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vcls.v8i16(<8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vcls.v4i32(<4 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vclz.ll b/test/CodeGen/ARM/vclz.ll
new file mode 100644
index 0000000000..575ea7d2e8
--- /dev/null
+++ b/test/CodeGen/ARM/vclz.ll
@@ -0,0 +1,48 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vclz\\.i8} %t | count 2
+; RUN: grep {vclz\\.i16} %t | count 2
+; RUN: grep {vclz\\.i32} %t | count 2
+
+define <8 x i8> @vclz8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vclz.v8i8(<8 x i8> %tmp1)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vclz16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vclz.v4i16(<4 x i16> %tmp1)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vclz32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vclz.v2i32(<2 x i32> %tmp1)
+	ret <2 x i32> %tmp2
+}
+
+define <16 x i8> @vclzQ8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vclz.v16i8(<16 x i8> %tmp1)
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vclzQ16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vclz.v8i16(<8 x i16> %tmp1)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vclzQ32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vclz.v4i32(<4 x i32> %tmp1)
+	ret <4 x i32> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vclz.v8i8(<8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vclz.v4i16(<4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vclz.v2i32(<2 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vclz.v16i8(<16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vclz.v8i16(<8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vclz.v4i32(<4 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vcnt.ll b/test/CodeGen/ARM/vcnt.ll
new file mode 100644
index 0000000000..9817168958
--- /dev/null
+++ b/test/CodeGen/ARM/vcnt.ll
@@ -0,0 +1,17 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vcnt\\.8} %t | count 2
+
+define <8 x i8> @vcnt8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vcnt.v8i8(<8 x i8> %tmp1)
+	ret <8 x i8> %tmp2
+}
+
+define <16 x i8> @vcntQ8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vcnt.v16i8(<16 x i8> %tmp1)
+	ret <16 x i8> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vcnt.v8i8(<8 x i8>) nounwind readnone
+declare <16 x i8> @llvm.arm.neon.vcnt.v16i8(<16 x i8>) nounwind readnone
diff --git a/test/CodeGen/ARM/vcvt.ll b/test/CodeGen/ARM/vcvt.ll
new file mode 100644
index 0000000000..1cb42bf155
--- /dev/null
+++ b/test/CodeGen/ARM/vcvt.ll
@@ -0,0 +1,53 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vcvt\\.s32\\.f32} %t | count 2
+; RUN: grep {vcvt\\.u32\\.f32} %t | count 2
+; RUN: grep {vcvt\\.f32\\.s32} %t | count 2
+; RUN: grep {vcvt\\.f32\\.u32} %t | count 2
+
+define <2 x i32> @vcvt_f32tos32(<2 x float>* %A) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = fptosi <2 x float> %tmp1 to <2 x i32>
+	ret <2 x i32> %tmp2
+}
+
+define <2 x i32> @vcvt_f32tou32(<2 x float>* %A) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = fptoui <2 x float> %tmp1 to <2 x i32>
+	ret <2 x i32> %tmp2
+}
+
+define <2 x float> @vcvt_s32tof32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = sitofp <2 x i32> %tmp1 to <2 x float>
+	ret <2 x float> %tmp2
+}
+
+define <2 x float> @vcvt_u32tof32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = uitofp <2 x i32> %tmp1 to <2 x float>
+	ret <2 x float> %tmp2
+}
+
+define <4 x i32> @vcvtQ_f32tos32(<4 x float>* %A) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = fptosi <4 x float> %tmp1 to <4 x i32>
+	ret <4 x i32> %tmp2
+}
+
+define <4 x i32> @vcvtQ_f32tou32(<4 x float>* %A) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = fptoui <4 x float> %tmp1 to <4 x i32>
+	ret <4 x i32> %tmp2
+}
+
+define <4 x float> @vcvtQ_s32tof32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = sitofp <4 x i32> %tmp1 to <4 x float>
+	ret <4 x float> %tmp2
+}
+
+define <4 x float> @vcvtQ_u32tof32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = uitofp <4 x i32> %tmp1 to <4 x float>
+	ret <4 x float> %tmp2
+}
diff --git a/test/CodeGen/ARM/vcvt_n.ll b/test/CodeGen/ARM/vcvt_n.ll
new file mode 100644
index 0000000000..ac86b73d8d
--- /dev/null
+++ b/test/CodeGen/ARM/vcvt_n.ll
@@ -0,0 +1,64 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vcvt\\.s32\\.f32} %t | count 2
+; RUN: grep {vcvt\\.u32\\.f32} %t | count 2
+; RUN: grep {vcvt\\.f32\\.s32} %t | count 2
+; RUN: grep {vcvt\\.f32\\.u32} %t | count 2
+
+define <2 x i32> @vcvt_f32tos32(<2 x float>* %A) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vcvtfp2fxs.v2i32.v2f32(<2 x float> %tmp1, i32 1)
+	ret <2 x i32> %tmp2
+}
+
+define <2 x i32> @vcvt_f32tou32(<2 x float>* %A) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vcvtfp2fxu.v2i32.v2f32(<2 x float> %tmp1, i32 1)
+	ret <2 x i32> %tmp2
+}
+
+define <2 x float> @vcvt_s32tof32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x float> @llvm.arm.neon.vcvtfxs2fp.v2f32.v2i32(<2 x i32> %tmp1, i32 1)
+	ret <2 x float> %tmp2
+}
+
+define <2 x float> @vcvt_u32tof32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x float> @llvm.arm.neon.vcvtfxu2fp.v2f32.v2i32(<2 x i32> %tmp1, i32 1)
+	ret <2 x float> %tmp2
+}
+
+declare <2 x i32> @llvm.arm.neon.vcvtfp2fxs.v2i32.v2f32(<2 x float>, i32) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vcvtfp2fxu.v2i32.v2f32(<2 x float>, i32) nounwind readnone
+declare <2 x float> @llvm.arm.neon.vcvtfxs2fp.v2f32.v2i32(<2 x i32>, i32) nounwind readnone
+declare <2 x float> @llvm.arm.neon.vcvtfxu2fp.v2f32.v2i32(<2 x i32>, i32) nounwind readnone
+
+define <4 x i32> @vcvtQ_f32tos32(<4 x float>* %A) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vcvtfp2fxs.v4i32.v4f32(<4 x float> %tmp1, i32 1)
+	ret <4 x i32> %tmp2
+}
+
+define <4 x i32> @vcvtQ_f32tou32(<4 x float>* %A) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vcvtfp2fxu.v4i32.v4f32(<4 x float> %tmp1, i32 1)
+	ret <4 x i32> %tmp2
+}
+
+define <4 x float> @vcvtQ_s32tof32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x float> @llvm.arm.neon.vcvtfxs2fp.v4f32.v4i32(<4 x i32> %tmp1, i32 1)
+	ret <4 x float> %tmp2
+}
+
+define <4 x float> @vcvtQ_u32tof32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x float> @llvm.arm.neon.vcvtfxu2fp.v4f32.v4i32(<4 x i32> %tmp1, i32 1)
+	ret <4 x float> %tmp2
+}
+
+declare <4 x i32> @llvm.arm.neon.vcvtfp2fxs.v4i32.v4f32(<4 x float>, i32) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vcvtfp2fxu.v4i32.v4f32(<4 x float>, i32) nounwind readnone
+declare <4 x float> @llvm.arm.neon.vcvtfxs2fp.v4f32.v4i32(<4 x i32>, i32) nounwind readnone
+declare <4 x float> @llvm.arm.neon.vcvtfxu2fp.v4f32.v4i32(<4 x i32>, i32) nounwind readnone
+
diff --git a/test/CodeGen/ARM/vdup.ll b/test/CodeGen/ARM/vdup.ll
new file mode 100644
index 0000000000..1c0887a249
--- /dev/null
+++ b/test/CodeGen/ARM/vdup.ll
@@ -0,0 +1,134 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep vdup.8 %t | count 4
+; RUN: grep vdup.16 %t | count 4
+; RUN: grep vdup.32 %t | count 8
+
+define <8 x i8> @v_dup8(i8 %A) nounwind {
+	%tmp1 = insertelement <8 x i8> zeroinitializer, i8 %A, i32 0
+	%tmp2 = insertelement <8 x i8> %tmp1, i8 %A, i32 1
+	%tmp3 = insertelement <8 x i8> %tmp2, i8 %A, i32 2
+	%tmp4 = insertelement <8 x i8> %tmp3, i8 %A, i32 3
+	%tmp5 = insertelement <8 x i8> %tmp4, i8 %A, i32 4
+	%tmp6 = insertelement <8 x i8> %tmp5, i8 %A, i32 5
+	%tmp7 = insertelement <8 x i8> %tmp6, i8 %A, i32 6
+	%tmp8 = insertelement <8 x i8> %tmp7, i8 %A, i32 7
+	ret <8 x i8> %tmp8
+}
+
+define <4 x i16> @v_dup16(i16 %A) nounwind {
+	%tmp1 = insertelement <4 x i16> zeroinitializer, i16 %A, i32 0
+	%tmp2 = insertelement <4 x i16> %tmp1, i16 %A, i32 1
+	%tmp3 = insertelement <4 x i16> %tmp2, i16 %A, i32 2
+	%tmp4 = insertelement <4 x i16> %tmp3, i16 %A, i32 3
+	ret <4 x i16> %tmp4
+}
+
+define <2 x i32> @v_dup32(i32 %A) nounwind {
+	%tmp1 = insertelement <2 x i32> zeroinitializer, i32 %A, i32 0
+	%tmp2 = insertelement <2 x i32> %tmp1, i32 %A, i32 1
+	ret <2 x i32> %tmp2
+}
+
+define <2 x float> @v_dupfloat(float %A) nounwind {
+	%tmp1 = insertelement <2 x float> zeroinitializer, float %A, i32 0
+	%tmp2 = insertelement <2 x float> %tmp1, float %A, i32 1
+	ret <2 x float> %tmp2
+}
+
+define <16 x i8> @v_dupQ8(i8 %A) nounwind {
+	%tmp1 = insertelement <16 x i8> zeroinitializer, i8 %A, i32 0
+	%tmp2 = insertelement <16 x i8> %tmp1, i8 %A, i32 1
+	%tmp3 = insertelement <16 x i8> %tmp2, i8 %A, i32 2
+	%tmp4 = insertelement <16 x i8> %tmp3, i8 %A, i32 3
+	%tmp5 = insertelement <16 x i8> %tmp4, i8 %A, i32 4
+	%tmp6 = insertelement <16 x i8> %tmp5, i8 %A, i32 5
+	%tmp7 = insertelement <16 x i8> %tmp6, i8 %A, i32 6
+	%tmp8 = insertelement <16 x i8> %tmp7, i8 %A, i32 7
+	%tmp9 = insertelement <16 x i8> %tmp8, i8 %A, i32 8
+	%tmp10 = insertelement <16 x i8> %tmp9, i8 %A, i32 9
+	%tmp11 = insertelement <16 x i8> %tmp10, i8 %A, i32 10
+	%tmp12 = insertelement <16 x i8> %tmp11, i8 %A, i32 11
+	%tmp13 = insertelement <16 x i8> %tmp12, i8 %A, i32 12
+	%tmp14 = insertelement <16 x i8> %tmp13, i8 %A, i32 13
+	%tmp15 = insertelement <16 x i8> %tmp14, i8 %A, i32 14
+	%tmp16 = insertelement <16 x i8> %tmp15, i8 %A, i32 15
+	ret <16 x i8> %tmp16
+}
+
+define <8 x i16> @v_dupQ16(i16 %A) nounwind {
+	%tmp1 = insertelement <8 x i16> zeroinitializer, i16 %A, i32 0
+	%tmp2 = insertelement <8 x i16> %tmp1, i16 %A, i32 1
+	%tmp3 = insertelement <8 x i16> %tmp2, i16 %A, i32 2
+	%tmp4 = insertelement <8 x i16> %tmp3, i16 %A, i32 3
+	%tmp5 = insertelement <8 x i16> %tmp4, i16 %A, i32 4
+	%tmp6 = insertelement <8 x i16> %tmp5, i16 %A, i32 5
+	%tmp7 = insertelement <8 x i16> %tmp6, i16 %A, i32 6
+	%tmp8 = insertelement <8 x i16> %tmp7, i16 %A, i32 7
+	ret <8 x i16> %tmp8
+}
+
+define <4 x i32> @v_dupQ32(i32 %A) nounwind {
+	%tmp1 = insertelement <4 x i32> zeroinitializer, i32 %A, i32 0
+	%tmp2 = insertelement <4 x i32> %tmp1, i32 %A, i32 1
+	%tmp3 = insertelement <4 x i32> %tmp2, i32 %A, i32 2
+	%tmp4 = insertelement <4 x i32> %tmp3, i32 %A, i32 3
+	ret <4 x i32> %tmp4
+}
+
+define <4 x float> @v_dupQfloat(float %A) nounwind {
+	%tmp1 = insertelement <4 x float> zeroinitializer, float %A, i32 0
+	%tmp2 = insertelement <4 x float> %tmp1, float %A, i32 1
+	%tmp3 = insertelement <4 x float> %tmp2, float %A, i32 2
+	%tmp4 = insertelement <4 x float> %tmp3, float %A, i32 3
+	ret <4 x float> %tmp4
+}
+
+; Check to make sure it works with shuffles, too.
+
+define <8 x i8> @v_shuffledup8(i8 %A) nounwind {
+	%tmp1 = insertelement <8 x i8> undef, i8 %A, i32 0
+	%tmp2 = shufflevector <8 x i8> %tmp1, <8 x i8> undef, <8 x i32> zeroinitializer
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @v_shuffledup16(i16 %A) nounwind {
+	%tmp1 = insertelement <4 x i16> undef, i16 %A, i32 0
+	%tmp2 = shufflevector <4 x i16> %tmp1, <4 x i16> undef, <4 x i32> zeroinitializer
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @v_shuffledup32(i32 %A) nounwind {
+	%tmp1 = insertelement <2 x i32> undef, i32 %A, i32 0
+	%tmp2 = shufflevector <2 x i32> %tmp1, <2 x i32> undef, <2 x i32> zeroinitializer
+	ret <2 x i32> %tmp2
+}
+
+define <2 x float> @v_shuffledupfloat(float %A) nounwind {
+	%tmp1 = insertelement <2 x float> undef, float %A, i32 0
+	%tmp2 = shufflevector <2 x float> %tmp1, <2 x float> undef, <2 x i32> zeroinitializer
+	ret <2 x float> %tmp2
+}
+
+define <16 x i8> @v_shuffledupQ8(i8 %A) nounwind {
+	%tmp1 = insertelement <16 x i8> undef, i8 %A, i32 0
+	%tmp2 = shufflevector <16 x i8> %tmp1, <16 x i8> undef, <16 x i32> zeroinitializer
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @v_shuffledupQ16(i16 %A) nounwind {
+	%tmp1 = insertelement <8 x i16> undef, i16 %A, i32 0
+	%tmp2 = shufflevector <8 x i16> %tmp1, <8 x i16> undef, <8 x i32> zeroinitializer
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @v_shuffledupQ32(i32 %A) nounwind {
+	%tmp1 = insertelement <4 x i32> undef, i32 %A, i32 0
+	%tmp2 = shufflevector <4 x i32> %tmp1, <4 x i32> undef, <4 x i32> zeroinitializer
+	ret <4 x i32> %tmp2
+}
+
+define <4 x float> @v_shuffledupQfloat(float %A) nounwind {
+	%tmp1 = insertelement <4 x float> undef, float %A, i32 0
+	%tmp2 = shufflevector <4 x float> %tmp1, <4 x float> undef, <4 x i32> zeroinitializer
+	ret <4 x float> %tmp2
+}
diff --git a/test/CodeGen/ARM/vdup_lane.ll b/test/CodeGen/ARM/vdup_lane.ll
new file mode 100644
index 0000000000..adadc9f5d3
--- /dev/null
+++ b/test/CodeGen/ARM/vdup_lane.ll
@@ -0,0 +1,52 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep vdup.8 %t | count 2
+; RUN: grep vdup.16 %t | count 2
+; RUN: grep vdup.32 %t | count 4
+
+define <8 x i8> @vduplane8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = shufflevector <8 x i8> %tmp1, <8 x i8> undef, <8 x i32> < i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1 >
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vduplane16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = shufflevector <4 x i16> %tmp1, <4 x i16> undef, <4 x i32> < i32 1, i32 1, i32 1, i32 1 >
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vduplane32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = shufflevector <2 x i32> %tmp1, <2 x i32> undef, <2 x i32> < i32 1, i32 1 >
+	ret <2 x i32> %tmp2
+}
+
+define <2 x float> @vduplanefloat(<2 x float>* %A) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = shufflevector <2 x float> %tmp1, <2 x float> undef, <2 x i32> < i32 1, i32 1 >
+	ret <2 x float> %tmp2
+}
+
+define <16 x i8> @vduplaneQ8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = shufflevector <8 x i8> %tmp1, <8 x i8> undef, <16 x i32> < i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1 >
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vduplaneQ16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = shufflevector <4 x i16> %tmp1, <4 x i16> undef, <8 x i32> < i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1 >
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vduplaneQ32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = shufflevector <2 x i32> %tmp1, <2 x i32> undef, <4 x i32> < i32 1, i32 1, i32 1, i32 1 >
+	ret <4 x i32> %tmp2
+}
+
+define <4 x float> @vduplaneQfloat(<2 x float>* %A) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = shufflevector <2 x float> %tmp1, <2 x float> undef, <4 x i32> < i32 1, i32 1, i32 1, i32 1 >
+	ret <4 x float> %tmp2
+}
diff --git a/test/CodeGen/ARM/veor.ll b/test/CodeGen/ARM/veor.ll
new file mode 100644
index 0000000000..47a5f3f9f7
--- /dev/null
+++ b/test/CodeGen/ARM/veor.ll
@@ -0,0 +1,59 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep veor %t | count 8
+; Note: function names do not include "veor" to allow simple grep for opcodes
+
+define <8 x i8> @v_eori8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = xor <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @v_eori16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = xor <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @v_eori32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = xor <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @v_eori64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = xor <1 x i64> %tmp1, %tmp2
+	ret <1 x i64> %tmp3
+}
+
+define <16 x i8> @v_eorQi8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = xor <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @v_eorQi16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = xor <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @v_eorQi32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = xor <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @v_eorQi64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = xor <2 x i64> %tmp1, %tmp2
+	ret <2 x i64> %tmp3
+}
diff --git a/test/CodeGen/ARM/vfcmp.ll b/test/CodeGen/ARM/vfcmp.ll
new file mode 100644
index 0000000000..58c2068bc8
--- /dev/null
+++ b/test/CodeGen/ARM/vfcmp.ll
@@ -0,0 +1,96 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vceq\\.f32} %t | count 1
+; RUN: grep {vcgt\\.f32} %t | count 9
+; RUN: grep {vcge\\.f32} %t | count 5
+; RUN: grep vorr %t | count 4
+; RUN: grep vmvn %t | count 7
+
+; This tests vfcmp operations that do not map directly to NEON instructions.
+
+; une is implemented with VCEQ/VMVN
+define <2 x i32> @vcunef32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp une <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+; olt is implemented with VCGT
+define <2 x i32> @vcoltf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp olt <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+; ole is implemented with VCGE
+define <2 x i32> @vcolef32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp ole <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+; uge is implemented with VCGT/VMVN
+define <2 x i32> @vcugef32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp uge <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+; ule is implemented with VCGT/VMVN
+define <2 x i32> @vculef32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp ule <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+; ugt is implemented with VCGE/VMVN
+define <2 x i32> @vcugtf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp ugt <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+; ult is implemented with VCGE/VMVN
+define <2 x i32> @vcultf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp ult <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+; ueq is implemented with VCGT/VCGT/VORR/VMVN
+define <2 x i32> @vcueqf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp ueq <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+; one is implemented with VCGT/VCGT/VORR
+define <2 x i32> @vconef32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp one <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+; uno is implemented with VCGT/VCGE/VORR/VMVN
+define <2 x i32> @vcunof32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp uno <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+; ord is implemented with VCGT/VCGE/VORR
+define <2 x i32> @vcordf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = vfcmp ord <2 x float> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
diff --git a/test/CodeGen/ARM/vget_lane.ll b/test/CodeGen/ARM/vget_lane.ll
new file mode 100644
index 0000000000..a361ba2ba9
--- /dev/null
+++ b/test/CodeGen/ARM/vget_lane.ll
@@ -0,0 +1,78 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vmov\\.s8} %t | count 2
+; RUN: grep {vmov\\.s16} %t | count 2
+; RUN: grep {vmov\\.u8} %t | count 2
+; RUN: grep {vmov\\.u16} %t | count 2
+; RUN: grep {vmov\\.32} %t | count 2
+
+define i32 @vget_lanes8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = extractelement <8 x i8> %tmp1, i32 1
+	%tmp3 = sext i8 %tmp2 to i32
+	ret i32 %tmp3
+}
+
+define i32 @vget_lanes16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = extractelement <4 x i16> %tmp1, i32 1
+	%tmp3 = sext i16 %tmp2 to i32
+	ret i32 %tmp3
+}
+
+define i32 @vget_laneu8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = extractelement <8 x i8> %tmp1, i32 1
+	%tmp3 = zext i8 %tmp2 to i32
+	ret i32 %tmp3
+}
+
+define i32 @vget_laneu16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = extractelement <4 x i16> %tmp1, i32 1
+	%tmp3 = zext i16 %tmp2 to i32
+	ret i32 %tmp3
+}
+
+; Do a vector add to keep the extraction from being done directly from memory.
+define i32 @vget_lanei32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = add <2 x i32> %tmp1, %tmp1
+	%tmp3 = extractelement <2 x i32> %tmp2, i32 1
+	ret i32 %tmp3
+}
+
+define i32 @vgetQ_lanes8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = extractelement <16 x i8> %tmp1, i32 1
+	%tmp3 = sext i8 %tmp2 to i32
+	ret i32 %tmp3
+}
+
+define i32 @vgetQ_lanes16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = extractelement <8 x i16> %tmp1, i32 1
+	%tmp3 = sext i16 %tmp2 to i32
+	ret i32 %tmp3
+}
+
+define i32 @vgetQ_laneu8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = extractelement <16 x i8> %tmp1, i32 1
+	%tmp3 = zext i8 %tmp2 to i32
+	ret i32 %tmp3
+}
+
+define i32 @vgetQ_laneu16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = extractelement <8 x i16> %tmp1, i32 1
+	%tmp3 = zext i16 %tmp2 to i32
+	ret i32 %tmp3
+}
+
+; Do a vector add to keep the extraction from being done directly from memory.
+define i32 @vgetQ_lanei32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = add <4 x i32> %tmp1, %tmp1
+	%tmp3 = extractelement <4 x i32> %tmp2, i32 1
+	ret i32 %tmp3
+}
diff --git a/test/CodeGen/ARM/vhadd.ll b/test/CodeGen/ARM/vhadd.ll
new file mode 100644
index 0000000000..5e7503dc71
--- /dev/null
+++ b/test/CodeGen/ARM/vhadd.ll
@@ -0,0 +1,107 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vhadd\\.s8} %t | count 2
+; RUN: grep {vhadd\\.s16} %t | count 2
+; RUN: grep {vhadd\\.s32} %t | count 2
+; RUN: grep {vhadd\\.u8} %t | count 2
+; RUN: grep {vhadd\\.u16} %t | count 2
+; RUN: grep {vhadd\\.u32} %t | count 2
+
+define <8 x i8> @vhadds8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vhadds.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vhadds16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vhadds.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vhadds32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vhadds.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <8 x i8> @vhaddu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vhaddu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vhaddu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vhaddu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vhaddu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vhaddu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <16 x i8> @vhaddQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vhadds.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vhaddQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vhadds.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vhaddQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vhadds.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <16 x i8> @vhaddQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vhaddu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vhaddQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vhaddu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vhaddQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vhaddu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vhadds.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vhadds.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vhadds.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vhaddu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vhaddu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vhaddu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vhadds.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vhadds.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vhadds.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vhaddu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vhaddu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vhaddu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vhsub.ll b/test/CodeGen/ARM/vhsub.ll
new file mode 100644
index 0000000000..32a66e5479
--- /dev/null
+++ b/test/CodeGen/ARM/vhsub.ll
@@ -0,0 +1,107 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vhsub\\.s8} %t | count 2
+; RUN: grep {vhsub\\.s16} %t | count 2
+; RUN: grep {vhsub\\.s32} %t | count 2
+; RUN: grep {vhsub\\.u8} %t | count 2
+; RUN: grep {vhsub\\.u16} %t | count 2
+; RUN: grep {vhsub\\.u32} %t | count 2
+
+define <8 x i8> @vhsubs8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vhsubs.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vhsubs16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vhsubs.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vhsubs32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vhsubs.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <8 x i8> @vhsubu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vhsubu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vhsubu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vhsubu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vhsubu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vhsubu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <16 x i8> @vhsubQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vhsubs.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vhsubQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vhsubs.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vhsubQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vhsubs.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <16 x i8> @vhsubQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vhsubu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vhsubQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vhsubu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vhsubQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vhsubu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vhsubs.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vhsubs.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vhsubs.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vhsubu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vhsubu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vhsubu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vhsubs.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vhsubs.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vhsubs.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vhsubu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vhsubu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vhsubu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vicmp.ll b/test/CodeGen/ARM/vicmp.ll
new file mode 100644
index 0000000000..86858f9293
--- /dev/null
+++ b/test/CodeGen/ARM/vicmp.ll
@@ -0,0 +1,85 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vceq\\.i8} %t | count 2
+; RUN: grep {vceq\\.i16} %t | count 2
+; RUN: grep {vceq\\.i32} %t | count 2
+; RUN: grep vmvn %t | count 6
+; RUN: grep {vcgt\\.s8} %t | count 1
+; RUN: grep {vcge\\.s16} %t | count 1
+; RUN: grep {vcgt\\.u16} %t | count 1
+; RUN: grep {vcge\\.u32} %t | count 1
+
+; This tests vicmp operations that do not map directly to NEON instructions.
+; Not-equal (ne) operations are implemented by VCEQ/VMVN.  Less-than (lt/ult)
+; and less-than-or-equal (le/ule) are implemented by swapping the arguments
+; to VCGT and VCGE.  Test all the operand types for not-equal but only sample
+; the other operations.
+
+define <8 x i8> @vcnei8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = vicmp ne <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vcnei16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = vicmp ne <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vcnei32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = vicmp ne <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <16 x i8> @vcneQi8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = vicmp ne <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vcneQi16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = vicmp ne <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vcneQi32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = vicmp ne <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <16 x i8> @vcltQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = vicmp slt <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <4 x i16> @vcles16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = vicmp sle <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <4 x i16> @vcltu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = vicmp ult <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <4 x i32> @vcleQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = vicmp ule <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
diff --git a/test/CodeGen/ARM/vmax.ll b/test/CodeGen/ARM/vmax.ll
new file mode 100644
index 0000000000..60322f85d3
--- /dev/null
+++ b/test/CodeGen/ARM/vmax.ll
@@ -0,0 +1,126 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vmax\\.s8} %t | count 2
+; RUN: grep {vmax\\.s16} %t | count 2
+; RUN: grep {vmax\\.s32} %t | count 2
+; RUN: grep {vmax\\.u8} %t | count 2
+; RUN: grep {vmax\\.u16} %t | count 2
+; RUN: grep {vmax\\.u32} %t | count 2
+; RUN: grep {vmax\\.f32} %t | count 2
+
+define <8 x i8> @vmaxs8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vmaxs.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vmaxs16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vmaxs.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vmaxs32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vmaxs.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <8 x i8> @vmaxu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vmaxu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vmaxu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vmaxu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vmaxu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vmaxu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <2 x float> @vmaxf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = call <2 x float> @llvm.arm.neon.vmaxf.v2f32(<2 x float> %tmp1, <2 x float> %tmp2)
+	ret <2 x float> %tmp3
+}
+
+define <16 x i8> @vmaxQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vmaxs.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vmaxQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vmaxs.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vmaxQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vmaxs.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <16 x i8> @vmaxQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vmaxu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vmaxQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vmaxu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vmaxQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vmaxu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <4 x float> @vmaxQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = call <4 x float> @llvm.arm.neon.vmaxf.v4f32(<4 x float> %tmp1, <4 x float> %tmp2)
+	ret <4 x float> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vmaxs.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vmaxs.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vmaxs.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vmaxu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vmaxu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vmaxu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <2 x float> @llvm.arm.neon.vmaxf.v2f32(<2 x float>, <2 x float>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vmaxs.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vmaxs.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmaxs.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vmaxu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vmaxu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmaxu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+
+declare <4 x float> @llvm.arm.neon.vmaxf.v4f32(<4 x float>, <4 x float>) nounwind readnone
diff --git a/test/CodeGen/ARM/vmin.ll b/test/CodeGen/ARM/vmin.ll
new file mode 100644
index 0000000000..a6936937c7
--- /dev/null
+++ b/test/CodeGen/ARM/vmin.ll
@@ -0,0 +1,126 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vmin\\.s8} %t | count 2
+; RUN: grep {vmin\\.s16} %t | count 2
+; RUN: grep {vmin\\.s32} %t | count 2
+; RUN: grep {vmin\\.u8} %t | count 2
+; RUN: grep {vmin\\.u16} %t | count 2
+; RUN: grep {vmin\\.u32} %t | count 2
+; RUN: grep {vmin\\.f32} %t | count 2
+
+define <8 x i8> @vmins8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vmins.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vmins16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vmins.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vmins32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vmins.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <8 x i8> @vminu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vminu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vminu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vminu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vminu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vminu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <2 x float> @vminf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = call <2 x float> @llvm.arm.neon.vminf.v2f32(<2 x float> %tmp1, <2 x float> %tmp2)
+	ret <2 x float> %tmp3
+}
+
+define <16 x i8> @vminQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vmins.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vminQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vmins.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vminQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vmins.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <16 x i8> @vminQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vminu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vminQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vminu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vminQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vminu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <4 x float> @vminQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = call <4 x float> @llvm.arm.neon.vminf.v4f32(<4 x float> %tmp1, <4 x float> %tmp2)
+	ret <4 x float> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vmins.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vmins.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vmins.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vminu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vminu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vminu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <2 x float> @llvm.arm.neon.vminf.v2f32(<2 x float>, <2 x float>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vmins.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vmins.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmins.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vminu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vminu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vminu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+
+declare <4 x float> @llvm.arm.neon.vminf.v4f32(<4 x float>, <4 x float>) nounwind readnone
diff --git a/test/CodeGen/ARM/vmla.ll b/test/CodeGen/ARM/vmla.ll
new file mode 100644
index 0000000000..ed77e11a7c
--- /dev/null
+++ b/test/CodeGen/ARM/vmla.ll
@@ -0,0 +1,77 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vmla\\.i8} %t | count 2
+; RUN: grep {vmla\\.i16} %t | count 2
+; RUN: grep {vmla\\.i32} %t | count 2
+; RUN: grep {vmla\\.f32} %t | count 2
+
+define <8 x i8> @vmlai8(<8 x i8>* %A, <8 x i8>* %B, <8 x i8> * %C) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = load <8 x i8>* %C
+	%tmp4 = mul <8 x i8> %tmp2, %tmp3
+	%tmp5 = add <8 x i8> %tmp1, %tmp4
+	ret <8 x i8> %tmp5
+}
+
+define <4 x i16> @vmlai16(<4 x i16>* %A, <4 x i16>* %B, <4 x i16>* %C) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = load <4 x i16>* %C
+	%tmp4 = mul <4 x i16> %tmp2, %tmp3
+	%tmp5 = add <4 x i16> %tmp1, %tmp4
+	ret <4 x i16> %tmp5
+}
+
+define <2 x i32> @vmlai32(<2 x i32>* %A, <2 x i32>* %B, <2 x i32>* %C) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = load <2 x i32>* %C
+	%tmp4 = mul <2 x i32> %tmp2, %tmp3
+	%tmp5 = add <2 x i32> %tmp1, %tmp4
+	ret <2 x i32> %tmp5
+}
+
+define <2 x float> @vmlaf32(<2 x float>* %A, <2 x float>* %B, <2 x float>* %C) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = load <2 x float>* %C
+	%tmp4 = mul <2 x float> %tmp2, %tmp3
+	%tmp5 = add <2 x float> %tmp1, %tmp4
+	ret <2 x float> %tmp5
+}
+
+define <16 x i8> @vmlaQi8(<16 x i8>* %A, <16 x i8>* %B, <16 x i8> * %C) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = load <16 x i8>* %C
+	%tmp4 = mul <16 x i8> %tmp2, %tmp3
+	%tmp5 = add <16 x i8> %tmp1, %tmp4
+	ret <16 x i8> %tmp5
+}
+
+define <8 x i16> @vmlaQi16(<8 x i16>* %A, <8 x i16>* %B, <8 x i16>* %C) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = load <8 x i16>* %C
+	%tmp4 = mul <8 x i16> %tmp2, %tmp3
+	%tmp5 = add <8 x i16> %tmp1, %tmp4
+	ret <8 x i16> %tmp5
+}
+
+define <4 x i32> @vmlaQi32(<4 x i32>* %A, <4 x i32>* %B, <4 x i32>* %C) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = load <4 x i32>* %C
+	%tmp4 = mul <4 x i32> %tmp2, %tmp3
+	%tmp5 = add <4 x i32> %tmp1, %tmp4
+	ret <4 x i32> %tmp5
+}
+
+define <4 x float> @vmlaQf32(<4 x float>* %A, <4 x float>* %B, <4 x float>* %C) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = load <4 x float>* %C
+	%tmp4 = mul <4 x float> %tmp2, %tmp3
+	%tmp5 = add <4 x float> %tmp1, %tmp4
+	ret <4 x float> %tmp5
+}
diff --git a/test/CodeGen/ARM/vmlal.ll b/test/CodeGen/ARM/vmlal.ll
new file mode 100644
index 0000000000..7fd00bac5e
--- /dev/null
+++ b/test/CodeGen/ARM/vmlal.ll
@@ -0,0 +1,63 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vmlal\\.s8} %t | count 1
+; RUN: grep {vmlal\\.s16} %t | count 1
+; RUN: grep {vmlal\\.s32} %t | count 1
+; RUN: grep {vmlal\\.u8} %t | count 1
+; RUN: grep {vmlal\\.u16} %t | count 1
+; RUN: grep {vmlal\\.u32} %t | count 1
+
+define <8 x i16> @vmlals8(<8 x i16>* %A, <8 x i8>* %B, <8 x i8>* %C) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = load <8 x i8>* %C
+	%tmp4 = call <8 x i16> @llvm.arm.neon.vmlals.v8i16(<8 x i16> %tmp1, <8 x i8> %tmp2, <8 x i8> %tmp3)
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vmlals16(<4 x i32>* %A, <4 x i16>* %B, <4 x i16>* %C) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = load <4 x i16>* %C
+	%tmp4 = call <4 x i32> @llvm.arm.neon.vmlals.v4i32(<4 x i32> %tmp1, <4 x i16> %tmp2, <4 x i16> %tmp3)
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @vmlals32(<2 x i64>* %A, <2 x i32>* %B, <2 x i32>* %C) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = load <2 x i32>* %C
+	%tmp4 = call <2 x i64> @llvm.arm.neon.vmlals.v2i64(<2 x i64> %tmp1, <2 x i32> %tmp2, <2 x i32> %tmp3)
+	ret <2 x i64> %tmp4
+}
+
+define <8 x i16> @vmlalu8(<8 x i16>* %A, <8 x i8>* %B, <8 x i8>* %C) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = load <8 x i8>* %C
+	%tmp4 = call <8 x i16> @llvm.arm.neon.vmlalu.v8i16(<8 x i16> %tmp1, <8 x i8> %tmp2, <8 x i8> %tmp3)
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vmlalu16(<4 x i32>* %A, <4 x i16>* %B, <4 x i16>* %C) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = load <4 x i16>* %C
+	%tmp4 = call <4 x i32> @llvm.arm.neon.vmlalu.v4i32(<4 x i32> %tmp1, <4 x i16> %tmp2, <4 x i16> %tmp3)
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @vmlalu32(<2 x i64>* %A, <2 x i32>* %B, <2 x i32>* %C) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = load <2 x i32>* %C
+	%tmp4 = call <2 x i64> @llvm.arm.neon.vmlalu.v2i64(<2 x i64> %tmp1, <2 x i32> %tmp2, <2 x i32> %tmp3)
+	ret <2 x i64> %tmp4
+}
+
+declare <8 x i16> @llvm.arm.neon.vmlals.v8i16(<8 x i16>, <8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmlals.v4i32(<4 x i32>, <4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vmlals.v2i64(<2 x i64>, <2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vmlalu.v8i16(<8 x i16>, <8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmlalu.v4i32(<4 x i32>, <4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vmlalu.v2i64(<2 x i64>, <2 x i32>, <2 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vmls.ll b/test/CodeGen/ARM/vmls.ll
new file mode 100644
index 0000000000..d519b7e70e
--- /dev/null
+++ b/test/CodeGen/ARM/vmls.ll
@@ -0,0 +1,77 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vmls\\.i8} %t | count 2
+; RUN: grep {vmls\\.i16} %t | count 2
+; RUN: grep {vmls\\.i32} %t | count 2
+; RUN: grep {vmls\\.f32} %t | count 2
+
+define <8 x i8> @vmlsi8(<8 x i8>* %A, <8 x i8>* %B, <8 x i8> * %C) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = load <8 x i8>* %C
+	%tmp4 = mul <8 x i8> %tmp2, %tmp3
+	%tmp5 = sub <8 x i8> %tmp1, %tmp4
+	ret <8 x i8> %tmp5
+}
+
+define <4 x i16> @vmlsi16(<4 x i16>* %A, <4 x i16>* %B, <4 x i16>* %C) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = load <4 x i16>* %C
+	%tmp4 = mul <4 x i16> %tmp2, %tmp3
+	%tmp5 = sub <4 x i16> %tmp1, %tmp4
+	ret <4 x i16> %tmp5
+}
+
+define <2 x i32> @vmlsi32(<2 x i32>* %A, <2 x i32>* %B, <2 x i32>* %C) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = load <2 x i32>* %C
+	%tmp4 = mul <2 x i32> %tmp2, %tmp3
+	%tmp5 = sub <2 x i32> %tmp1, %tmp4
+	ret <2 x i32> %tmp5
+}
+
+define <2 x float> @vmlsf32(<2 x float>* %A, <2 x float>* %B, <2 x float>* %C) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = load <2 x float>* %C
+	%tmp4 = mul <2 x float> %tmp2, %tmp3
+	%tmp5 = sub <2 x float> %tmp1, %tmp4
+	ret <2 x float> %tmp5
+}
+
+define <16 x i8> @vmlsQi8(<16 x i8>* %A, <16 x i8>* %B, <16 x i8> * %C) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = load <16 x i8>* %C
+	%tmp4 = mul <16 x i8> %tmp2, %tmp3
+	%tmp5 = sub <16 x i8> %tmp1, %tmp4
+	ret <16 x i8> %tmp5
+}
+
+define <8 x i16> @vmlsQi16(<8 x i16>* %A, <8 x i16>* %B, <8 x i16>* %C) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = load <8 x i16>* %C
+	%tmp4 = mul <8 x i16> %tmp2, %tmp3
+	%tmp5 = sub <8 x i16> %tmp1, %tmp4
+	ret <8 x i16> %tmp5
+}
+
+define <4 x i32> @vmlsQi32(<4 x i32>* %A, <4 x i32>* %B, <4 x i32>* %C) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = load <4 x i32>* %C
+	%tmp4 = mul <4 x i32> %tmp2, %tmp3
+	%tmp5 = sub <4 x i32> %tmp1, %tmp4
+	ret <4 x i32> %tmp5
+}
+
+define <4 x float> @vmlsQf32(<4 x float>* %A, <4 x float>* %B, <4 x float>* %C) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = load <4 x float>* %C
+	%tmp4 = mul <4 x float> %tmp2, %tmp3
+	%tmp5 = sub <4 x float> %tmp1, %tmp4
+	ret <4 x float> %tmp5
+}
diff --git a/test/CodeGen/ARM/vmlsl.ll b/test/CodeGen/ARM/vmlsl.ll
new file mode 100644
index 0000000000..94910dc77b
--- /dev/null
+++ b/test/CodeGen/ARM/vmlsl.ll
@@ -0,0 +1,63 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vmlsl\\.s8} %t | count 1
+; RUN: grep {vmlsl\\.s16} %t | count 1
+; RUN: grep {vmlsl\\.s32} %t | count 1
+; RUN: grep {vmlsl\\.u8} %t | count 1
+; RUN: grep {vmlsl\\.u16} %t | count 1
+; RUN: grep {vmlsl\\.u32} %t | count 1
+
+define <8 x i16> @vmlsls8(<8 x i16>* %A, <8 x i8>* %B, <8 x i8>* %C) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = load <8 x i8>* %C
+	%tmp4 = call <8 x i16> @llvm.arm.neon.vmlsls.v8i16(<8 x i16> %tmp1, <8 x i8> %tmp2, <8 x i8> %tmp3)
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vmlsls16(<4 x i32>* %A, <4 x i16>* %B, <4 x i16>* %C) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = load <4 x i16>* %C
+	%tmp4 = call <4 x i32> @llvm.arm.neon.vmlsls.v4i32(<4 x i32> %tmp1, <4 x i16> %tmp2, <4 x i16> %tmp3)
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @vmlsls32(<2 x i64>* %A, <2 x i32>* %B, <2 x i32>* %C) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = load <2 x i32>* %C
+	%tmp4 = call <2 x i64> @llvm.arm.neon.vmlsls.v2i64(<2 x i64> %tmp1, <2 x i32> %tmp2, <2 x i32> %tmp3)
+	ret <2 x i64> %tmp4
+}
+
+define <8 x i16> @vmlslu8(<8 x i16>* %A, <8 x i8>* %B, <8 x i8>* %C) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = load <8 x i8>* %C
+	%tmp4 = call <8 x i16> @llvm.arm.neon.vmlslu.v8i16(<8 x i16> %tmp1, <8 x i8> %tmp2, <8 x i8> %tmp3)
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vmlslu16(<4 x i32>* %A, <4 x i16>* %B, <4 x i16>* %C) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = load <4 x i16>* %C
+	%tmp4 = call <4 x i32> @llvm.arm.neon.vmlslu.v4i32(<4 x i32> %tmp1, <4 x i16> %tmp2, <4 x i16> %tmp3)
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @vmlslu32(<2 x i64>* %A, <2 x i32>* %B, <2 x i32>* %C) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = load <2 x i32>* %C
+	%tmp4 = call <2 x i64> @llvm.arm.neon.vmlslu.v2i64(<2 x i64> %tmp1, <2 x i32> %tmp2, <2 x i32> %tmp3)
+	ret <2 x i64> %tmp4
+}
+
+declare <8 x i16> @llvm.arm.neon.vmlsls.v8i16(<8 x i16>, <8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmlsls.v4i32(<4 x i32>, <4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vmlsls.v2i64(<2 x i64>, <2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vmlslu.v8i16(<8 x i16>, <8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmlslu.v4i32(<4 x i32>, <4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vmlslu.v2i64(<2 x i64>, <2 x i32>, <2 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vmov.ll b/test/CodeGen/ARM/vmov.ll
new file mode 100644
index 0000000000..af9c8e2598
--- /dev/null
+++ b/test/CodeGen/ARM/vmov.ll
@@ -0,0 +1,101 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep vmov.i8 %t | count 2
+; RUN: grep vmov.i16 %t | count 4
+; RUN: grep vmov.i32 %t | count 12
+; RUN: grep vmov.i64 %t | count 2
+; Note: function names do not include "vmov" to allow simple grep for opcodes
+
+define <8 x i8> @v_movi8() nounwind {
+	ret <8 x i8> < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
+}
+
+define <4 x i16> @v_movi16a() nounwind {
+	ret <4 x i16> < i16 16, i16 16, i16 16, i16 16 >
+}
+
+; 0x1000 = 4096
+define <4 x i16> @v_movi16b() nounwind {
+	ret <4 x i16> < i16 4096, i16 4096, i16 4096, i16 4096 >
+}
+
+define <2 x i32> @v_movi32a() nounwind {
+	ret <2 x i32> < i32 32, i32 32 >
+}
+
+; 0x2000 = 8192
+define <2 x i32> @v_movi32b() nounwind {
+	ret <2 x i32> < i32 8192, i32 8192 >
+}
+
+; 0x200000 = 2097152
+define <2 x i32> @v_movi32c() nounwind {
+	ret <2 x i32> < i32 2097152, i32 2097152 >
+}
+
+; 0x20000000 = 536870912
+define <2 x i32> @v_movi32d() nounwind {
+	ret <2 x i32> < i32 536870912, i32 536870912 >
+}
+
+; 0x20ff = 8447
+define <2 x i32> @v_movi32e() nounwind {
+	ret <2 x i32> < i32 8447, i32 8447 >
+}
+
+; 0x20ffff = 2162687
+define <2 x i32> @v_movi32f() nounwind {
+	ret <2 x i32> < i32 2162687, i32 2162687 >
+}
+
+; 0xff0000ff0000ffff = 18374687574888349695
+define <1 x i64> @v_movi64() nounwind {
+	ret <1 x i64> < i64 18374687574888349695 >
+}
+
+define <16 x i8> @v_movQi8() nounwind {
+	ret <16 x i8> < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
+}
+
+define <8 x i16> @v_movQi16a() nounwind {
+	ret <8 x i16> < i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16 >
+}
+
+; 0x1000 = 4096
+define <8 x i16> @v_movQi16b() nounwind {
+	ret <8 x i16> < i16 4096, i16 4096, i16 4096, i16 4096, i16 4096, i16 4096, i16 4096, i16 4096 >
+}
+
+define <4 x i32> @v_movQi32a() nounwind {
+	ret <4 x i32> < i32 32, i32 32, i32 32, i32 32 >
+}
+
+; 0x2000 = 8192
+define <4 x i32> @v_movQi32b() nounwind {
+	ret <4 x i32> < i32 8192, i32 8192, i32 8192, i32 8192 >
+}
+
+; 0x200000 = 2097152
+define <4 x i32> @v_movQi32c() nounwind {
+	ret <4 x i32> < i32 2097152, i32 2097152, i32 2097152, i32 2097152 >
+}
+
+; 0x20000000 = 536870912
+define <4 x i32> @v_movQi32d() nounwind {
+	ret <4 x i32> < i32 536870912, i32 536870912, i32 536870912, i32 536870912 >
+}
+
+; 0x20ff = 8447
+define <4 x i32> @v_movQi32e() nounwind {
+	ret <4 x i32> < i32 8447, i32 8447, i32 8447, i32 8447 >
+}
+
+; 0x20ffff = 2162687
+define <4 x i32> @v_movQi32f() nounwind {
+	ret <4 x i32> < i32 2162687, i32 2162687, i32 2162687, i32 2162687 >
+}
+
+; 0xff0000ff0000ffff = 18374687574888349695
+define <2 x i64> @v_movQi64() nounwind {
+	ret <2 x i64> < i64 18374687574888349695, i64 18374687574888349695 >
+}
+
diff --git a/test/CodeGen/ARM/vmovl.ll b/test/CodeGen/ARM/vmovl.ll
new file mode 100644
index 0000000000..09a77af0be
--- /dev/null
+++ b/test/CodeGen/ARM/vmovl.ll
@@ -0,0 +1,51 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vmovl\\.s8} %t | count 1
+; RUN: grep {vmovl\\.s16} %t | count 1
+; RUN: grep {vmovl\\.s32} %t | count 1
+; RUN: grep {vmovl\\.u8} %t | count 1
+; RUN: grep {vmovl\\.u16} %t | count 1
+; RUN: grep {vmovl\\.u32} %t | count 1
+
+define <8 x i16> @vmovls8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vmovls.v8i16(<8 x i8> %tmp1)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vmovls16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vmovls.v4i32(<4 x i16> %tmp1)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vmovls32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vmovls.v2i64(<2 x i32> %tmp1)
+	ret <2 x i64> %tmp2
+}
+
+define <8 x i16> @vmovlu8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vmovlu.v8i16(<8 x i8> %tmp1)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vmovlu16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vmovlu.v4i32(<4 x i16> %tmp1)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vmovlu32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vmovlu.v2i64(<2 x i32> %tmp1)
+	ret <2 x i64> %tmp2
+}
+
+declare <8 x i16> @llvm.arm.neon.vmovls.v8i16(<8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmovls.v4i32(<4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vmovls.v2i64(<2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vmovlu.v8i16(<8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmovlu.v4i32(<4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vmovlu.v2i64(<2 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vmovn.ll b/test/CodeGen/ARM/vmovn.ll
new file mode 100644
index 0000000000..d4d027c26c
--- /dev/null
+++ b/test/CodeGen/ARM/vmovn.ll
@@ -0,0 +1,26 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vmovn\\.i16} %t | count 1
+; RUN: grep {vmovn\\.i32} %t | count 1
+; RUN: grep {vmovn\\.i64} %t | count 1
+
+define <8 x i8> @vmovni16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vmovn.v8i8(<8 x i16> %tmp1)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vmovni32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vmovn.v4i16(<4 x i32> %tmp1)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vmovni64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vmovn.v2i32(<2 x i64> %tmp1)
+	ret <2 x i32> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vmovn.v8i8(<8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vmovn.v4i16(<4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vmovn.v2i32(<2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vmul.ll b/test/CodeGen/ARM/vmul.ll
new file mode 100644
index 0000000000..eb9ae7b95c
--- /dev/null
+++ b/test/CodeGen/ARM/vmul.ll
@@ -0,0 +1,79 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vmul\\.i8} %t | count 2
+; RUN: grep {vmul\\.i16} %t | count 2
+; RUN: grep {vmul\\.i32} %t | count 2
+; RUN: grep {vmul\\.f32} %t | count 2
+; RUN: grep {vmul\\.p8} %t | count 2
+
+define <8 x i8> @vmuli8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = mul <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vmuli16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = mul <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vmuli32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = mul <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <2 x float> @vmulf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = mul <2 x float> %tmp1, %tmp2
+	ret <2 x float> %tmp3
+}
+
+define <8 x i8> @vmulp8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vmulp.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <16 x i8> @vmulQi8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = mul <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vmulQi16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = mul <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vmulQi32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = mul <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <4 x float> @vmulQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = mul <4 x float> %tmp1, %tmp2
+	ret <4 x float> %tmp3
+}
+
+define <16 x i8> @vmulQp8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vmulp.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vmulp.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <16 x i8>  @llvm.arm.neon.vmulp.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
diff --git a/test/CodeGen/ARM/vmull.ll b/test/CodeGen/ARM/vmull.ll
new file mode 100644
index 0000000000..2fe6f37411
--- /dev/null
+++ b/test/CodeGen/ARM/vmull.ll
@@ -0,0 +1,67 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vmull\\.s8} %t | count 1
+; RUN: grep {vmull\\.s16} %t | count 1
+; RUN: grep {vmull\\.s32} %t | count 1
+; RUN: grep {vmull\\.u8} %t | count 1
+; RUN: grep {vmull\\.u16} %t | count 1
+; RUN: grep {vmull\\.u32} %t | count 1
+; RUN: grep {vmull\\.p8} %t | count 1
+
+define <8 x i16> @vmulls8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vmulls.v8i16(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vmulls16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vmulls.v4i32(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vmulls32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vmulls.v2i64(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <8 x i16> @vmullu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vmullu.v8i16(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vmullu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vmullu.v4i32(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vmullu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vmullu.v2i64(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <8 x i16> @vmullp8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vmullp.v8i16(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+declare <8 x i16> @llvm.arm.neon.vmulls.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmulls.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vmulls.v2i64(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vmullu.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vmullu.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vmullu.v2i64(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i16>  @llvm.arm.neon.vmullp.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
diff --git a/test/CodeGen/ARM/vmvn.ll b/test/CodeGen/ARM/vmvn.ll
new file mode 100644
index 0000000000..9bc135241e
--- /dev/null
+++ b/test/CodeGen/ARM/vmvn.ll
@@ -0,0 +1,51 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep vmvn %t | count 8
+; Note: function names do not include "vmvn" to allow simple grep for opcodes
+
+define <8 x i8> @v_mvni8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = xor <8 x i8> %tmp1, < i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1 >
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @v_mvni16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = xor <4 x i16> %tmp1, < i16 -1, i16 -1, i16 -1, i16 -1 >
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @v_mvni32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = xor <2 x i32> %tmp1, < i32 -1, i32 -1 >
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @v_mvni64(<1 x i64>* %A) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = xor <1 x i64> %tmp1, < i64 -1 >
+	ret <1 x i64> %tmp2
+}
+
+define <16 x i8> @v_mvnQi8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = xor <16 x i8> %tmp1, < i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1 >
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @v_mvnQi16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = xor <8 x i16> %tmp1, < i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1 >
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @v_mvnQi32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = xor <4 x i32> %tmp1, < i32 -1, i32 -1, i32 -1, i32 -1 >
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @v_mvnQi64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = xor <2 x i64> %tmp1, < i64 -1, i64 -1 >
+	ret <2 x i64> %tmp2
+}
diff --git a/test/CodeGen/ARM/vneg.ll b/test/CodeGen/ARM/vneg.ll
new file mode 100644
index 0000000000..9fa527f52f
--- /dev/null
+++ b/test/CodeGen/ARM/vneg.ll
@@ -0,0 +1,53 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vneg\\.s8} %t | count 2
+; RUN: grep {vneg\\.s16} %t | count 2
+; RUN: grep {vneg\\.s32} %t | count 2
+; RUN: grep {vneg\\.f32} %t | count 2
+
+define <8 x i8> @vnegs8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = sub <8 x i8> zeroinitializer, %tmp1
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vnegs16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = sub <4 x i16> zeroinitializer, %tmp1
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vnegs32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = sub <2 x i32> zeroinitializer, %tmp1
+	ret <2 x i32> %tmp2
+}
+
+define <2 x float> @vnegf32(<2 x float>* %A) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = sub <2 x float> < float -0.000000e+00, float -0.000000e+00 >, %tmp1
+	ret <2 x float> %tmp2
+}
+
+define <16 x i8> @vnegQs8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = sub <16 x i8> zeroinitializer, %tmp1
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vnegQs16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = sub <8 x i16> zeroinitializer, %tmp1
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vnegQs32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = sub <4 x i32> zeroinitializer, %tmp1
+	ret <4 x i32> %tmp2
+}
+
+define <4 x float> @vnegQf32(<4 x float>* %A) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = sub <4 x float> < float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00 >, %tmp1
+	ret <4 x float> %tmp2
+}
diff --git a/test/CodeGen/ARM/vorn.ll b/test/CodeGen/ARM/vorn.ll
new file mode 100644
index 0000000000..cee744ccb1
--- /dev/null
+++ b/test/CodeGen/ARM/vorn.ll
@@ -0,0 +1,67 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep vorn %t | count 8
+; Note: function names do not include "vorn" to allow simple grep for opcodes
+
+define <8 x i8> @v_orni8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = xor <8 x i8> %tmp2, < i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1 >
+	%tmp4 = or <8 x i8> %tmp1, %tmp3
+	ret <8 x i8> %tmp4
+}
+
+define <4 x i16> @v_orni16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = xor <4 x i16> %tmp2, < i16 -1, i16 -1, i16 -1, i16 -1 >
+	%tmp4 = or <4 x i16> %tmp1, %tmp3
+	ret <4 x i16> %tmp4
+}
+
+define <2 x i32> @v_orni32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = xor <2 x i32> %tmp2, < i32 -1, i32 -1 >
+	%tmp4 = or <2 x i32> %tmp1, %tmp3
+	ret <2 x i32> %tmp4
+}
+
+define <1 x i64> @v_orni64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = xor <1 x i64> %tmp2, < i64 -1 >
+	%tmp4 = or <1 x i64> %tmp1, %tmp3
+	ret <1 x i64> %tmp4
+}
+
+define <16 x i8> @v_ornQi8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = xor <16 x i8> %tmp2, < i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1 >
+	%tmp4 = or <16 x i8> %tmp1, %tmp3
+	ret <16 x i8> %tmp4
+}
+
+define <8 x i16> @v_ornQi16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = xor <8 x i16> %tmp2, < i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1 >
+	%tmp4 = or <8 x i16> %tmp1, %tmp3
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @v_ornQi32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = xor <4 x i32> %tmp2, < i32 -1, i32 -1, i32 -1, i32 -1 >
+	%tmp4 = or <4 x i32> %tmp1, %tmp3
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @v_ornQi64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = xor <2 x i64> %tmp2, < i64 -1, i64 -1 >
+	%tmp4 = or <2 x i64> %tmp1, %tmp3
+	ret <2 x i64> %tmp4
+}
diff --git a/test/CodeGen/ARM/vorr.ll b/test/CodeGen/ARM/vorr.ll
new file mode 100644
index 0000000000..39f4814c66
--- /dev/null
+++ b/test/CodeGen/ARM/vorr.ll
@@ -0,0 +1,59 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep vorr %t | count 8
+; Note: function names do not include "vorr" to allow simple grep for opcodes
+
+define <8 x i8> @v_orri8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = or <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @v_orri16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = or <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @v_orri32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = or <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @v_orri64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = or <1 x i64> %tmp1, %tmp2
+	ret <1 x i64> %tmp3
+}
+
+define <16 x i8> @v_orrQi8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = or <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @v_orrQi16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = or <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @v_orrQi32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = or <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @v_orrQi64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = or <2 x i64> %tmp1, %tmp2
+	ret <2 x i64> %tmp3
+}
diff --git a/test/CodeGen/ARM/vpadal.ll b/test/CodeGen/ARM/vpadal.ll
new file mode 100644
index 0000000000..c41c532988
--- /dev/null
+++ b/test/CodeGen/ARM/vpadal.ll
@@ -0,0 +1,107 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vpadal\\.s8} %t | count 2
+; RUN: grep {vpadal\\.s16} %t | count 2
+; RUN: grep {vpadal\\.s32} %t | count 2
+; RUN: grep {vpadal\\.u8} %t | count 2
+; RUN: grep {vpadal\\.u16} %t | count 2
+; RUN: grep {vpadal\\.u32} %t | count 2
+
+define <4 x i16> @vpadals8(<4 x i16>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vpadals.v4i16.v8i8(<4 x i16> %tmp1, <8 x i8> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vpadals16(<2 x i32>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vpadals.v2i32.v4i16(<2 x i32> %tmp1, <4 x i16> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vpadals32(<1 x i64>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vpadals.v1i64.v2i32(<1 x i64> %tmp1, <2 x i32> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <4 x i16> @vpadalu8(<4 x i16>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vpadalu.v4i16.v8i8(<4 x i16> %tmp1, <8 x i8> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vpadalu16(<2 x i32>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vpadalu.v2i32.v4i16(<2 x i32> %tmp1, <4 x i16> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vpadalu32(<1 x i64>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vpadalu.v1i64.v2i32(<1 x i64> %tmp1, <2 x i32> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <8 x i16> @vpadalQs8(<8 x i16>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vpadals.v8i16.v16i8(<8 x i16> %tmp1, <16 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vpadalQs16(<4 x i32>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vpadals.v4i32.v8i16(<4 x i32> %tmp1, <8 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vpadalQs32(<2 x i64>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vpadals.v2i64.v4i32(<2 x i64> %tmp1, <4 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <8 x i16> @vpadalQu8(<8 x i16>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vpadalu.v8i16.v16i8(<8 x i16> %tmp1, <16 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vpadalQu16(<4 x i32>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vpadalu.v4i32.v8i16(<4 x i32> %tmp1, <8 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vpadalQu32(<2 x i64>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vpadalu.v2i64.v4i32(<2 x i64> %tmp1, <4 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+declare <4 x i16> @llvm.arm.neon.vpadals.v4i16.v8i8(<4 x i16>, <8 x i8>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vpadals.v2i32.v4i16(<2 x i32>, <4 x i16>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vpadals.v1i64.v2i32(<1 x i64>, <2 x i32>) nounwind readnone
+
+declare <4 x i16> @llvm.arm.neon.vpadalu.v4i16.v8i8(<4 x i16>, <8 x i8>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vpadalu.v2i32.v4i16(<2 x i32>, <4 x i16>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vpadalu.v1i64.v2i32(<1 x i64>, <2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vpadals.v8i16.v16i8(<8 x i16>, <16 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vpadals.v4i32.v8i16(<4 x i32>, <8 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vpadals.v2i64.v4i32(<2 x i64>, <4 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vpadalu.v8i16.v16i8(<8 x i16>, <16 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vpadalu.v4i32.v8i16(<4 x i32>, <8 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vpadalu.v2i64.v4i32(<2 x i64>, <4 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vpadd.ll b/test/CodeGen/ARM/vpadd.ll
new file mode 100644
index 0000000000..baff49227e
--- /dev/null
+++ b/test/CodeGen/ARM/vpadd.ll
@@ -0,0 +1,39 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vpadd\\.i8} %t | count 1
+; RUN: grep {vpadd\\.i16} %t | count 1
+; RUN: grep {vpadd\\.i32} %t | count 1
+; RUN: grep {vpadd\\.f32} %t | count 1
+
+define <8 x i8> @vpaddi8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vpaddi.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vpaddi16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vpaddi.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vpaddi32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vpaddi.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <2 x float> @vpaddf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = call <2 x float> @llvm.arm.neon.vpaddf.v2f32(<2 x float> %tmp1, <2 x float> %tmp2)
+	ret <2 x float> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vpaddi.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vpaddi.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vpaddi.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <2 x float> @llvm.arm.neon.vpaddf.v2f32(<2 x float>, <2 x float>) nounwind readnone
diff --git a/test/CodeGen/ARM/vpaddl.ll b/test/CodeGen/ARM/vpaddl.ll
new file mode 100644
index 0000000000..babb49515a
--- /dev/null
+++ b/test/CodeGen/ARM/vpaddl.ll
@@ -0,0 +1,95 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vpaddl\\.s8} %t | count 2
+; RUN: grep {vpaddl\\.s16} %t | count 2
+; RUN: grep {vpaddl\\.s32} %t | count 2
+; RUN: grep {vpaddl\\.u8} %t | count 2
+; RUN: grep {vpaddl\\.u16} %t | count 2
+; RUN: grep {vpaddl\\.u32} %t | count 2
+
+define <4 x i16> @vpaddls8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vpaddls.v4i16.v8i8(<8 x i8> %tmp1)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vpaddls16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vpaddls.v2i32.v4i16(<4 x i16> %tmp1)
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @vpaddls32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <1 x i64> @llvm.arm.neon.vpaddls.v1i64.v2i32(<2 x i32> %tmp1)
+	ret <1 x i64> %tmp2
+}
+
+define <4 x i16> @vpaddlu8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vpaddlu.v4i16.v8i8(<8 x i8> %tmp1)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vpaddlu16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vpaddlu.v2i32.v4i16(<4 x i16> %tmp1)
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @vpaddlu32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <1 x i64> @llvm.arm.neon.vpaddlu.v1i64.v2i32(<2 x i32> %tmp1)
+	ret <1 x i64> %tmp2
+}
+
+define <8 x i16> @vpaddlQs8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vpaddls.v8i16.v16i8(<16 x i8> %tmp1)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vpaddlQs16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vpaddls.v4i32.v8i16(<8 x i16> %tmp1)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vpaddlQs32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vpaddls.v2i64.v4i32(<4 x i32> %tmp1)
+	ret <2 x i64> %tmp2
+}
+
+define <8 x i16> @vpaddlQu8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vpaddlu.v8i16.v16i8(<16 x i8> %tmp1)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vpaddlQu16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vpaddlu.v4i32.v8i16(<8 x i16> %tmp1)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vpaddlQu32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vpaddlu.v2i64.v4i32(<4 x i32> %tmp1)
+	ret <2 x i64> %tmp2
+}
+
+declare <4 x i16> @llvm.arm.neon.vpaddls.v4i16.v8i8(<8 x i8>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vpaddls.v2i32.v4i16(<4 x i16>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vpaddls.v1i64.v2i32(<2 x i32>) nounwind readnone
+
+declare <4 x i16> @llvm.arm.neon.vpaddlu.v4i16.v8i8(<8 x i8>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vpaddlu.v2i32.v4i16(<4 x i16>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vpaddlu.v1i64.v2i32(<2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vpaddls.v8i16.v16i8(<16 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vpaddls.v4i32.v8i16(<8 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vpaddls.v2i64.v4i32(<4 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vpaddlu.v8i16.v16i8(<16 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vpaddlu.v4i32.v8i16(<8 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vpaddlu.v2i64.v4i32(<4 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vpmax.ll b/test/CodeGen/ARM/vpmax.ll
new file mode 100644
index 0000000000..9878ca8c7b
--- /dev/null
+++ b/test/CodeGen/ARM/vpmax.ll
@@ -0,0 +1,67 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vpmax\\.s8} %t | count 1
+; RUN: grep {vpmax\\.s16} %t | count 1
+; RUN: grep {vpmax\\.s32} %t | count 1
+; RUN: grep {vpmax\\.u8} %t | count 1
+; RUN: grep {vpmax\\.u16} %t | count 1
+; RUN: grep {vpmax\\.u32} %t | count 1
+; RUN: grep {vpmax\\.f32} %t | count 1
+
+define <8 x i8> @vpmaxs8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vpmaxs.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vpmaxs16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vpmaxs.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vpmaxs32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vpmaxs.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <8 x i8> @vpmaxu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vpmaxu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vpmaxu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vpmaxu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vpmaxu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vpmaxu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <2 x float> @vpmaxf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = call <2 x float> @llvm.arm.neon.vpmaxf.v2f32(<2 x float> %tmp1, <2 x float> %tmp2)
+	ret <2 x float> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vpmaxs.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vpmaxs.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vpmaxs.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vpmaxu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vpmaxu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vpmaxu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <2 x float> @llvm.arm.neon.vpmaxf.v2f32(<2 x float>, <2 x float>) nounwind readnone
diff --git a/test/CodeGen/ARM/vpmin.ll b/test/CodeGen/ARM/vpmin.ll
new file mode 100644
index 0000000000..7b5348baa5
--- /dev/null
+++ b/test/CodeGen/ARM/vpmin.ll
@@ -0,0 +1,67 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vpmin\\.s8} %t | count 1
+; RUN: grep {vpmin\\.s16} %t | count 1
+; RUN: grep {vpmin\\.s32} %t | count 1
+; RUN: grep {vpmin\\.u8} %t | count 1
+; RUN: grep {vpmin\\.u16} %t | count 1
+; RUN: grep {vpmin\\.u32} %t | count 1
+; RUN: grep {vpmin\\.f32} %t | count 1
+
+define <8 x i8> @vpmins8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vpmins.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vpmins16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vpmins.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vpmins32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vpmins.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <8 x i8> @vpminu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vpminu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vpminu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vpminu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vpminu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vpminu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <2 x float> @vpminf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = call <2 x float> @llvm.arm.neon.vpminf.v2f32(<2 x float> %tmp1, <2 x float> %tmp2)
+	ret <2 x float> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vpmins.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vpmins.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vpmins.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vpminu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vpminu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vpminu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <2 x float> @llvm.arm.neon.vpminf.v2f32(<2 x float>, <2 x float>) nounwind readnone
diff --git a/test/CodeGen/ARM/vqabs.ll b/test/CodeGen/ARM/vqabs.ll
new file mode 100644
index 0000000000..04b8ad995b
--- /dev/null
+++ b/test/CodeGen/ARM/vqabs.ll
@@ -0,0 +1,48 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vqabs\\.s8} %t | count 2
+; RUN: grep {vqabs\\.s16} %t | count 2
+; RUN: grep {vqabs\\.s32} %t | count 2
+
+define <8 x i8> @vqabss8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqabs.v8i8(<8 x i8> %tmp1)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqabss16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqabs.v4i16(<4 x i16> %tmp1)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqabss32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqabs.v2i32(<2 x i32> %tmp1)
+	ret <2 x i32> %tmp2
+}
+
+define <16 x i8> @vqabsQs8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vqabs.v16i8(<16 x i8> %tmp1)
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vqabsQs16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vqabs.v8i16(<8 x i16> %tmp1)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vqabsQs32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vqabs.v4i32(<4 x i32> %tmp1)
+	ret <4 x i32> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vqabs.v8i8(<8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqabs.v4i16(<4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqabs.v2i32(<2 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vqabs.v16i8(<16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vqabs.v8i16(<8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vqabs.v4i32(<4 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vqadd.ll b/test/CodeGen/ARM/vqadd.ll
new file mode 100644
index 0000000000..c9e2359953
--- /dev/null
+++ b/test/CodeGen/ARM/vqadd.ll
@@ -0,0 +1,141 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vqadd\\.s8} %t | count 2
+; RUN: grep {vqadd\\.s16} %t | count 2
+; RUN: grep {vqadd\\.s32} %t | count 2
+; RUN: grep {vqadd\\.s64} %t | count 2
+; RUN: grep {vqadd\\.u8} %t | count 2
+; RUN: grep {vqadd\\.u16} %t | count 2
+; RUN: grep {vqadd\\.u32} %t | count 2
+; RUN: grep {vqadd\\.u64} %t | count 2
+
+define <8 x i8> @vqadds8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vqadds.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vqadds16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vqadds.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vqadds32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vqadds.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vqadds64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vqadds.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <8 x i8> @vqaddu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vqaddu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vqaddu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vqaddu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vqaddu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vqaddu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vqaddu64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vqaddu.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <16 x i8> @vqaddQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vqadds.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vqaddQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vqadds.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vqaddQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vqadds.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vqaddQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vqadds.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <16 x i8> @vqaddQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vqaddu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vqaddQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vqaddu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vqaddQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vqaddu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vqaddQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vqaddu.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vqadds.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqadds.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqadds.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vqadds.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vqaddu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqaddu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqaddu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vqaddu.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vqadds.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vqadds.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vqadds.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vqadds.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vqaddu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vqaddu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vqaddu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vqaddu.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vqdmlal.ll b/test/CodeGen/ARM/vqdmlal.ll
new file mode 100644
index 0000000000..f05bf530c4
--- /dev/null
+++ b/test/CodeGen/ARM/vqdmlal.ll
@@ -0,0 +1,22 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vqdmlal\\.s16} %t | count 1
+; RUN: grep {vqdmlal\\.s32} %t | count 1
+
+define <4 x i32> @vqdmlals16(<4 x i32>* %A, <4 x i16>* %B, <4 x i16>* %C) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = load <4 x i16>* %C
+	%tmp4 = call <4 x i32> @llvm.arm.neon.vqdmlal.v4i32(<4 x i32> %tmp1, <4 x i16> %tmp2, <4 x i16> %tmp3)
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @vqdmlals32(<2 x i64>* %A, <2 x i32>* %B, <2 x i32>* %C) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = load <2 x i32>* %C
+	%tmp4 = call <2 x i64> @llvm.arm.neon.vqdmlal.v2i64(<2 x i64> %tmp1, <2 x i32> %tmp2, <2 x i32> %tmp3)
+	ret <2 x i64> %tmp4
+}
+
+declare <4 x i32>  @llvm.arm.neon.vqdmlal.v4i32(<4 x i32>, <4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64>  @llvm.arm.neon.vqdmlal.v2i64(<2 x i64>, <2 x i32>, <2 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vqdmlsl.ll b/test/CodeGen/ARM/vqdmlsl.ll
new file mode 100644
index 0000000000..2b71ed5563
--- /dev/null
+++ b/test/CodeGen/ARM/vqdmlsl.ll
@@ -0,0 +1,22 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vqdmlsl\\.s16} %t | count 1
+; RUN: grep {vqdmlsl\\.s32} %t | count 1
+
+define <4 x i32> @vqdmlsls16(<4 x i32>* %A, <4 x i16>* %B, <4 x i16>* %C) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = load <4 x i16>* %C
+	%tmp4 = call <4 x i32> @llvm.arm.neon.vqdmlsl.v4i32(<4 x i32> %tmp1, <4 x i16> %tmp2, <4 x i16> %tmp3)
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @vqdmlsls32(<2 x i64>* %A, <2 x i32>* %B, <2 x i32>* %C) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = load <2 x i32>* %C
+	%tmp4 = call <2 x i64> @llvm.arm.neon.vqdmlsl.v2i64(<2 x i64> %tmp1, <2 x i32> %tmp2, <2 x i32> %tmp3)
+	ret <2 x i64> %tmp4
+}
+
+declare <4 x i32>  @llvm.arm.neon.vqdmlsl.v4i32(<4 x i32>, <4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64>  @llvm.arm.neon.vqdmlsl.v2i64(<2 x i64>, <2 x i32>, <2 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vqdmulh.ll b/test/CodeGen/ARM/vqdmulh.ll
new file mode 100644
index 0000000000..dd686ddd93
--- /dev/null
+++ b/test/CodeGen/ARM/vqdmulh.ll
@@ -0,0 +1,73 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vqdmulh\\.s16} %t | count 2
+; RUN: grep {vqdmulh\\.s32} %t | count 2
+; RUN: grep {vqrdmulh\\.s16} %t | count 2
+; RUN: grep {vqrdmulh\\.s32} %t | count 2
+
+define <4 x i16> @vqdmulhs16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vqdmulh.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vqdmulhs32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vqdmulh.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <8 x i16> @vqdmulhQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vqdmulh.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vqdmulhQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vqdmulh.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+declare <4 x i16> @llvm.arm.neon.vqdmulh.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqdmulh.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vqdmulh.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vqdmulh.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+
+define <4 x i16> @vqrdmulhs16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vqrdmulh.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vqrdmulhs32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vqrdmulh.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <8 x i16> @vqrdmulhQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vqrdmulh.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vqrdmulhQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vqrdmulh.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+declare <4 x i16> @llvm.arm.neon.vqrdmulh.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqrdmulh.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vqrdmulh.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vqrdmulh.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vqdmull.ll b/test/CodeGen/ARM/vqdmull.ll
new file mode 100644
index 0000000000..3ff90f795d
--- /dev/null
+++ b/test/CodeGen/ARM/vqdmull.ll
@@ -0,0 +1,20 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vqdmull\\.s16} %t | count 1
+; RUN: grep {vqdmull\\.s32} %t | count 1
+
+define <4 x i32> @vqdmulls16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vqdmull.v4i32(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vqdmulls32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vqdmull.v2i64(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+declare <4 x i32>  @llvm.arm.neon.vqdmull.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64>  @llvm.arm.neon.vqdmull.v2i64(<2 x i32>, <2 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vqmovn.ll b/test/CodeGen/ARM/vqmovn.ll
new file mode 100644
index 0000000000..78e6d23416
--- /dev/null
+++ b/test/CodeGen/ARM/vqmovn.ll
@@ -0,0 +1,76 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vqmovn\\.s16} %t | count 1
+; RUN: grep {vqmovn\\.s32} %t | count 1
+; RUN: grep {vqmovn\\.s64} %t | count 1
+; RUN: grep {vqmovn\\.u16} %t | count 1
+; RUN: grep {vqmovn\\.u32} %t | count 1
+; RUN: grep {vqmovn\\.u64} %t | count 1
+; RUN: grep {vqmovun\\.s16} %t | count 1
+; RUN: grep {vqmovun\\.s32} %t | count 1
+; RUN: grep {vqmovun\\.s64} %t | count 1
+
+define <8 x i8> @vqmovns16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqmovns.v8i8(<8 x i16> %tmp1)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqmovns32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqmovns.v4i16(<4 x i32> %tmp1)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqmovns64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqmovns.v2i32(<2 x i64> %tmp1)
+	ret <2 x i32> %tmp2
+}
+
+define <8 x i8> @vqmovnu16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqmovnu.v8i8(<8 x i16> %tmp1)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqmovnu32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqmovnu.v4i16(<4 x i32> %tmp1)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqmovnu64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqmovnu.v2i32(<2 x i64> %tmp1)
+	ret <2 x i32> %tmp2
+}
+
+define <8 x i8> @vqmovuns16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqmovnsu.v8i8(<8 x i16> %tmp1)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqmovuns32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqmovnsu.v4i16(<4 x i32> %tmp1)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqmovuns64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqmovnsu.v2i32(<2 x i64> %tmp1)
+	ret <2 x i32> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vqmovns.v8i8(<8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqmovns.v4i16(<4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqmovns.v2i32(<2 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vqmovnu.v8i8(<8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqmovnu.v4i16(<4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqmovnu.v2i32(<2 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vqmovnsu.v8i8(<8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqmovnsu.v4i16(<4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqmovnsu.v2i32(<2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vqneg.ll b/test/CodeGen/ARM/vqneg.ll
new file mode 100644
index 0000000000..72035f024f
--- /dev/null
+++ b/test/CodeGen/ARM/vqneg.ll
@@ -0,0 +1,48 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vqneg\\.s8} %t | count 2
+; RUN: grep {vqneg\\.s16} %t | count 2
+; RUN: grep {vqneg\\.s32} %t | count 2
+
+define <8 x i8> @vqnegs8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqneg.v8i8(<8 x i8> %tmp1)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqnegs16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqneg.v4i16(<4 x i16> %tmp1)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqnegs32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqneg.v2i32(<2 x i32> %tmp1)
+	ret <2 x i32> %tmp2
+}
+
+define <16 x i8> @vqnegQs8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vqneg.v16i8(<16 x i8> %tmp1)
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vqnegQs16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vqneg.v8i16(<8 x i16> %tmp1)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vqnegQs32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vqneg.v4i32(<4 x i32> %tmp1)
+	ret <4 x i32> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vqneg.v8i8(<8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqneg.v4i16(<4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqneg.v2i32(<2 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vqneg.v16i8(<16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vqneg.v8i16(<8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vqneg.v4i32(<4 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vqrshl.ll b/test/CodeGen/ARM/vqrshl.ll
new file mode 100644
index 0000000000..63e03c1245
--- /dev/null
+++ b/test/CodeGen/ARM/vqrshl.ll
@@ -0,0 +1,141 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vqrshl\\.s8} %t | count 2
+; RUN: grep {vqrshl\\.s16} %t | count 2
+; RUN: grep {vqrshl\\.s32} %t | count 2
+; RUN: grep {vqrshl\\.s64} %t | count 2
+; RUN: grep {vqrshl\\.u8} %t | count 2
+; RUN: grep {vqrshl\\.u16} %t | count 2
+; RUN: grep {vqrshl\\.u32} %t | count 2
+; RUN: grep {vqrshl\\.u64} %t | count 2
+
+define <8 x i8> @vqrshls8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vqrshifts.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vqrshls16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vqrshifts.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vqrshls32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vqrshifts.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vqrshls64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vqrshifts.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <8 x i8> @vqrshlu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vqrshiftu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vqrshlu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vqrshiftu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vqrshlu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vqrshiftu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vqrshlu64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vqrshiftu.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <16 x i8> @vqrshlQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vqrshifts.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vqrshlQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vqrshifts.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vqrshlQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vqrshifts.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vqrshlQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vqrshifts.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <16 x i8> @vqrshlQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vqrshiftu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vqrshlQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vqrshiftu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vqrshlQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vqrshiftu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vqrshlQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vqrshiftu.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vqrshifts.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqrshifts.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqrshifts.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vqrshifts.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vqrshiftu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqrshiftu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqrshiftu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vqrshiftu.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vqrshifts.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vqrshifts.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vqrshifts.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vqrshifts.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vqrshiftu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vqrshiftu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vqrshiftu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vqrshiftu.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vqrshrn.ll b/test/CodeGen/ARM/vqrshrn.ll
new file mode 100644
index 0000000000..38f98f2a2c
--- /dev/null
+++ b/test/CodeGen/ARM/vqrshrn.ll
@@ -0,0 +1,76 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vqrshrn\\.s16} %t | count 1
+; RUN: grep {vqrshrn\\.s32} %t | count 1
+; RUN: grep {vqrshrn\\.s64} %t | count 1
+; RUN: grep {vqrshrn\\.u16} %t | count 1
+; RUN: grep {vqrshrn\\.u32} %t | count 1
+; RUN: grep {vqrshrn\\.u64} %t | count 1
+; RUN: grep {vqrshrun\\.s16} %t | count 1
+; RUN: grep {vqrshrun\\.s32} %t | count 1
+; RUN: grep {vqrshrun\\.s64} %t | count 1
+
+define <8 x i8> @vqrshrns8(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqrshiftns.v8i8(<8 x i16> %tmp1, <8 x i16> < i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqrshrns16(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqrshiftns.v4i16(<4 x i32> %tmp1, <4 x i32> < i32 -16, i32 -16, i32 -16, i32 -16 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqrshrns32(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqrshiftns.v2i32(<2 x i64> %tmp1, <2 x i64> < i64 -32, i64 -32 >)
+	ret <2 x i32> %tmp2
+}
+
+define <8 x i8> @vqrshrnu8(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqrshiftnu.v8i8(<8 x i16> %tmp1, <8 x i16> < i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqrshrnu16(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqrshiftnu.v4i16(<4 x i32> %tmp1, <4 x i32> < i32 -16, i32 -16, i32 -16, i32 -16 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqrshrnu32(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqrshiftnu.v2i32(<2 x i64> %tmp1, <2 x i64> < i64 -32, i64 -32 >)
+	ret <2 x i32> %tmp2
+}
+
+define <8 x i8> @vqrshruns8(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqrshiftnsu.v8i8(<8 x i16> %tmp1, <8 x i16> < i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqrshruns16(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqrshiftnsu.v4i16(<4 x i32> %tmp1, <4 x i32> < i32 -16, i32 -16, i32 -16, i32 -16 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqrshruns32(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqrshiftnsu.v2i32(<2 x i64> %tmp1, <2 x i64> < i64 -32, i64 -32 >)
+	ret <2 x i32> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vqrshiftns.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqrshiftns.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqrshiftns.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vqrshiftnu.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqrshiftnu.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqrshiftnu.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vqrshiftnsu.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqrshiftnsu.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqrshiftnsu.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vqshl.ll b/test/CodeGen/ARM/vqshl.ll
new file mode 100644
index 0000000000..60b04bd583
--- /dev/null
+++ b/test/CodeGen/ARM/vqshl.ll
@@ -0,0 +1,307 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vqshl\\.s8} %t | count 4
+; RUN: grep {vqshl\\.s16} %t | count 4
+; RUN: grep {vqshl\\.s32} %t | count 4
+; RUN: grep {vqshl\\.s64} %t | count 4
+; RUN: grep {vqshl\\.u8} %t | count 4
+; RUN: grep {vqshl\\.u16} %t | count 4
+; RUN: grep {vqshl\\.u32} %t | count 4
+; RUN: grep {vqshl\\.u64} %t | count 4
+; RUN: grep {vqshl\\.s8.*#7} %t | count 2
+; RUN: grep {vqshl\\.s16.*#15} %t | count 2
+; RUN: grep {vqshl\\.s32.*#31} %t | count 2
+; RUN: grep {vqshl\\.s64.*#63} %t | count 2
+; RUN: grep {vqshl\\.u8.*#7} %t | count 2
+; RUN: grep {vqshl\\.u16.*#15} %t | count 2
+; RUN: grep {vqshl\\.u32.*#31} %t | count 2
+; RUN: grep {vqshl\\.u64.*#63} %t | count 2
+; RUN: grep {vqshlu\\.s8} %t | count 2
+; RUN: grep {vqshlu\\.s16} %t | count 2
+; RUN: grep {vqshlu\\.s32} %t | count 2
+; RUN: grep {vqshlu\\.s64} %t | count 2
+
+define <8 x i8> @vqshls8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vqshifts.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vqshls16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vqshifts.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vqshls32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vqshifts.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vqshls64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vqshifts.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <8 x i8> @vqshlu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vqshiftu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vqshlu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vqshiftu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vqshlu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vqshiftu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vqshlu64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vqshiftu.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <16 x i8> @vqshlQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vqshifts.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vqshlQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vqshifts.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vqshlQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vqshifts.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vqshlQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vqshifts.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <16 x i8> @vqshlQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vqshiftu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vqshlQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vqshiftu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vqshlQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vqshiftu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vqshlQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vqshiftu.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <8 x i8> @vqshls_n8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqshifts.v8i8(<8 x i8> %tmp1, <8 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqshls_n16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqshifts.v4i16(<4 x i16> %tmp1, <4 x i16> < i16 15, i16 15, i16 15, i16 15 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqshls_n32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqshifts.v2i32(<2 x i32> %tmp1, <2 x i32> < i32 31, i32 31 >)
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @vqshls_n64(<1 x i64>* %A) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = call <1 x i64> @llvm.arm.neon.vqshifts.v1i64(<1 x i64> %tmp1, <1 x i64> < i64 63 >)
+	ret <1 x i64> %tmp2
+}
+
+define <8 x i8> @vqshlu_n8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqshiftu.v8i8(<8 x i8> %tmp1, <8 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqshlu_n16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqshiftu.v4i16(<4 x i16> %tmp1, <4 x i16> < i16 15, i16 15, i16 15, i16 15 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqshlu_n32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqshiftu.v2i32(<2 x i32> %tmp1, <2 x i32> < i32 31, i32 31 >)
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @vqshlu_n64(<1 x i64>* %A) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = call <1 x i64> @llvm.arm.neon.vqshiftu.v1i64(<1 x i64> %tmp1, <1 x i64> < i64 63 >)
+	ret <1 x i64> %tmp2
+}
+
+define <8 x i8> @vqshlsu_n8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqshiftsu.v8i8(<8 x i8> %tmp1, <8 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqshlsu_n16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqshiftsu.v4i16(<4 x i16> %tmp1, <4 x i16> < i16 15, i16 15, i16 15, i16 15 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqshlsu_n32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqshiftsu.v2i32(<2 x i32> %tmp1, <2 x i32> < i32 31, i32 31 >)
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @vqshlsu_n64(<1 x i64>* %A) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = call <1 x i64> @llvm.arm.neon.vqshiftsu.v1i64(<1 x i64> %tmp1, <1 x i64> < i64 63 >)
+	ret <1 x i64> %tmp2
+}
+
+define <16 x i8> @vqshlQs_n8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vqshifts.v16i8(<16 x i8> %tmp1, <16 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >)
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vqshlQs_n16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vqshifts.v8i16(<8 x i16> %tmp1, <8 x i16> < i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15 >)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vqshlQs_n32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vqshifts.v4i32(<4 x i32> %tmp1, <4 x i32> < i32 31, i32 31, i32 31, i32 31 >)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vqshlQs_n64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vqshifts.v2i64(<2 x i64> %tmp1, <2 x i64> < i64 63, i64 63 >)
+	ret <2 x i64> %tmp2
+}
+
+define <16 x i8> @vqshlQu_n8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vqshiftu.v16i8(<16 x i8> %tmp1, <16 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >)
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vqshlQu_n16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vqshiftu.v8i16(<8 x i16> %tmp1, <8 x i16> < i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15 >)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vqshlQu_n32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vqshiftu.v4i32(<4 x i32> %tmp1, <4 x i32> < i32 31, i32 31, i32 31, i32 31 >)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vqshlQu_n64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vqshiftu.v2i64(<2 x i64> %tmp1, <2 x i64> < i64 63, i64 63 >)
+	ret <2 x i64> %tmp2
+}
+
+define <16 x i8> @vqshlQsu_n8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vqshiftsu.v16i8(<16 x i8> %tmp1, <16 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >)
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vqshlQsu_n16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vqshiftsu.v8i16(<8 x i16> %tmp1, <8 x i16> < i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15 >)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vqshlQsu_n32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vqshiftsu.v4i32(<4 x i32> %tmp1, <4 x i32> < i32 31, i32 31, i32 31, i32 31 >)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vqshlQsu_n64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vqshiftsu.v2i64(<2 x i64> %tmp1, <2 x i64> < i64 63, i64 63 >)
+	ret <2 x i64> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vqshifts.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqshifts.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqshifts.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vqshifts.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vqshiftu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqshiftu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqshiftu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vqshiftu.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vqshiftsu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqshiftsu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqshiftsu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vqshiftsu.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vqshifts.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vqshifts.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vqshifts.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vqshifts.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vqshiftu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vqshiftu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vqshiftu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vqshiftu.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vqshiftsu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vqshiftsu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vqshiftsu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vqshiftsu.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vqshrn.ll b/test/CodeGen/ARM/vqshrn.ll
new file mode 100644
index 0000000000..6bd607abb4
--- /dev/null
+++ b/test/CodeGen/ARM/vqshrn.ll
@@ -0,0 +1,76 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vqshrn\\.s16} %t | count 1
+; RUN: grep {vqshrn\\.s32} %t | count 1
+; RUN: grep {vqshrn\\.s64} %t | count 1
+; RUN: grep {vqshrn\\.u16} %t | count 1
+; RUN: grep {vqshrn\\.u32} %t | count 1
+; RUN: grep {vqshrn\\.u64} %t | count 1
+; RUN: grep {vqshrun\\.s16} %t | count 1
+; RUN: grep {vqshrun\\.s32} %t | count 1
+; RUN: grep {vqshrun\\.s64} %t | count 1
+
+define <8 x i8> @vqshrns8(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqshiftns.v8i8(<8 x i16> %tmp1, <8 x i16> < i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqshrns16(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqshiftns.v4i16(<4 x i32> %tmp1, <4 x i32> < i32 -16, i32 -16, i32 -16, i32 -16 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqshrns32(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqshiftns.v2i32(<2 x i64> %tmp1, <2 x i64> < i64 -32, i64 -32 >)
+	ret <2 x i32> %tmp2
+}
+
+define <8 x i8> @vqshrnu8(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqshiftnu.v8i8(<8 x i16> %tmp1, <8 x i16> < i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqshrnu16(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqshiftnu.v4i16(<4 x i32> %tmp1, <4 x i32> < i32 -16, i32 -16, i32 -16, i32 -16 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqshrnu32(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqshiftnu.v2i32(<2 x i64> %tmp1, <2 x i64> < i64 -32, i64 -32 >)
+	ret <2 x i32> %tmp2
+}
+
+define <8 x i8> @vqshruns8(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vqshiftnsu.v8i8(<8 x i16> %tmp1, <8 x i16> < i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vqshruns16(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vqshiftnsu.v4i16(<4 x i32> %tmp1, <4 x i32> < i32 -16, i32 -16, i32 -16, i32 -16 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vqshruns32(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vqshiftnsu.v2i32(<2 x i64> %tmp1, <2 x i64> < i64 -32, i64 -32 >)
+	ret <2 x i32> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vqshiftns.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqshiftns.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqshiftns.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vqshiftnu.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqshiftnu.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqshiftnu.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vqshiftnsu.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqshiftnsu.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqshiftnsu.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vqsub.ll b/test/CodeGen/ARM/vqsub.ll
new file mode 100644
index 0000000000..07052f78db
--- /dev/null
+++ b/test/CodeGen/ARM/vqsub.ll
@@ -0,0 +1,141 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vqsub\\.s8} %t | count 2
+; RUN: grep {vqsub\\.s16} %t | count 2
+; RUN: grep {vqsub\\.s32} %t | count 2
+; RUN: grep {vqsub\\.s64} %t | count 2
+; RUN: grep {vqsub\\.u8} %t | count 2
+; RUN: grep {vqsub\\.u16} %t | count 2
+; RUN: grep {vqsub\\.u32} %t | count 2
+; RUN: grep {vqsub\\.u64} %t | count 2
+
+define <8 x i8> @vqsubs8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vqsubs.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vqsubs16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vqsubs.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vqsubs32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vqsubs.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vqsubs64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vqsubs.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <8 x i8> @vqsubu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vqsubu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vqsubu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vqsubu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vqsubu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vqsubu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vqsubu64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vqsubu.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <16 x i8> @vqsubQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vqsubs.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vqsubQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vqsubs.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vqsubQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vqsubs.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vqsubQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vqsubs.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <16 x i8> @vqsubQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vqsubu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vqsubQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vqsubu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vqsubQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vqsubu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vqsubQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vqsubu.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vqsubs.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqsubs.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqsubs.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vqsubs.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vqsubu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vqsubu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vqsubu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vqsubu.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vqsubs.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vqsubs.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vqsubs.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vqsubs.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vqsubu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vqsubu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vqsubu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vqsubu.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vraddhn.ll b/test/CodeGen/ARM/vraddhn.ll
new file mode 100644
index 0000000000..d69e54554c
--- /dev/null
+++ b/test/CodeGen/ARM/vraddhn.ll
@@ -0,0 +1,29 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vraddhn\\.i16} %t | count 1
+; RUN: grep {vraddhn\\.i32} %t | count 1
+; RUN: grep {vraddhn\\.i64} %t | count 1
+
+define <8 x i8> @vraddhni16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vraddhn.v8i8(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vraddhni32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vraddhn.v4i16(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vraddhni64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vraddhn.v2i32(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vraddhn.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vraddhn.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vraddhn.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vrecpe.ll b/test/CodeGen/ARM/vrecpe.ll
new file mode 100644
index 0000000000..79cb595bc8
--- /dev/null
+++ b/test/CodeGen/ARM/vrecpe.ll
@@ -0,0 +1,33 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vrecpe\\.u32} %t | count 2
+; RUN: grep {vrecpe\\.f32} %t | count 2
+
+define <2 x i32> @vrecpei32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vrecpe.v2i32(<2 x i32> %tmp1)
+	ret <2 x i32> %tmp2
+}
+
+define <4 x i32> @vrecpeQi32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vrecpe.v4i32(<4 x i32> %tmp1)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x float> @vrecpef32(<2 x float>* %A) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = call <2 x float> @llvm.arm.neon.vrecpef.v2f32(<2 x float> %tmp1)
+	ret <2 x float> %tmp2
+}
+
+define <4 x float> @vrecpeQf32(<4 x float>* %A) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = call <4 x float> @llvm.arm.neon.vrecpef.v4f32(<4 x float> %tmp1)
+	ret <4 x float> %tmp2
+}
+
+declare <2 x i32> @llvm.arm.neon.vrecpe.v2i32(<2 x i32>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vrecpe.v4i32(<4 x i32>) nounwind readnone
+
+declare <2 x float> @llvm.arm.neon.vrecpef.v2f32(<2 x float>) nounwind readnone
+declare <4 x float> @llvm.arm.neon.vrecpef.v4f32(<4 x float>) nounwind readnone
diff --git a/test/CodeGen/ARM/vrecps.ll b/test/CodeGen/ARM/vrecps.ll
new file mode 100644
index 0000000000..7ded4157f0
--- /dev/null
+++ b/test/CodeGen/ARM/vrecps.ll
@@ -0,0 +1,19 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vrecps\\.f32} %t | count 2
+
+define <2 x float> @vrecpsf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = call <2 x float> @llvm.arm.neon.vrecps.v2f32(<2 x float> %tmp1, <2 x float> %tmp2)
+	ret <2 x float> %tmp3
+}
+
+define <4 x float> @vrecpsQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = call <4 x float> @llvm.arm.neon.vrecps.v4f32(<4 x float> %tmp1, <4 x float> %tmp2)
+	ret <4 x float> %tmp3
+}
+
+declare <2 x float> @llvm.arm.neon.vrecps.v2f32(<2 x float>, <2 x float>) nounwind readnone
+declare <4 x float> @llvm.arm.neon.vrecps.v4f32(<4 x float>, <4 x float>) nounwind readnone
diff --git a/test/CodeGen/ARM/vrhadd.ll b/test/CodeGen/ARM/vrhadd.ll
new file mode 100644
index 0000000000..4f6437caea
--- /dev/null
+++ b/test/CodeGen/ARM/vrhadd.ll
@@ -0,0 +1,107 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vrhadd\\.s8} %t | count 2
+; RUN: grep {vrhadd\\.s16} %t | count 2
+; RUN: grep {vrhadd\\.s32} %t | count 2
+; RUN: grep {vrhadd\\.u8} %t | count 2
+; RUN: grep {vrhadd\\.u16} %t | count 2
+; RUN: grep {vrhadd\\.u32} %t | count 2
+
+define <8 x i8> @vrhadds8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vrhadds.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vrhadds16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vrhadds.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vrhadds32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vrhadds.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <8 x i8> @vrhaddu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vrhaddu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vrhaddu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vrhaddu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vrhaddu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vrhaddu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <16 x i8> @vrhaddQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vrhadds.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vrhaddQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vrhadds.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vrhaddQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vrhadds.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <16 x i8> @vrhaddQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vrhaddu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vrhaddQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vrhaddu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vrhaddQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vrhaddu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vrhadds.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vrhadds.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vrhadds.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vrhaddu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vrhaddu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vrhaddu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vrhadds.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vrhadds.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vrhadds.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vrhaddu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vrhaddu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vrhaddu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vrshl.ll b/test/CodeGen/ARM/vrshl.ll
new file mode 100644
index 0000000000..fbbf5489c7
--- /dev/null
+++ b/test/CodeGen/ARM/vrshl.ll
@@ -0,0 +1,245 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vrshl\\.s8} %t | count 2
+; RUN: grep {vrshl\\.s16} %t | count 2
+; RUN: grep {vrshl\\.s32} %t | count 2
+; RUN: grep {vrshl\\.s64} %t | count 2
+; RUN: grep {vrshl\\.u8} %t | count 2
+; RUN: grep {vrshl\\.u16} %t | count 2
+; RUN: grep {vrshl\\.u32} %t | count 2
+; RUN: grep {vrshl\\.u64} %t | count 2
+; RUN: grep {vrshr\\.s8} %t | count 2
+; RUN: grep {vrshr\\.s16} %t | count 2
+; RUN: grep {vrshr\\.s32} %t | count 2
+; RUN: grep {vrshr\\.s64} %t | count 2
+; RUN: grep {vrshr\\.u8} %t | count 2
+; RUN: grep {vrshr\\.u16} %t | count 2
+; RUN: grep {vrshr\\.u32} %t | count 2
+; RUN: grep {vrshr\\.u64} %t | count 2
+
+define <8 x i8> @vrshls8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vrshifts.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vrshls16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vrshifts.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vrshls32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vrshifts.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vrshls64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vrshifts.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <8 x i8> @vrshlu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vrshiftu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vrshlu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vrshiftu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vrshlu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vrshiftu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vrshlu64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vrshiftu.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <16 x i8> @vrshlQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vrshifts.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vrshlQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vrshifts.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vrshlQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vrshifts.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vrshlQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vrshifts.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <16 x i8> @vrshlQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vrshiftu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vrshlQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vrshiftu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vrshlQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vrshiftu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vrshlQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vrshiftu.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <8 x i8> @vrshrs8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vrshifts.v8i8(<8 x i8> %tmp1, <8 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vrshrs16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vrshifts.v4i16(<4 x i16> %tmp1, <4 x i16> < i16 -16, i16 -16, i16 -16, i16 -16 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vrshrs32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vrshifts.v2i32(<2 x i32> %tmp1, <2 x i32> < i32 -32, i32 -32 >)
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @vrshrs64(<1 x i64>* %A) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = call <1 x i64> @llvm.arm.neon.vrshifts.v1i64(<1 x i64> %tmp1, <1 x i64> < i64 -64 >)
+	ret <1 x i64> %tmp2
+}
+
+define <8 x i8> @vrshru8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vrshiftu.v8i8(<8 x i8> %tmp1, <8 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vrshru16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vrshiftu.v4i16(<4 x i16> %tmp1, <4 x i16> < i16 -16, i16 -16, i16 -16, i16 -16 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vrshru32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vrshiftu.v2i32(<2 x i32> %tmp1, <2 x i32> < i32 -32, i32 -32 >)
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @vrshru64(<1 x i64>* %A) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = call <1 x i64> @llvm.arm.neon.vrshiftu.v1i64(<1 x i64> %tmp1, <1 x i64> < i64 -64 >)
+	ret <1 x i64> %tmp2
+}
+
+define <16 x i8> @vrshrQs8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vrshifts.v16i8(<16 x i8> %tmp1, <16 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vrshrQs16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vrshifts.v8i16(<8 x i16> %tmp1, <8 x i16> < i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16 >)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vrshrQs32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vrshifts.v4i32(<4 x i32> %tmp1, <4 x i32> < i32 -32, i32 -32, i32 -32, i32 -32 >)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vrshrQs64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vrshifts.v2i64(<2 x i64> %tmp1, <2 x i64> < i64 -64, i64 -64 >)
+	ret <2 x i64> %tmp2
+}
+
+define <16 x i8> @vrshrQu8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vrshiftu.v16i8(<16 x i8> %tmp1, <16 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vrshrQu16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vrshiftu.v8i16(<8 x i16> %tmp1, <8 x i16> < i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16 >)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vrshrQu32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vrshiftu.v4i32(<4 x i32> %tmp1, <4 x i32> < i32 -32, i32 -32, i32 -32, i32 -32 >)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vrshrQu64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vrshiftu.v2i64(<2 x i64> %tmp1, <2 x i64> < i64 -64, i64 -64 >)
+	ret <2 x i64> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vrshifts.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vrshifts.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vrshifts.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vrshifts.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vrshiftu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vrshiftu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vrshiftu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vrshiftu.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vrshifts.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vrshifts.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vrshifts.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vrshifts.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vrshiftu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vrshiftu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vrshiftu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vrshiftu.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vrshrn.ll b/test/CodeGen/ARM/vrshrn.ll
new file mode 100644
index 0000000000..8af24fd4d6
--- /dev/null
+++ b/test/CodeGen/ARM/vrshrn.ll
@@ -0,0 +1,26 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vrshrn\\.i16} %t | count 1
+; RUN: grep {vrshrn\\.i32} %t | count 1
+; RUN: grep {vrshrn\\.i64} %t | count 1
+
+define <8 x i8> @vrshrns8(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vrshiftn.v8i8(<8 x i16> %tmp1, <8 x i16> < i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vrshrns16(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vrshiftn.v4i16(<4 x i32> %tmp1, <4 x i32> < i32 -16, i32 -16, i32 -16, i32 -16 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vrshrns32(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vrshiftn.v2i32(<2 x i64> %tmp1, <2 x i64> < i64 -32, i64 -32 >)
+	ret <2 x i32> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vrshiftn.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vrshiftn.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vrshiftn.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vrsqrte.ll b/test/CodeGen/ARM/vrsqrte.ll
new file mode 100644
index 0000000000..10529f61b5
--- /dev/null
+++ b/test/CodeGen/ARM/vrsqrte.ll
@@ -0,0 +1,33 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vrsqrte\\.u32} %t | count 2
+; RUN: grep {vrsqrte\\.f32} %t | count 2
+
+define <2 x i32> @vrsqrtei32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vrsqrte.v2i32(<2 x i32> %tmp1)
+	ret <2 x i32> %tmp2
+}
+
+define <4 x i32> @vrsqrteQi32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vrsqrte.v4i32(<4 x i32> %tmp1)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x float> @vrsqrtef32(<2 x float>* %A) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = call <2 x float> @llvm.arm.neon.vrsqrtef.v2f32(<2 x float> %tmp1)
+	ret <2 x float> %tmp2
+}
+
+define <4 x float> @vrsqrteQf32(<4 x float>* %A) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = call <4 x float> @llvm.arm.neon.vrsqrtef.v4f32(<4 x float> %tmp1)
+	ret <4 x float> %tmp2
+}
+
+declare <2 x i32> @llvm.arm.neon.vrsqrte.v2i32(<2 x i32>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vrsqrte.v4i32(<4 x i32>) nounwind readnone
+
+declare <2 x float> @llvm.arm.neon.vrsqrtef.v2f32(<2 x float>) nounwind readnone
+declare <4 x float> @llvm.arm.neon.vrsqrtef.v4f32(<4 x float>) nounwind readnone
diff --git a/test/CodeGen/ARM/vrsqrts.ll b/test/CodeGen/ARM/vrsqrts.ll
new file mode 100644
index 0000000000..3fd0567842
--- /dev/null
+++ b/test/CodeGen/ARM/vrsqrts.ll
@@ -0,0 +1,19 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vrsqrts\\.f32} %t | count 2
+
+define <2 x float> @vrsqrtsf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = call <2 x float> @llvm.arm.neon.vrsqrts.v2f32(<2 x float> %tmp1, <2 x float> %tmp2)
+	ret <2 x float> %tmp3
+}
+
+define <4 x float> @vrsqrtsQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = call <4 x float> @llvm.arm.neon.vrsqrts.v4f32(<4 x float> %tmp1, <4 x float> %tmp2)
+	ret <4 x float> %tmp3
+}
+
+declare <2 x float> @llvm.arm.neon.vrsqrts.v2f32(<2 x float>, <2 x float>) nounwind readnone
+declare <4 x float> @llvm.arm.neon.vrsqrts.v4f32(<4 x float>, <4 x float>) nounwind readnone
diff --git a/test/CodeGen/ARM/vrsubhn.ll b/test/CodeGen/ARM/vrsubhn.ll
new file mode 100644
index 0000000000..587092397c
--- /dev/null
+++ b/test/CodeGen/ARM/vrsubhn.ll
@@ -0,0 +1,29 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vrsubhn\\.i16} %t | count 1
+; RUN: grep {vrsubhn\\.i32} %t | count 1
+; RUN: grep {vrsubhn\\.i64} %t | count 1
+
+define <8 x i8> @vrsubhni16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vrsubhn.v8i8(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vrsubhni32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vrsubhn.v4i16(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vrsubhni64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vrsubhn.v2i32(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vrsubhn.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vrsubhn.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vrsubhn.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vset_lane.ll b/test/CodeGen/ARM/vset_lane.ll
new file mode 100644
index 0000000000..279c628aed
--- /dev/null
+++ b/test/CodeGen/ARM/vset_lane.ll
@@ -0,0 +1,40 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vmov\\.8} %t | count 2
+; RUN: grep {vmov\\.16} %t | count 2
+; RUN: grep {vmov\\.32} %t | count 2
+
+define <8 x i8> @vset_lane8(<8 x i8>* %A, i8 %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = insertelement <8 x i8> %tmp1, i8 %B, i32 1
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vset_lane16(<4 x i16>* %A, i16 %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = insertelement <4 x i16> %tmp1, i16 %B, i32 1
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vset_lane32(<2 x i32>* %A, i32 %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = insertelement <2 x i32> %tmp1, i32 %B, i32 1
+	ret <2 x i32> %tmp2
+}
+
+define <16 x i8> @vsetQ_lane8(<16 x i8>* %A, i8 %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = insertelement <16 x i8> %tmp1, i8 %B, i32 1
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vsetQ_lane16(<8 x i16>* %A, i16 %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = insertelement <8 x i16> %tmp1, i16 %B, i32 1
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vsetQ_lane32(<4 x i32>* %A, i32 %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = insertelement <4 x i32> %tmp1, i32 %B, i32 1
+	ret <4 x i32> %tmp2
+}
diff --git a/test/CodeGen/ARM/vshift.ll b/test/CodeGen/ARM/vshift.ll
new file mode 100644
index 0000000000..8c5c4aad18
--- /dev/null
+++ b/test/CodeGen/ARM/vshift.ll
@@ -0,0 +1,337 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vshl\\.s8} %t | count 2
+; RUN: grep {vshl\\.s16} %t | count 2
+; RUN: grep {vshl\\.s32} %t | count 2
+; RUN: grep {vshl\\.s64} %t | count 2
+; RUN: grep {vshl\\.u8} %t | count 4
+; RUN: grep {vshl\\.u16} %t | count 4
+; RUN: grep {vshl\\.u32} %t | count 4
+; RUN: grep {vshl\\.u64} %t | count 4
+; RUN: grep {vshl\\.i8} %t | count 2
+; RUN: grep {vshl\\.i16} %t | count 2
+; RUN: grep {vshl\\.i32} %t | count 2
+; RUN: grep {vshl\\.i64} %t | count 2
+; RUN: grep {vshr\\.u8} %t | count 2
+; RUN: grep {vshr\\.u16} %t | count 2
+; RUN: grep {vshr\\.u32} %t | count 2
+; RUN: grep {vshr\\.u64} %t | count 2
+; RUN: grep {vshr\\.s8} %t | count 2
+; RUN: grep {vshr\\.s16} %t | count 2
+; RUN: grep {vshr\\.s32} %t | count 2
+; RUN: grep {vshr\\.s64} %t | count 2
+; RUN: grep {vneg\\.s8} %t | count 4
+; RUN: grep {vneg\\.s16} %t | count 4
+; RUN: grep {vneg\\.s32} %t | count 4
+; RUN: grep {vsub\\.i64} %t | count 4
+
+define <8 x i8> @vshls8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = shl <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vshls16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = shl <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vshls32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = shl <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vshls64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = shl <1 x i64> %tmp1, %tmp2
+	ret <1 x i64> %tmp3
+}
+
+define <8 x i8> @vshli8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = shl <8 x i8> %tmp1, < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vshli16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = shl <4 x i16> %tmp1, < i16 15, i16 15, i16 15, i16 15 >
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vshli32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = shl <2 x i32> %tmp1, < i32 31, i32 31 >
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @vshli64(<1 x i64>* %A) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = shl <1 x i64> %tmp1, < i64 63 >
+	ret <1 x i64> %tmp2
+}
+
+define <16 x i8> @vshlQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = shl <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vshlQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = shl <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vshlQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = shl <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vshlQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = shl <2 x i64> %tmp1, %tmp2
+	ret <2 x i64> %tmp3
+}
+
+define <16 x i8> @vshlQi8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = shl <16 x i8> %tmp1, < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vshlQi16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = shl <8 x i16> %tmp1, < i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15 >
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vshlQi32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = shl <4 x i32> %tmp1, < i32 31, i32 31, i32 31, i32 31 >
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vshlQi64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = shl <2 x i64> %tmp1, < i64 63, i64 63 >
+	ret <2 x i64> %tmp2
+}
+
+define <8 x i8> @vlshru8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = lshr <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vlshru16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = lshr <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vlshru32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = lshr <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vlshru64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = lshr <1 x i64> %tmp1, %tmp2
+	ret <1 x i64> %tmp3
+}
+
+define <8 x i8> @vlshri8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = lshr <8 x i8> %tmp1, < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vlshri16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = lshr <4 x i16> %tmp1, < i16 16, i16 16, i16 16, i16 16 >
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vlshri32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = lshr <2 x i32> %tmp1, < i32 32, i32 32 >
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @vlshri64(<1 x i64>* %A) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = lshr <1 x i64> %tmp1, < i64 64 >
+	ret <1 x i64> %tmp2
+}
+
+define <16 x i8> @vlshrQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = lshr <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vlshrQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = lshr <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vlshrQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = lshr <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vlshrQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = lshr <2 x i64> %tmp1, %tmp2
+	ret <2 x i64> %tmp3
+}
+
+define <16 x i8> @vlshrQi8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = lshr <16 x i8> %tmp1, < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vlshrQi16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = lshr <8 x i16> %tmp1, < i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16 >
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vlshrQi32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = lshr <4 x i32> %tmp1, < i32 32, i32 32, i32 32, i32 32 >
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vlshrQi64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = lshr <2 x i64> %tmp1, < i64 64, i64 64 >
+	ret <2 x i64> %tmp2
+}
+
+define <8 x i8> @vashrs8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = ashr <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vashrs16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = ashr <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vashrs32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = ashr <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vashrs64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = ashr <1 x i64> %tmp1, %tmp2
+	ret <1 x i64> %tmp3
+}
+
+define <8 x i8> @vashri8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = ashr <8 x i8> %tmp1, < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vashri16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = ashr <4 x i16> %tmp1, < i16 16, i16 16, i16 16, i16 16 >
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vashri32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = ashr <2 x i32> %tmp1, < i32 32, i32 32 >
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @vashri64(<1 x i64>* %A) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = ashr <1 x i64> %tmp1, < i64 64 >
+	ret <1 x i64> %tmp2
+}
+
+define <16 x i8> @vashrQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = ashr <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vashrQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = ashr <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vashrQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = ashr <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vashrQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = ashr <2 x i64> %tmp1, %tmp2
+	ret <2 x i64> %tmp3
+}
+
+define <16 x i8> @vashrQi8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = ashr <16 x i8> %tmp1, < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vashrQi16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = ashr <8 x i16> %tmp1, < i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16 >
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vashrQi32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = ashr <4 x i32> %tmp1, < i32 32, i32 32, i32 32, i32 32 >
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vashrQi64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = ashr <2 x i64> %tmp1, < i64 64, i64 64 >
+	ret <2 x i64> %tmp2
+}
diff --git a/test/CodeGen/ARM/vshiftins.ll b/test/CodeGen/ARM/vshiftins.ll
new file mode 100644
index 0000000000..cb7cbb89ec
--- /dev/null
+++ b/test/CodeGen/ARM/vshiftins.ll
@@ -0,0 +1,131 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vsli\\.8} %t | count 2
+; RUN: grep {vsli\\.16} %t | count 2
+; RUN: grep {vsli\\.32} %t | count 2
+; RUN: grep {vsli\\.64} %t | count 2
+; RUN: grep {vsri\\.8} %t | count 2
+; RUN: grep {vsri\\.16} %t | count 2
+; RUN: grep {vsri\\.32} %t | count 2
+; RUN: grep {vsri\\.64} %t | count 2
+
+define <8 x i8> @vsli8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vshiftins.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2, <8 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vsli16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vshiftins.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2, <4 x i16> < i16 15, i16 15, i16 15, i16 15 >)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vsli32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vshiftins.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2, <2 x i32> < i32 31, i32 31 >)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vsli64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vshiftins.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2, <1 x i64> < i64 63 >)
+	ret <1 x i64> %tmp3
+}
+
+define <16 x i8> @vsliQ8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vshiftins.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2, <16 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vsliQ16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vshiftins.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2, <8 x i16> < i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15 >)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vsliQ32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vshiftins.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2, <4 x i32> < i32 31, i32 31, i32 31, i32 31 >)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vsliQ64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vshiftins.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2, <2 x i64> < i64 63, i64 63 >)
+	ret <2 x i64> %tmp3
+}
+
+define <8 x i8> @vsri8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vshiftins.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2, <8 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vsri16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vshiftins.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2, <4 x i16> < i16 -16, i16 -16, i16 -16, i16 -16 >)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vsri32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vshiftins.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2, <2 x i32> < i32 -32, i32 -32 >)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vsri64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vshiftins.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2, <1 x i64> < i64 -64 >)
+	ret <1 x i64> %tmp3
+}
+
+define <16 x i8> @vsriQ8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vshiftins.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2, <16 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vsriQ16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vshiftins.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2, <8 x i16> < i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16 >)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vsriQ32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vshiftins.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2, <4 x i32> < i32 -32, i32 -32, i32 -32, i32 -32 >)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vsriQ64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vshiftins.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2, <2 x i64> < i64 -64, i64 -64 >)
+	ret <2 x i64> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vshiftins.v8i8(<8 x i8>, <8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vshiftins.v4i16(<4 x i16>, <4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vshiftins.v2i32(<2 x i32>, <2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vshiftins.v1i64(<1 x i64>, <1 x i64>, <1 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vshiftins.v16i8(<16 x i8>, <16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vshiftins.v8i16(<8 x i16>, <8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vshiftins.v4i32(<4 x i32>, <4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vshiftins.v2i64(<2 x i64>, <2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vshl.ll b/test/CodeGen/ARM/vshl.ll
new file mode 100644
index 0000000000..993126ea57
--- /dev/null
+++ b/test/CodeGen/ARM/vshl.ll
@@ -0,0 +1,302 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vshl\\.s8} %t | count 2
+; RUN: grep {vshl\\.s16} %t | count 2
+; RUN: grep {vshl\\.s32} %t | count 2
+; RUN: grep {vshl\\.s64} %t | count 2
+; RUN: grep {vshl\\.u8} %t | count 2
+; RUN: grep {vshl\\.u16} %t | count 2
+; RUN: grep {vshl\\.u32} %t | count 2
+; RUN: grep {vshl\\.u64} %t | count 2
+; RUN: grep {vshl\\.i8} %t | count 2
+; RUN: grep {vshl\\.i16} %t | count 2
+; RUN: grep {vshl\\.i32} %t | count 2
+; RUN: grep {vshl\\.i64} %t | count 2
+; RUN: grep {vshr\\.s8} %t | count 2
+; RUN: grep {vshr\\.s16} %t | count 2
+; RUN: grep {vshr\\.s32} %t | count 2
+; RUN: grep {vshr\\.s64} %t | count 2
+; RUN: grep {vshr\\.u8} %t | count 2
+; RUN: grep {vshr\\.u16} %t | count 2
+; RUN: grep {vshr\\.u32} %t | count 2
+; RUN: grep {vshr\\.u64} %t | count 2
+
+define <8 x i8> @vshls8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vshifts.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vshls16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vshifts.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vshls32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vshifts.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vshls64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vshifts.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <8 x i8> @vshlu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vshiftu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vshlu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vshiftu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vshlu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vshiftu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vshlu64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vshiftu.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2)
+	ret <1 x i64> %tmp3
+}
+
+define <16 x i8> @vshlQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vshifts.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vshlQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vshifts.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vshlQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vshifts.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vshlQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vshifts.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <16 x i8> @vshlQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vshiftu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2)
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vshlQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vshiftu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vshlQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vshiftu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vshlQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vshiftu.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+; For left shifts by immediates, the signedness is irrelevant.
+; Test a mix of both signed and unsigned intrinsics.
+
+define <8 x i8> @vshli8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vshifts.v8i8(<8 x i8> %tmp1, <8 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vshli16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vshiftu.v4i16(<4 x i16> %tmp1, <4 x i16> < i16 15, i16 15, i16 15, i16 15 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vshli32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vshifts.v2i32(<2 x i32> %tmp1, <2 x i32> < i32 31, i32 31 >)
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @vshli64(<1 x i64>* %A) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = call <1 x i64> @llvm.arm.neon.vshiftu.v1i64(<1 x i64> %tmp1, <1 x i64> < i64 63 >)
+	ret <1 x i64> %tmp2
+}
+
+define <16 x i8> @vshlQi8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vshifts.v16i8(<16 x i8> %tmp1, <16 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >)
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vshlQi16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vshiftu.v8i16(<8 x i16> %tmp1, <8 x i16> < i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15 >)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vshlQi32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vshifts.v4i32(<4 x i32> %tmp1, <4 x i32> < i32 31, i32 31, i32 31, i32 31 >)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vshlQi64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vshiftu.v2i64(<2 x i64> %tmp1, <2 x i64> < i64 63, i64 63 >)
+	ret <2 x i64> %tmp2
+}
+
+; Right shift by immediate:
+
+define <8 x i8> @vshrs8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vshifts.v8i8(<8 x i8> %tmp1, <8 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vshrs16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vshifts.v4i16(<4 x i16> %tmp1, <4 x i16> < i16 -16, i16 -16, i16 -16, i16 -16 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vshrs32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vshifts.v2i32(<2 x i32> %tmp1, <2 x i32> < i32 -32, i32 -32 >)
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @vshrs64(<1 x i64>* %A) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = call <1 x i64> @llvm.arm.neon.vshifts.v1i64(<1 x i64> %tmp1, <1 x i64> < i64 -64 >)
+	ret <1 x i64> %tmp2
+}
+
+define <8 x i8> @vshru8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vshiftu.v8i8(<8 x i8> %tmp1, <8 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vshru16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vshiftu.v4i16(<4 x i16> %tmp1, <4 x i16> < i16 -16, i16 -16, i16 -16, i16 -16 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vshru32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vshiftu.v2i32(<2 x i32> %tmp1, <2 x i32> < i32 -32, i32 -32 >)
+	ret <2 x i32> %tmp2
+}
+
+define <1 x i64> @vshru64(<1 x i64>* %A) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = call <1 x i64> @llvm.arm.neon.vshiftu.v1i64(<1 x i64> %tmp1, <1 x i64> < i64 -64 >)
+	ret <1 x i64> %tmp2
+}
+
+define <16 x i8> @vshrQs8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vshifts.v16i8(<16 x i8> %tmp1, <16 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vshrQs16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vshifts.v8i16(<8 x i16> %tmp1, <8 x i16> < i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16 >)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vshrQs32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vshifts.v4i32(<4 x i32> %tmp1, <4 x i32> < i32 -32, i32 -32, i32 -32, i32 -32 >)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vshrQs64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vshifts.v2i64(<2 x i64> %tmp1, <2 x i64> < i64 -64, i64 -64 >)
+	ret <2 x i64> %tmp2
+}
+
+define <16 x i8> @vshrQu8(<16 x i8>* %A) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = call <16 x i8> @llvm.arm.neon.vshiftu.v16i8(<16 x i8> %tmp1, <16 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+	ret <16 x i8> %tmp2
+}
+
+define <8 x i16> @vshrQu16(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vshiftu.v8i16(<8 x i16> %tmp1, <8 x i16> < i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16 >)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vshrQu32(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vshiftu.v4i32(<4 x i32> %tmp1, <4 x i32> < i32 -32, i32 -32, i32 -32, i32 -32 >)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vshrQu64(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vshiftu.v2i64(<2 x i64> %tmp1, <2 x i64> < i64 -64, i64 -64 >)
+	ret <2 x i64> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vshifts.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vshifts.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vshifts.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vshifts.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vshiftu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vshiftu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vshiftu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vshiftu.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vshifts.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vshifts.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vshifts.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vshifts.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vshiftu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vshiftu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vshiftu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vshiftu.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vshll.ll b/test/CodeGen/ARM/vshll.ll
new file mode 100644
index 0000000000..f81c09a7b9
--- /dev/null
+++ b/test/CodeGen/ARM/vshll.ll
@@ -0,0 +1,74 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vshll\\.s8} %t | count 1
+; RUN: grep {vshll\\.s16} %t | count 1
+; RUN: grep {vshll\\.s32} %t | count 1
+; RUN: grep {vshll\\.u8} %t | count 1
+; RUN: grep {vshll\\.u16} %t | count 1
+; RUN: grep {vshll\\.u32} %t | count 1
+; RUN: grep {vshll\\.i8} %t | count 1
+; RUN: grep {vshll\\.i16} %t | count 1
+; RUN: grep {vshll\\.i32} %t | count 1
+
+define <8 x i16> @vshlls8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vshiftls.v8i16(<8 x i8> %tmp1, <8 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vshlls16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vshiftls.v4i32(<4 x i16> %tmp1, <4 x i16> < i16 15, i16 15, i16 15, i16 15 >)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vshlls32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vshiftls.v2i64(<2 x i32> %tmp1, <2 x i32> < i32 31, i32 31 >)
+	ret <2 x i64> %tmp2
+}
+
+define <8 x i16> @vshllu8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vshiftlu.v8i16(<8 x i8> %tmp1, <8 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vshllu16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vshiftlu.v4i32(<4 x i16> %tmp1, <4 x i16> < i16 15, i16 15, i16 15, i16 15 >)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vshllu32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vshiftlu.v2i64(<2 x i32> %tmp1, <2 x i32> < i32 31, i32 31 >)
+	ret <2 x i64> %tmp2
+}
+
+; The following tests use the maximum shift count, so the signedness is
+; irrelevant.  Test both signed and unsigned versions.
+define <8 x i16> @vshlli8(<8 x i8>* %A) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = call <8 x i16> @llvm.arm.neon.vshiftls.v8i16(<8 x i8> %tmp1, <8 x i8> < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >)
+	ret <8 x i16> %tmp2
+}
+
+define <4 x i32> @vshlli16(<4 x i16>* %A) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = call <4 x i32> @llvm.arm.neon.vshiftlu.v4i32(<4 x i16> %tmp1, <4 x i16> < i16 16, i16 16, i16 16, i16 16 >)
+	ret <4 x i32> %tmp2
+}
+
+define <2 x i64> @vshlli32(<2 x i32>* %A) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = call <2 x i64> @llvm.arm.neon.vshiftls.v2i64(<2 x i32> %tmp1, <2 x i32> < i32 32, i32 32 >)
+	ret <2 x i64> %tmp2
+}
+
+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
+
+declare <8 x i16> @llvm.arm.neon.vshiftlu.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vshiftlu.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vshiftlu.v2i64(<2 x i32>, <2 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vshrn.ll b/test/CodeGen/ARM/vshrn.ll
new file mode 100644
index 0000000000..bc640cbbca
--- /dev/null
+++ b/test/CodeGen/ARM/vshrn.ll
@@ -0,0 +1,26 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vshrn\\.i16} %t | count 1
+; RUN: grep {vshrn\\.i32} %t | count 1
+; RUN: grep {vshrn\\.i64} %t | count 1
+
+define <8 x i8> @vshrns8(<8 x i16>* %A) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = call <8 x i8> @llvm.arm.neon.vshiftn.v8i8(<8 x i16> %tmp1, <8 x i16> < i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8, i16 -8 >)
+	ret <8 x i8> %tmp2
+}
+
+define <4 x i16> @vshrns16(<4 x i32>* %A) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = call <4 x i16> @llvm.arm.neon.vshiftn.v4i16(<4 x i32> %tmp1, <4 x i32> < i32 -16, i32 -16, i32 -16, i32 -16 >)
+	ret <4 x i16> %tmp2
+}
+
+define <2 x i32> @vshrns32(<2 x i64>* %A) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = call <2 x i32> @llvm.arm.neon.vshiftn.v2i32(<2 x i64> %tmp1, <2 x i64> < i64 -32, i64 -32 >)
+	ret <2 x i32> %tmp2
+}
+
+declare <8 x i8>  @llvm.arm.neon.vshiftn.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vshiftn.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vshiftn.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vsra.ll b/test/CodeGen/ARM/vsra.ll
new file mode 100644
index 0000000000..e2829dcdda
--- /dev/null
+++ b/test/CodeGen/ARM/vsra.ll
@@ -0,0 +1,293 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vsra\\.s8} %t | count 2
+; RUN: grep {vsra\\.s16} %t | count 2
+; RUN: grep {vsra\\.s32} %t | count 2
+; RUN: grep {vsra\\.s64} %t | count 2
+; RUN: grep {vsra\\.u8} %t | count 2
+; RUN: grep {vsra\\.u16} %t | count 2
+; RUN: grep {vsra\\.u32} %t | count 2
+; RUN: grep {vsra\\.u64} %t | count 2
+; RUN: grep {vrsra\\.s8} %t | count 2
+; RUN: grep {vrsra\\.s16} %t | count 2
+; RUN: grep {vrsra\\.s32} %t | count 2
+; RUN: grep {vrsra\\.s64} %t | count 2
+; RUN: grep {vrsra\\.u8} %t | count 2
+; RUN: grep {vrsra\\.u16} %t | count 2
+; RUN: grep {vrsra\\.u32} %t | count 2
+; RUN: grep {vrsra\\.u64} %t | count 2
+
+define <8 x i8> @vsras8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = ashr <8 x i8> %tmp2, < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
+        %tmp4 = add <8 x i8> %tmp1, %tmp3
+	ret <8 x i8> %tmp4
+}
+
+define <4 x i16> @vsras16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = ashr <4 x i16> %tmp2, < i16 16, i16 16, i16 16, i16 16 >
+        %tmp4 = add <4 x i16> %tmp1, %tmp3
+	ret <4 x i16> %tmp4
+}
+
+define <2 x i32> @vsras32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = ashr <2 x i32> %tmp2, < i32 32, i32 32 >
+        %tmp4 = add <2 x i32> %tmp1, %tmp3
+	ret <2 x i32> %tmp4
+}
+
+define <1 x i64> @vsras64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = ashr <1 x i64> %tmp2, < i64 64 >
+        %tmp4 = add <1 x i64> %tmp1, %tmp3
+	ret <1 x i64> %tmp4
+}
+
+define <16 x i8> @vsraQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = ashr <16 x i8> %tmp2, < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
+        %tmp4 = add <16 x i8> %tmp1, %tmp3
+	ret <16 x i8> %tmp4
+}
+
+define <8 x i16> @vsraQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = ashr <8 x i16> %tmp2, < i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16 >
+        %tmp4 = add <8 x i16> %tmp1, %tmp3
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vsraQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = ashr <4 x i32> %tmp2, < i32 32, i32 32, i32 32, i32 32 >
+        %tmp4 = add <4 x i32> %tmp1, %tmp3
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @vsraQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = ashr <2 x i64> %tmp2, < i64 64, i64 64 >
+        %tmp4 = add <2 x i64> %tmp1, %tmp3
+	ret <2 x i64> %tmp4
+}
+
+define <8 x i8> @vsrau8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = lshr <8 x i8> %tmp2, < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
+        %tmp4 = add <8 x i8> %tmp1, %tmp3
+	ret <8 x i8> %tmp4
+}
+
+define <4 x i16> @vsrau16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = lshr <4 x i16> %tmp2, < i16 16, i16 16, i16 16, i16 16 >
+        %tmp4 = add <4 x i16> %tmp1, %tmp3
+	ret <4 x i16> %tmp4
+}
+
+define <2 x i32> @vsrau32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = lshr <2 x i32> %tmp2, < i32 32, i32 32 >
+        %tmp4 = add <2 x i32> %tmp1, %tmp3
+	ret <2 x i32> %tmp4
+}
+
+define <1 x i64> @vsrau64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = lshr <1 x i64> %tmp2, < i64 64 >
+        %tmp4 = add <1 x i64> %tmp1, %tmp3
+	ret <1 x i64> %tmp4
+}
+
+define <16 x i8> @vsraQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = lshr <16 x i8> %tmp2, < i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8 >
+        %tmp4 = add <16 x i8> %tmp1, %tmp3
+	ret <16 x i8> %tmp4
+}
+
+define <8 x i16> @vsraQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = lshr <8 x i16> %tmp2, < i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16 >
+        %tmp4 = add <8 x i16> %tmp1, %tmp3
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vsraQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = lshr <4 x i32> %tmp2, < i32 32, i32 32, i32 32, i32 32 >
+        %tmp4 = add <4 x i32> %tmp1, %tmp3
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @vsraQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = lshr <2 x i64> %tmp2, < i64 64, i64 64 >
+        %tmp4 = add <2 x i64> %tmp1, %tmp3
+	ret <2 x i64> %tmp4
+}
+
+define <8 x i8> @vrsras8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vrshifts.v8i8(<8 x i8> %tmp2, <8 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+        %tmp4 = add <8 x i8> %tmp1, %tmp3
+	ret <8 x i8> %tmp4
+}
+
+define <4 x i16> @vrsras16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vrshifts.v4i16(<4 x i16> %tmp2, <4 x i16> < i16 -16, i16 -16, i16 -16, i16 -16 >)
+        %tmp4 = add <4 x i16> %tmp1, %tmp3
+	ret <4 x i16> %tmp4
+}
+
+define <2 x i32> @vrsras32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vrshifts.v2i32(<2 x i32> %tmp2, <2 x i32> < i32 -32, i32 -32 >)
+        %tmp4 = add <2 x i32> %tmp1, %tmp3
+	ret <2 x i32> %tmp4
+}
+
+define <1 x i64> @vrsras64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vrshifts.v1i64(<1 x i64> %tmp2, <1 x i64> < i64 -64 >)
+        %tmp4 = add <1 x i64> %tmp1, %tmp3
+	ret <1 x i64> %tmp4
+}
+
+define <8 x i8> @vrsrau8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vrshiftu.v8i8(<8 x i8> %tmp2, <8 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+        %tmp4 = add <8 x i8> %tmp1, %tmp3
+	ret <8 x i8> %tmp4
+}
+
+define <4 x i16> @vrsrau16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vrshiftu.v4i16(<4 x i16> %tmp2, <4 x i16> < i16 -16, i16 -16, i16 -16, i16 -16 >)
+        %tmp4 = add <4 x i16> %tmp1, %tmp3
+	ret <4 x i16> %tmp4
+}
+
+define <2 x i32> @vrsrau32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vrshiftu.v2i32(<2 x i32> %tmp2, <2 x i32> < i32 -32, i32 -32 >)
+        %tmp4 = add <2 x i32> %tmp1, %tmp3
+	ret <2 x i32> %tmp4
+}
+
+define <1 x i64> @vrsrau64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = call <1 x i64> @llvm.arm.neon.vrshiftu.v1i64(<1 x i64> %tmp2, <1 x i64> < i64 -64 >)
+        %tmp4 = add <1 x i64> %tmp1, %tmp3
+	ret <1 x i64> %tmp4
+}
+
+define <16 x i8> @vrsraQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vrshifts.v16i8(<16 x i8> %tmp2, <16 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+        %tmp4 = add <16 x i8> %tmp1, %tmp3
+	ret <16 x i8> %tmp4
+}
+
+define <8 x i16> @vrsraQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vrshifts.v8i16(<8 x i16> %tmp2, <8 x i16> < i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16 >)
+        %tmp4 = add <8 x i16> %tmp1, %tmp3
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vrsraQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vrshifts.v4i32(<4 x i32> %tmp2, <4 x i32> < i32 -32, i32 -32, i32 -32, i32 -32 >)
+        %tmp4 = add <4 x i32> %tmp1, %tmp3
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @vrsraQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vrshifts.v2i64(<2 x i64> %tmp2, <2 x i64> < i64 -64, i64 -64 >)
+        %tmp4 = add <2 x i64> %tmp1, %tmp3
+	ret <2 x i64> %tmp4
+}
+
+define <16 x i8> @vrsraQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = call <16 x i8> @llvm.arm.neon.vrshiftu.v16i8(<16 x i8> %tmp2, <16 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >)
+        %tmp4 = add <16 x i8> %tmp1, %tmp3
+	ret <16 x i8> %tmp4
+}
+
+define <8 x i16> @vrsraQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vrshiftu.v8i16(<8 x i16> %tmp2, <8 x i16> < i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16 >)
+        %tmp4 = add <8 x i16> %tmp1, %tmp3
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vrsraQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vrshiftu.v4i32(<4 x i32> %tmp2, <4 x i32> < i32 -32, i32 -32, i32 -32, i32 -32 >)
+        %tmp4 = add <4 x i32> %tmp1, %tmp3
+	ret <4 x i32> %tmp4
+}
+
+define <2 x i64> @vrsraQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vrshiftu.v2i64(<2 x i64> %tmp2, <2 x i64> < i64 -64, i64 -64 >)
+        %tmp4 = add <2 x i64> %tmp1, %tmp3
+	ret <2 x i64> %tmp4
+}
+
+declare <8 x i8>  @llvm.arm.neon.vrshifts.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vrshifts.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vrshifts.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vrshifts.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <8 x i8>  @llvm.arm.neon.vrshiftu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vrshiftu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vrshiftu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vrshiftu.v1i64(<1 x i64>, <1 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vrshifts.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vrshifts.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vrshifts.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vrshifts.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
+
+declare <16 x i8> @llvm.arm.neon.vrshiftu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vrshiftu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vrshiftu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vrshiftu.v2i64(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vsub.ll b/test/CodeGen/ARM/vsub.ll
new file mode 100644
index 0000000000..85dea41835
--- /dev/null
+++ b/test/CodeGen/ARM/vsub.ll
@@ -0,0 +1,76 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vsub\\.i8} %t | count 2
+; RUN: grep {vsub\\.i16} %t | count 2
+; RUN: grep {vsub\\.i32} %t | count 2
+; RUN: grep {vsub\\.i64} %t | count 2
+; RUN: grep {vsub\\.f32} %t | count 2
+
+define <8 x i8> @vsubi8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = sub <8 x i8> %tmp1, %tmp2
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vsubi16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = sub <4 x i16> %tmp1, %tmp2
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vsubi32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = sub <2 x i32> %tmp1, %tmp2
+	ret <2 x i32> %tmp3
+}
+
+define <1 x i64> @vsubi64(<1 x i64>* %A, <1 x i64>* %B) nounwind {
+	%tmp1 = load <1 x i64>* %A
+	%tmp2 = load <1 x i64>* %B
+	%tmp3 = sub <1 x i64> %tmp1, %tmp2
+	ret <1 x i64> %tmp3
+}
+
+define <2 x float> @vsubf32(<2 x float>* %A, <2 x float>* %B) nounwind {
+	%tmp1 = load <2 x float>* %A
+	%tmp2 = load <2 x float>* %B
+	%tmp3 = sub <2 x float> %tmp1, %tmp2
+	ret <2 x float> %tmp3
+}
+
+define <16 x i8> @vsubQi8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = sub <16 x i8> %tmp1, %tmp2
+	ret <16 x i8> %tmp3
+}
+
+define <8 x i16> @vsubQi16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = sub <8 x i16> %tmp1, %tmp2
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vsubQi32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = sub <4 x i32> %tmp1, %tmp2
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vsubQi64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = sub <2 x i64> %tmp1, %tmp2
+	ret <2 x i64> %tmp3
+}
+
+define <4 x float> @vsubQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
+	%tmp1 = load <4 x float>* %A
+	%tmp2 = load <4 x float>* %B
+	%tmp3 = sub <4 x float> %tmp1, %tmp2
+	ret <4 x float> %tmp3
+}
diff --git a/test/CodeGen/ARM/vsubhn.ll b/test/CodeGen/ARM/vsubhn.ll
new file mode 100644
index 0000000000..5adc9ed97d
--- /dev/null
+++ b/test/CodeGen/ARM/vsubhn.ll
@@ -0,0 +1,29 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vsubhn\\.i16} %t | count 1
+; RUN: grep {vsubhn\\.i32} %t | count 1
+; RUN: grep {vsubhn\\.i64} %t | count 1
+
+define <8 x i8> @vsubhni16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = call <8 x i8> @llvm.arm.neon.vsubhn.v8i8(<8 x i16> %tmp1, <8 x i16> %tmp2)
+	ret <8 x i8> %tmp3
+}
+
+define <4 x i16> @vsubhni32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = call <4 x i16> @llvm.arm.neon.vsubhn.v4i16(<4 x i32> %tmp1, <4 x i32> %tmp2)
+	ret <4 x i16> %tmp3
+}
+
+define <2 x i32> @vsubhni64(<2 x i64>* %A, <2 x i64>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i64>* %B
+	%tmp3 = call <2 x i32> @llvm.arm.neon.vsubhn.v2i32(<2 x i64> %tmp1, <2 x i64> %tmp2)
+	ret <2 x i32> %tmp3
+}
+
+declare <8 x i8>  @llvm.arm.neon.vsubhn.v8i8(<8 x i16>, <8 x i16>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vsubhn.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vsubhn.v2i32(<2 x i64>, <2 x i64>) nounwind readnone
diff --git a/test/CodeGen/ARM/vsubl.ll b/test/CodeGen/ARM/vsubl.ll
new file mode 100644
index 0000000000..d2af455843
--- /dev/null
+++ b/test/CodeGen/ARM/vsubl.ll
@@ -0,0 +1,57 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vsubl\\.s8} %t | count 1
+; RUN: grep {vsubl\\.s16} %t | count 1
+; RUN: grep {vsubl\\.s32} %t | count 1
+; RUN: grep {vsubl\\.u8} %t | count 1
+; RUN: grep {vsubl\\.u16} %t | count 1
+; RUN: grep {vsubl\\.u32} %t | count 1
+
+define <8 x i16> @vsubls8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vsubls.v8i16(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vsubls16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vsubls.v4i32(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vsubls32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vsubls.v2i64(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <8 x i16> @vsublu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vsublu.v8i16(<8 x i8> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vsublu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vsublu.v4i32(<4 x i16> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vsublu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vsublu.v2i64(<2 x i32> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+declare <8 x i16> @llvm.arm.neon.vsubls.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vsubls.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vsubls.v2i64(<2 x i32>, <2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vsublu.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vsublu.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vsublu.v2i64(<2 x i32>, <2 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vsubw.ll b/test/CodeGen/ARM/vsubw.ll
new file mode 100644
index 0000000000..30bec5532a
--- /dev/null
+++ b/test/CodeGen/ARM/vsubw.ll
@@ -0,0 +1,57 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vsubw\\.s8} %t | count 1
+; RUN: grep {vsubw\\.s16} %t | count 1
+; RUN: grep {vsubw\\.s32} %t | count 1
+; RUN: grep {vsubw\\.u8} %t | count 1
+; RUN: grep {vsubw\\.u16} %t | count 1
+; RUN: grep {vsubw\\.u32} %t | count 1
+
+define <8 x i16> @vsubws8(<8 x i16>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vsubws.v8i16(<8 x i16> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vsubws16(<4 x i32>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vsubws.v4i32(<4 x i32> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vsubws32(<2 x i64>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vsubws.v2i64(<2 x i64> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+define <8 x i16> @vsubwu8(<8 x i16>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = call <8 x i16> @llvm.arm.neon.vsubwu.v8i16(<8 x i16> %tmp1, <8 x i8> %tmp2)
+	ret <8 x i16> %tmp3
+}
+
+define <4 x i32> @vsubwu16(<4 x i32>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = call <4 x i32> @llvm.arm.neon.vsubwu.v4i32(<4 x i32> %tmp1, <4 x i16> %tmp2)
+	ret <4 x i32> %tmp3
+}
+
+define <2 x i64> @vsubwu32(<2 x i64>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i64>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = call <2 x i64> @llvm.arm.neon.vsubwu.v2i64(<2 x i64> %tmp1, <2 x i32> %tmp2)
+	ret <2 x i64> %tmp3
+}
+
+declare <8 x i16> @llvm.arm.neon.vsubws.v8i16(<8 x i16>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vsubws.v4i32(<4 x i32>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vsubws.v2i64(<2 x i64>, <2 x i32>) nounwind readnone
+
+declare <8 x i16> @llvm.arm.neon.vsubwu.v8i16(<8 x i16>, <8 x i8>) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vsubwu.v4i32(<4 x i32>, <4 x i16>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vsubwu.v2i64(<2 x i64>, <2 x i32>) nounwind readnone
diff --git a/test/CodeGen/ARM/vtst.ll b/test/CodeGen/ARM/vtst.ll
new file mode 100644
index 0000000000..d4f7d2aef1
--- /dev/null
+++ b/test/CodeGen/ARM/vtst.ll
@@ -0,0 +1,52 @@
+; RUN: llvm-as < %s | llc -march=arm -mattr=+neon > %t
+; RUN: grep {vtst\\.i8} %t | count 2
+; RUN: grep {vtst\\.i16} %t | count 2
+; RUN: grep {vtst\\.i32} %t | count 2
+
+define <8 x i8> @vtsti8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
+	%tmp1 = load <8 x i8>* %A
+	%tmp2 = load <8 x i8>* %B
+	%tmp3 = and <8 x i8> %tmp1, %tmp2
+	%tmp4 = vicmp ne <8 x i8> %tmp3, zeroinitializer
+	ret <8 x i8> %tmp4
+}
+
+define <4 x i16> @vtsti16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
+	%tmp1 = load <4 x i16>* %A
+	%tmp2 = load <4 x i16>* %B
+	%tmp3 = and <4 x i16> %tmp1, %tmp2
+	%tmp4 = vicmp ne <4 x i16> %tmp3, zeroinitializer
+	ret <4 x i16> %tmp4
+}
+
+define <2 x i32> @vtsti32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
+	%tmp1 = load <2 x i32>* %A
+	%tmp2 = load <2 x i32>* %B
+	%tmp3 = and <2 x i32> %tmp1, %tmp2
+	%tmp4 = vicmp ne <2 x i32> %tmp3, zeroinitializer
+	ret <2 x i32> %tmp4
+}
+
+define <16 x i8> @vtstQi8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
+	%tmp1 = load <16 x i8>* %A
+	%tmp2 = load <16 x i8>* %B
+	%tmp3 = and <16 x i8> %tmp1, %tmp2
+	%tmp4 = vicmp ne <16 x i8> %tmp3, zeroinitializer
+	ret <16 x i8> %tmp4
+}
+
+define <8 x i16> @vtstQi16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
+	%tmp1 = load <8 x i16>* %A
+	%tmp2 = load <8 x i16>* %B
+	%tmp3 = and <8 x i16> %tmp1, %tmp2
+	%tmp4 = vicmp ne <8 x i16> %tmp3, zeroinitializer
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vtstQi32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
+	%tmp1 = load <4 x i32>* %A
+	%tmp2 = load <4 x i32>* %B
+	%tmp3 = and <4 x i32> %tmp1, %tmp2
+	%tmp4 = vicmp ne <4 x i32> %tmp3, zeroinitializer
+	ret <4 x i32> %tmp4
+}