1 ; RUN: llc < %s -march=arm -mattr=+neon > %t
2 ; RUN: grep {vsubl\\.s8} %t | count 1
3 ; RUN: grep {vsubl\\.s16} %t | count 1
4 ; RUN: grep {vsubl\\.s32} %t | count 1
5 ; RUN: grep {vsubl\\.u8} %t | count 1
6 ; RUN: grep {vsubl\\.u16} %t | count 1
7 ; RUN: grep {vsubl\\.u32} %t | count 1
9 define <8 x i16> @vsubls8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
10 %tmp1 = load <8 x i8>* %A
11 %tmp2 = load <8 x i8>* %B
12 %tmp3 = call <8 x i16> @llvm.arm.neon.vsubls.v8i16(<8 x i8> %tmp1, <8 x i8> %tmp2)
16 define <4 x i32> @vsubls16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
17 %tmp1 = load <4 x i16>* %A
18 %tmp2 = load <4 x i16>* %B
19 %tmp3 = call <4 x i32> @llvm.arm.neon.vsubls.v4i32(<4 x i16> %tmp1, <4 x i16> %tmp2)
23 define <2 x i64> @vsubls32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
24 %tmp1 = load <2 x i32>* %A
25 %tmp2 = load <2 x i32>* %B
26 %tmp3 = call <2 x i64> @llvm.arm.neon.vsubls.v2i64(<2 x i32> %tmp1, <2 x i32> %tmp2)
30 define <8 x i16> @vsublu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
31 %tmp1 = load <8 x i8>* %A
32 %tmp2 = load <8 x i8>* %B
33 %tmp3 = call <8 x i16> @llvm.arm.neon.vsublu.v8i16(<8 x i8> %tmp1, <8 x i8> %tmp2)
37 define <4 x i32> @vsublu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
38 %tmp1 = load <4 x i16>* %A
39 %tmp2 = load <4 x i16>* %B
40 %tmp3 = call <4 x i32> @llvm.arm.neon.vsublu.v4i32(<4 x i16> %tmp1, <4 x i16> %tmp2)
44 define <2 x i64> @vsublu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
45 %tmp1 = load <2 x i32>* %A
46 %tmp2 = load <2 x i32>* %B
47 %tmp3 = call <2 x i64> @llvm.arm.neon.vsublu.v2i64(<2 x i32> %tmp1, <2 x i32> %tmp2)
51 declare <8 x i16> @llvm.arm.neon.vsubls.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
52 declare <4 x i32> @llvm.arm.neon.vsubls.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
53 declare <2 x i64> @llvm.arm.neon.vsubls.v2i64(<2 x i32>, <2 x i32>) nounwind readnone
55 declare <8 x i16> @llvm.arm.neon.vsublu.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
56 declare <4 x i32> @llvm.arm.neon.vsublu.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
57 declare <2 x i64> @llvm.arm.neon.vsublu.v2i64(<2 x i32>, <2 x i32>) nounwind readnone