1 ; RUN: llc < %s -march=arm -mattr=+neon > %t
2 ; RUN: grep {vshll\\.s8} %t | count 1
3 ; RUN: grep {vshll\\.s16} %t | count 1
4 ; RUN: grep {vshll\\.s32} %t | count 1
5 ; RUN: grep {vshll\\.u8} %t | count 1
6 ; RUN: grep {vshll\\.u16} %t | count 1
7 ; RUN: grep {vshll\\.u32} %t | count 1
8 ; RUN: grep {vshll\\.i8} %t | count 1
9 ; RUN: grep {vshll\\.i16} %t | count 1
10 ; RUN: grep {vshll\\.i32} %t | count 1
12 define <8 x i16> @vshlls8(<8 x i8>* %A) nounwind {
13 %tmp1 = load <8 x i8>* %A
14 %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 >)
18 define <4 x i32> @vshlls16(<4 x i16>* %A) nounwind {
19 %tmp1 = load <4 x i16>* %A
20 %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 >)
24 define <2 x i64> @vshlls32(<2 x i32>* %A) nounwind {
25 %tmp1 = load <2 x i32>* %A
26 %tmp2 = call <2 x i64> @llvm.arm.neon.vshiftls.v2i64(<2 x i32> %tmp1, <2 x i32> < i32 31, i32 31 >)
30 define <8 x i16> @vshllu8(<8 x i8>* %A) nounwind {
31 %tmp1 = load <8 x i8>* %A
32 %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 >)
36 define <4 x i32> @vshllu16(<4 x i16>* %A) nounwind {
37 %tmp1 = load <4 x i16>* %A
38 %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 >)
42 define <2 x i64> @vshllu32(<2 x i32>* %A) nounwind {
43 %tmp1 = load <2 x i32>* %A
44 %tmp2 = call <2 x i64> @llvm.arm.neon.vshiftlu.v2i64(<2 x i32> %tmp1, <2 x i32> < i32 31, i32 31 >)
48 ; The following tests use the maximum shift count, so the signedness is
49 ; irrelevant. Test both signed and unsigned versions.
50 define <8 x i16> @vshlli8(<8 x i8>* %A) nounwind {
51 %tmp1 = load <8 x i8>* %A
52 %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 >)
56 define <4 x i32> @vshlli16(<4 x i16>* %A) nounwind {
57 %tmp1 = load <4 x i16>* %A
58 %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 >)
62 define <2 x i64> @vshlli32(<2 x i32>* %A) nounwind {
63 %tmp1 = load <2 x i32>* %A
64 %tmp2 = call <2 x i64> @llvm.arm.neon.vshiftls.v2i64(<2 x i32> %tmp1, <2 x i32> < i32 32, i32 32 >)
68 declare <8 x i16> @llvm.arm.neon.vshiftls.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
69 declare <4 x i32> @llvm.arm.neon.vshiftls.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
70 declare <2 x i64> @llvm.arm.neon.vshiftls.v2i64(<2 x i32>, <2 x i32>) nounwind readnone
72 declare <8 x i16> @llvm.arm.neon.vshiftlu.v8i16(<8 x i8>, <8 x i8>) nounwind readnone
73 declare <4 x i32> @llvm.arm.neon.vshiftlu.v4i32(<4 x i16>, <4 x i16>) nounwind readnone
74 declare <2 x i64> @llvm.arm.neon.vshiftlu.v2i64(<2 x i32>, <2 x i32>) nounwind readnone