1; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii -slp-vectorizer %s | FileCheck -check-prefixes=GCN,GFX7 %s
2; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -slp-vectorizer %s | FileCheck -check-prefixes=GCN,GFX8 %s
3; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -slp-vectorizer %s | FileCheck -check-prefixes=GCN,GFX8 %s
4
5define <2 x i16> @bswap_v2i16(<2 x i16> %arg) {
6; GFX7-LABEL: @bswap_v2i16(
7; GFX7-NEXT:  bb:
8; GFX7-NEXT:    [[T:%.*]] = extractelement <2 x i16> [[ARG:%.*]], i64 0
9; GFX7-NEXT:    [[T1:%.*]] = tail call i16 @llvm.bswap.i16(i16 [[T]])
10; GFX7-NEXT:    [[T2:%.*]] = insertelement <2 x i16> undef, i16 [[T1]], i64 0
11; GFX7-NEXT:    [[T3:%.*]] = extractelement <2 x i16> [[ARG]], i64 1
12; GFX7-NEXT:    [[T4:%.*]] = tail call i16 @llvm.bswap.i16(i16 [[T3]])
13; GFX7-NEXT:    [[T5:%.*]] = insertelement <2 x i16> [[T2]], i16 [[T4]], i64 1
14; GFX7-NEXT:    ret <2 x i16> [[T5]]
15;
16; GFX8-LABEL: @bswap_v2i16(
17; GFX8-NEXT:  bb:
18; GFX8-NEXT:    [[TMP0:%.*]] = call <2 x i16> @llvm.bswap.v2i16(<2 x i16> [[ARG:%.*]])
19; GFX8-NEXT:    [[TMP1:%.*]] = extractelement <2 x i16> [[TMP0]], i32 0
20; GFX8-NEXT:    [[T2:%.*]] = insertelement <2 x i16> undef, i16 [[TMP1]], i64 0
21; GFX8-NEXT:    [[TMP2:%.*]] = extractelement <2 x i16> [[TMP0]], i32 1
22; GFX8-NEXT:    [[T5:%.*]] = insertelement <2 x i16> [[T2]], i16 [[TMP2]], i64 1
23; GFX8-NEXT:    ret <2 x i16> [[T5]]
24;
25bb:
26  %t = extractelement <2 x i16> %arg, i64 0
27  %t1 = tail call i16 @llvm.bswap.i16(i16 %t)
28  %t2 = insertelement <2 x i16> undef, i16 %t1, i64 0
29  %t3 = extractelement <2 x i16> %arg, i64 1
30  %t4 = tail call i16 @llvm.bswap.i16(i16 %t3)
31  %t5 = insertelement <2 x i16> %t2, i16 %t4, i64 1
32  ret <2 x i16> %t5
33}
34
35define <2 x i32> @bswap_v2i32(<2 x i32> %arg) {
36; GCN-LABEL: @bswap_v2i32(
37; GCN-NEXT:  bb:
38; GCN-NEXT:    [[T:%.*]] = extractelement <2 x i32> [[ARG:%.*]], i64 0
39; GCN-NEXT:    [[T1:%.*]] = tail call i32 @llvm.bswap.i32(i32 [[T]])
40; GCN-NEXT:    [[T2:%.*]] = insertelement <2 x i32> undef, i32 [[T1]], i64 0
41; GCN-NEXT:    [[T3:%.*]] = extractelement <2 x i32> [[ARG]], i64 1
42; GCN-NEXT:    [[T4:%.*]] = tail call i32 @llvm.bswap.i32(i32 [[T3]])
43; GCN-NEXT:    [[T5:%.*]] = insertelement <2 x i32> [[T2]], i32 [[T4]], i64 1
44; GCN-NEXT:    ret <2 x i32> [[T5]]
45;
46bb:
47  %t = extractelement <2 x i32> %arg, i64 0
48  %t1 = tail call i32 @llvm.bswap.i32(i32 %t)
49  %t2 = insertelement <2 x i32> undef, i32 %t1, i64 0
50  %t3 = extractelement <2 x i32> %arg, i64 1
51  %t4 = tail call i32 @llvm.bswap.i32(i32 %t3)
52  %t5 = insertelement <2 x i32> %t2, i32 %t4, i64 1
53  ret <2 x i32> %t5
54}
55
56declare i16 @llvm.bswap.i16(i16) #0
57declare i32 @llvm.bswap.i32(i32) #0
58
59attributes #0 = { nounwind readnone speculatable willreturn }
60