1; RUN: llc -march=hexagon -O2 -disable-hexagon-shuffle=1 < %s | FileCheck %s
2; CHECK: vmemu(r{{[0-9]+}}) = v{{[0-9]*}};
3
4target triple = "hexagon"
5
6; Function Attrs: nounwind
7define void @f0(i8* %a0, i32 %a1, i8* %a2, i32 %a3, i8* %a4) #0 {
8b0:
9  %v0 = alloca i8*, align 4
10  %v1 = alloca i32, align 4
11  %v2 = alloca i8*, align 4
12  %v3 = alloca i32, align 4
13  %v4 = alloca i8*, align 4
14  %v5 = alloca i32, align 4
15  %v6 = alloca i32, align 4
16  %v7 = alloca i32, align 4
17  %v8 = alloca i32, align 4
18  %v9 = alloca i32, align 4
19  %v10 = alloca <16 x i32>, align 64
20  %v11 = alloca <16 x i32>, align 64
21  %v12 = alloca <16 x i32>, align 64
22  %v13 = alloca <16 x i32>, align 64
23  %v14 = alloca <16 x i32>, align 64
24  %v15 = alloca <16 x i32>, align 64
25  %v16 = alloca <16 x i32>, align 64
26  %v17 = alloca <16 x i32>, align 64
27  %v18 = alloca <16 x i32>, align 64
28  %v19 = alloca <16 x i32>, align 64
29  %v20 = alloca <16 x i32>, align 64
30  store i8* %a0, i8** %v0, align 4
31  store i32 %a1, i32* %v1, align 4
32  store i8* %a2, i8** %v2, align 4
33  store i32 %a3, i32* %v3, align 4
34  store i8* %a4, i8** %v4, align 4
35  %v21 = load i32, i32* %v1, align 4
36  %v22 = ashr i32 %v21, 16
37  %v23 = and i32 65535, %v22
38  store i32 %v23, i32* %v8, align 4
39  %v24 = load i32, i32* %v1, align 4
40  %v25 = and i32 65535, %v24
41  store i32 %v25, i32* %v5, align 4
42  %v26 = load i32, i32* %v3, align 4
43  %v27 = and i32 65535, %v26
44  store i32 %v27, i32* %v6, align 4
45  %v28 = load i32, i32* %v3, align 4
46  %v29 = ashr i32 %v28, 16
47  %v30 = and i32 65535, %v29
48  store i32 %v30, i32* %v9, align 4
49  %v31 = load i8*, i8** %v4, align 4
50  %v32 = bitcast i8* %v31 to <16 x i32>*
51  %v33 = load <16 x i32>, <16 x i32>* %v32, align 64
52  store <16 x i32> %v33, <16 x i32>* %v10, align 64
53  %v34 = load i8*, i8** %v4, align 4
54  %v35 = getelementptr inbounds i8, i8* %v34, i32 64
55  %v36 = bitcast i8* %v35 to <16 x i32>*
56  %v37 = load <16 x i32>, <16 x i32>* %v36, align 64
57  store <16 x i32> %v37, <16 x i32>* %v12, align 64
58  %v38 = load i32, i32* %v9, align 4
59  store i32 %v38, i32* %v7, align 4
60  br label %b1
61
62b1:                                               ; preds = %b3, %b0
63  %v39 = load i32, i32* %v7, align 4
64  %v40 = icmp sge i32 %v39, 0
65  br i1 %v40, label %b2, label %b4
66
67b2:                                               ; preds = %b1
68  %v41 = load i8*, i8** %v0, align 4
69  %v42 = bitcast i8* %v41 to <16 x i32>*
70  %v43 = load <16 x i32>, <16 x i32>* %v42, align 4
71  store <16 x i32> %v43, <16 x i32>* %v14, align 64
72  %v44 = load i32, i32* %v5, align 4
73  %v45 = load i8*, i8** %v0, align 4
74  %v46 = getelementptr inbounds i8, i8* %v45, i32 %v44
75  store i8* %v46, i8** %v0, align 4
76  %v47 = load <16 x i32>, <16 x i32>* %v14, align 64
77  %v48 = load <16 x i32>, <16 x i32>* %v10, align 64
78  %v49 = call <16 x i32> @llvm.hexagon.V6.vrdelta(<16 x i32> %v47, <16 x i32> %v48)
79  store <16 x i32> %v49, <16 x i32>* %v15, align 64
80  %v50 = load <16 x i32>, <16 x i32>* %v14, align 64
81  %v51 = load <16 x i32>, <16 x i32>* %v12, align 64
82  %v52 = call <16 x i32> @llvm.hexagon.V6.vrdelta(<16 x i32> %v50, <16 x i32> %v51)
83  store <16 x i32> %v52, <16 x i32>* %v17, align 64
84  %v53 = load <16 x i32>, <16 x i32>* %v15, align 64
85  %v54 = load <16 x i32>, <16 x i32>* %v17, align 64
86  %v55 = call <16 x i32> @llvm.hexagon.V6.vavgub(<16 x i32> %v53, <16 x i32> %v54)
87  store <16 x i32> %v55, <16 x i32>* %v19, align 64
88  %v56 = load i8*, i8** %v2, align 4
89  %v57 = load <16 x i32>, <16 x i32>* %v19, align 64
90  call void asm sideeffect "  vmemu($0) = $1;\0A", "r,v,~{memory}"(i8* %v56, <16 x i32> %v57) #2, !srcloc !0
91  br label %b3
92
93b3:                                               ; preds = %b2
94  %v58 = load i32, i32* %v6, align 4
95  %v59 = load i32, i32* %v7, align 4
96  %v60 = sub nsw i32 %v59, %v58
97  store i32 %v60, i32* %v7, align 4
98  br label %b1
99
100b4:                                               ; preds = %b1
101  ret void
102}
103
104; Function Attrs: nounwind readnone
105declare <16 x i32> @llvm.hexagon.V6.vrdelta(<16 x i32>, <16 x i32>) #1
106
107; Function Attrs: nounwind readnone
108declare <16 x i32> @llvm.hexagon.V6.vavgub(<16 x i32>, <16 x i32>) #1
109
110attributes #0 = { nounwind "target-cpu"="hexagonv60" "target-features"="+hvxv60,+hvx-length64b" }
111attributes #1 = { nounwind readnone }
112attributes #2 = { nounwind }
113
114!0 = !{i32 1708}
115