1; RUN: opt -S -loop-vectorize -force-vector-interleave=1 -instcombine -mattr=+sve -mtriple aarch64-unknown-linux-gnu < %s | FileCheck %s
2
3define void @vec_load(i64 %N, double* nocapture %a, double* nocapture readonly %b) {
4; CHECK-LABEL: @vec_load
5; CHECK: vector.body:
6; CHECK: %[[LOAD:.*]] = load <vscale x 2 x double>, <vscale x 2 x double>*
7; CHECK: call <vscale x 2 x double> @foo_vec(<vscale x 2 x double> %[[LOAD]])
8entry:
9  %cmp7 = icmp sgt i64 %N, 0
10  br i1 %cmp7, label %for.body, label %for.end
11
12for.body:                                         ; preds = %for.body.preheader, %for.body
13  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
14  %arrayidx = getelementptr inbounds double, double* %b, i64 %iv
15  %0 = load double, double* %arrayidx, align 8
16  %1 = call double @foo(double %0) #0
17  %add = fadd double %1, 1.000000e+00
18  %arrayidx2 = getelementptr inbounds double, double* %a, i64 %iv
19  store double %add, double* %arrayidx2, align 8
20  %iv.next = add nuw nsw i64 %iv, 1
21  %exitcond.not = icmp eq i64 %iv.next, %N
22  br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !1
23
24for.end:                                 ; preds = %for.body, %entry
25  ret void
26}
27
28define void @vec_scalar(i64 %N, double* nocapture %a) {
29; CHECK-LABEL: @vec_scalar
30; CHECK: vector.body:
31; CHECK: call <vscale x 2 x double> @foo_vec(<vscale x 2 x double> shufflevector (<vscale x 2 x double> insertelement (<vscale x 2 x double> poison, double 1.000000e+01, i32 0), <vscale x 2 x double> poison, <vscale x 2 x i32> zeroinitializer))
32entry:
33  %cmp7 = icmp sgt i64 %N, 0
34  br i1 %cmp7, label %for.body, label %for.end
35
36for.body:                                         ; preds = %for.body.preheader, %for.body
37  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
38  %0 = call double @foo(double 10.0) #0
39  %sub = fsub double %0, 1.000000e+00
40  %arrayidx = getelementptr inbounds double, double* %a, i64 %iv
41  store double %sub, double* %arrayidx, align 8
42  %iv.next = add nuw nsw i64 %iv, 1
43  %exitcond.not = icmp eq i64 %iv.next, %N
44  br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !1
45
46for.end:                                 ; preds = %for.body, %entry
47  ret void
48}
49
50define void @vec_ptr(i64 %N, i64* noalias %a, i64** readnone %b) {
51; CHECK-LABEL: @vec_ptr
52; CHECK: vector.body:
53; CHECK: %[[LOAD:.*]] = load <vscale x 2 x i64*>, <vscale x 2 x i64*>*
54; CHECK: call <vscale x 2 x i64> @bar_vec(<vscale x 2 x i64*> %[[LOAD]])
55entry:
56  %cmp7 = icmp sgt i64 %N, 0
57  br i1 %cmp7, label %for.body, label %for.end
58
59for.body:
60  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
61  %gep = getelementptr i64*, i64** %b, i64 %iv
62  %load = load i64*, i64** %gep
63  %call = call i64 @bar(i64* %load) #1
64  %arrayidx = getelementptr inbounds i64, i64* %a, i64 %iv
65  store i64 %call, i64* %arrayidx
66  %iv.next = add nuw nsw i64 %iv, 1
67  %exitcond = icmp eq i64 %iv.next, 1024
68  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !1
69
70for.end:
71  ret void
72}
73
74define void @vec_intrinsic(i64 %N, double* nocapture readonly %a) {
75; CHECK-LABEL: @vec_intrinsic
76; CHECK: vector.body:
77; CHECK: %[[LOAD:.*]] = load <vscale x 2 x double>, <vscale x 2 x double>*
78; CHECK: call fast <vscale x 2 x double> @sin_vec(<vscale x 2 x double> %[[LOAD]])
79entry:
80  %cmp7 = icmp sgt i64 %N, 0
81  br i1 %cmp7, label %for.body, label %for.end
82
83for.body:
84  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
85  %arrayidx = getelementptr inbounds double, double* %a, i64 %iv
86  %0 = load double, double* %arrayidx, align 8
87  %1 = call fast double @llvm.sin.f64(double %0) #2
88  %add = fadd fast double %1, 1.000000e+00
89  %iv.next = add nuw nsw i64 %iv, 1
90  %exitcond = icmp eq i64 %iv.next, %N
91  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !1
92
93for.end:
94  ret void
95}
96
97declare double @foo(double)
98declare i64 @bar(i64*)
99declare double @llvm.sin.f64(double)
100
101declare <vscale x 2 x double> @foo_vec(<vscale x 2 x double>)
102declare <vscale x 2 x i64> @bar_vec(<vscale x 2 x i64*>)
103declare <vscale x 2 x double> @sin_vec(<vscale x 2 x double>)
104
105attributes #0 = { "vector-function-abi-variant"="_ZGV_LLVM_Nxv_foo(foo_vec)" }
106attributes #1 = { "vector-function-abi-variant"="_ZGV_LLVM_Nxv_bar(bar_vec)" }
107attributes #2 = { "vector-function-abi-variant"="_ZGV_LLVM_Nxv_llvm.sin.f64(sin_vec)" }
108
109!1 = distinct !{!1, !2, !3}
110!2 = !{!"llvm.loop.vectorize.width", i32 2}
111!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}
112