1; This is the loop in c++ being vectorize in this file with
2;experimental.vector.reverse
3;  #pragma clang loop vectorize_width(8, scalable)
4;  for (int i = N-1; i >= 0; --i)
5;    a[i] = b[i] + 1.0;
6
7; RUN: opt -loop-vectorize -dce -instcombine -mtriple aarch64-linux-gnu -S < %s 2>%t | FileCheck %s
8
9; RUN: FileCheck --check-prefix=WARN --allow-empty %s <%t
10
11; If this check fails please read test/CodeGen/AArch64/README for instructions on how to resolve it.
12; WARN-NOT: warning
13
14define void @vector_reverse_f64(i64 %N, double* %a, double* %b) #0{
15; CHECK-LABEL: @vector_reverse_f64
16; CHECK-LABEL: vector.body:
17; CHECK: %[[ADD:.*]] = add i64 %{{.*}}, %N
18; CHECK-NEXT: %[[GEP:.*]] = getelementptr inbounds double, double* %b, i64 %[[ADD]]
19; CHECK-NEXT: %[[VSCALE:.*]] = call i32 @llvm.vscale.i32()
20; CHECK-NEXT: %[[MUL:.*]] = mul i32 %[[VSCALE]], -8
21; CHECK-NEXT: %[[OR:.*]] = or i32 %[[MUL]], 1
22; CHECK-NEXT: %[[SEXT:.*]] = sext i32 %[[OR]] to i64
23; CHECK-NEXT: %[[GEP1:.*]] = getelementptr inbounds double, double* %[[GEP]], i64 %[[SEXT]]
24; CHECK-NEXT: %[[CAST:.*]] = bitcast double* %[[GEP1]] to <vscale x 8 x double>*
25; CHECK-NEXT: %[[WIDE:.*]] = load <vscale x 8 x double>, <vscale x 8 x double>* %[[CAST]], align 8
26; CHECK-NEXT: %[[REVERSE:.*]] = call <vscale x 8 x double> @llvm.experimental.vector.reverse.nxv8f64(<vscale x 8 x double> %[[WIDE]])
27; CHECK-NEXT: %[[FADD:.*]] = fadd <vscale x 8 x double> %[[REVERSE]], shufflevector
28; CHECK-NEXT: %[[GEP2:.*]] = getelementptr inbounds double, double* %a, i64 %[[ADD]]
29; CHECK-NEXT: %[[REVERSE6:.*]] = call <vscale x 8 x double> @llvm.experimental.vector.reverse.nxv8f64(<vscale x 8 x double> %[[FADD]])
30; CHECK-NEXT: %[[VSCALE1:.*]] = call i32 @llvm.vscale.i32()
31; CHECK-NEXT: %[[MUL1:.*]] = mul i32 %[[VSCALE1]], -8
32; CHECK-NEXT: %[[OR1:.*]] = or i32 %[[MUL1]], 1
33; CHECK-NEXT: %[[SEXT1:.*]] = sext i32 %[[OR1]] to i64
34; CHECK-NEXT: %[[GEP3:.*]] = getelementptr inbounds double, double* %[[GEP2]], i64 %[[SEXT1]]
35; CHECK-NEXT: %[[CAST1:.*]] = bitcast double* %[[GEP3]] to <vscale x 8 x double>*
36; CHECK-NEXT: store <vscale x 8 x double> %[[REVERSE6]], <vscale x 8 x double>* %[[CAST1]], align 8
37
38entry:
39  %cmp7 = icmp sgt i64 %N, 0
40  br i1 %cmp7, label %for.body, label %for.cond.cleanup
41
42for.cond.cleanup:                                 ; preds = %for.body
43  ret void
44
45for.body:                                         ; preds = %entry, %for.body
46  %i.08.in = phi i64 [ %i.08, %for.body ], [ %N, %entry ]
47  %i.08 = add nsw i64 %i.08.in, -1
48  %arrayidx = getelementptr inbounds double, double* %b, i64 %i.08
49  %0 = load double, double* %arrayidx, align 8
50  %add = fadd double %0, 1.000000e+00
51  %arrayidx1 = getelementptr inbounds double, double* %a, i64 %i.08
52  store double %add, double* %arrayidx1, align 8
53  %cmp = icmp sgt i64 %i.08.in, 1
54  br i1 %cmp, label %for.body, label %for.cond.cleanup, !llvm.loop !0
55}
56
57
58define void @vector_reverse_i64(i64 %N, i64* %a, i64* %b) #0 {
59; CHECK-LABEL: vector_reverse_i64
60; CHECK-LABEL: vector.body:
61; CHECK: %[[ADD:.*]] = add i64 %{{.*}}, %N
62; CHECK-NEXT: %[[GEP:.*]] = getelementptr inbounds i64, i64* %b, i64 %[[ADD]]
63; CHECK-NEXT: %[[VSCALE:.*]] = call i32 @llvm.vscale.i32()
64; CHECK-NEXT: %[[MUL:.*]] = mul i32 %[[VSCALE]], -8
65; CHECK-NEXT: %[[OR:.*]] = or i32 %[[MUL]], 1
66; CHECK-NEXT: %[[SEXT:.*]] = sext i32 %[[OR]] to i64
67; CHECK-NEXT: %[[GEP1:.*]] = getelementptr inbounds i64, i64* %[[GEP]], i64 %[[SEXT]]
68; CHECK-NEXT: %[[CAST:.*]] = bitcast i64* %[[GEP1]] to <vscale x 8 x i64>*
69; CHECK-NEXT: %[[WIDE:.*]] = load <vscale x 8 x i64>, <vscale x 8 x i64>* %[[CAST]], align 8
70; CHECK-NEXT: %[[REVERSE:.*]] = call <vscale x 8 x i64> @llvm.experimental.vector.reverse.nxv8i64(<vscale x 8 x i64> %[[WIDE]])
71; CHECK-NEXT: %[[ADD1:.*]] = add <vscale x 8 x i64> %[[REVERSE]]
72; CHECK-NEXT: %[[GEP2:.*]] = getelementptr inbounds i64, i64* %a, i64 %[[ADD]]
73; CHECK-NEXT: %[[REVERSE6]] = call <vscale x 8 x i64> @llvm.experimental.vector.reverse.nxv8i64(<vscale x 8 x i64> %[[ADD1]])
74; CHECK-NEXT: %[[VSCALE:.*]] = call i32 @llvm.vscale.i32()
75; CHECK-NEXT: %[[MUL1:.*]] = mul i32 %[[VSCALE]], -8
76; CHECK-NEXT: %[[OR1:.*]] = or i32 %[[MUL1]], 1
77; CHECK-NEXT: %[[SEXT1:.*]] = sext i32 %[[OR1]] to i64
78; CHECK-NEXT: %[[GEP3:.*]] = getelementptr inbounds i64, i64* %[[GEP2]], i64 %[[SEXT1]]
79; CHECK-NEXT: %[[CAST1:.*]] = bitcast i64* %[[GEP3]] to <vscale x 8 x i64>*
80; CHECK-NEXT:  store <vscale x 8 x i64> %[[REVERSE6]], <vscale x 8 x i64>* %[[CAST1]], align 8
81
82entry:
83  %cmp8 = icmp sgt i64 %N, 0
84  br i1 %cmp8, label %for.body, label %for.cond.cleanup
85
86for.cond.cleanup:                                 ; preds = %for.body
87  ret void
88
89for.body:                                         ; preds = %entry, %for.body
90  %i.09.in = phi i64 [ %i.09, %for.body ], [ %N, %entry ]
91  %i.09 = add nsw i64 %i.09.in, -1
92  %arrayidx = getelementptr inbounds i64, i64* %b, i64 %i.09
93  %0 = load i64, i64* %arrayidx, align 8
94  %add = add i64 %0, 1
95  %arrayidx2 = getelementptr inbounds i64, i64* %a, i64 %i.09
96  store i64 %add, i64* %arrayidx2, align 8
97  %cmp = icmp sgt i64 %i.09.in, 1
98  br i1 %cmp, label %for.body, label %for.cond.cleanup, !llvm.loop !0
99}
100
101attributes #0 = { "target-cpu"="generic" "target-features"="+neon,+sve" }
102
103!0 = distinct !{!0, !1, !2, !3, !4}
104!1 = !{!"llvm.loop.mustprogress"}
105!2 = !{!"llvm.loop.vectorize.width", i32 8}
106!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}
107!4 = !{!"llvm.loop.vectorize.enable", i1 true}
108
109