1 ; Test VLA for reverse with fixed size vector
2 ; This is the loop in c++ being vectorize in this file with
4 ; #pragma clang loop vectorize_width(8, fixed)
5 ; for (int i = N-1; i >= 0; --i)
8 ; RUN: opt -loop-vectorize -dce -mtriple aarch64-linux-gnu -S < %s | FileCheck %s
10 define void @vector_reverse_f64(i64 %N, double* %a, double* %b) #0 {
11 ; CHECK-LABEL: vector_reverse_f64
12 ; CHECK-LABEL: vector.body
13 ; CHECK: %[[GEP:.*]] = getelementptr inbounds double, double* %{{.*}}, i32 0
14 ; CHECK-NEXT: %[[GEP1:.*]] = getelementptr inbounds double, double* %[[GEP]], i32 -7
15 ; CHECK-NEXT: %[[CAST:.*]] = bitcast double* %[[GEP1]] to <8 x double>*
16 ; CHECK-NEXT: %[[WIDE:.*]] = load <8 x double>, <8 x double>* %[[CAST]], align 8
17 ; CHECK-NEXT: %[[REVERSE:.*]] = shufflevector <8 x double> %[[WIDE]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
18 ; CHECK-NEXT: %[[FADD:.*]] = fadd <8 x double> %[[REVERSE]]
19 ; CHECK-NEXT: %[[GEP2:.*]] = getelementptr inbounds double, double* {{.*}}, i64 {{.*}}
20 ; CHECK-NEXT: %[[REVERSE6:.*]] = shufflevector <8 x double> %[[FADD]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
21 ; CHECK-NEXT: %[[GEP3:.*]] = getelementptr inbounds double, double* %[[GEP2]], i32 0
22 ; CHECK-NEXT: %[[GEP4:.*]] = getelementptr inbounds double, double* %[[GEP3]], i32 -7
23 ; CHECK-NEXT: %[[CAST:.*]] = bitcast double* %[[GEP4]] to <8 x double>*
24 ; CHECK-NEXT: store <8 x double> %[[REVERSE6]], <8 x double>* %[[CAST]], align 8
27 %cmp7 = icmp sgt i64 %N, 0
28 br i1 %cmp7, label %for.body, label %for.cond.cleanup
30 for.cond.cleanup: ; preds = %for.cond.cleanup, %entry
33 for.body: ; preds = %entry, %for.body
34 %i.08.in = phi i64 [ %i.08, %for.body ], [ %N, %entry ]
35 %i.08 = add nsw i64 %i.08.in, -1
36 %arrayidx = getelementptr inbounds double, double* %b, i64 %i.08
37 %0 = load double, double* %arrayidx, align 8
38 %add = fadd double %0, 1.000000e+00
39 %arrayidx1 = getelementptr inbounds double, double* %a, i64 %i.08
40 store double %add, double* %arrayidx1, align 8
41 %cmp = icmp sgt i64 %i.08.in, 1
42 br i1 %cmp, label %for.body, label %for.cond.cleanup, !llvm.loop !0
45 define void @vector_reverse_i64(i64 %N, i64* %a, i64* %b) #0 {
46 ; CHECK-LABEL: vector_reverse_i64
47 ; CHECK-LABEL: vector.body
48 ; CHECK: %[[GEP:.*]] = getelementptr inbounds i64, i64* %{{.*}}, i32 0
49 ; CHECK-NEXT: %[[GEP1:.*]] = getelementptr inbounds i64, i64* %[[GEP]], i32 -7
50 ; CHECK-NEXT: %[[CAST:.*]] = bitcast i64* %[[GEP1]] to <8 x i64>*
51 ; CHECK-NEXT: %[[WIDE:.*]] = load <8 x i64>, <8 x i64>* %[[CAST]], align 8
52 ; CHECK-NEXT: %[[REVERSE:.*]] = shufflevector <8 x i64> %[[WIDE]], <8 x i64> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
53 ; CHECK-NEXT: %[[FADD:.*]] = add <8 x i64> %[[REVERSE]]
54 ; CHECK-NEXT: %[[GEP2:.*]] = getelementptr inbounds i64, i64* {{.*}}, i64 {{.*}}
55 ; CHECK-NEXT: %[[REVERSE6:.*]] = shufflevector <8 x i64> %[[FADD]], <8 x i64> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
56 ; CHECK-NEXT: %[[GEP3:.*]] = getelementptr inbounds i64, i64* %[[GEP2]], i32 0
57 ; CHECK-NEXT: %[[GEP4:.*]] = getelementptr inbounds i64, i64* %[[GEP3]], i32 -7
58 ; CHECK-NEXT: %[[CAST1:.*]] = bitcast i64* %[[GEP4]] to <8 x i64>*
59 ; CHECK-NEXT: store <8 x i64> %[[REVERSE6]], <8 x i64>* %[[CAST1]], align 8
62 %cmp8 = icmp sgt i64 %N, 0
63 br i1 %cmp8, label %for.body, label %for.cond.cleanup
65 for.cond.cleanup: ; preds = %for.cond.cleanup, %entry
68 for.body: ; preds = %entry, %for.body
69 %i.09.in = phi i64 [ %i.09, %for.body ], [ %N, %entry ]
70 %i.09 = add nsw i64 %i.09.in, -1
71 %arrayidx = getelementptr inbounds i64, i64* %b, i64 %i.09
72 %0 = load i64, i64* %arrayidx, align 8
74 %arrayidx2 = getelementptr inbounds i64, i64* %a, i64 %i.09
75 store i64 %add, i64* %arrayidx2, align 8
76 %cmp = icmp sgt i64 %i.09.in, 1
77 br i1 %cmp, label %for.body, label %for.cond.cleanup, !llvm.loop !0
80 attributes #0 = { "target-cpu"="generic" "target-features"="+neon,+sve" }
82 !0 = distinct !{!0, !1, !2, !3, !4}
83 !1 = !{!"llvm.loop.mustprogress"}
84 !2 = !{!"llvm.loop.vectorize.width", i32 8}
85 !3 = !{!"llvm.loop.vectorize.scalable.enable", i1 false}
86 !4 = !{!"llvm.loop.vectorize.enable", i1 true}