1; RUN: opt < %s  -loop-vectorize -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx -debug-only=loop-vectorize -stats -S 2>&1 | FileCheck %s
2; REQUIRES: asserts
3
4; CHECK: LV: Loop hints: force=enabled
5; CHECK: LV: Loop hints: force=?
6; No more loops in the module
7; CHECK-NOT: LV: Loop hints: force=
8; CHECK: 2 loop-vectorize               - Number of loops analyzed for vectorization
9; CHECK: 1 loop-vectorize               - Number of loops vectorized
10
11target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
12target triple = "x86_64-apple-macosx10.8.0"
13
14;
15; The source code for the test:
16;
17; #include <math.h>
18; void foo(float* restrict A, float * restrict B)
19; {
20;   for (int i = 0; i < 1000; i+=2) A[i] = sinf(B[i]);
21; }
22;
23
24;
25; This loop will be vectorized, although the scalar cost is lower than any of vector costs, but vectorization is explicitly forced in metadata.
26;
27
28define void @vectorized(float* noalias nocapture %A, float* noalias nocapture %B) {
29entry:
30  br label %for.body
31
32for.body:
33  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
34  %arrayidx = getelementptr inbounds float, float* %B, i64 %indvars.iv
35  %0 = load float, float* %arrayidx, align 4, !llvm.mem.parallel_loop_access !1
36  %call = tail call float @llvm.sin.f32(float %0)
37  %arrayidx2 = getelementptr inbounds float, float* %A, i64 %indvars.iv
38  store float %call, float* %arrayidx2, align 4, !llvm.mem.parallel_loop_access !1
39  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 2
40  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
41  %exitcond = icmp eq i32 %lftr.wideiv, 1000
42  br i1 %exitcond, label %for.end.loopexit, label %for.body, !llvm.loop !1
43
44for.end.loopexit:
45  br label %for.end
46
47for.end:
48  ret void
49}
50
51!1 = !{!1, !2}
52!2 = !{!"llvm.loop.vectorize.enable", i1 true}
53
54;
55; This method will not be vectorized, as scalar cost is lower than any of vector costs.
56;
57
58define void @not_vectorized(float* noalias nocapture %A, float* noalias nocapture %B) {
59entry:
60  br label %for.body
61
62for.body:
63  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
64  %arrayidx = getelementptr inbounds float, float* %B, i64 %indvars.iv
65  %0 = load float, float* %arrayidx, align 4, !llvm.mem.parallel_loop_access !3
66  %call = tail call float @llvm.sin.f32(float %0)
67  %arrayidx2 = getelementptr inbounds float, float* %A, i64 %indvars.iv
68  store float %call, float* %arrayidx2, align 4, !llvm.mem.parallel_loop_access !3
69  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 2
70  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
71  %exitcond = icmp eq i32 %lftr.wideiv, 1000
72  br i1 %exitcond, label %for.end.loopexit, label %for.body, !llvm.loop !3
73
74for.end.loopexit:
75  br label %for.end
76
77for.end:
78  ret void
79}
80
81declare float @llvm.sin.f32(float) nounwind readnone
82
83; Dummy metadata
84!3 = !{!3}
85
86