1; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
2
3target 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"
4target triple = "x86_64-apple-macosx10.8.0"
5
6; double foo(double * restrict b,  double * restrict a, int n, int m) {
7;   double r=a[1];
8;   double g=a[0];
9;   double x;
10;   for (int i=0; i < 100; i++) {
11;     r += 10;
12;     g += 10;
13;     r *= 4;
14;     g *= 4;
15;     x = g; <----- external user!
16;     r += 4;
17;     g += 4;
18;   }
19;   b[0] = g;
20;   b[1] = r;
21;
22;   return x; <-- must extract here!
23; }
24
25;CHECK: ext_user
26;CHECK: phi <2 x double>
27;CHECK: fadd <2 x double>
28;CHECK: fmul <2 x double>
29;CHECK: br
30;CHECK: store <2 x double>
31;CHECK: extractelement <2 x double>
32;CHECK: ret double
33
34define double @ext_user(double* noalias nocapture %B, double* noalias nocapture %A, i32 %n, i32 %m) {
35entry:
36  %arrayidx = getelementptr inbounds double, double* %A, i64 1
37  %0 = load double, double* %arrayidx, align 8
38  %1 = load double, double* %A, align 8
39  br label %for.body
40
41for.body:                                         ; preds = %for.body, %entry
42  %i.020 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
43  %G.019 = phi double [ %1, %entry ], [ %add5, %for.body ]
44  %R.018 = phi double [ %0, %entry ], [ %add4, %for.body ]
45  %add = fadd double %R.018, 1.000000e+01
46  %add2 = fadd double %G.019, 1.000000e+01
47  %mul = fmul double %add, 4.000000e+00
48  %mul3 = fmul double %add2, 4.000000e+00
49  %add4 = fadd double %mul, 4.000000e+00
50  %add5 = fadd double %mul3, 4.000000e+00
51  %inc = add nsw i32 %i.020, 1
52  %exitcond = icmp eq i32 %inc, 100
53  br i1 %exitcond, label %for.end, label %for.body
54
55for.end:                                          ; preds = %for.body
56  store double %add5, double* %B, align 8
57  %arrayidx7 = getelementptr inbounds double, double* %B, i64 1
58  store double %add4, double* %arrayidx7, align 8
59  ret double %mul3
60}
61
62; A need-to-gather entry cannot be an external use of the scalar element.
63; Instead the insertelement instructions of the need-to-gather entry are the
64; external users.
65; This test would assert because we would keep the scalar fpext and fadd alive.
66; PR18129
67
68; CHECK-LABEL: needtogather
69define i32 @needtogather(double *noalias %a, i32 *noalias %b,  float * noalias %c,
70                i32 * noalias %d) {
71entry:
72  %0 = load i32, i32* %d, align 4
73  %conv = sitofp i32 %0 to float
74  %1 = load float, float* %c
75  %sub = fsub float 0.000000e+00, %1
76  %mul = fmul float %sub, 0.000000e+00
77  %add = fadd float %conv, %mul
78  %conv1 = fpext float %add to double
79  %sub3 = fsub float 1.000000e+00, %1
80  %mul4 = fmul float %sub3, 0.000000e+00
81  %add5 = fadd float %conv, %mul4
82  %conv6 = fpext float %add5 to double
83  %tobool = fcmp une float %add, 0.000000e+00
84  br i1 %tobool, label %if.then, label %if.end
85
86if.then:
87  br label %if.end
88
89if.end:
90  %storemerge = phi double [ %conv6, %if.then ], [ %conv1, %entry ]
91  %e.0 = phi double [ %conv1, %if.then ], [ %conv6, %entry ]
92  store double %storemerge, double* %a, align 8
93  %conv7 = fptosi double %e.0 to i32
94  store i32 %conv7, i32* %b, align 4
95  ret i32 undef
96}
97