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-macosx10.7.0"
5
6; int bar(double *A, int d) {
7;   double A0 = A[0];
8;   double A1 = A[1];
9;   float F0 = A0;
10;   float F1 = A1;
11;   if (d) foo(); <----- This splits the blocks
12;   F0+=4.0;
13;   F1+=5.0;
14;   A[8] = 9.0 + F0;
15;   A[9] = 5.0 + F1;
16; }
17
18
19;CHECK-LABEL: @bar(
20;CHECK: load <2 x double>
21;CHECK: fptrunc <2 x double>
22;CHECK: call i32
23;CHECK: fadd <2 x float>
24;CHECK: fpext <2 x float>
25;CHECK: store <2 x double>
26;CHECK: ret
27define i32 @bar(double* nocapture %A, i32 %d) {
28  %1 = load double* %A, align 8
29  %2 = getelementptr inbounds double* %A, i64 1
30  %3 = load double* %2, align 8
31  %4 = fptrunc double %1 to float
32  %5 = fptrunc double %3 to float
33  %6 = icmp eq i32 %d, 0
34  br i1 %6, label %9, label %7
35
36; <label>:7                                       ; preds = %0
37  %8 = tail call i32 (...)* @foo()
38  br label %9
39
40; <label>:9                                       ; preds = %0, %7
41  %10 = fadd float %4, 4.000000e+00
42  %11 = fadd float %5, 5.000000e+00
43  %12 = fpext float %10 to double
44  %13 = fadd double %12, 9.000000e+00
45  %14 = getelementptr inbounds double* %A, i64 8
46  store double %13, double* %14, align 8
47  %15 = fpext float %11 to double
48  %16 = fadd double %15, 5.000000e+00
49  %17 = getelementptr inbounds double* %A, i64 9
50  store double %16, double* %17, align 8
51  ret i32 undef
52}
53
54declare i32 @foo(...)
55
56