1 ; RUN: opt < %s -loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -enable-if-conversion -dce -instcombine -S | FileCheck %s
3 target 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"
4 target triple = "x86_64-apple-macosx10.9.0"
6 ; This is the loop in this example:
8 ;int function0(int *a, int *b, int start, int end) {
10 ; for (int i=start; i<end; ++i) {
13 ; if (a[i] > b[i]) <------ notice the IF inside the loop.
16 ; a[i] = k; <---- K is a phi node that becomes vector-select.
20 ;CHECK-LABEL: @function0(
21 ;CHECK: load <4 x i32>
22 ;CHECK: icmp sgt <4 x i32>
25 ;CHECK: select <4 x i1>
27 define i32 @function0(i32* nocapture %a, i32* nocapture %b, i32 %start, i32 %end) nounwind uwtable ssp {
29 %cmp16 = icmp slt i32 %start, %end
30 br i1 %cmp16, label %for.body.lr.ph, label %for.end
33 %0 = sext i32 %start to i64
37 %indvars.iv = phi i64 [ %0, %for.body.lr.ph ], [ %indvars.iv.next, %if.end ]
38 %arrayidx = getelementptr inbounds i32* %a, i64 %indvars.iv
39 %1 = load i32* %arrayidx, align 4
40 %arrayidx4 = getelementptr inbounds i32* %b, i64 %indvars.iv
41 %2 = load i32* %arrayidx4, align 4
42 %cmp5 = icmp sgt i32 %1, %2
43 br i1 %cmp5, label %if.then, label %if.end
47 %add = add i32 %mul, 3
51 %k.0 = phi i32 [ %add, %if.then ], [ %1, %for.body ]
52 store i32 %k.0, i32* %arrayidx, align 4
53 %indvars.iv.next = add i64 %indvars.iv, 1
54 %3 = trunc i64 %indvars.iv.next to i32
55 %cmp = icmp slt i32 %3, %end
56 br i1 %cmp, label %for.body, label %for.end
64 ; int func(int *A, int n) {
66 ; for (int i = 0; i < n; ++i)
73 ;CHECK-LABEL: @reduction_func(
74 ;CHECK: load <4 x i32>
75 ;CHECK: icmp sgt <4 x i32>
77 ;CHECK: select <4 x i1>
79 define i32 @reduction_func(i32* nocapture %A, i32 %n) nounwind uwtable readonly ssp {
81 %cmp10 = icmp sgt i32 %n, 0
82 br i1 %cmp10, label %for.body, label %for.end
84 for.body: ; preds = %entry, %for.inc
85 %indvars.iv = phi i64 [ %indvars.iv.next, %for.inc ], [ 0, %entry ]
86 %sum.011 = phi i32 [ %sum.1, %for.inc ], [ 0, %entry ]
87 %arrayidx = getelementptr inbounds i32* %A, i64 %indvars.iv
88 %0 = load i32* %arrayidx, align 4
89 %cmp1 = icmp sgt i32 %0, 30
90 br i1 %cmp1, label %if.then, label %for.inc
92 if.then: ; preds = %for.body
93 %add = add i32 %sum.011, 2
94 %add4 = add i32 %add, %0
97 for.inc: ; preds = %for.body, %if.then
98 %sum.1 = phi i32 [ %add4, %if.then ], [ %sum.011, %for.body ]
99 %indvars.iv.next = add i64 %indvars.iv, 1
100 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
101 %exitcond = icmp eq i32 %lftr.wideiv, %n
102 br i1 %exitcond, label %for.end, label %for.body
104 for.end: ; preds = %for.inc, %entry
105 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %sum.1, %for.inc ]
109 @a = common global [1 x i32*] zeroinitializer, align 8
110 @c = common global i32* null, align 8
112 ; We use to if convert this loop. This is not safe because there is a trapping
113 ; constant expression.
116 ; CHECK-LABEL: trapping_constant_expression
117 ; CHECK-NOT: or <4 x i32>
119 define i32 @trapping_constant_expression() {
124 %inc3 = phi i32 [ 0, %entry ], [ %inc, %cond.end ]
125 %or2 = phi i32 [ 0, %entry ], [ %or, %cond.end ]
126 br i1 icmp eq (i32** getelementptr inbounds ([1 x i32*]* @a, i64 0, i64 0), i32** @c), label %cond.false, label %cond.end
132 %cond = phi i32 [ sdiv (i32 1, i32 zext (i1 icmp eq (i32** getelementptr inbounds ([1 x i32*]* @a, i64 0, i64 0), i32** @c) to i32)), %cond.false ], [ 0, %for.body ]
133 %or = or i32 %or2, %cond
134 %inc = add nsw i32 %inc3, 1
135 %cmp = icmp slt i32 %inc, 128
136 br i1 %cmp, label %for.body, label %for.end
142 ; Neither should we if-convert if there is an instruction operand that is a
143 ; trapping constant expression.
146 ; CHECK-LABEL: trapping_constant_expression2
147 ; CHECK-NOT: or <4 x i32>
149 define i32 @trapping_constant_expression2() {
154 %inc3 = phi i32 [ 0, %entry ], [ %inc, %cond.end ]
155 %or2 = phi i32 [ 0, %entry ], [ %or, %cond.end ]
156 br i1 icmp eq (i32** getelementptr inbounds ([1 x i32*]* @a, i64 0, i64 0), i32** @c), label %cond.false, label %cond.end
159 %cond.1 = or i32 %inc3, sdiv (i32 1, i32 zext (i1 icmp eq (i32** getelementptr inbounds ([1 x i32*]* @a, i64 0, i64 1), i32** @c) to i32))
163 %cond = phi i32 [ %cond.1, %cond.false ], [ %inc3, %for.body ]
164 %or = or i32 %or2, %cond
165 %inc = add nsw i32 %inc3, 1
166 %cmp = icmp slt i32 %inc, 128
167 br i1 %cmp, label %for.body, label %for.end