1 ; RUN: opt < %s -loop-vectorize -mcpu=core2 -debug-only=loop-vectorize 2>&1 -S | FileCheck %s
3 ; Make sure we use the right select kind when querying select costs.
5 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"
6 target triple = "x86_64-apple-macosx10.8.0"
8 @a = common global [2048 x i32] zeroinitializer, align 16
9 @b = common global [2048 x i32] zeroinitializer, align 16
10 @c = common global [2048 x i32] zeroinitializer, align 16
12 ; CHECK: Checking a loop in "scalarselect"
13 define void @scalarselect(i1 %cond) nounwind uwtable ssp {
16 ; <label>:1 ; preds = %1, %0
17 %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ]
18 %2 = getelementptr inbounds [2048 x i32]* @b, i64 0, i64 %indvars.iv
19 %3 = load i32* %2, align 4
20 %4 = getelementptr inbounds [2048 x i32]* @c, i64 0, i64 %indvars.iv
21 %5 = load i32* %4, align 4
22 %6 = add nsw i32 %5, %3
23 %7 = getelementptr inbounds [2048 x i32]* @a, i64 0, i64 %indvars.iv
24 ; A scalar select has a cost of 1 on core2
25 ; CHECK: cost of 1 for VF 2 {{.*}} select i1 %cond, i32 %6, i32 0
26 %sel = select i1 %cond, i32 %6, i32 zeroinitializer
27 store i32 %sel, i32* %7, align 4
28 %indvars.iv.next = add i64 %indvars.iv, 1
29 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
30 %exitcond = icmp eq i32 %lftr.wideiv, 256
31 br i1 %exitcond, label %8, label %1
33 ; <label>:8 ; preds = %1
37 ; CHECK: Checking a loop in "vectorselect"
38 define void @vectorselect(i1 %cond) nounwind uwtable ssp {
41 ; <label>:1 ; preds = %1, %0
42 %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ]
43 %2 = getelementptr inbounds [2048 x i32]* @b, i64 0, i64 %indvars.iv
44 %3 = load i32* %2, align 4
45 %4 = getelementptr inbounds [2048 x i32]* @c, i64 0, i64 %indvars.iv
46 %5 = load i32* %4, align 4
47 %6 = add nsw i32 %5, %3
48 %7 = getelementptr inbounds [2048 x i32]* @a, i64 0, i64 %indvars.iv
49 %8 = icmp ult i64 %indvars.iv, 8
50 ; A vector select has a cost of 4 on core2
51 ; CHECK: cost of 4 for VF 2 {{.*}} select i1 %8, i32 %6, i32 0
52 %sel = select i1 %8, i32 %6, i32 zeroinitializer
53 store i32 %sel, i32* %7, align 4
54 %indvars.iv.next = add i64 %indvars.iv, 1
55 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
56 %exitcond = icmp eq i32 %lftr.wideiv, 256
57 br i1 %exitcond, label %9, label %1
59 ; <label>:9 ; preds = %1