1 ; RUN: opt < %s -debug-only=loop-vectorize -loop-vectorize -vectorizer-maximize-bandwidth -O2 -mtriple=powerpc64-unknown-linux -S -mcpu=pwr8 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-PWR8
2 ; RUN: opt < %s -debug-only=loop-vectorize -loop-vectorize -vectorizer-maximize-bandwidth -O2 -mtriple=powerpc64le-unknown-linux -S -mcpu=pwr9 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-PWR9
5 @a = global [1024 x i8] zeroinitializer, align 16
6 @b = global [1024 x i8] zeroinitializer, align 16
11 ; CHECK-PWR8: Setting best plan to VF=16, UF=4
13 ; CHECK-PWR9: Setting best plan to VF=8, UF=8
20 %add.lcssa = phi i32 [ %add, %for.body ]
24 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
25 %s.015 = phi i32 [ 0, %entry ], [ %add, %for.body ]
26 %arrayidx = getelementptr inbounds [1024 x i8], [1024 x i8]* @a, i64 0, i64 %indvars.iv
27 %0 = load i8, i8* %arrayidx, align 1
28 %conv = zext i8 %0 to i32
29 %arrayidx2 = getelementptr inbounds [1024 x i8], [1024 x i8]* @b, i64 0, i64 %indvars.iv
30 %1 = load i8, i8* %arrayidx2, align 1
31 %conv3 = zext i8 %1 to i32
32 %sub = sub nsw i32 %conv, %conv3
33 %ispos = icmp sgt i32 %sub, -1
34 %neg = sub nsw i32 0, %sub
35 %2 = select i1 %ispos, i32 %sub, i32 %neg
36 %add = add nsw i32 %2, %s.015
37 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
38 %exitcond = icmp eq i64 %indvars.iv.next, 1024
39 br i1 %exitcond, label %for.cond.cleanup, label %for.body
43 ; For indvars.iv used in a computating chain only feeding into getelementptr or cmp,
44 ; it will not have vector version and the vector register usage will not exceed the
45 ; available vector register number.
49 ; CHECK: Setting best plan to VF=16, UF=4
54 for.cond.cleanup: ; preds = %for.body
55 %add.lcssa = phi i32 [ %add, %for.body ]
58 for.body: ; preds = %for.body, %entry
59 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
60 %s.015 = phi i32 [ 0, %entry ], [ %add, %for.body ]
61 %tmp1 = add nsw i64 %indvars.iv, 3
62 %arrayidx = getelementptr inbounds [1024 x i8], [1024 x i8]* @a, i64 0, i64 %tmp1
63 %tmp = load i8, i8* %arrayidx, align 1
64 %conv = zext i8 %tmp to i32
65 %tmp2 = add nsw i64 %indvars.iv, 2
66 %arrayidx2 = getelementptr inbounds [1024 x i8], [1024 x i8]* @b, i64 0, i64 %tmp2
67 %tmp3 = load i8, i8* %arrayidx2, align 1
68 %conv3 = zext i8 %tmp3 to i32
69 %sub = sub nsw i32 %conv, %conv3
70 %ispos = icmp sgt i32 %sub, -1
71 %neg = sub nsw i32 0, %sub
72 %tmp4 = select i1 %ispos, i32 %sub, i32 %neg
73 %add = add nsw i32 %tmp4, %s.015
74 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
75 %exitcond = icmp eq i64 %indvars.iv.next, 1024
76 br i1 %exitcond, label %for.cond.cleanup, label %for.body
79 define i64 @bar(i64* nocapture %a) {
82 ; CHECK: Setting best plan to VF=2, UF=12
88 %add2.lcssa = phi i64 [ %add2, %for.body ]
92 %i.012 = phi i64 [ 0, %entry ], [ %inc, %for.body ]
93 %s.011 = phi i64 [ 0, %entry ], [ %add2, %for.body ]
94 %arrayidx = getelementptr inbounds i64, i64* %a, i64 %i.012
95 %0 = load i64, i64* %arrayidx, align 8
96 %add = add nsw i64 %0, %i.012
97 store i64 %add, i64* %arrayidx, align 8
98 %add2 = add nsw i64 %add, %s.011
99 %inc = add nuw nsw i64 %i.012, 1
100 %exitcond = icmp eq i64 %inc, 1024
101 br i1 %exitcond, label %for.cond.cleanup, label %for.body
104 @d = external global [0 x i64], align 8
105 @e = external global [0 x i32], align 4
106 @c = external global [0 x i32], align 4
108 define void @hoo(i32 %n) {
110 ; CHECK: Setting best plan to VF=1, UF=12
115 for.body: ; preds = %for.body, %entry
116 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
117 %arrayidx = getelementptr inbounds [0 x i64], [0 x i64]* @d, i64 0, i64 %indvars.iv
118 %tmp = load i64, i64* %arrayidx, align 8
119 %arrayidx1 = getelementptr inbounds [0 x i32], [0 x i32]* @e, i64 0, i64 %tmp
120 %tmp1 = load i32, i32* %arrayidx1, align 4
121 %arrayidx3 = getelementptr inbounds [0 x i32], [0 x i32]* @c, i64 0, i64 %indvars.iv
122 store i32 %tmp1, i32* %arrayidx3, align 4
123 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
124 %exitcond = icmp eq i64 %indvars.iv.next, 10000
125 br i1 %exitcond, label %for.end, label %for.body
127 for.end: ; preds = %for.body