• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: opt -S -loop-fusion -loop-fusion-peel-max-count=3 < %s | FileCheck %s
2
3; Tests if we are able to fuse two guarded loops which have constant but
4; different trip counts. The first two iterations of the first loop should be
5; peeled off, and then the loops should be fused together.
6
7@B = common global [1024 x i32] zeroinitializer, align 16
8
9; CHECK-LABEL: void @main(i32* noalias %A)
10; CHECK-NEXT:  entry:
11; CHECK:         br i1 %cmp4, label %for.first.entry, label %for.end
12; CHECK:       for.first.entry
13; CHECK-NEXT:    br label %for.first.peel.begin
14; CHECK:       for.first.peel.begin:
15; CHECK-NEXT:    br label %for.first.peel
16; CHECK:       for.first.peel:
17; CHECK:         br label %for.first.peel.next
18; CHECK:       for.first.peel.next:
19; CHECK-NEXT:    br label %for.first.peel2
20; CHECK:       for.first.peel2:
21; CHECK:         br label %for.first.peel.next1
22; CHECK:       for.first.peel.next1:
23; CHECK-NEXT:    br label %for.first.peel.next11
24; CHECK:       for.first.peel.next11:
25; CHECK-NEXT:    br label %for.first.entry.peel.newph
26; CHECK:       for.first.entry.peel.newph:
27; CHECK:         br label %for.first
28; CHECK:       for.first:
29; CHECK:         br i1 %cmp3, label %for.first, label %for.second.exit
30; CHECK:       for.second.exit:
31; CHECK:         br label %for.end
32; CHECK:       for.end:
33; CHECK-NEXT:    ret void
34
35define void @main(i32* noalias %A) {
36entry:
37  %cmp4 = icmp slt i64 0, 45
38  br i1 %cmp4, label %for.first.entry, label %for.second.guard
39
40for.first.entry:                                 ; preds = %entry
41  br label %for.first
42
43for.first:                                       ; preds = %for.first.entry, %for.first
44  %i.05 = phi i64 [ %inc, %for.first ], [ 0, %for.first.entry ]
45  %sub = sub nsw i64 %i.05, 3
46  %add = add nsw i64 %i.05, 3
47  %mul = mul nsw i64 %sub, %add
48  %rem = srem i64 %mul, %i.05
49  %conv = trunc i64 %rem to i32
50  %arrayidx = getelementptr inbounds i32, i32* %A, i64 %i.05
51  store i32 %conv, i32* %arrayidx, align 4
52  %inc = add nsw i64 %i.05, 1
53  %cmp = icmp slt i64 %inc, 45
54  br i1 %cmp, label %for.first, label %for.first.exit
55
56for.first.exit:                                  ; preds = %for.first
57  br label %for.second.guard
58
59for.second.guard:                                ; preds = %for.first.exit, %entry
60  %cmp31 = icmp slt i64 2, 45
61  br i1 %cmp31, label %for.second.entry, label %for.end
62
63for.second.entry:                                ; preds = %for.second.guard
64  br label %for.second
65
66for.second:                                      ; preds = %for.second.entry, %for.second
67  %i1.02 = phi i64 [ %inc14, %for.second ], [ 2, %for.second.entry ]
68  %sub7 = sub nsw i64 %i1.02, 3
69  %add8 = add nsw i64 %i1.02, 3
70  %mul9 = mul nsw i64 %sub7, %add8
71  %rem10 = srem i64 %mul9, %i1.02
72  %conv11 = trunc i64 %rem10 to i32
73  %arrayidx12 = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 %i1.02
74  store i32 %conv11, i32* %arrayidx12, align 4
75  %inc14 = add nsw i64 %i1.02, 1
76  %cmp3 = icmp slt i64 %inc14, 45
77  br i1 %cmp3, label %for.second, label %for.second.exit
78
79for.second.exit:                                 ; preds = %for.second
80  br label %for.end
81
82for.end:                                         ; preds = %for.second.exit, %for.second.guard
83  ret void
84}
85