Debian Codegraph @

llvm/test/Transforms/LoopInterchange/inner-only-reductions.ll

raw
1 ; RUN: opt < %s -passes=loop-interchange -cache-line-size=64 -pass-remarks-missed='loop-interchange' -pass-remarks-output=%t -S \
2 ; RUN: -verify-dom-info -verify-loop-info -verify-loop-lcssa 2>&1 | FileCheck -check-prefix=IR %s
3 ; RUN: FileCheck --input-file=%t %s
4
5 ; Both tests should be rejected as interchange candidates. For now, they are
6 ; rejected for dependence analysis reasons, but that's because support for 'S'
7 ; scalar dependencies was removed. When that is properly, the inner loop only
8 ; reductions should still not be supported currently, see discussion at D53027
9 ; for more information on the required checks.
10
11 @A = common global [500 x [500 x i32]] zeroinitializer
12 @X = common global i32 0
13 @B = common global [500 x [500 x i32]] zeroinitializer
14 @Y = common global i32 0
15
16 ;; global X
17
18 ;; for( int i=1;i<N;i++)
19 ;; for( int j=1;j<N;j++)
20 ;; X+=A[j][i];
21
22 ; CHECK: --- !Missed
23 ; CHECK-NEXT: Pass: loop-interchange
24 ; CHECK-NEXT: Name: Dependence
25 ; CHECK-NEXT: Function: reduction_01
26
27 ; IR-LABEL: @reduction_01(
28 ; IR-NOT: split
29
30 define void @reduction_01(i32 %N) {
31 entry:
32 %cmp16 = icmp sgt i32 %N, 1
33 br i1 %cmp16, label %for.body3.lr.ph, label %for.end8
34
35 for.body3.lr.ph: ; preds = %for.cond1.for.inc6_crit_edge, %entry
36 %indvars.iv18 = phi i64 [ %indvars.iv.next19, %for.cond1.for.inc6_crit_edge ], [ 1, %entry ]
37 %X.promoted = load i32, ptr @X
38 br label %for.body3
39
40 for.body3: ; preds = %for.body3, %for.body3.lr.ph
41 %indvars.iv = phi i64 [ 1, %for.body3.lr.ph ], [ %indvars.iv.next, %for.body3 ]
42 %add15 = phi i32 [ %X.promoted, %for.body3.lr.ph ], [ %add, %for.body3 ]
43 %arrayidx5 = getelementptr inbounds [500 x [500 x i32]], ptr @A, i64 0, i64 %indvars.iv, i64 %indvars.iv18
44 %0 = load i32, ptr %arrayidx5
45 %add = add nsw i32 %add15, %0
46 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
47 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
48 %exitcond = icmp eq i32 %lftr.wideiv, %N
49 br i1 %exitcond, label %for.cond1.for.inc6_crit_edge, label %for.body3
50
51 for.cond1.for.inc6_crit_edge: ; preds = %for.body3
52 %add.lcssa = phi i32 [ %add, %for.body3 ]
53 store i32 %add.lcssa, ptr @X
54 %indvars.iv.next19 = add nuw nsw i64 %indvars.iv18, 1
55 %lftr.wideiv20 = trunc i64 %indvars.iv.next19 to i32
56 %exitcond21 = icmp eq i32 %lftr.wideiv20, %N
57 br i1 %exitcond21, label %for.end8, label %for.body3.lr.ph
58
59 for.end8: ; preds = %for.cond1.for.inc6_crit_edge, %entry
60 ret void
61 }
62
63 ;; Not tightly nested. Do not interchange.
64 ;; for( int i=1;i<N;i++)
65 ;; for( int j=1;j<N;j++) {
66 ;; for( int k=1;k<N;k++) {
67 ;; X+=A[k][j];
68 ;; }
69 ;; Y+=B[j][i];
70 ;; }
71
72 ;; Not tightly nested. Do not interchange.
73 ;; Not interchanged hence the phi's in the inner loop will not be split.
74
75 ; CHECK: --- !Missed
76 ; CHECK-NEXT: Pass: loop-interchange
77 ; CHECK-NEXT: Name: UnsupportedPHIOuter
78 ; CHECK-NEXT: Function: reduction_03
79
80 ; IR-LABEL: @reduction_03(
81 ; IR-NOT: split
82
83 define void @reduction_03(i32 %N) {
84 entry:
85 %cmp35 = icmp sgt i32 %N, 1
86 br i1 %cmp35, label %for.cond4.preheader.lr.ph, label %for.end19
87
88 for.cond4.preheader.lr.ph: ; preds = %for.cond1.for.inc17_crit_edge, %entry
89 %indvars.iv41 = phi i64 [ %indvars.iv.next42, %for.cond1.for.inc17_crit_edge ], [ 1, %entry ]
90 %Y.promoted = load i32, ptr @Y
91 br label %for.body6.lr.ph
92
93 for.body6.lr.ph: ; preds = %for.cond4.for.end_crit_edge, %for.cond4.preheader.lr.ph
94 %indvars.iv37 = phi i64 [ 1, %for.cond4.preheader.lr.ph ], [ %indvars.iv.next38, %for.cond4.for.end_crit_edge ]
95 %add1334 = phi i32 [ %Y.promoted, %for.cond4.preheader.lr.ph ], [ %add13, %for.cond4.for.end_crit_edge ]
96 %X.promoted = load i32, ptr @X
97 br label %for.body6
98
99 for.body6: ; preds = %for.body6, %for.body6.lr.ph
100 %indvars.iv = phi i64 [ 1, %for.body6.lr.ph ], [ %indvars.iv.next, %for.body6 ]
101 %arrayidx8 = getelementptr inbounds [500 x [500 x i32]], ptr @A, i64 0, i64 %indvars.iv, i64 %indvars.iv37
102 %0 = load i32, ptr %arrayidx8
103 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
104 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
105 %exitcond = icmp eq i32 %lftr.wideiv, %N
106 br i1 %exitcond, label %for.cond4.for.end_crit_edge, label %for.body6
107
108 for.cond4.for.end_crit_edge: ; preds = %for.body6
109 %arrayidx12 = getelementptr inbounds [500 x [500 x i32]], ptr @B, i64 0, i64 %indvars.iv37, i64 %indvars.iv41
110 %1 = load i32, ptr %arrayidx12
111 %add13 = add nsw i32 %add1334, %1
112 %indvars.iv.next38 = add nuw nsw i64 %indvars.iv37, 1
113 %lftr.wideiv39 = trunc i64 %indvars.iv.next38 to i32
114 %exitcond40 = icmp eq i32 %lftr.wideiv39, %N
115 br i1 %exitcond40, label %for.cond1.for.inc17_crit_edge, label %for.body6.lr.ph
116
117 for.cond1.for.inc17_crit_edge: ; preds = %for.cond4.for.end_crit_edge
118 %add13.lcssa = phi i32 [ %add13, %for.cond4.for.end_crit_edge ]
119 store i32 %add13.lcssa, ptr @Y
120 %indvars.iv.next42 = add nuw nsw i64 %indvars.iv41, 1
121 %lftr.wideiv43 = trunc i64 %indvars.iv.next42 to i32
122 %exitcond44 = icmp eq i32 %lftr.wideiv43, %N
123 br i1 %exitcond44, label %for.end19, label %for.cond4.preheader.lr.ph
124
125 for.end19: ; preds = %for.cond1.for.inc17_crit_edge, %entry
126 ret void
127 }