1; RUN: opt < %s -basicaa -loop-interchange -S | FileCheck %s
2
3@A = common global [500 x [500 x i32]] zeroinitializer
4@X = common global i32 0
5@B = common global [500 x [500 x i32]] zeroinitializer
6@Y = common global i32 0
7
8;;  for( int i=1;i<N;i++)
9;;    for( int j=1;j<N;j++)
10;;      X+=A[j][i];
11
12define void @reduction_01(i32 %N) {
13entry:
14  %cmp16 = icmp sgt i32 %N, 1
15  br i1 %cmp16, label %for.body3.lr.ph, label %for.end8
16
17for.body3.lr.ph:                                  ; preds = %entry, %for.cond1.for.inc6_crit_edge
18  %indvars.iv18 = phi i64 [ %indvars.iv.next19, %for.cond1.for.inc6_crit_edge ], [ 1, %entry ]
19  %X.promoted = load i32, i32* @X
20  br label %for.body3
21
22for.body3:                                        ; preds = %for.body3, %for.body3.lr.ph
23  %indvars.iv = phi i64 [ 1, %for.body3.lr.ph ], [ %indvars.iv.next, %for.body3 ]
24  %add15 = phi i32 [ %X.promoted, %for.body3.lr.ph ], [ %add, %for.body3 ]
25  %arrayidx5 = getelementptr inbounds [500 x [500 x i32]], [500 x [500 x i32]]* @A, i64 0, i64 %indvars.iv, i64 %indvars.iv18
26  %0 = load i32, i32* %arrayidx5
27  %add = add nsw i32 %add15, %0
28  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
29  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
30  %exitcond = icmp eq i32 %lftr.wideiv, %N
31  br i1 %exitcond, label %for.cond1.for.inc6_crit_edge, label %for.body3
32
33for.cond1.for.inc6_crit_edge:                     ; preds = %for.body3
34  store i32 %add, i32* @X
35  %indvars.iv.next19 = add nuw nsw i64 %indvars.iv18, 1
36  %lftr.wideiv20 = trunc i64 %indvars.iv.next19 to i32
37  %exitcond21 = icmp eq i32 %lftr.wideiv20, %N
38  br i1 %exitcond21, label %for.end8, label %for.body3.lr.ph
39
40for.end8:                                         ; preds = %for.cond1.for.inc6_crit_edge, %entry
41  ret void
42}
43
44;; Loop is interchanged check that the phi nodes are split and the promoted value is used instead of the reduction phi.
45; CHECK-LABEL: @reduction_01
46; CHECK:  for.body3:                                        ; preds = %for.body3.preheader, %for.body3.split
47; CHECK:    %indvars.iv = phi i64 [ %indvars.iv.next, %for.body3.split ], [ 1, %for.body3.preheader ]
48; CHECK:    br label %for.body3.lr.ph.preheader
49; CHECK:    %add = add nsw i32 %X.promoted
50
51
52;; Test for more than 1 reductions inside a loop.
53;;  for( int i=1;i<N;i++)
54;;    for( int j=1;j<N;j++)
55;;      for( int k=1;k<N;k++) {
56;;        X+=A[k][j];
57;;        Y+=B[k][i];
58;;      }
59
60define void @reduction_02(i32 %N)  {
61entry:
62  %cmp34 = icmp sgt i32 %N, 1
63  br i1 %cmp34, label %for.cond4.preheader.preheader, label %for.end19
64
65for.cond4.preheader.preheader:                    ; preds = %entry, %for.inc17
66  %indvars.iv40 = phi i64 [ %indvars.iv.next41, %for.inc17 ], [ 1, %entry ]
67  br label %for.body6.lr.ph
68
69for.body6.lr.ph:                                  ; preds = %for.cond4.for.inc14_crit_edge, %for.cond4.preheader.preheader
70  %indvars.iv36 = phi i64 [ %indvars.iv.next37, %for.cond4.for.inc14_crit_edge ], [ 1, %for.cond4.preheader.preheader ]
71  %X.promoted = load i32, i32* @X
72  %Y.promoted = load i32, i32* @Y
73  br label %for.body6
74
75for.body6:                                        ; preds = %for.body6, %for.body6.lr.ph
76  %indvars.iv = phi i64 [ 1, %for.body6.lr.ph ], [ %indvars.iv.next, %for.body6 ]
77  %add1331 = phi i32 [ %Y.promoted, %for.body6.lr.ph ], [ %add13, %for.body6 ]
78  %add30 = phi i32 [ %X.promoted, %for.body6.lr.ph ], [ %add, %for.body6 ]
79  %arrayidx8 = getelementptr inbounds [500 x [500 x i32]], [500 x [500 x i32]]* @A, i64 0, i64 %indvars.iv, i64 %indvars.iv36
80  %0 = load i32, i32* %arrayidx8
81  %add = add nsw i32 %add30, %0
82  %arrayidx12 = getelementptr inbounds [500 x [500 x i32]], [500 x [500 x i32]]* @B, i64 0, i64 %indvars.iv, i64 %indvars.iv40
83  %1 = load i32, i32* %arrayidx12
84  %add13 = add nsw i32 %add1331, %1
85  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
86  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
87  %exitcond = icmp eq i32 %lftr.wideiv, %N
88  br i1 %exitcond, label %for.cond4.for.inc14_crit_edge, label %for.body6
89
90for.cond4.for.inc14_crit_edge:                    ; preds = %for.body6
91  store i32 %add, i32* @X
92  store i32 %add13, i32* @Y
93  %indvars.iv.next37 = add nuw nsw i64 %indvars.iv36, 1
94  %lftr.wideiv38 = trunc i64 %indvars.iv.next37 to i32
95  %exitcond39 = icmp eq i32 %lftr.wideiv38, %N
96  br i1 %exitcond39, label %for.inc17, label %for.body6.lr.ph
97
98for.inc17:                                        ; preds = %for.cond4.for.inc14_crit_edge
99  %indvars.iv.next41 = add nuw nsw i64 %indvars.iv40, 1
100  %lftr.wideiv42 = trunc i64 %indvars.iv.next41 to i32
101  %exitcond43 = icmp eq i32 %lftr.wideiv42, %N
102  br i1 %exitcond43, label %for.end19, label %for.cond4.preheader.preheader
103
104for.end19:                                        ; preds = %for.inc17, %entry
105  ret void
106}
107
108;; Loop is interchanged check that the phi nodes are split and the promoted value is used instead of the reduction phi.
109; CHECK-LABEL: @reduction_02
110; CHECK:  for.body6:                                        ; preds = %for.body6.preheader, %for.body6.split
111; CHECK:    %indvars.iv = phi i64 [ %indvars.iv.next, %for.body6.split ], [ 1, %for.body6.preheader ]
112; CHECK:    br label %for.cond4.preheader.preheader.preheader
113; CHECK:    %add13 = add nsw i32 %Y.promoted
114
115
116;; Not tightly nested. Do not interchange.
117;;  for( int i=1;i<N;i++)
118;;    for( int j=1;j<N;j++) {
119;;      for( int k=1;k<N;k++) {
120;;        X+=A[k][j];
121;;      }
122;;      Y+=B[j][i];
123;;    }
124define void @reduction_03(i32 %N)  {
125entry:
126  %cmp35 = icmp sgt i32 %N, 1
127  br i1 %cmp35, label %for.cond4.preheader.lr.ph, label %for.end19
128
129for.cond4.preheader.lr.ph:                        ; preds = %entry, %for.cond1.for.inc17_crit_edge
130  %indvars.iv41 = phi i64 [ %indvars.iv.next42, %for.cond1.for.inc17_crit_edge ], [ 1, %entry ]
131  %Y.promoted = load i32, i32* @Y
132  br label %for.body6.lr.ph
133
134for.body6.lr.ph:                                  ; preds = %for.cond4.preheader.lr.ph, %for.cond4.for.end_crit_edge
135  %indvars.iv37 = phi i64 [ 1, %for.cond4.preheader.lr.ph ], [ %indvars.iv.next38, %for.cond4.for.end_crit_edge ]
136  %add1334 = phi i32 [ %Y.promoted, %for.cond4.preheader.lr.ph ], [ %add13, %for.cond4.for.end_crit_edge ]
137  %X.promoted = load i32, i32* @X
138  br label %for.body6
139
140for.body6:                                        ; preds = %for.body6, %for.body6.lr.ph
141  %indvars.iv = phi i64 [ 1, %for.body6.lr.ph ], [ %indvars.iv.next, %for.body6 ]
142  %add31 = phi i32 [ %X.promoted, %for.body6.lr.ph ], [ %add, %for.body6 ]
143  %arrayidx8 = getelementptr inbounds [500 x [500 x i32]], [500 x [500 x i32]]* @A, i64 0, i64 %indvars.iv, i64 %indvars.iv37
144  %0 = load i32, i32* %arrayidx8
145  %add = add nsw i32 %add31, %0
146  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
147  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
148  %exitcond = icmp eq i32 %lftr.wideiv, %N
149  br i1 %exitcond, label %for.cond4.for.end_crit_edge, label %for.body6
150
151for.cond4.for.end_crit_edge:                      ; preds = %for.body6
152  store i32 %add, i32* @X
153  %arrayidx12 = getelementptr inbounds [500 x [500 x i32]], [500 x [500 x i32]]* @B, i64 0, i64 %indvars.iv37, i64 %indvars.iv41
154  %1 = load i32, i32* %arrayidx12
155  %add13 = add nsw i32 %add1334, %1
156  %indvars.iv.next38 = add nuw nsw i64 %indvars.iv37, 1
157  %lftr.wideiv39 = trunc i64 %indvars.iv.next38 to i32
158  %exitcond40 = icmp eq i32 %lftr.wideiv39, %N
159  br i1 %exitcond40, label %for.cond1.for.inc17_crit_edge, label %for.body6.lr.ph
160
161for.cond1.for.inc17_crit_edge:                    ; preds = %for.cond4.for.end_crit_edge
162  store i32 %add13, i32* @Y
163  %indvars.iv.next42 = add nuw nsw i64 %indvars.iv41, 1
164  %lftr.wideiv43 = trunc i64 %indvars.iv.next42 to i32
165  %exitcond44 = icmp eq i32 %lftr.wideiv43, %N
166  br i1 %exitcond44, label %for.end19, label %for.cond4.preheader.lr.ph
167
168for.end19:                                        ; preds = %for.cond1.for.inc17_crit_edge, %entry
169  ret void
170}
171;; Not tightly nested. Do not interchange.
172;; Not interchanged hence the phi's in the inner loop will not be split. Check for the same.
173; CHECK-LABEL: @reduction_03
174; CHECK:  for.body6:                                        ; preds = %for.body6.preheader, %for.body6
175; CHECK:    %indvars.iv = phi i64 [ %indvars.iv.next, %for.body6 ], [ 1, %for.body6.preheader ]
176; CHECK:    %add31 = phi i32 [ %add, %for.body6 ], [ %X.promoted, %for.body6.preheader ]
177
178
179
180;; Multiple use of reduction not safe. Do not interchange.
181;;  for( int i=1;i<N;i++)
182;;    for( int j=1;j<N;j++)
183;;      for( int k=1;k<N;k++) {
184;;        X+=A[k][j];
185;;        Y+=X;
186;;      }
187define void @reduction_04(i32 %N) {
188entry:
189  %cmp28 = icmp sgt i32 %N, 1
190  br i1 %cmp28, label %for.cond4.preheader.preheader, label %for.end15
191
192for.cond4.preheader.preheader:                    ; preds = %entry, %for.inc13
193  %i.029 = phi i32 [ %inc14, %for.inc13 ], [ 1, %entry ]
194  br label %for.body6.lr.ph
195
196for.body6.lr.ph:                                  ; preds = %for.cond4.for.inc10_crit_edge, %for.cond4.preheader.preheader
197  %indvars.iv30 = phi i64 [ %indvars.iv.next31, %for.cond4.for.inc10_crit_edge ], [ 1, %for.cond4.preheader.preheader ]
198  %X.promoted = load i32, i32* @X
199  %Y.promoted = load i32, i32* @Y
200  br label %for.body6
201
202for.body6:                                        ; preds = %for.body6, %for.body6.lr.ph
203  %indvars.iv = phi i64 [ 1, %for.body6.lr.ph ], [ %indvars.iv.next, %for.body6 ]
204  %add925 = phi i32 [ %Y.promoted, %for.body6.lr.ph ], [ %add9, %for.body6 ]
205  %add24 = phi i32 [ %X.promoted, %for.body6.lr.ph ], [ %add, %for.body6 ]
206  %arrayidx8 = getelementptr inbounds [500 x [500 x i32]], [500 x [500 x i32]]* @A, i64 0, i64 %indvars.iv, i64 %indvars.iv30
207  %0 = load i32, i32* %arrayidx8
208  %add = add nsw i32 %add24, %0
209  %add9 = add nsw i32 %add925, %add
210  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
211  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
212  %exitcond = icmp eq i32 %lftr.wideiv, %N
213  br i1 %exitcond, label %for.cond4.for.inc10_crit_edge, label %for.body6
214
215for.cond4.for.inc10_crit_edge:                    ; preds = %for.body6
216  store i32 %add, i32* @X
217  store i32 %add9, i32* @Y
218  %indvars.iv.next31 = add nuw nsw i64 %indvars.iv30, 1
219  %lftr.wideiv32 = trunc i64 %indvars.iv.next31 to i32
220  %exitcond33 = icmp eq i32 %lftr.wideiv32, %N
221  br i1 %exitcond33, label %for.inc13, label %for.body6.lr.ph
222
223for.inc13:                                        ; preds = %for.cond4.for.inc10_crit_edge
224  %inc14 = add nuw nsw i32 %i.029, 1
225  %exitcond34 = icmp eq i32 %inc14, %N
226  br i1 %exitcond34, label %for.end15, label %for.cond4.preheader.preheader
227
228for.end15:                                        ; preds = %for.inc13, %entry
229  ret void
230}
231;; Not interchanged hence the phi's in the inner loop will not be split. Check for the same.
232; CHECK-LABEL: @reduction_04
233; CHECK:  for.body6:                                        ; preds = %for.body6.preheader, %for.body6
234; CHECK:    %indvars.iv = phi i64 [ %indvars.iv.next, %for.body6 ], [ 1, %for.body6.preheader ]
235; CHECK:    %add925 = phi i32 [ %add9, %for.body6 ], [ %Y.promoted, %for.body6.preheader ]
236