1 ; Test optimization remarks generated by the LoopInterchange pass.
3 ; RUN: opt < %s -passes=loop-interchange -cache-line-size=64 -verify-dom-info -verify-loop-info \
4 ; RUN: -pass-remarks-output=%t -pass-remarks-missed='loop-interchange' \
5 ; RUN: -pass-remarks='loop-interchange' -S
6 ; RUN: cat %t | FileCheck %s
8 ; RUN: opt < %s -passes=loop-interchange -cache-line-size=64 -verify-dom-info -verify-loop-info \
9 ; RUN: -pass-remarks-output=%t -pass-remarks-missed='loop-interchange' \
10 ; RUN: -pass-remarks='loop-interchange' -S -da-disable-delinearization-checks
11 ; RUN: cat %t | FileCheck --check-prefix=DELIN %s
13 @A = common global [100 x [100 x i32]] zeroinitializer
14 @B = common global [100 x [100 x i32]] zeroinitializer
15 @C = common global [100 x i32] zeroinitializer
17 ;;---------------------------------------Test case 01---------------------------------
18 ;; Loops interchange is not profitable.
19 ;; for(int i=1;i<N;i++)
20 ;; for(int j=1;j<N;j++)
21 ;; A[i-1][j-1] = A[i - 1][j-1] + B[i][j];
23 define void @test01(i32 %N){
25 %cmp31 = icmp sgt i32 %N, 1
26 br i1 %cmp31, label %for.cond1.preheader.lr.ph, label %for.end19
28 for.cond1.preheader.lr.ph:
30 br label %for.body3.lr.ph
33 %indvars.iv34 = phi i64 [ 1, %for.cond1.preheader.lr.ph ], [ %indvars.iv.next35, %for.inc17 ]
34 %1 = add nsw i64 %indvars.iv34, -1
38 %indvars.iv = phi i64 [ 1, %for.body3.lr.ph ], [ %indvars.iv.next, %for.body3 ]
39 %2 = add nsw i64 %indvars.iv, -1
40 %arrayidx6 = getelementptr inbounds [100 x [100 x i32]], ptr @A, i64 0, i64 %1, i64 %2
41 %3 = load i32, ptr %arrayidx6
42 %arrayidx10 = getelementptr inbounds [100 x [100 x i32]], ptr @B, i64 0, i64 %indvars.iv34, i64 %indvars.iv
43 %4 = load i32, ptr %arrayidx10
44 %add = add nsw i32 %4, %3
45 store i32 %add, ptr %arrayidx6
46 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
47 %lftr.wideiv = trunc i64 %indvars.iv to i32
48 %exitcond = icmp eq i32 %lftr.wideiv, %0
49 br i1 %exitcond, label %for.inc17, label %for.body3
52 %indvars.iv.next35 = add nuw nsw i64 %indvars.iv34, 1
53 %lftr.wideiv37 = trunc i64 %indvars.iv34 to i32
54 %exitcond38 = icmp eq i32 %lftr.wideiv37, %0
55 br i1 %exitcond38, label %for.end19, label %for.body3.lr.ph
62 ; CHECK-NEXT: Pass: loop-interchange
63 ; CHECK-NEXT: Name: Dependence
64 ; CHECK-NEXT: Function: test01
66 ; CHECK-NEXT: - String: Cannot interchange loops due to dependences.
70 ; DELIN-NEXT: Pass: loop-interchange
71 ; DELIN-NEXT: Name: InterchangeNotProfitable
72 ; DELIN-NEXT: Function: test01
74 ; DELIN-NEXT: - String: Interchanging loops is not considered to improve cache locality nor vectorization.
77 ;;--------------------------------------Test case 02------------------------------------
78 ;; [FIXME] This loop though valid is currently not interchanged due to the
79 ;; limitation that we cannot split the inner loop latch due to multiple use of inner induction
80 ;; variable.(used to increment the loop counter and to access A[j+1][i+1]
81 ;; for(int i=0;i<N-1;i++)
82 ;; for(int j=1;j<N-1;j++)
83 ;; A[j+1][i+1] = A[j+1][i+1] + k;
85 define void @test02(i32 %k, i32 %N) {
87 %sub = add nsw i32 %N, -1
88 %cmp26 = icmp sgt i32 %N, 1
89 br i1 %cmp26, label %for.cond1.preheader.lr.ph, label %for.end17
91 for.cond1.preheader.lr.ph:
92 %cmp324 = icmp sgt i32 %sub, 1
94 %1 = sext i32 %sub to i64
95 br label %for.cond1.preheader
98 %cmp = icmp slt i64 %indvars.iv.next29, %1
99 br i1 %cmp, label %for.cond1.preheader, label %for.end17
102 %indvars.iv28 = phi i64 [ 0, %for.cond1.preheader.lr.ph ], [ %indvars.iv.next29, %for.cond.loopexit ]
103 %indvars.iv.next29 = add nuw nsw i64 %indvars.iv28, 1
104 br i1 %cmp324, label %for.body4, label %for.cond.loopexit
107 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body4 ], [ 1, %for.cond1.preheader ]
108 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
109 %arrayidx7 = getelementptr inbounds [100 x [100 x i32]], ptr @A, i64 0, i64 %indvars.iv.next, i64 %indvars.iv.next29
110 %2 = load i32, ptr %arrayidx7
111 %add8 = add nsw i32 %2, %k
112 store i32 %add8, ptr %arrayidx7
113 %lftr.wideiv = trunc i64 %indvars.iv to i32
114 %exitcond = icmp eq i32 %lftr.wideiv, %0
115 br i1 %exitcond, label %for.cond.loopexit, label %for.body4
122 ; CHECK-NEXT: Pass: loop-interchange
123 ; CHECK-NEXT: Name: Dependence
124 ; CHECK-NEXT: Function: test02
126 ; CHECK-NEXT: - String: Cannot interchange loops due to dependences.
130 ; DELIN-NEXT: Pass: loop-interchange
131 ; DELIN-NEXT: Name: Interchanged
132 ; DELIN-NEXT: Function: test02
134 ; DELIN-NEXT: - String: Loop interchanged with enclosing loop.
137 ;;-----------------------------------Test case 03-------------------------------
138 ;; Test to make sure we can handle output dependencies.
140 ;; for (int i = 0; i < 2; ++i)
141 ;; for(int j = 0; j < 3; ++j) {
146 @A10 = local_unnamed_addr global [3 x [3 x i32]] zeroinitializer, align 16
148 define void @test03() {
150 br label %for.cond1.preheader
152 for.cond.loopexit: ; preds = %for.body4
153 %exitcond28 = icmp ne i64 %indvars.iv.next27, 2
154 br i1 %exitcond28, label %for.cond1.preheader, label %for.cond.cleanup
156 for.cond1.preheader: ; preds = %for.cond.loopexit, %entry
157 %indvars.iv26 = phi i64 [ 0, %entry ], [ %indvars.iv.next27, %for.cond.loopexit ]
158 %indvars.iv.next27 = add nuw nsw i64 %indvars.iv26, 1
161 for.cond.cleanup: ; preds = %for.cond.loopexit
164 for.body4: ; preds = %for.body4, %for.cond1.preheader
165 %indvars.iv = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next, %for.body4 ]
166 %arrayidx6 = getelementptr inbounds [3 x [3 x i32]], ptr @A10, i64 0, i64 %indvars.iv, i64 %indvars.iv26
167 %tmp = trunc i64 %indvars.iv26 to i32
168 store i32 %tmp, ptr %arrayidx6, align 4
169 %arrayidx10 = getelementptr inbounds [3 x [3 x i32]], ptr @A10, i64 0, i64 %indvars.iv, i64 %indvars.iv.next27
170 %tmp1 = trunc i64 %indvars.iv to i32
171 store i32 %tmp1, ptr %arrayidx10, align 4
172 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
173 %exitcond = icmp ne i64 %indvars.iv.next, 3
174 br i1 %exitcond, label %for.body4, label %for.cond.loopexit
178 ; CHECK-NEXT: Pass: loop-interchange
179 ; CHECK-NEXT: Name: Interchanged
180 ; CHECK-NEXT: Function: test03
182 ; CHECK-NEXT: - String: Loop interchanged with enclosing loop.
186 ; DELIN-NEXT: Pass: loop-interchange
187 ; DELIN-NEXT: Name: Interchanged
188 ; DELIN-NEXT: Function: test03
190 ; DELIN-NEXT: - String: Loop interchanged with enclosing loop.
193 ;;--------------------------------------Test case 04-------------------------------------
194 ;; Loops not tightly nested are not interchanged
195 ;; for(int j=0;j<N;j++) {
197 ;; for(int i=0;i<N;i++)
198 ;; A[j][i] = A[j][i]+B[j];
201 define void @test04(i32 %k, i32 %N){
203 %cmp30 = icmp sgt i32 %N, 0
204 br i1 %cmp30, label %for.body.lr.ph, label %for.end17
208 %1 = zext i32 %k to i64
212 %indvars.iv32 = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next33, %for.inc15 ]
213 %2 = add nsw i64 %indvars.iv32, %1
214 %arrayidx = getelementptr inbounds [100 x i32], ptr @C, i64 0, i64 %indvars.iv32
215 %3 = trunc i64 %2 to i32
216 store i32 %3, ptr %arrayidx
220 %indvars.iv = phi i64 [ 0, %for.body ], [ %indvars.iv.next, %for.body3 ]
221 %arrayidx7 = getelementptr inbounds [100 x [100 x i32]], ptr @A, i64 0, i64 %indvars.iv32, i64 %indvars.iv
222 %4 = load i32, ptr %arrayidx7
223 %add10 = add nsw i32 %3, %4
224 store i32 %add10, ptr %arrayidx7
225 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
226 %lftr.wideiv = trunc i64 %indvars.iv to i32
227 %exitcond = icmp eq i32 %lftr.wideiv, %0
228 br i1 %exitcond, label %for.inc15, label %for.body3
231 %indvars.iv.next33 = add nuw nsw i64 %indvars.iv32, 1
232 %lftr.wideiv35 = trunc i64 %indvars.iv32 to i32
233 %exitcond36 = icmp eq i32 %lftr.wideiv35, %0
234 br i1 %exitcond36, label %for.end17, label %for.body
241 ; CHECK-NEXT: Pass: loop-interchange
242 ; CHECK-NEXT: Name: Dependence
243 ; CHECK-NEXT: Function: test04
245 ; CHECK-NEXT: - String: Cannot interchange loops due to dependences.
249 ; DELIN-NEXT: Pass: loop-interchange
250 ; DELIN-NEXT: Name: NotTightlyNested
251 ; DELIN-NEXT: Function: test04
253 ; DELIN-NEXT: - String: Cannot interchange loops because they are not tightly nested.