1; Test optimization remarks generated by the LoopInterchange pass. 2; 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 7 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 12 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 16 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]; 22 23define void @test01(i32 %N){ 24entry: 25 %cmp31 = icmp sgt i32 %N, 1 26 br i1 %cmp31, label %for.cond1.preheader.lr.ph, label %for.end19 27 28for.cond1.preheader.lr.ph: 29 %0 = add i32 %N, -1 30 br label %for.body3.lr.ph 31 32for.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 35 br label %for.body3 36 37for.body3: 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 50 51for.inc17: 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 56 57for.end19: 58 ret void 59} 60 61; CHECK: --- !Missed 62; CHECK-NEXT: Pass: loop-interchange 63; CHECK-NEXT: Name: Dependence 64; CHECK-NEXT: Function: test01 65; CHECK-NEXT: Args: 66; CHECK-NEXT: - String: Cannot interchange loops due to dependences. 67; CHECK-NEXT: ... 68 69; DELIN: --- !Missed 70; DELIN-NEXT: Pass: loop-interchange 71; DELIN-NEXT: Name: InterchangeNotProfitable 72; DELIN-NEXT: Function: test01 73; DELIN-NEXT: Args: 74; DELIN-NEXT: - String: Interchanging loops is not considered to improve cache locality nor vectorization. 75; DELIN-NEXT: ... 76 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; 84 85define void @test02(i32 %k, i32 %N) { 86 entry: 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 90 91 for.cond1.preheader.lr.ph: 92 %cmp324 = icmp sgt i32 %sub, 1 93 %0 = add i32 %N, -2 94 %1 = sext i32 %sub to i64 95 br label %for.cond1.preheader 96 97 for.cond.loopexit: 98 %cmp = icmp slt i64 %indvars.iv.next29, %1 99 br i1 %cmp, label %for.cond1.preheader, label %for.end17 100 101 for.cond1.preheader: 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 105 106 for.body4: 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 116 117 for.end17: 118 ret void 119} 120 121; CHECK: --- !Missed 122; CHECK-NEXT: Pass: loop-interchange 123; CHECK-NEXT: Name: Dependence 124; CHECK-NEXT: Function: test02 125; CHECK-NEXT: Args: 126; CHECK-NEXT: - String: Cannot interchange loops due to dependences. 127; CHECK-NEXT: ... 128 129; DELIN: --- !Passed 130; DELIN-NEXT: Pass: loop-interchange 131; DELIN-NEXT: Name: Interchanged 132; DELIN-NEXT: Function: test02 133; DELIN-NEXT: Args: 134; DELIN-NEXT: - String: Loop interchanged with enclosing loop. 135; DELIN-NEXT: ... 136 137;;-----------------------------------Test case 03------------------------------- 138;; Test to make sure we can handle output dependencies. 139;; 140;; for (int i = 0; i < 2; ++i) 141;; for(int j = 0; j < 3; ++j) { 142;; A[j][i] = i; 143;; A[j][i+1] = j; 144;; } 145 146@A10 = local_unnamed_addr global [3 x [3 x i32]] zeroinitializer, align 16 147 148define void @test03() { 149entry: 150 br label %for.cond1.preheader 151 152for.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 155 156for.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, 1 159 br label %for.body4 160 161for.cond.cleanup: ; preds = %for.cond.loopexit 162 ret void 163 164for.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 175} 176 177; CHECK: --- !Passed 178; CHECK-NEXT: Pass: loop-interchange 179; CHECK-NEXT: Name: Interchanged 180; CHECK-NEXT: Function: test03 181; CHECK-NEXT: Args: 182; CHECK-NEXT: - String: Loop interchanged with enclosing loop. 183; CHECK-NEXT: ... 184 185; DELIN: --- !Passed 186; DELIN-NEXT: Pass: loop-interchange 187; DELIN-NEXT: Name: Interchanged 188; DELIN-NEXT: Function: test03 189; DELIN-NEXT: Args: 190; DELIN-NEXT: - String: Loop interchanged with enclosing loop. 191; DELIN-NEXT: ... 192 193;;--------------------------------------Test case 04------------------------------------- 194;; Loops not tightly nested are not interchanged 195;; for(int j=0;j<N;j++) { 196;; B[j] = j+k; 197;; for(int i=0;i<N;i++) 198;; A[j][i] = A[j][i]+B[j]; 199;; } 200 201define void @test04(i32 %k, i32 %N){ 202entry: 203 %cmp30 = icmp sgt i32 %N, 0 204 br i1 %cmp30, label %for.body.lr.ph, label %for.end17 205 206for.body.lr.ph: 207 %0 = add i32 %N, -1 208 %1 = zext i32 %k to i64 209 br label %for.body 210 211for.body: 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 217 br label %for.body3 218 219for.body3: 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 229 230for.inc15: 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 235 236for.end17: 237 ret void 238} 239 240; CHECK: --- !Missed 241; CHECK-NEXT: Pass: loop-interchange 242; CHECK-NEXT: Name: Dependence 243; CHECK-NEXT: Function: test04 244; CHECK-NEXT: Args: 245; CHECK-NEXT: - String: Cannot interchange loops due to dependences. 246; CHECK-NEXT: ... 247 248; DELIN: --- !Missed 249; DELIN-NEXT: Pass: loop-interchange 250; DELIN-NEXT: Name: NotTightlyNested 251; DELIN-NEXT: Function: test04 252; DELIN-NEXT: Args: 253; DELIN-NEXT: - String: Cannot interchange loops because they are not tightly nested. 254; DELIN-NEXT: ... 255