1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5 2; RUN: opt -passes=loop-distribute -enable-loop-distribute -verify-loop-info -verify-dom-info -S < %s \ 3; RUN: | FileCheck %s 4 5; Check that definitions used outside the loop are handled correctly: (1) they 6; are not dropped (2) when version the loop, a phi is added to merge the value 7; from the non-distributed loop and the distributed loop. 8; 9; for (i = 0; i < n; i++) { 10; A[i + 1] = A[i] * B[i]; 11; ========================== 12; sum += C[i]; 13; } 14 15@B = common global ptr null, align 8 16@A = common global ptr null, align 8 17@C = common global ptr null, align 8 18@D = common global ptr null, align 8 19@E = common global ptr null, align 8 20@SUM = common global i32 0, align 8 21 22define void @f() { 23; CHECK-LABEL: define void @f() { 24; CHECK-NEXT: [[ENTRY:.*:]] 25; CHECK-NEXT: [[A:%.*]] = load ptr, ptr @A, align 8 26; CHECK-NEXT: [[B:%.*]] = load ptr, ptr @B, align 8 27; CHECK-NEXT: [[C:%.*]] = load ptr, ptr @C, align 8 28; CHECK-NEXT: [[D:%.*]] = load ptr, ptr @D, align 8 29; CHECK-NEXT: [[E:%.*]] = load ptr, ptr @E, align 8 30; CHECK-NEXT: br label %[[FOR_BODY_LVER_CHECK:.*]] 31; CHECK: [[FOR_BODY_LVER_CHECK]]: 32; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[A]], i64 84 33; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[C]], i64 80 34; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[A]], [[SCEVGEP1]] 35; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[C]], [[SCEVGEP]] 36; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] 37; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label %[[FOR_BODY_PH_LVER_ORIG:.*]], label %[[FOR_BODY_PH_LDIST1:.*]] 38; CHECK: [[FOR_BODY_PH_LVER_ORIG]]: 39; CHECK-NEXT: br label %[[FOR_BODY_LVER_ORIG:.*]] 40; CHECK: [[FOR_BODY_LVER_ORIG]]: 41; CHECK-NEXT: [[IND_LVER_ORIG:%.*]] = phi i64 [ 0, %[[FOR_BODY_PH_LVER_ORIG]] ], [ [[ADD_LVER_ORIG:%.*]], %[[FOR_BODY_LVER_ORIG]] ] 42; CHECK-NEXT: [[SUM_LVER_ORIG:%.*]] = phi i32 [ 0, %[[FOR_BODY_PH_LVER_ORIG]] ], [ [[SUM_ADD_LVER_ORIG:%.*]], %[[FOR_BODY_LVER_ORIG]] ] 43; CHECK-NEXT: [[ARRAYIDXA_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IND_LVER_ORIG]] 44; CHECK-NEXT: [[LOADA_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDXA_LVER_ORIG]], align 4 45; CHECK-NEXT: [[ARRAYIDXB_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IND_LVER_ORIG]] 46; CHECK-NEXT: [[LOADB_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDXB_LVER_ORIG]], align 4 47; CHECK-NEXT: [[MULA_LVER_ORIG:%.*]] = mul i32 [[LOADB_LVER_ORIG]], [[LOADA_LVER_ORIG]] 48; CHECK-NEXT: [[ADD_LVER_ORIG]] = add nuw nsw i64 [[IND_LVER_ORIG]], 1 49; CHECK-NEXT: [[ARRAYIDXA_PLUS_4_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[ADD_LVER_ORIG]] 50; CHECK-NEXT: store i32 [[MULA_LVER_ORIG]], ptr [[ARRAYIDXA_PLUS_4_LVER_ORIG]], align 4 51; CHECK-NEXT: [[ARRAYIDXC_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[IND_LVER_ORIG]] 52; CHECK-NEXT: [[LOADC_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDXC_LVER_ORIG]], align 4 53; CHECK-NEXT: [[SUM_ADD_LVER_ORIG]] = add nuw nsw i32 [[SUM_LVER_ORIG]], [[LOADC_LVER_ORIG]] 54; CHECK-NEXT: [[EXITCOND_LVER_ORIG:%.*]] = icmp eq i64 [[ADD_LVER_ORIG]], 20 55; CHECK-NEXT: br i1 [[EXITCOND_LVER_ORIG]], label %[[FOR_END_LOOPEXIT:.*]], label %[[FOR_BODY_LVER_ORIG]] 56; CHECK: [[FOR_BODY_PH_LDIST1]]: 57; CHECK-NEXT: br label %[[FOR_BODY_LDIST1:.*]] 58; CHECK: [[FOR_BODY_LDIST1]]: 59; CHECK-NEXT: [[IND_LDIST1:%.*]] = phi i64 [ 0, %[[FOR_BODY_PH_LDIST1]] ], [ [[ADD_LDIST1:%.*]], %[[FOR_BODY_LDIST1]] ] 60; CHECK-NEXT: [[ARRAYIDXA_LDIST1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IND_LDIST1]] 61; CHECK-NEXT: [[LOADA_LDIST1:%.*]] = load i32, ptr [[ARRAYIDXA_LDIST1]], align 4, !alias.scope [[META0:![0-9]+]], !noalias [[META3:![0-9]+]] 62; CHECK-NEXT: [[ARRAYIDXB_LDIST1:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IND_LDIST1]] 63; CHECK-NEXT: [[LOADB_LDIST1:%.*]] = load i32, ptr [[ARRAYIDXB_LDIST1]], align 4, !alias.scope [[META5:![0-9]+]] 64; CHECK-NEXT: [[MULA_LDIST1:%.*]] = mul i32 [[LOADB_LDIST1]], [[LOADA_LDIST1]] 65; CHECK-NEXT: [[ADD_LDIST1]] = add nuw nsw i64 [[IND_LDIST1]], 1 66; CHECK-NEXT: [[ARRAYIDXA_PLUS_4_LDIST1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[ADD_LDIST1]] 67; CHECK-NEXT: store i32 [[MULA_LDIST1]], ptr [[ARRAYIDXA_PLUS_4_LDIST1]], align 4, !alias.scope [[META0]], !noalias [[META3]] 68; CHECK-NEXT: [[EXITCOND_LDIST1:%.*]] = icmp eq i64 [[ADD_LDIST1]], 20 69; CHECK-NEXT: br i1 [[EXITCOND_LDIST1]], label %[[FOR_BODY_PH:.*]], label %[[FOR_BODY_LDIST1]] 70; CHECK: [[FOR_BODY_PH]]: 71; CHECK-NEXT: br label %[[FOR_BODY:.*]] 72; CHECK: [[FOR_BODY]]: 73; CHECK-NEXT: [[IND:%.*]] = phi i64 [ 0, %[[FOR_BODY_PH]] ], [ [[ADD:%.*]], %[[FOR_BODY]] ] 74; CHECK-NEXT: [[SUM:%.*]] = phi i32 [ 0, %[[FOR_BODY_PH]] ], [ [[SUM_ADD:%.*]], %[[FOR_BODY]] ] 75; CHECK-NEXT: [[ADD]] = add nuw nsw i64 [[IND]], 1 76; CHECK-NEXT: [[ARRAYIDXC:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[IND]] 77; CHECK-NEXT: [[LOADC:%.*]] = load i32, ptr [[ARRAYIDXC]], align 4, !alias.scope [[META3]] 78; CHECK-NEXT: [[SUM_ADD]] = add nuw nsw i32 [[SUM]], [[LOADC]] 79; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[ADD]], 20 80; CHECK-NEXT: br i1 [[EXITCOND]], label %[[FOR_END_LOOPEXIT2:.*]], label %[[FOR_BODY]] 81; CHECK: [[FOR_END_LOOPEXIT]]: 82; CHECK-NEXT: [[SUM_ADD_LVER_PH:%.*]] = phi i32 [ [[SUM_ADD_LVER_ORIG]], %[[FOR_BODY_LVER_ORIG]] ] 83; CHECK-NEXT: br label %[[FOR_END:.*]] 84; CHECK: [[FOR_END_LOOPEXIT2]]: 85; CHECK-NEXT: [[SUM_ADD_LVER_PH3:%.*]] = phi i32 [ [[SUM_ADD]], %[[FOR_BODY]] ] 86; CHECK-NEXT: br label %[[FOR_END]] 87; CHECK: [[FOR_END]]: 88; CHECK-NEXT: [[SUM_ADD_LVER:%.*]] = phi i32 [ [[SUM_ADD_LVER_PH]], %[[FOR_END_LOOPEXIT]] ], [ [[SUM_ADD_LVER_PH3]], %[[FOR_END_LOOPEXIT2]] ] 89; CHECK-NEXT: store i32 [[SUM_ADD_LVER]], ptr @SUM, align 4 90; CHECK-NEXT: ret void 91; 92entry: 93 %a = load ptr, ptr @A, align 8 94 %b = load ptr, ptr @B, align 8 95 %c = load ptr, ptr @C, align 8 96 %d = load ptr, ptr @D, align 8 97 %e = load ptr, ptr @E, align 8 98 99 br label %for.body 100 101for.body: ; preds = %for.body, %entry 102 %ind = phi i64 [ 0, %entry ], [ %add, %for.body ] 103 %sum = phi i32 [ 0, %entry ], [ %sum_add, %for.body ] 104 105 %arrayidxA = getelementptr inbounds i32, ptr %a, i64 %ind 106 %loadA = load i32, ptr %arrayidxA, align 4 107 108 %arrayidxB = getelementptr inbounds i32, ptr %b, i64 %ind 109 %loadB = load i32, ptr %arrayidxB, align 4 110 111 %mulA = mul i32 %loadB, %loadA 112 113 %add = add nuw nsw i64 %ind, 1 114 %arrayidxA_plus_4 = getelementptr inbounds i32, ptr %a, i64 %add 115 store i32 %mulA, ptr %arrayidxA_plus_4, align 4 116 117 %arrayidxC = getelementptr inbounds i32, ptr %c, i64 %ind 118 %loadC = load i32, ptr %arrayidxC, align 4 119 120 %sum_add = add nuw nsw i32 %sum, %loadC 121 122 %exitcond = icmp eq i64 %add, 20 123 br i1 %exitcond, label %for.end, label %for.body 124 125for.end: ; preds = %for.body 126 store i32 %sum_add, ptr @SUM, align 4 127 ret void 128} 129;. 130; CHECK: [[META0]] = !{[[META1:![0-9]+]]} 131; CHECK: [[META1]] = distinct !{[[META1]], [[META2:![0-9]+]]} 132; CHECK: [[META2]] = distinct !{[[META2]], !"LVerDomain"} 133; CHECK: [[META3]] = !{[[META4:![0-9]+]]} 134; CHECK: [[META4]] = distinct !{[[META4]], [[META2]]} 135; CHECK: [[META5]] = !{[[META6:![0-9]+]]} 136; CHECK: [[META6]] = distinct !{[[META6]], [[META2]]} 137;. 138