1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -loop-vectorize -force-vector-width=2 -S %s | FileCheck %s 3 4; Tests where the indices of some accesses are clamped to a small range. 5 6; FIXME: At the moment, the runtime checks require that the indices do not wrap 7; and runtime checks are emitted to ensure that. The clamped indices do 8; wrap, so the vector loops are dead at the moment. But it is still 9; possible to compute the bounds of the accesses and generate proper 10; runtime checks. 11 12; The relevant bounds for %gep.A are [%A, %A+4). 13define void @load_clamped_index(i32* %A, i32* %B, i32 %N) { 14; CHECK-LABEL: @load_clamped_index( 15; CHECK-NEXT: entry: 16; CHECK-NEXT: [[B1:%.*]] = bitcast i32* [[B:%.*]] to i8* 17; CHECK-NEXT: [[A3:%.*]] = bitcast i32* [[A:%.*]] to i8* 18; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N:%.*]], 2 19; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]] 20; CHECK: vector.scevcheck: 21; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[N]], -1 22; CHECK-NEXT: [[TMP7:%.*]] = icmp ugt i32 [[TMP0]], 3 23; CHECK-NEXT: br i1 [[TMP7]], label [[SCALAR_PH]], label [[VECTOR_MEMCHECK:%.*]] 24; CHECK: vector.memcheck: 25; CHECK-NEXT: [[TMP10:%.*]] = add i32 [[N]], -1 26; CHECK-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 27; CHECK-NEXT: [[TMP12:%.*]] = add nuw nsw i64 [[TMP11]], 1 28; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i32, i32* [[B]], i64 [[TMP12]] 29; CHECK-NEXT: [[SCEVGEP2:%.*]] = bitcast i32* [[SCEVGEP]] to i8* 30; CHECK-NEXT: [[SCEVGEP4:%.*]] = getelementptr i32, i32* [[A]], i64 [[TMP12]] 31; CHECK-NEXT: [[SCEVGEP45:%.*]] = bitcast i32* [[SCEVGEP4]] to i8* 32; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult i8* [[B1]], [[SCEVGEP45]] 33; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult i8* [[A3]], [[SCEVGEP2]] 34; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] 35; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] 36; CHECK: vector.ph: 37; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[N]], 2 38; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[N]], [[N_MOD_VF]] 39; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 40; CHECK: vector.body: 41; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 42; CHECK-NEXT: [[TMP13:%.*]] = add i32 [[INDEX]], 0 43; CHECK-NEXT: [[TMP14:%.*]] = urem i32 [[TMP13]], 4 44; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[TMP14]] 45; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds i32, i32* [[TMP15]], i32 0 46; CHECK-NEXT: [[TMP17:%.*]] = bitcast i32* [[TMP16]] to <2 x i32>* 47; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, <2 x i32>* [[TMP17]], align 4, !alias.scope !0 48; CHECK-NEXT: [[TMP18:%.*]] = add <2 x i32> [[WIDE_LOAD]], <i32 10, i32 10> 49; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 [[TMP13]] 50; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds i32, i32* [[TMP19]], i32 0 51; CHECK-NEXT: [[TMP21:%.*]] = bitcast i32* [[TMP20]] to <2 x i32>* 52; CHECK-NEXT: store <2 x i32> [[TMP18]], <2 x i32>* [[TMP21]], align 4, !alias.scope !3, !noalias !0 53; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 54; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 55; CHECK-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] 56; CHECK: middle.block: 57; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[N]], [[N_VEC]] 58; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] 59; CHECK: scalar.ph: 60; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ] 61; CHECK-NEXT: br label [[LOOP:%.*]] 62; CHECK: loop: 63; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ] 64; CHECK-NEXT: [[CLAMPED_INDEX:%.*]] = urem i32 [[IV]], 4 65; CHECK-NEXT: [[GEP_A:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[CLAMPED_INDEX]] 66; CHECK-NEXT: [[LV:%.*]] = load i32, i32* [[GEP_A]], align 4 67; CHECK-NEXT: [[ADD:%.*]] = add i32 [[LV]], 10 68; CHECK-NEXT: [[GEP_B:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 [[IV]] 69; CHECK-NEXT: store i32 [[ADD]], i32* [[GEP_B]], align 4 70; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1 71; CHECK-NEXT: [[COND:%.*]] = icmp eq i32 [[IV_NEXT]], [[N]] 72; CHECK-NEXT: br i1 [[COND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP7:![0-9]+]] 73; CHECK: exit: 74; CHECK-NEXT: ret void 75; 76entry: 77 br label %loop 78 79loop: 80 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ] 81 %clamped.index = urem i32 %iv, 4 82 %gep.A = getelementptr inbounds i32, i32* %A, i32 %clamped.index 83 %lv = load i32, i32* %gep.A 84 %add = add i32 %lv, 10 85 %gep.B = getelementptr inbounds i32, i32* %B, i32 %iv 86 store i32 %add, i32* %gep.B 87 %iv.next = add nuw nsw i32 %iv, 1 88 %cond = icmp eq i32 %iv.next, %N 89 br i1 %cond, label %exit, label %loop 90 91exit: 92 ret void 93} 94 95; The relevant bounds for %gep.A are [%A, %A+4). 96define void @store_clamped_index(i32* %A, i32* %B, i32 %N) { 97; CHECK-LABEL: @store_clamped_index( 98; CHECK-NEXT: entry: 99; CHECK-NEXT: [[B1:%.*]] = bitcast i32* [[B:%.*]] to i8* 100; CHECK-NEXT: [[A3:%.*]] = bitcast i32* [[A:%.*]] to i8* 101; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N:%.*]], 2 102; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]] 103; CHECK: vector.scevcheck: 104; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[N]], -1 105; CHECK-NEXT: [[TMP7:%.*]] = icmp ugt i32 [[TMP0]], 3 106; CHECK-NEXT: br i1 [[TMP7]], label [[SCALAR_PH]], label [[VECTOR_MEMCHECK:%.*]] 107; CHECK: vector.memcheck: 108; CHECK-NEXT: [[TMP10:%.*]] = add i32 [[N]], -1 109; CHECK-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 110; CHECK-NEXT: [[TMP12:%.*]] = add nuw nsw i64 [[TMP11]], 1 111; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i32, i32* [[B]], i64 [[TMP12]] 112; CHECK-NEXT: [[SCEVGEP2:%.*]] = bitcast i32* [[SCEVGEP]] to i8* 113; CHECK-NEXT: [[SCEVGEP4:%.*]] = getelementptr i32, i32* [[A]], i64 [[TMP12]] 114; CHECK-NEXT: [[SCEVGEP45:%.*]] = bitcast i32* [[SCEVGEP4]] to i8* 115; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult i8* [[B1]], [[SCEVGEP45]] 116; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult i8* [[A3]], [[SCEVGEP2]] 117; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] 118; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] 119; CHECK: vector.ph: 120; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[N]], 2 121; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[N]], [[N_MOD_VF]] 122; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 123; CHECK: vector.body: 124; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 125; CHECK-NEXT: [[TMP13:%.*]] = add i32 [[INDEX]], 0 126; CHECK-NEXT: [[TMP14:%.*]] = urem i32 [[TMP13]], 4 127; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 [[TMP13]] 128; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds i32, i32* [[TMP15]], i32 0 129; CHECK-NEXT: [[TMP17:%.*]] = bitcast i32* [[TMP16]] to <2 x i32>* 130; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, <2 x i32>* [[TMP17]], align 4, !alias.scope !8, !noalias !11 131; CHECK-NEXT: [[TMP18:%.*]] = add <2 x i32> [[WIDE_LOAD]], <i32 10, i32 10> 132; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[TMP14]] 133; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds i32, i32* [[TMP19]], i32 0 134; CHECK-NEXT: [[TMP21:%.*]] = bitcast i32* [[TMP20]] to <2 x i32>* 135; CHECK-NEXT: store <2 x i32> [[TMP18]], <2 x i32>* [[TMP21]], align 4, !alias.scope !11 136; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 137; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 138; CHECK-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]] 139; CHECK: middle.block: 140; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[N]], [[N_VEC]] 141; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] 142; CHECK: scalar.ph: 143; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ] 144; CHECK-NEXT: br label [[LOOP:%.*]] 145; CHECK: loop: 146; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ] 147; CHECK-NEXT: [[CLAMPED_INDEX:%.*]] = urem i32 [[IV]], 4 148; CHECK-NEXT: [[GEP_B:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 [[IV]] 149; CHECK-NEXT: [[LV:%.*]] = load i32, i32* [[GEP_B]], align 4 150; CHECK-NEXT: [[ADD:%.*]] = add i32 [[LV]], 10 151; CHECK-NEXT: [[GEP_A:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[CLAMPED_INDEX]] 152; CHECK-NEXT: store i32 [[ADD]], i32* [[GEP_A]], align 4 153; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1 154; CHECK-NEXT: [[COND:%.*]] = icmp eq i32 [[IV_NEXT]], [[N]] 155; CHECK-NEXT: br i1 [[COND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP14:![0-9]+]] 156; CHECK: exit: 157; CHECK-NEXT: ret void 158; 159entry: 160 br label %loop 161 162loop: 163 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ] 164 %clamped.index = urem i32 %iv, 4 165 %gep.B = getelementptr inbounds i32, i32* %B, i32 %iv 166 %lv = load i32, i32* %gep.B 167 %add = add i32 %lv, 10 168 %gep.A = getelementptr inbounds i32, i32* %A, i32 %clamped.index 169 store i32 %add, i32* %gep.A 170 %iv.next = add nuw nsw i32 %iv, 1 171 %cond = icmp eq i32 %iv.next, %N 172 br i1 %cond, label %exit, label %loop 173 174exit: 175 ret void 176} 177 178define void @clamped_index_dependence_non_clamped(i32* %A, i32* %B, i32 %N) { 179; CHECK-LABEL: @clamped_index_dependence_non_clamped( 180; CHECK-NEXT: entry: 181; CHECK-NEXT: br label [[LOOP:%.*]] 182; CHECK: loop: 183; CHECK-NEXT: [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ] 184; CHECK-NEXT: [[GEP_B:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i32 [[IV]] 185; CHECK-NEXT: [[LV:%.*]] = load i32, i32* [[GEP_B]], align 4 186; CHECK-NEXT: [[GEP_A_1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[IV]] 187; CHECK-NEXT: [[LV_A:%.*]] = load i32, i32* [[GEP_A_1]], align 4 188; CHECK-NEXT: [[ADD:%.*]] = add i32 [[LV]], [[LV_A]] 189; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1 190; CHECK-NEXT: [[CLAMPED_INDEX:%.*]] = urem i32 [[IV_NEXT]], 4 191; CHECK-NEXT: [[GEP_A:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[CLAMPED_INDEX]] 192; CHECK-NEXT: store i32 [[ADD]], i32* [[GEP_A]], align 4 193; CHECK-NEXT: [[COND:%.*]] = icmp eq i32 [[IV_NEXT]], [[N:%.*]] 194; CHECK-NEXT: br i1 [[COND]], label [[EXIT:%.*]], label [[LOOP]] 195; CHECK: exit: 196; CHECK-NEXT: ret void 197; 198entry: 199 br label %loop 200 201loop: 202 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ] 203 %gep.B = getelementptr inbounds i32, i32* %B, i32 %iv 204 %lv = load i32, i32* %gep.B 205 %gep.A.1 = getelementptr inbounds i32, i32* %A, i32 %iv 206 %lv.A = load i32, i32* %gep.A.1 207 %add = add i32 %lv, %lv.A 208 209 %iv.next = add nuw nsw i32 %iv, 1 210 %clamped.index = urem i32 %iv.next, 4 211 %gep.A = getelementptr inbounds i32, i32* %A, i32 %clamped.index 212 store i32 %add, i32* %gep.A 213 %cond = icmp eq i32 %iv.next, %N 214 br i1 %cond, label %exit, label %loop 215 216exit: 217 ret void 218} 219 220define void @clamped_index_dependence_clamped_index(i32* %A, i32* %B, i32 %N) { 221; CHECK-LABEL: @clamped_index_dependence_clamped_index( 222; CHECK-NEXT: entry: 223; CHECK-NEXT: br label [[LOOP:%.*]] 224; CHECK: loop: 225; CHECK-NEXT: [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ] 226; CHECK-NEXT: [[CLAMPED_INDEX_1:%.*]] = urem i32 [[IV]], 4 227; CHECK-NEXT: [[GEP_A_1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[CLAMPED_INDEX_1]] 228; CHECK-NEXT: [[LV_A:%.*]] = load i32, i32* [[GEP_A_1]], align 4 229; CHECK-NEXT: [[ADD:%.*]] = add i32 [[LV_A]], 10 230; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1 231; CHECK-NEXT: [[CLAMPED_INDEX:%.*]] = urem i32 [[IV_NEXT]], 4 232; CHECK-NEXT: [[GEP_A:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[CLAMPED_INDEX]] 233; CHECK-NEXT: store i32 [[ADD]], i32* [[GEP_A]], align 4 234; CHECK-NEXT: [[COND:%.*]] = icmp eq i32 [[IV_NEXT]], [[N:%.*]] 235; CHECK-NEXT: br i1 [[COND]], label [[EXIT:%.*]], label [[LOOP]] 236; CHECK: exit: 237; CHECK-NEXT: ret void 238; 239entry: 240 br label %loop 241 242loop: 243 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ] 244 %clamped.index.1 = urem i32 %iv, 4 245 %gep.A.1 = getelementptr inbounds i32, i32* %A, i32 %clamped.index.1 246 %lv.A = load i32, i32* %gep.A.1 247 %add = add i32 %lv.A, 10 248 249 %iv.next = add nuw nsw i32 %iv, 1 250 %clamped.index = urem i32 %iv.next, 4 251 %gep.A = getelementptr inbounds i32, i32* %A, i32 %clamped.index 252 store i32 %add, i32* %gep.A 253 %cond = icmp eq i32 %iv.next, %N 254 br i1 %cond, label %exit, label %loop 255 256exit: 257 ret void 258} 259 260define void @clamped_index_equal_dependence(i32* %A, i32* %B, i32 %N) { 261; CHECK-LABEL: @clamped_index_equal_dependence( 262; CHECK-NEXT: entry: 263; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N:%.*]], 2 264; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]] 265; CHECK: vector.scevcheck: 266; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[N]], -1 267; CHECK-NEXT: [[TMP7:%.*]] = icmp ugt i32 [[TMP0]], 3 268; CHECK-NEXT: br i1 [[TMP7]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] 269; CHECK: vector.ph: 270; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[N]], 2 271; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[N]], [[N_MOD_VF]] 272; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 273; CHECK: vector.body: 274; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 275; CHECK-NEXT: [[TMP10:%.*]] = add i32 [[INDEX]], 0 276; CHECK-NEXT: [[TMP11:%.*]] = urem i32 [[TMP10]], 4 277; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[TMP11]] 278; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, i32* [[TMP12]], i32 0 279; CHECK-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to <2 x i32>* 280; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, <2 x i32>* [[TMP14]], align 4 281; CHECK-NEXT: [[TMP15:%.*]] = add <2 x i32> [[WIDE_LOAD]], <i32 10, i32 10> 282; CHECK-NEXT: [[TMP16:%.*]] = bitcast i32* [[TMP13]] to <2 x i32>* 283; CHECK-NEXT: store <2 x i32> [[TMP15]], <2 x i32>* [[TMP16]], align 4 284; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 285; CHECK-NEXT: [[TMP17:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 286; CHECK-NEXT: br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]] 287; CHECK: middle.block: 288; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[N]], [[N_VEC]] 289; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] 290; CHECK: scalar.ph: 291; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_SCEVCHECK]] ] 292; CHECK-NEXT: br label [[LOOP:%.*]] 293; CHECK: loop: 294; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ] 295; CHECK-NEXT: [[CLAMPED_INDEX:%.*]] = urem i32 [[IV]], 4 296; CHECK-NEXT: [[GEP_A:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[CLAMPED_INDEX]] 297; CHECK-NEXT: [[LV_A:%.*]] = load i32, i32* [[GEP_A]], align 4 298; CHECK-NEXT: [[ADD:%.*]] = add i32 [[LV_A]], 10 299; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1 300; CHECK-NEXT: store i32 [[ADD]], i32* [[GEP_A]], align 4 301; CHECK-NEXT: [[COND:%.*]] = icmp eq i32 [[IV_NEXT]], [[N]] 302; CHECK-NEXT: br i1 [[COND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP16:![0-9]+]] 303; CHECK: exit: 304; CHECK-NEXT: ret void 305; 306entry: 307 br label %loop 308 309loop: 310 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ] 311 %clamped.index = urem i32 %iv, 4 312 %gep.A = getelementptr inbounds i32, i32* %A, i32 %clamped.index 313 %lv.A = load i32, i32* %gep.A 314 %add = add i32 %lv.A, 10 315 316 %iv.next = add nuw nsw i32 %iv, 1 317 store i32 %add, i32* %gep.A 318 %cond = icmp eq i32 %iv.next, %N 319 br i1 %cond, label %exit, label %loop 320 321exit: 322 ret void 323} 324