1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -loop-vectorize -scalable-vectorization=on -riscv-v-vector-bits-min=128 -mtriple riscv64-linux-gnu -mattr=+v,+f -S 2>%t | FileCheck %s -check-prefix=CHECK 3 4; A collection of fairly basic functional tests when both fixed and scalable vectorization is 5; allowed. The primary goal of this is check for crashes during cost modeling, but it also 6; exercises the default heuristics in a useful way. 7 8target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n64-S128" 9target triple = "riscv64" 10 11define void @vector_add(ptr noalias nocapture %a, i64 %v, i64 %n) { 12; CHECK-LABEL: @vector_add( 13; CHECK-NEXT: entry: 14; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 15; CHECK: vector.ph: 16; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[V:%.*]], i32 0 17; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer 18; CHECK-NEXT: [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <2 x i64> poison, i64 [[V]], i32 0 19; CHECK-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT2]], <2 x i64> poison, <2 x i32> zeroinitializer 20; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 21; CHECK: vector.body: 22; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 23; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 24; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 2 25; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP0]] 26; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP1]] 27; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 0 28; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP4]], align 8 29; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 2 30; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <2 x i64>, ptr [[TMP5]], align 8 31; CHECK-NEXT: [[TMP6:%.*]] = add <2 x i64> [[WIDE_LOAD]], [[BROADCAST_SPLAT]] 32; CHECK-NEXT: [[TMP7:%.*]] = add <2 x i64> [[WIDE_LOAD1]], [[BROADCAST_SPLAT3]] 33; CHECK-NEXT: store <2 x i64> [[TMP6]], ptr [[TMP4]], align 8 34; CHECK-NEXT: store <2 x i64> [[TMP7]], ptr [[TMP5]], align 8 35; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 36; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 37; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 38; CHECK: middle.block: 39; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, 1024 40; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 41; CHECK: scalar.ph: 42; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 43; CHECK-NEXT: br label [[FOR_BODY:%.*]] 44; CHECK: for.body: 45; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 46; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] 47; CHECK-NEXT: [[ELEM:%.*]] = load i64, ptr [[ARRAYIDX]], align 8 48; CHECK-NEXT: [[ADD:%.*]] = add i64 [[ELEM]], [[V]] 49; CHECK-NEXT: store i64 [[ADD]], ptr [[ARRAYIDX]], align 8 50; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 51; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 52; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]] 53; CHECK: for.end: 54; CHECK-NEXT: ret void 55; 56entry: 57 br label %for.body 58 59for.body: 60 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 61 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv 62 %elem = load i64, ptr %arrayidx 63 %add = add i64 %elem, %v 64 store i64 %add, ptr %arrayidx 65 %iv.next = add nuw nsw i64 %iv, 1 66 %exitcond.not = icmp eq i64 %iv.next, 1024 67 br i1 %exitcond.not, label %for.end, label %for.body 68 69for.end: 70 ret void 71} 72 73; a[b[i]] += v, mostly to exercise scatter/gather costing 74; TODO: Currently fails to vectorize due to a memory conflict 75define void @indexed_add(ptr noalias nocapture %a, ptr noalias nocapture %b, i64 %v, i64 %n) { 76; CHECK-LABEL: @indexed_add( 77; CHECK-NEXT: entry: 78; CHECK-NEXT: br label [[FOR_BODY:%.*]] 79; CHECK: for.body: 80; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 81; CHECK-NEXT: [[BADDR:%.*]] = getelementptr inbounds i64, ptr [[B:%.*]], i64 [[IV]] 82; CHECK-NEXT: [[AIDX:%.*]] = load i64, ptr [[BADDR]], align 8 83; CHECK-NEXT: [[AADDR:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[AIDX]] 84; CHECK-NEXT: [[ELEM:%.*]] = load i64, ptr [[AADDR]], align 8 85; CHECK-NEXT: [[ADD:%.*]] = add i64 [[ELEM]], [[V:%.*]] 86; CHECK-NEXT: store i64 [[ADD]], ptr [[AADDR]], align 8 87; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 88; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 89; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END:%.*]], label [[FOR_BODY]] 90; CHECK: for.end: 91; CHECK-NEXT: ret void 92; 93entry: 94 br label %for.body 95 96for.body: 97 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 98 %baddr = getelementptr inbounds i64, ptr %b, i64 %iv 99 %aidx = load i64, ptr %baddr 100 %aaddr = getelementptr inbounds i64, ptr %a, i64 %aidx 101 %elem = load i64, ptr %aaddr 102 %add = add i64 %elem, %v 103 store i64 %add, ptr %aaddr 104 %iv.next = add nuw nsw i64 %iv, 1 105 %exitcond.not = icmp eq i64 %iv.next, 1024 106 br i1 %exitcond.not, label %for.end, label %for.body 107 108for.end: 109 ret void 110} 111 112; a[b[i]] = v, exercise scatter support 113define void @indexed_store(ptr noalias nocapture %a, ptr noalias nocapture %b, i64 %v, i64 %n) { 114; CHECK-LABEL: @indexed_store( 115; CHECK-NEXT: entry: 116; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 117; CHECK: vector.ph: 118; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[V:%.*]], i32 0 119; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer 120; CHECK-NEXT: [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <2 x i64> poison, i64 [[V]], i32 0 121; CHECK-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT2]], <2 x i64> poison, <2 x i32> zeroinitializer 122; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 123; CHECK: vector.body: 124; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 125; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 126; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 2 127; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[B:%.*]], i64 [[TMP0]] 128; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[TMP1]] 129; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 0 130; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP4]], align 8 131; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 2 132; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <2 x i64>, ptr [[TMP5]], align 8 133; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], <2 x i64> [[WIDE_LOAD]] 134; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[A]], <2 x i64> [[WIDE_LOAD1]] 135; CHECK-NEXT: call void @llvm.masked.scatter.v2i64.v2p0(<2 x i64> [[BROADCAST_SPLAT]], <2 x ptr> [[TMP6]], i32 8, <2 x i1> <i1 true, i1 true>) 136; CHECK-NEXT: call void @llvm.masked.scatter.v2i64.v2p0(<2 x i64> [[BROADCAST_SPLAT3]], <2 x ptr> [[TMP7]], i32 8, <2 x i1> <i1 true, i1 true>) 137; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 138; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 139; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] 140; CHECK: middle.block: 141; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, 1024 142; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 143; CHECK: scalar.ph: 144; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 145; CHECK-NEXT: br label [[FOR_BODY:%.*]] 146; CHECK: for.body: 147; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 148; CHECK-NEXT: [[BADDR:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[IV]] 149; CHECK-NEXT: [[AIDX:%.*]] = load i64, ptr [[BADDR]], align 8 150; CHECK-NEXT: [[AADDR:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[AIDX]] 151; CHECK-NEXT: store i64 [[V]], ptr [[AADDR]], align 8 152; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 153; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 154; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] 155; CHECK: for.end: 156; CHECK-NEXT: ret void 157; 158entry: 159 br label %for.body 160 161for.body: 162 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 163 %baddr = getelementptr inbounds i64, ptr %b, i64 %iv 164 %aidx = load i64, ptr %baddr 165 %aaddr = getelementptr inbounds i64, ptr %a, i64 %aidx 166 store i64 %v, ptr %aaddr 167 %iv.next = add nuw nsw i64 %iv, 1 168 %exitcond.not = icmp eq i64 %iv.next, 1024 169 br i1 %exitcond.not, label %for.end, label %for.body 170 171for.end: 172 ret void 173} 174 175define i64 @indexed_load(ptr noalias nocapture %a, ptr noalias nocapture %b, i64 %v, i64 %n) { 176; CHECK-LABEL: @indexed_load( 177; CHECK-NEXT: entry: 178; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 179; CHECK: vector.ph: 180; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 181; CHECK: vector.body: 182; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 183; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP8:%.*]], [[VECTOR_BODY]] ] 184; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP9:%.*]], [[VECTOR_BODY]] ] 185; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 186; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 2 187; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[B:%.*]], i64 [[TMP0]] 188; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[TMP1]] 189; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 0 190; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP4]], align 8 191; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 2 192; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <2 x i64>, ptr [[TMP5]], align 8 193; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], <2 x i64> [[WIDE_LOAD]] 194; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[A]], <2 x i64> [[WIDE_LOAD2]] 195; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <2 x i64> @llvm.masked.gather.v2i64.v2p0(<2 x ptr> [[TMP6]], i32 8, <2 x i1> <i1 true, i1 true>, <2 x i64> undef) 196; CHECK-NEXT: [[WIDE_MASKED_GATHER3:%.*]] = call <2 x i64> @llvm.masked.gather.v2i64.v2p0(<2 x ptr> [[TMP7]], i32 8, <2 x i1> <i1 true, i1 true>, <2 x i64> undef) 197; CHECK-NEXT: [[TMP8]] = add <2 x i64> [[VEC_PHI]], [[WIDE_MASKED_GATHER]] 198; CHECK-NEXT: [[TMP9]] = add <2 x i64> [[VEC_PHI1]], [[WIDE_MASKED_GATHER3]] 199; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 200; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 201; CHECK-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] 202; CHECK: middle.block: 203; CHECK-NEXT: [[BIN_RDX:%.*]] = add <2 x i64> [[TMP9]], [[TMP8]] 204; CHECK-NEXT: [[TMP11:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX]]) 205; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, 1024 206; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 207; CHECK: scalar.ph: 208; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 209; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[TMP11]], [[MIDDLE_BLOCK]] ] 210; CHECK-NEXT: br label [[FOR_BODY:%.*]] 211; CHECK: for.body: 212; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 213; CHECK-NEXT: [[SUM:%.*]] = phi i64 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[SUM_NEXT:%.*]], [[FOR_BODY]] ] 214; CHECK-NEXT: [[BADDR:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[IV]] 215; CHECK-NEXT: [[AIDX:%.*]] = load i64, ptr [[BADDR]], align 8 216; CHECK-NEXT: [[AADDR:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[AIDX]] 217; CHECK-NEXT: [[ELEM:%.*]] = load i64, ptr [[AADDR]], align 8 218; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 219; CHECK-NEXT: [[SUM_NEXT]] = add i64 [[SUM]], [[ELEM]] 220; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 221; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] 222; CHECK: for.end: 223; CHECK-NEXT: [[SUM_NEXT_LCSSA:%.*]] = phi i64 [ [[SUM_NEXT]], [[FOR_BODY]] ], [ [[TMP11]], [[MIDDLE_BLOCK]] ] 224; CHECK-NEXT: ret i64 [[SUM_NEXT_LCSSA]] 225; 226entry: 227 br label %for.body 228 229for.body: 230 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 231 %sum = phi i64 [0, %entry], [%sum.next, %for.body] 232 %baddr = getelementptr inbounds i64, ptr %b, i64 %iv 233 %aidx = load i64, ptr %baddr 234 %aaddr = getelementptr inbounds i64, ptr %a, i64 %aidx 235 %elem = load i64, ptr %aaddr 236 %iv.next = add nuw nsw i64 %iv, 1 237 %sum.next = add i64 %sum, %elem 238 %exitcond.not = icmp eq i64 %iv.next, 1024 239 br i1 %exitcond.not, label %for.end, label %for.body 240 241for.end: 242 ret i64 %sum.next 243} 244 245define void @splat_int(ptr noalias nocapture %a, i64 %v, i64 %n) { 246; CHECK-LABEL: @splat_int( 247; CHECK-NEXT: entry: 248; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 249; CHECK: vector.ph: 250; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[V:%.*]], i32 0 251; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer 252; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x i64> poison, i64 [[V]], i32 0 253; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT1]], <2 x i64> poison, <2 x i32> zeroinitializer 254; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 255; CHECK: vector.body: 256; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 257; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 258; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 2 259; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP0]] 260; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP1]] 261; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 0 262; CHECK-NEXT: store <2 x i64> [[BROADCAST_SPLAT]], ptr [[TMP4]], align 8 263; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 2 264; CHECK-NEXT: store <2 x i64> [[BROADCAST_SPLAT2]], ptr [[TMP5]], align 8 265; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 266; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 267; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] 268; CHECK: middle.block: 269; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, 1024 270; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 271; CHECK: scalar.ph: 272; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 273; CHECK-NEXT: br label [[FOR_BODY:%.*]] 274; CHECK: for.body: 275; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 276; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] 277; CHECK-NEXT: store i64 [[V]], ptr [[ARRAYIDX]], align 8 278; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 279; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 280; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] 281; CHECK: for.end: 282; CHECK-NEXT: ret void 283; 284entry: 285 br label %for.body 286 287for.body: 288 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 289 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv 290 store i64 %v, ptr %arrayidx 291 %iv.next = add nuw nsw i64 %iv, 1 292 %exitcond.not = icmp eq i64 %iv.next, 1024 293 br i1 %exitcond.not, label %for.end, label %for.body 294 295for.end: 296 ret void 297} 298 299define void @splat_ptr(ptr noalias nocapture %a, ptr %v, i64 %n) { 300; CHECK-LABEL: @splat_ptr( 301; CHECK-NEXT: entry: 302; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 303; CHECK: vector.ph: 304; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x ptr> poison, ptr [[V:%.*]], i32 0 305; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x ptr> [[BROADCAST_SPLATINSERT]], <2 x ptr> poison, <2 x i32> zeroinitializer 306; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x ptr> poison, ptr [[V]], i32 0 307; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x ptr> [[BROADCAST_SPLATINSERT1]], <2 x ptr> poison, <2 x i32> zeroinitializer 308; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 309; CHECK: vector.body: 310; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 311; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 312; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 2 313; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP0]] 314; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP1]] 315; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds ptr, ptr [[TMP2]], i32 0 316; CHECK-NEXT: store <2 x ptr> [[BROADCAST_SPLAT]], ptr [[TMP4]], align 8 317; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds ptr, ptr [[TMP2]], i32 2 318; CHECK-NEXT: store <2 x ptr> [[BROADCAST_SPLAT2]], ptr [[TMP5]], align 8 319; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 320; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 321; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] 322; CHECK: middle.block: 323; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, 1024 324; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 325; CHECK: scalar.ph: 326; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 327; CHECK-NEXT: br label [[FOR_BODY:%.*]] 328; CHECK: for.body: 329; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 330; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] 331; CHECK-NEXT: store ptr [[V]], ptr [[ARRAYIDX]], align 8 332; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 333; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 334; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]] 335; CHECK: for.end: 336; CHECK-NEXT: ret void 337; 338entry: 339 br label %for.body 340 341for.body: 342 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 343 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv 344 store ptr %v, ptr %arrayidx 345 %iv.next = add nuw nsw i64 %iv, 1 346 %exitcond.not = icmp eq i64 %iv.next, 1024 347 br i1 %exitcond.not, label %for.end, label %for.body 348 349for.end: 350 ret void 351} 352 353define void @uniform_store(ptr noalias nocapture %a, ptr noalias nocapture %b, i64 %v, i64 %n) { 354; CHECK-LABEL: @uniform_store( 355; CHECK-NEXT: entry: 356; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 357; CHECK: vector.ph: 358; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[V:%.*]], i32 0 359; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer 360; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x i64> poison, i64 [[V]], i32 0 361; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT1]], <2 x i64> poison, <2 x i32> zeroinitializer 362; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 363; CHECK: vector.body: 364; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 365; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 366; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 2 367; CHECK-NEXT: store i64 [[V]], ptr [[B:%.*]], align 8 368; CHECK-NEXT: store i64 [[V]], ptr [[B]], align 8 369; CHECK-NEXT: store i64 [[V]], ptr [[B]], align 8 370; CHECK-NEXT: store i64 [[V]], ptr [[B]], align 8 371; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP0]] 372; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP1]] 373; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 0 374; CHECK-NEXT: store <2 x i64> [[BROADCAST_SPLAT]], ptr [[TMP4]], align 8 375; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 2 376; CHECK-NEXT: store <2 x i64> [[BROADCAST_SPLAT2]], ptr [[TMP5]], align 8 377; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 378; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 379; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]] 380; CHECK: middle.block: 381; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, 1024 382; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 383; CHECK: scalar.ph: 384; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 385; CHECK-NEXT: br label [[FOR_BODY:%.*]] 386; CHECK: for.body: 387; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 388; CHECK-NEXT: store i64 [[V]], ptr [[B]], align 8 389; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] 390; CHECK-NEXT: store i64 [[V]], ptr [[ARRAYIDX]], align 8 391; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 392; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 393; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]] 394; CHECK: for.end: 395; CHECK-NEXT: ret void 396; 397entry: 398 br label %for.body 399 400for.body: 401 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 402 store i64 %v, ptr %b, align 8 403 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv 404 store i64 %v, ptr %arrayidx 405 %iv.next = add nuw nsw i64 %iv, 1 406 %exitcond.not = icmp eq i64 %iv.next, 1024 407 br i1 %exitcond.not, label %for.end, label %for.body 408 409for.end: 410 ret void 411} 412 413define void @uniform_store_unaligned(ptr noalias nocapture %a, ptr noalias nocapture %b, i64 %v, i64 %n) { 414; CHECK-LABEL: @uniform_store_unaligned( 415; CHECK-NEXT: entry: 416; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 417; CHECK: vector.ph: 418; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[V:%.*]], i32 0 419; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer 420; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x i64> poison, i64 [[V]], i32 0 421; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT1]], <2 x i64> poison, <2 x i32> zeroinitializer 422; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 423; CHECK: vector.body: 424; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 425; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 426; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 2 427; CHECK-NEXT: store i64 [[V]], ptr [[B:%.*]], align 1 428; CHECK-NEXT: store i64 [[V]], ptr [[B]], align 1 429; CHECK-NEXT: store i64 [[V]], ptr [[B]], align 1 430; CHECK-NEXT: store i64 [[V]], ptr [[B]], align 1 431; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP0]] 432; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP1]] 433; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 0 434; CHECK-NEXT: store <2 x i64> [[BROADCAST_SPLAT]], ptr [[TMP4]], align 8 435; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 2 436; CHECK-NEXT: store <2 x i64> [[BROADCAST_SPLAT2]], ptr [[TMP5]], align 8 437; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 438; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 439; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]] 440; CHECK: middle.block: 441; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, 1024 442; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 443; CHECK: scalar.ph: 444; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 445; CHECK-NEXT: br label [[FOR_BODY:%.*]] 446; CHECK: for.body: 447; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 448; CHECK-NEXT: store i64 [[V]], ptr [[B]], align 1 449; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] 450; CHECK-NEXT: store i64 [[V]], ptr [[ARRAYIDX]], align 8 451; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 452; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 453; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]] 454; CHECK: for.end: 455; CHECK-NEXT: ret void 456; 457entry: 458 br label %for.body 459 460for.body: 461 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 462 store i64 %v, ptr %b, align 1 463 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv 464 store i64 %v, ptr %arrayidx 465 %iv.next = add nuw nsw i64 %iv, 1 466 %exitcond.not = icmp eq i64 %iv.next, 1024 467 br i1 %exitcond.not, label %for.end, label %for.body 468 469for.end: 470 ret void 471} 472 473define i64 @uniform_load(ptr noalias nocapture %a, ptr noalias nocapture %b, i64 %n) { 474; CHECK-LABEL: @uniform_load( 475; CHECK-NEXT: entry: 476; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 477; CHECK: vector.ph: 478; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 479; CHECK: vector.body: 480; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 481; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 482; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 2 483; CHECK-NEXT: [[TMP2:%.*]] = load i64, ptr [[B:%.*]], align 8 484; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[TMP2]], i32 0 485; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer 486; CHECK-NEXT: [[TMP3:%.*]] = load i64, ptr [[B]], align 8 487; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x i64> poison, i64 [[TMP3]], i32 0 488; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT1]], <2 x i64> poison, <2 x i32> zeroinitializer 489; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP0]] 490; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP1]] 491; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[TMP4]], i32 0 492; CHECK-NEXT: store <2 x i64> [[BROADCAST_SPLAT]], ptr [[TMP6]], align 8 493; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[TMP4]], i32 2 494; CHECK-NEXT: store <2 x i64> [[BROADCAST_SPLAT2]], ptr [[TMP7]], align 8 495; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 496; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 497; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]] 498; CHECK: middle.block: 499; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, 1024 500; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 501; CHECK: scalar.ph: 502; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 503; CHECK-NEXT: br label [[FOR_BODY:%.*]] 504; CHECK: for.body: 505; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 506; CHECK-NEXT: [[V:%.*]] = load i64, ptr [[B]], align 8 507; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] 508; CHECK-NEXT: store i64 [[V]], ptr [[ARRAYIDX]], align 8 509; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 510; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 511; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP17:![0-9]+]] 512; CHECK: for.end: 513; CHECK-NEXT: [[V_LCSSA:%.*]] = phi i64 [ [[V]], [[FOR_BODY]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ] 514; CHECK-NEXT: ret i64 [[V_LCSSA]] 515; 516entry: 517 br label %for.body 518 519for.body: 520 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 521 %v = load i64, ptr %b, align 8 522 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv 523 store i64 %v, ptr %arrayidx 524 %iv.next = add nuw nsw i64 %iv, 1 525 %exitcond.not = icmp eq i64 %iv.next, 1024 526 br i1 %exitcond.not, label %for.end, label %for.body 527 528for.end: 529 ret i64 %v 530} 531 532define i64 @uniform_load_unaligned(ptr noalias nocapture %a, ptr noalias nocapture %b, i64 %n) { 533; CHECK-LABEL: @uniform_load_unaligned( 534; CHECK-NEXT: entry: 535; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 536; CHECK: vector.ph: 537; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 538; CHECK: vector.body: 539; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 540; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 541; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 2 542; CHECK-NEXT: [[TMP2:%.*]] = load i64, ptr [[B:%.*]], align 1 543; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[TMP2]], i32 0 544; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer 545; CHECK-NEXT: [[TMP3:%.*]] = load i64, ptr [[B]], align 1 546; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x i64> poison, i64 [[TMP3]], i32 0 547; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT1]], <2 x i64> poison, <2 x i32> zeroinitializer 548; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP0]] 549; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP1]] 550; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[TMP4]], i32 0 551; CHECK-NEXT: store <2 x i64> [[BROADCAST_SPLAT]], ptr [[TMP6]], align 8 552; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[TMP4]], i32 2 553; CHECK-NEXT: store <2 x i64> [[BROADCAST_SPLAT2]], ptr [[TMP7]], align 8 554; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 555; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 556; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]] 557; CHECK: middle.block: 558; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, 1024 559; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 560; CHECK: scalar.ph: 561; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 562; CHECK-NEXT: br label [[FOR_BODY:%.*]] 563; CHECK: for.body: 564; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 565; CHECK-NEXT: [[V:%.*]] = load i64, ptr [[B]], align 1 566; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] 567; CHECK-NEXT: store i64 [[V]], ptr [[ARRAYIDX]], align 8 568; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 569; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1024 570; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP19:![0-9]+]] 571; CHECK: for.end: 572; CHECK-NEXT: [[V_LCSSA:%.*]] = phi i64 [ [[V]], [[FOR_BODY]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ] 573; CHECK-NEXT: ret i64 [[V_LCSSA]] 574; 575entry: 576 br label %for.body 577 578for.body: 579 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 580 %v = load i64, ptr %b, align 1 581 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv 582 store i64 %v, ptr %arrayidx 583 %iv.next = add nuw nsw i64 %iv, 1 584 %exitcond.not = icmp eq i64 %iv.next, 1024 585 br i1 %exitcond.not, label %for.end, label %for.body 586 587for.end: 588 ret i64 %v 589} 590