1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --replace-value-regex "!llvm.loop ![0-9]+" --version 5 2; RUN: opt -S -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=2 < %s | FileCheck --check-prefixes=CHECK,VEC %s 3; RUN: opt -S -passes=loop-vectorize -force-vector-interleave=2 -force-vector-width=1 < %s | FileCheck --check-prefixes=CHECK,INTERLEAVE %s 4 5define i32 @postinc(i32 %k) { 6; CHECK-LABEL: define i32 @postinc( 7; CHECK-SAME: i32 [[K:%.*]]) { 8; CHECK-NEXT: [[ENTRY:.*]]: 9; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[K]], 2 10; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 11; CHECK: [[VECTOR_PH]]: 12; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[K]], 2 13; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[K]], [[N_MOD_VF]] 14; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] 15; CHECK: [[VECTOR_BODY]]: 16; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 17; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 18; CHECK-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 19; CHECK-NEXT: br i1 [[TMP0]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 20; CHECK: [[MIDDLE_BLOCK]]: 21; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[K]], [[N_VEC]] 22; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_END:.*]], label %[[SCALAR_PH]] 23; CHECK: [[SCALAR_PH]]: 24; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 25; CHECK-NEXT: br label %[[FOR_BODY:.*]] 26; CHECK: [[FOR_BODY]]: 27; CHECK-NEXT: [[INC_PHI:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY]] ] 28; CHECK-NEXT: [[INC]] = add nsw i32 [[INC_PHI]], 1 29; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[INC]], [[K]] 30; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_END]], label %[[FOR_BODY]], {{!llvm.loop ![0-9]+}} 31; CHECK: [[FOR_END]]: 32; CHECK-NEXT: [[INC_LCSSA:%.*]] = phi i32 [ [[INC]], %[[FOR_BODY]] ], [ [[N_VEC]], %[[MIDDLE_BLOCK]] ] 33; CHECK-NEXT: ret i32 [[INC_LCSSA]] 34; 35entry: 36 br label %for.body 37 38for.body: 39 %inc.phi = phi i32 [ 0, %entry ], [ %inc, %for.body ] 40 %inc = add nsw i32 %inc.phi, 1 41 %cmp = icmp eq i32 %inc, %k 42 br i1 %cmp, label %for.end, label %for.body 43 44for.end: 45 ret i32 %inc 46} 47 48define i32 @preinc(i32 %k) { 49; CHECK-LABEL: define i32 @preinc( 50; CHECK-SAME: i32 [[K:%.*]]) { 51; CHECK-NEXT: [[ENTRY:.*]]: 52; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[K]], 2 53; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 54; CHECK: [[VECTOR_PH]]: 55; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[K]], 2 56; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[K]], [[N_MOD_VF]] 57; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] 58; CHECK: [[VECTOR_BODY]]: 59; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 60; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 61; CHECK-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 62; CHECK-NEXT: br i1 [[TMP0]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 63; CHECK: [[MIDDLE_BLOCK]]: 64; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[K]], [[N_VEC]] 65; CHECK-NEXT: [[IND_ESCAPE:%.*]] = sub i32 [[N_VEC]], 1 66; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_END:.*]], label %[[SCALAR_PH]] 67; CHECK: [[SCALAR_PH]]: 68; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 69; CHECK-NEXT: br label %[[FOR_BODY:.*]] 70; CHECK: [[FOR_BODY]]: 71; CHECK-NEXT: [[INC_PHI:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY]] ] 72; CHECK-NEXT: [[INC]] = add nsw i32 [[INC_PHI]], 1 73; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[INC]], [[K]] 74; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_END]], label %[[FOR_BODY]], {{!llvm.loop ![0-9]+}} 75; CHECK: [[FOR_END]]: 76; CHECK-NEXT: [[INC_PHI_LCSSA:%.*]] = phi i32 [ [[INC_PHI]], %[[FOR_BODY]] ], [ [[IND_ESCAPE]], %[[MIDDLE_BLOCK]] ] 77; CHECK-NEXT: ret i32 [[INC_PHI_LCSSA]] 78; 79entry: 80 br label %for.body 81 82for.body: 83 %inc.phi = phi i32 [ 0, %entry ], [ %inc, %for.body ] 84 %inc = add nsw i32 %inc.phi, 1 85 %cmp = icmp eq i32 %inc, %k 86 br i1 %cmp, label %for.end, label %for.body 87 88for.end: 89 ret i32 %inc.phi 90} 91 92define i32 @constpre() { 93; CHECK-LABEL: define i32 @constpre() { 94; CHECK-NEXT: [[ENTRY:.*]]: 95; CHECK-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 96; CHECK: [[VECTOR_PH]]: 97; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] 98; CHECK: [[VECTOR_BODY]]: 99; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 100; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 101; CHECK-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 16 102; CHECK-NEXT: br i1 [[TMP0]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 103; CHECK: [[MIDDLE_BLOCK]]: 104; CHECK-NEXT: br i1 true, label %[[FOR_END:.*]], label %[[SCALAR_PH]] 105; CHECK: [[SCALAR_PH]]: 106; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 0, %[[MIDDLE_BLOCK]] ], [ 32, %[[ENTRY]] ] 107; CHECK-NEXT: br label %[[FOR_BODY:.*]] 108; CHECK: [[FOR_BODY]]: 109; CHECK-NEXT: [[INC_PHI:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY]] ] 110; CHECK-NEXT: [[INC]] = sub nsw i32 [[INC_PHI]], 2 111; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[INC]], 0 112; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_END]], label %[[FOR_BODY]], {{!llvm.loop ![0-9]+}} 113; CHECK: [[FOR_END]]: 114; CHECK-NEXT: [[INC_PHI_LCSSA:%.*]] = phi i32 [ [[INC_PHI]], %[[FOR_BODY]] ], [ 2, %[[MIDDLE_BLOCK]] ] 115; CHECK-NEXT: ret i32 [[INC_PHI_LCSSA]] 116; 117entry: 118 br label %for.body 119 120for.body: 121 %inc.phi = phi i32 [ 32, %entry ], [ %inc, %for.body ] 122 %inc = sub nsw i32 %inc.phi, 2 123 %cmp = icmp eq i32 %inc, 0 124 br i1 %cmp, label %for.end, label %for.body 125 126for.end: 127 ret i32 %inc.phi 128} 129 130define ptr @geppre(ptr %ptr) { 131; CHECK-LABEL: define ptr @geppre( 132; CHECK-SAME: ptr [[PTR:%.*]]) { 133; CHECK-NEXT: [[ENTRY:.*]]: 134; CHECK-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 135; CHECK: [[VECTOR_PH]]: 136; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[PTR]], i64 512 137; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] 138; CHECK: [[VECTOR_BODY]]: 139; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 140; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 141; CHECK-NEXT: [[TMP0:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32 142; CHECK-NEXT: br i1 [[TMP0]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 143; CHECK: [[MIDDLE_BLOCK]]: 144; CHECK-NEXT: [[IND_ESCAPE:%.*]] = getelementptr i8, ptr [[IND_END]], i64 -16 145; CHECK-NEXT: br i1 true, label %[[FOR_END:.*]], label %[[SCALAR_PH]] 146; CHECK: [[SCALAR_PH]]: 147; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 32, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 148; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi ptr [ [[IND_END]], %[[MIDDLE_BLOCK]] ], [ [[PTR]], %[[ENTRY]] ] 149; CHECK-NEXT: br label %[[FOR_BODY:.*]] 150; CHECK: [[FOR_BODY]]: 151; CHECK-NEXT: [[INC_PHI:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY]] ] 152; CHECK-NEXT: [[PTR_PHI:%.*]] = phi ptr [ [[BC_RESUME_VAL1]], %[[SCALAR_PH]] ], [ [[INC_PTR:%.*]], %[[FOR_BODY]] ] 153; CHECK-NEXT: [[INC]] = add nsw i32 [[INC_PHI]], 1 154; CHECK-NEXT: [[INC_PTR]] = getelementptr i32, ptr [[PTR_PHI]], i32 4 155; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[INC]], 32 156; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_END]], label %[[FOR_BODY]], {{!llvm.loop ![0-9]+}} 157; CHECK: [[FOR_END]]: 158; CHECK-NEXT: [[PTR_PHI_LCSSA:%.*]] = phi ptr [ [[PTR_PHI]], %[[FOR_BODY]] ], [ [[IND_ESCAPE]], %[[MIDDLE_BLOCK]] ] 159; CHECK-NEXT: ret ptr [[PTR_PHI_LCSSA]] 160; 161entry: 162 br label %for.body 163 164for.body: 165 %inc.phi = phi i32 [ 0, %entry ], [ %inc, %for.body ] 166 %ptr.phi = phi ptr [ %ptr, %entry ], [ %inc.ptr, %for.body ] 167 %inc = add nsw i32 %inc.phi, 1 168 %inc.ptr = getelementptr i32, ptr %ptr.phi, i32 4 169 %cmp = icmp eq i32 %inc, 32 170 br i1 %cmp, label %for.end, label %for.body 171 172for.end: 173 ret ptr %ptr.phi 174} 175 176define ptr @both(i32 %k) { 177; CHECK-LABEL: define ptr @both( 178; CHECK-SAME: i32 [[K:%.*]]) { 179; CHECK-NEXT: [[ENTRY:.*]]: 180; CHECK-NEXT: [[BASE:%.*]] = getelementptr inbounds i32, ptr undef, i64 1 181; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[K]], -1 182; CHECK-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 183; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1 184; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 2 185; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 186; CHECK: [[VECTOR_PH]]: 187; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 2 188; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]] 189; CHECK-NEXT: [[IND_END:%.*]] = trunc i64 [[N_VEC]] to i32 190; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[N_VEC]], 4 191; CHECK-NEXT: [[IND_END1:%.*]] = getelementptr i8, ptr [[BASE]], i64 [[TMP3]] 192; CHECK-NEXT: [[TMP4:%.*]] = mul i64 [[N_VEC]], 4 193; CHECK-NEXT: [[IND_END2:%.*]] = getelementptr i8, ptr undef, i64 [[TMP4]] 194; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] 195; CHECK: [[VECTOR_BODY]]: 196; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 197; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 198; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 199; CHECK-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 200; CHECK: [[MIDDLE_BLOCK]]: 201; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]] 202; CHECK-NEXT: [[IND_ESCAPE:%.*]] = getelementptr i8, ptr [[IND_END1]], i64 -4 203; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_END:.*]], label %[[SCALAR_PH]] 204; CHECK: [[SCALAR_PH]]: 205; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 206; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi ptr [ [[IND_END1]], %[[MIDDLE_BLOCK]] ], [ [[BASE]], %[[ENTRY]] ] 207; CHECK-NEXT: [[BC_RESUME_VAL2:%.*]] = phi ptr [ [[IND_END2]], %[[MIDDLE_BLOCK]] ], [ undef, %[[ENTRY]] ] 208; CHECK-NEXT: br label %[[FOR_BODY:.*]] 209; CHECK: [[FOR_BODY]]: 210; CHECK-NEXT: [[INC_PHI:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY]] ] 211; CHECK-NEXT: [[INC_LAG1:%.*]] = phi ptr [ [[BC_RESUME_VAL1]], %[[SCALAR_PH]] ], [ [[TMP:%.*]], %[[FOR_BODY]] ] 212; CHECK-NEXT: [[INC_LAG2:%.*]] = phi ptr [ [[BC_RESUME_VAL2]], %[[SCALAR_PH]] ], [ [[INC_LAG1]], %[[FOR_BODY]] ] 213; CHECK-NEXT: [[TMP]] = getelementptr inbounds i32, ptr [[INC_LAG1]], i64 1 214; CHECK-NEXT: [[INC]] = add nsw i32 [[INC_PHI]], 1 215; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[INC]], [[K]] 216; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_END]], label %[[FOR_BODY]], {{!llvm.loop ![0-9]+}} 217; CHECK: [[FOR_END]]: 218; CHECK-NEXT: [[INC_LAG1_LCSSA:%.*]] = phi ptr [ [[INC_LAG1]], %[[FOR_BODY]] ], [ [[IND_ESCAPE]], %[[MIDDLE_BLOCK]] ] 219; CHECK-NEXT: ret ptr [[INC_LAG1_LCSSA]] 220; 221entry: 222 %base = getelementptr inbounds i32, ptr undef, i64 1 223 br label %for.body 224 225for.body: 226 %inc.phi = phi i32 [ 0, %entry ], [ %inc, %for.body ] 227 %inc.lag1 = phi ptr [ %base, %entry ], [ %tmp, %for.body] 228 %inc.lag2 = phi ptr [ undef, %entry ], [ %inc.lag1, %for.body] 229 %tmp = getelementptr inbounds i32, ptr %inc.lag1, i64 1 230 %inc = add nsw i32 %inc.phi, 1 231 %cmp = icmp eq i32 %inc, %k 232 br i1 %cmp, label %for.end, label %for.body 233 234for.end: 235 ret ptr %inc.lag1 236} 237 238define i32 @multiphi(i32 %k, ptr %p) { 239; CHECK-LABEL: define i32 @multiphi( 240; CHECK-SAME: i32 [[K:%.*]], ptr [[P:%.*]]) { 241; CHECK-NEXT: [[ENTRY:.*]]: 242; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[K]], 2 243; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 244; CHECK: [[VECTOR_PH]]: 245; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[K]], 2 246; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[K]], [[N_MOD_VF]] 247; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] 248; CHECK: [[VECTOR_BODY]]: 249; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 250; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 251; CHECK-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 252; CHECK-NEXT: br i1 [[TMP0]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 253; CHECK: [[MIDDLE_BLOCK]]: 254; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[K]], [[N_VEC]] 255; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_END:.*]], label %[[SCALAR_PH]] 256; CHECK: [[SCALAR_PH]]: 257; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 258; CHECK-NEXT: br label %[[FOR_BODY:.*]] 259; CHECK: [[FOR_BODY]]: 260; CHECK-NEXT: [[INC_PHI:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY]] ] 261; CHECK-NEXT: [[INC]] = add nsw i32 [[INC_PHI]], 1 262; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[INC]], [[K]] 263; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_END]], label %[[FOR_BODY]], {{!llvm.loop ![0-9]+}} 264; CHECK: [[FOR_END]]: 265; CHECK-NEXT: [[PHI:%.*]] = phi i32 [ [[INC]], %[[FOR_BODY]] ], [ [[N_VEC]], %[[MIDDLE_BLOCK]] ] 266; CHECK-NEXT: [[PHI2:%.*]] = phi i32 [ [[INC]], %[[FOR_BODY]] ], [ [[N_VEC]], %[[MIDDLE_BLOCK]] ] 267; CHECK-NEXT: store i32 [[PHI2]], ptr [[P]], align 4 268; CHECK-NEXT: ret i32 [[PHI]] 269; 270entry: 271 br label %for.body 272 273for.body: 274 %inc.phi = phi i32 [ 0, %entry ], [ %inc, %for.body ] 275 %inc = add nsw i32 %inc.phi, 1 276 %cmp = icmp eq i32 %inc, %k 277 br i1 %cmp, label %for.end, label %for.body 278 279for.end: 280 %phi = phi i32 [ %inc, %for.body ] 281 %phi2 = phi i32 [ %inc, %for.body ] 282 store i32 %phi2, ptr %p 283 ret i32 %phi 284} 285 286define void @PR30742() { 287; CHECK-LABEL: define void @PR30742() { 288; CHECK-NEXT: [[BB0:.*:]] 289; CHECK-NEXT: br label %[[BB1:.*]] 290; CHECK: [[BB1_LOOPEXIT:.*]]: 291; CHECK-NEXT: br label %[[BB1]] 292; CHECK: [[BB1]]: 293; CHECK-NEXT: [[TMP00:%.*]] = load i32, ptr undef, align 16 294; CHECK-NEXT: [[TMP01:%.*]] = sub i32 [[TMP00]], undef 295; CHECK-NEXT: [[TMP02:%.*]] = icmp slt i32 [[TMP01]], 1 296; CHECK-NEXT: [[TMP03:%.*]] = select i1 [[TMP02]], i32 1, i32 [[TMP01]] 297; CHECK-NEXT: [[TMP04:%.*]] = add nsw i32 [[TMP03]], -7 298; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[TMP03]], -8 299; CHECK-NEXT: [[TMP1:%.*]] = add nsw i32 [[TMP03]], -15 300; CHECK-NEXT: [[SMIN1:%.*]] = call i32 @llvm.smin.i32(i32 [[TMP1]], i32 0) 301; CHECK-NEXT: [[TMP2:%.*]] = sub i32 [[TMP0]], [[SMIN1]] 302; CHECK-NEXT: [[TMP3:%.*]] = lshr i32 [[TMP2]], 3 303; CHECK-NEXT: [[TMP4:%.*]] = add nuw nsw i32 [[TMP3]], 1 304; CHECK-NEXT: [[MIN_ITERS_CHECK4:%.*]] = icmp ult i32 [[TMP4]], 2 305; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK4]], label %[[SCALAR_PH3:.*]], label %[[VECTOR_PH5:.*]] 306; CHECK: [[VECTOR_PH5]]: 307; CHECK-NEXT: [[N_MOD_VF6:%.*]] = urem i32 [[TMP4]], 2 308; CHECK-NEXT: [[N_VEC7:%.*]] = sub i32 [[TMP4]], [[N_MOD_VF6]] 309; CHECK-NEXT: [[TMP5:%.*]] = mul i32 [[N_VEC7]], -8 310; CHECK-NEXT: [[IND_END8:%.*]] = add i32 [[TMP04]], [[TMP5]] 311; CHECK-NEXT: br label %[[VECTOR_BODY8:.*]] 312; CHECK: [[VECTOR_BODY8]]: 313; CHECK-NEXT: [[INDEX10:%.*]] = phi i32 [ 0, %[[VECTOR_PH5]] ], [ [[INDEX_NEXT11:%.*]], %[[VECTOR_BODY8]] ] 314; CHECK-NEXT: [[INDEX_NEXT11]] = add nuw i32 [[INDEX10]], 2 315; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT11]], [[N_VEC7]] 316; CHECK-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK2:.*]], label %[[VECTOR_BODY8]], {{!llvm.loop ![0-9]+}} 317; CHECK: [[MIDDLE_BLOCK2]]: 318; CHECK-NEXT: [[CMP_N12:%.*]] = icmp eq i32 [[TMP4]], [[N_VEC7]] 319; CHECK-NEXT: [[IND_ESCAPE:%.*]] = sub i32 [[IND_END8]], -8 320; CHECK-NEXT: br i1 [[CMP_N12]], label %[[BB3:.*]], label %[[SCALAR_PH3]] 321; CHECK: [[SCALAR_PH3]]: 322; CHECK-NEXT: [[BC_RESUME_VAL12:%.*]] = phi i32 [ [[IND_END8]], %[[MIDDLE_BLOCK2]] ], [ [[TMP04]], %[[BB1]] ] 323; CHECK-NEXT: br label %[[BB2:.*]] 324; CHECK: [[BB2]]: 325; CHECK-NEXT: [[TMP05:%.*]] = phi i32 [ [[BC_RESUME_VAL12]], %[[SCALAR_PH3]] ], [ [[TMP06:%.*]], %[[BB2]] ] 326; CHECK-NEXT: [[TMP06]] = add i32 [[TMP05]], -8 327; CHECK-NEXT: [[TMP07:%.*]] = icmp sgt i32 [[TMP06]], 0 328; CHECK-NEXT: br i1 [[TMP07]], label %[[BB2]], label %[[BB3]], {{!llvm.loop ![0-9]+}} 329; CHECK: [[BB3]]: 330; CHECK-NEXT: [[TMP08:%.*]] = phi i32 [ [[TMP05]], %[[BB2]] ], [ [[IND_ESCAPE]], %[[MIDDLE_BLOCK2]] ] 331; CHECK-NEXT: [[TMP09:%.*]] = sub i32 [[TMP00]], undef 332; CHECK-NEXT: [[TMP10:%.*]] = icmp slt i32 [[TMP09]], 1 333; CHECK-NEXT: [[TMP11:%.*]] = select i1 [[TMP10]], i32 1, i32 [[TMP09]] 334; CHECK-NEXT: [[TMP12:%.*]] = add nsw i32 [[TMP11]], -7 335; CHECK-NEXT: [[TMP7:%.*]] = add i32 [[TMP11]], -8 336; CHECK-NEXT: [[TMP8:%.*]] = add nsw i32 [[TMP11]], -15 337; CHECK-NEXT: [[SMIN:%.*]] = call i32 @llvm.smin.i32(i32 [[TMP8]], i32 0) 338; CHECK-NEXT: [[TMP10:%.*]] = sub i32 [[TMP7]], [[SMIN]] 339; CHECK-NEXT: [[TMP11:%.*]] = lshr i32 [[TMP10]], 3 340; CHECK-NEXT: [[TMP14:%.*]] = add nuw nsw i32 [[TMP11]], 1 341; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP14]], 2 342; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 343; CHECK: [[VECTOR_PH]]: 344; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP14]], 2 345; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP14]], [[N_MOD_VF]] 346; CHECK-NEXT: [[TMP16:%.*]] = mul i32 [[N_VEC]], -8 347; CHECK-NEXT: [[IND_END:%.*]] = add i32 [[TMP12]], [[TMP16]] 348; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] 349; CHECK: [[VECTOR_BODY]]: 350; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 351; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 352; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 353; CHECK-NEXT: br i1 [[TMP13]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 354; CHECK: [[MIDDLE_BLOCK]]: 355; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP14]], [[N_VEC]] 356; CHECK-NEXT: br i1 [[CMP_N]], label %[[BB1_LOOPEXIT]], label %[[SCALAR_PH]] 357; CHECK: [[SCALAR_PH]]: 358; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], %[[MIDDLE_BLOCK]] ], [ [[TMP12]], %[[BB3]] ] 359; CHECK-NEXT: br label %[[BB4:.*]] 360; CHECK: [[BB4]]: 361; CHECK-NEXT: [[TMP16:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IND_END:%.*]], %[[BB4]] ] 362; CHECK-NEXT: [[IND_END]] = add i32 [[TMP16]], -8 363; CHECK-NEXT: [[TMP13:%.*]] = icmp sgt i32 [[IND_END]], 0 364; CHECK-NEXT: br i1 [[TMP13]], label %[[BB4]], label %[[BB1_LOOPEXIT]], {{!llvm.loop ![0-9]+}} 365; 366BB0: 367 br label %BB1 368 369BB1: 370 %tmp00 = load i32, ptr undef, align 16 371 %tmp01 = sub i32 %tmp00, undef 372 %tmp02 = icmp slt i32 %tmp01, 1 373 %tmp03 = select i1 %tmp02, i32 1, i32 %tmp01 374 %tmp04 = add nsw i32 %tmp03, -7 375 br label %BB2 376 377BB2: 378 %tmp05 = phi i32 [ %tmp04, %BB1 ], [ %tmp06, %BB2 ] 379 %tmp06 = add i32 %tmp05, -8 380 %tmp07 = icmp sgt i32 %tmp06, 0 381 br i1 %tmp07, label %BB2, label %BB3 382 383BB3: 384 %tmp08 = phi i32 [ %tmp05, %BB2 ] 385 %tmp09 = sub i32 %tmp00, undef 386 %tmp10 = icmp slt i32 %tmp09, 1 387 %tmp11 = select i1 %tmp10, i32 1, i32 %tmp09 388 %tmp11.inc = add nsw i32 %tmp11, -7 389 br label %BB4 390 391BB4: 392 %tmp13 = phi i32 [ %tmp11.inc, %BB3 ], [ %tmp14, %BB4 ] 393 %tmp14 = add i32 %tmp13, -8 394 %tmp15 = icmp sgt i32 %tmp14, 0 395 br i1 %tmp15, label %BB4, label %BB1 396} 397 398; 399define i64 @iv_scalar_steps_and_outside_users(ptr %ptr) { 400; VEC-LABEL: define i64 @iv_scalar_steps_and_outside_users( 401; VEC-SAME: ptr [[PTR:%.*]]) { 402; VEC-NEXT: [[ENTRY:.*]]: 403; VEC-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 404; VEC: [[VECTOR_PH]]: 405; VEC-NEXT: br label %[[VECTOR_BODY:.*]] 406; VEC: [[VECTOR_BODY]]: 407; VEC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 408; VEC-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] 409; VEC-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 410; VEC-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[PTR]], i64 [[TMP0]] 411; VEC-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[TMP1]], i32 0 412; VEC-NEXT: store <2 x i64> [[VEC_IND]], ptr [[TMP2]], align 4 413; VEC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 414; VEC-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], splat (i64 2) 415; VEC-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1002 416; VEC-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 417; VEC: [[MIDDLE_BLOCK]]: 418; VEC-NEXT: br i1 true, label %[[EXIT:.*]], label %[[SCALAR_PH]] 419; VEC: [[SCALAR_PH]]: 420; VEC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1002, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 421; VEC-NEXT: br label %[[LOOP:.*]] 422; VEC: [[LOOP]]: 423; VEC-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 424; VEC-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1 425; VEC-NEXT: [[GEP_PTR:%.*]] = getelementptr inbounds i64, ptr [[PTR]], i64 [[IV]] 426; VEC-NEXT: store i64 [[IV]], ptr [[GEP_PTR]], align 4 427; VEC-NEXT: [[EXITCOND:%.*]] = icmp ugt i64 [[IV]], 1000 428; VEC-NEXT: br i1 [[EXITCOND]], label %[[EXIT]], label %[[LOOP]], {{!llvm.loop ![0-9]+}} 429; VEC: [[EXIT]]: 430; VEC-NEXT: [[IV_LCSSA:%.*]] = phi i64 [ [[IV]], %[[LOOP]] ], [ 1001, %[[MIDDLE_BLOCK]] ] 431; VEC-NEXT: ret i64 [[IV_LCSSA]] 432; 433; INTERLEAVE-LABEL: define i64 @iv_scalar_steps_and_outside_users( 434; INTERLEAVE-SAME: ptr [[PTR:%.*]]) { 435; INTERLEAVE-NEXT: [[ENTRY:.*]]: 436; INTERLEAVE-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 437; INTERLEAVE: [[VECTOR_PH]]: 438; INTERLEAVE-NEXT: br label %[[VECTOR_BODY:.*]] 439; INTERLEAVE: [[VECTOR_BODY]]: 440; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 441; INTERLEAVE-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 442; INTERLEAVE-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1 443; INTERLEAVE-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[PTR]], i64 [[TMP0]] 444; INTERLEAVE-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[PTR]], i64 [[TMP1]] 445; INTERLEAVE-NEXT: store i64 [[TMP0]], ptr [[TMP2]], align 4 446; INTERLEAVE-NEXT: store i64 [[TMP1]], ptr [[TMP3]], align 4 447; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 448; INTERLEAVE-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1002 449; INTERLEAVE-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 450; INTERLEAVE: [[MIDDLE_BLOCK]]: 451; INTERLEAVE-NEXT: br i1 true, label %[[EXIT:.*]], label %[[SCALAR_PH]] 452; INTERLEAVE: [[SCALAR_PH]]: 453; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1002, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 454; INTERLEAVE-NEXT: br label %[[LOOP:.*]] 455; INTERLEAVE: [[LOOP]]: 456; INTERLEAVE-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 457; INTERLEAVE-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1 458; INTERLEAVE-NEXT: [[GEP_PTR:%.*]] = getelementptr inbounds i64, ptr [[PTR]], i64 [[IV]] 459; INTERLEAVE-NEXT: store i64 [[IV]], ptr [[GEP_PTR]], align 4 460; INTERLEAVE-NEXT: [[EXITCOND:%.*]] = icmp ugt i64 [[IV]], 1000 461; INTERLEAVE-NEXT: br i1 [[EXITCOND]], label %[[EXIT]], label %[[LOOP]], {{!llvm.loop ![0-9]+}} 462; INTERLEAVE: [[EXIT]]: 463; INTERLEAVE-NEXT: [[IV_LCSSA:%.*]] = phi i64 [ [[IV]], %[[LOOP]] ], [ 1001, %[[MIDDLE_BLOCK]] ] 464; INTERLEAVE-NEXT: ret i64 [[IV_LCSSA]] 465; 466entry: 467 br label %loop 468 469loop: 470 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] 471 %iv.next = add nuw i64 %iv, 1 472 %gep.ptr = getelementptr inbounds i64, ptr %ptr, i64 %iv 473 store i64 %iv, ptr %gep.ptr 474 %exitcond = icmp ugt i64 %iv, 1000 475 br i1 %exitcond, label %exit, label %loop 476 477exit: 478 %iv.lcssa = phi i64 [ %iv, %loop ] 479 ret i64 %iv.lcssa 480} 481 482 483; %iv.2 is dead in the vector loop and only used outside the loop. 484define i32 @iv_2_dead_in_loop_only_used_outside(ptr %ptr) { 485; VEC-LABEL: define i32 @iv_2_dead_in_loop_only_used_outside( 486; VEC-SAME: ptr [[PTR:%.*]]) { 487; VEC-NEXT: [[ENTRY:.*]]: 488; VEC-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 489; VEC: [[VECTOR_PH]]: 490; VEC-NEXT: br label %[[VECTOR_BODY:.*]] 491; VEC: [[VECTOR_BODY]]: 492; VEC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 493; VEC-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] 494; VEC-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 495; VEC-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[PTR]], i64 [[TMP0]] 496; VEC-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[TMP1]], i32 0 497; VEC-NEXT: store <2 x i64> [[VEC_IND]], ptr [[TMP2]], align 4 498; VEC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 499; VEC-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], splat (i64 2) 500; VEC-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1002 501; VEC-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 502; VEC: [[MIDDLE_BLOCK]]: 503; VEC-NEXT: br i1 true, label %[[EXIT:.*]], label %[[SCALAR_PH]] 504; VEC: [[SCALAR_PH]]: 505; VEC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1002, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 506; VEC-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ 2004, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 507; VEC-NEXT: br label %[[LOOP:.*]] 508; VEC: [[LOOP]]: 509; VEC-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 510; VEC-NEXT: [[IV_2:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], %[[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], %[[LOOP]] ] 511; VEC-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1 512; VEC-NEXT: [[IV_2_NEXT]] = add nuw i32 [[IV_2]], 2 513; VEC-NEXT: [[GEP_PTR:%.*]] = getelementptr inbounds i64, ptr [[PTR]], i64 [[IV]] 514; VEC-NEXT: store i64 [[IV]], ptr [[GEP_PTR]], align 4 515; VEC-NEXT: [[EXITCOND:%.*]] = icmp ugt i64 [[IV]], 1000 516; VEC-NEXT: br i1 [[EXITCOND]], label %[[EXIT]], label %[[LOOP]], {{!llvm.loop ![0-9]+}} 517; VEC: [[EXIT]]: 518; VEC-NEXT: [[IV_2_LCSSA:%.*]] = phi i32 [ [[IV_2]], %[[LOOP]] ], [ 2002, %[[MIDDLE_BLOCK]] ] 519; VEC-NEXT: ret i32 [[IV_2_LCSSA]] 520; 521; INTERLEAVE-LABEL: define i32 @iv_2_dead_in_loop_only_used_outside( 522; INTERLEAVE-SAME: ptr [[PTR:%.*]]) { 523; INTERLEAVE-NEXT: [[ENTRY:.*]]: 524; INTERLEAVE-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 525; INTERLEAVE: [[VECTOR_PH]]: 526; INTERLEAVE-NEXT: br label %[[VECTOR_BODY:.*]] 527; INTERLEAVE: [[VECTOR_BODY]]: 528; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 529; INTERLEAVE-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 530; INTERLEAVE-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1 531; INTERLEAVE-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[PTR]], i64 [[TMP0]] 532; INTERLEAVE-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[PTR]], i64 [[TMP1]] 533; INTERLEAVE-NEXT: store i64 [[TMP0]], ptr [[TMP2]], align 4 534; INTERLEAVE-NEXT: store i64 [[TMP1]], ptr [[TMP3]], align 4 535; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 536; INTERLEAVE-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1002 537; INTERLEAVE-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 538; INTERLEAVE: [[MIDDLE_BLOCK]]: 539; INTERLEAVE-NEXT: br i1 true, label %[[EXIT:.*]], label %[[SCALAR_PH]] 540; INTERLEAVE: [[SCALAR_PH]]: 541; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1002, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 542; INTERLEAVE-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ 2004, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 543; INTERLEAVE-NEXT: br label %[[LOOP:.*]] 544; INTERLEAVE: [[LOOP]]: 545; INTERLEAVE-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 546; INTERLEAVE-NEXT: [[IV_2:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], %[[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], %[[LOOP]] ] 547; INTERLEAVE-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1 548; INTERLEAVE-NEXT: [[IV_2_NEXT]] = add nuw i32 [[IV_2]], 2 549; INTERLEAVE-NEXT: [[GEP_PTR:%.*]] = getelementptr inbounds i64, ptr [[PTR]], i64 [[IV]] 550; INTERLEAVE-NEXT: store i64 [[IV]], ptr [[GEP_PTR]], align 4 551; INTERLEAVE-NEXT: [[EXITCOND:%.*]] = icmp ugt i64 [[IV]], 1000 552; INTERLEAVE-NEXT: br i1 [[EXITCOND]], label %[[EXIT]], label %[[LOOP]], {{!llvm.loop ![0-9]+}} 553; INTERLEAVE: [[EXIT]]: 554; INTERLEAVE-NEXT: [[IV_2_LCSSA:%.*]] = phi i32 [ [[IV_2]], %[[LOOP]] ], [ 2002, %[[MIDDLE_BLOCK]] ] 555; INTERLEAVE-NEXT: ret i32 [[IV_2_LCSSA]] 556; 557entry: 558 br label %loop 559 560loop: 561 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] 562 %iv.2 = phi i32 [ 0, %entry ], [ %iv.2.next, %loop ] 563 %iv.next = add nuw i64 %iv, 1 564 %iv.2.next = add nuw i32 %iv.2, 2 565 %gep.ptr = getelementptr inbounds i64, ptr %ptr, i64 %iv 566 store i64 %iv, ptr %gep.ptr 567 %exitcond = icmp ugt i64 %iv, 1000 568 br i1 %exitcond, label %exit, label %loop 569 570exit: 571 %iv.2.lcssa = phi i32 [ %iv.2, %loop ] 572 ret i32 %iv.2.lcssa 573} 574 575define i32 @postinc_sub(i32 %k) { 576; CHECK-LABEL: define i32 @postinc_sub( 577; CHECK-SAME: i32 [[K:%.*]]) { 578; CHECK-NEXT: [[ENTRY:.*]]: 579; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[K]], 2 580; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 581; CHECK: [[VECTOR_PH]]: 582; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[K]], 2 583; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[K]], [[N_MOD_VF]] 584; CHECK-NEXT: [[TMP0:%.*]] = sub i32 [[K]], [[N_VEC]] 585; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] 586; CHECK: [[VECTOR_BODY]]: 587; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 588; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 589; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 590; CHECK-NEXT: br i1 [[TMP1]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 591; CHECK: [[MIDDLE_BLOCK]]: 592; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[K]], [[N_VEC]] 593; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_END:.*]], label %[[SCALAR_PH]] 594; CHECK: [[SCALAR_PH]]: 595; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[TMP0]], %[[MIDDLE_BLOCK]] ], [ [[K]], %[[ENTRY]] ] 596; CHECK-NEXT: br label %[[FOR_BODY:.*]] 597; CHECK: [[FOR_BODY]]: 598; CHECK-NEXT: [[INC_PHI:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY]] ] 599; CHECK-NEXT: [[INC]] = sub nsw i32 [[INC_PHI]], 1 600; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[INC]], 0 601; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_END]], label %[[FOR_BODY]], {{!llvm.loop ![0-9]+}} 602; CHECK: [[FOR_END]]: 603; CHECK-NEXT: [[INC_LCSSA:%.*]] = phi i32 [ [[INC]], %[[FOR_BODY]] ], [ [[TMP0]], %[[MIDDLE_BLOCK]] ] 604; CHECK-NEXT: ret i32 [[INC_LCSSA]] 605; 606entry: 607 br label %for.body 608 609for.body: 610 %inc.phi = phi i32 [ %k, %entry ], [ %inc, %for.body ] 611 %inc = sub nsw i32 %inc.phi, 1 612 %cmp = icmp eq i32 %inc, 0 613 br i1 %cmp, label %for.end, label %for.body 614 615for.end: 616 ret i32 %inc 617} 618 619define i32 @postinc_swapped_ops(i32 %k) { 620; CHECK-LABEL: define i32 @postinc_swapped_ops( 621; CHECK-SAME: i32 [[K:%.*]]) { 622; CHECK-NEXT: [[ENTRY:.*]]: 623; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[K]], 2 624; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 625; CHECK: [[VECTOR_PH]]: 626; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[K]], 2 627; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[K]], [[N_MOD_VF]] 628; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] 629; CHECK: [[VECTOR_BODY]]: 630; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 631; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 632; CHECK-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 633; CHECK-NEXT: br i1 [[TMP0]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 634; CHECK: [[MIDDLE_BLOCK]]: 635; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[K]], [[N_VEC]] 636; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_END:.*]], label %[[SCALAR_PH]] 637; CHECK: [[SCALAR_PH]]: 638; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 639; CHECK-NEXT: br label %[[FOR_BODY:.*]] 640; CHECK: [[FOR_BODY]]: 641; CHECK-NEXT: [[INC_PHI:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY]] ] 642; CHECK-NEXT: [[INC]] = add nsw i32 1, [[INC_PHI]] 643; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[INC]], [[K]] 644; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_END]], label %[[FOR_BODY]], {{!llvm.loop ![0-9]+}} 645; CHECK: [[FOR_END]]: 646; CHECK-NEXT: [[INC_LCSSA:%.*]] = phi i32 [ [[INC]], %[[FOR_BODY]] ], [ [[N_VEC]], %[[MIDDLE_BLOCK]] ] 647; CHECK-NEXT: ret i32 [[INC_LCSSA]] 648; 649entry: 650 br label %for.body 651 652for.body: 653 %inc.phi = phi i32 [ 0, %entry ], [ %inc, %for.body ] 654 %inc = add nsw i32 1, %inc.phi 655 %cmp = icmp eq i32 %inc, %k 656 br i1 %cmp, label %for.end, label %for.body 657 658for.end: 659 ret i32 %inc 660} 661 662define i32 @postinc_not_iv_backedge_value(i32 %k) { 663; VEC-LABEL: define i32 @postinc_not_iv_backedge_value( 664; VEC-SAME: i32 [[K:%.*]]) { 665; VEC-NEXT: [[ENTRY:.*]]: 666; VEC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[K]], 2 667; VEC-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 668; VEC: [[VECTOR_PH]]: 669; VEC-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[K]], 2 670; VEC-NEXT: [[N_VEC:%.*]] = sub i32 [[K]], [[N_MOD_VF]] 671; VEC-NEXT: br label %[[VECTOR_BODY:.*]] 672; VEC: [[VECTOR_BODY]]: 673; VEC-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 674; VEC-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] 675; VEC-NEXT: [[TMP0:%.*]] = add <2 x i32> [[VEC_IND]], splat (i32 2) 676; VEC-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 677; VEC-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2) 678; VEC-NEXT: [[TMP1:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 679; VEC-NEXT: br i1 [[TMP1]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 680; VEC: [[MIDDLE_BLOCK]]: 681; VEC-NEXT: [[TMP2:%.*]] = extractelement <2 x i32> [[TMP0]], i32 1 682; VEC-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[K]], [[N_VEC]] 683; VEC-NEXT: br i1 [[CMP_N]], label %[[FOR_END:.*]], label %[[SCALAR_PH]] 684; VEC: [[SCALAR_PH]]: 685; VEC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 686; VEC-NEXT: br label %[[FOR_BODY:.*]] 687; VEC: [[FOR_BODY]]: 688; VEC-NEXT: [[INC_PHI:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY]] ] 689; VEC-NEXT: [[INC]] = add nsw i32 [[INC_PHI]], 1 690; VEC-NEXT: [[INC_2:%.*]] = add i32 [[INC_PHI]], 2 691; VEC-NEXT: [[CMP:%.*]] = icmp eq i32 [[INC]], [[K]] 692; VEC-NEXT: br i1 [[CMP]], label %[[FOR_END]], label %[[FOR_BODY]], {{!llvm.loop ![0-9]+}} 693; VEC: [[FOR_END]]: 694; VEC-NEXT: [[INC_2_LCSSA:%.*]] = phi i32 [ [[INC_2]], %[[FOR_BODY]] ], [ [[TMP2]], %[[MIDDLE_BLOCK]] ] 695; VEC-NEXT: ret i32 [[INC_2_LCSSA]] 696; 697; INTERLEAVE-LABEL: define i32 @postinc_not_iv_backedge_value( 698; INTERLEAVE-SAME: i32 [[K:%.*]]) { 699; INTERLEAVE-NEXT: [[ENTRY:.*]]: 700; INTERLEAVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[K]], 2 701; INTERLEAVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 702; INTERLEAVE: [[VECTOR_PH]]: 703; INTERLEAVE-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[K]], 2 704; INTERLEAVE-NEXT: [[N_VEC:%.*]] = sub i32 [[K]], [[N_MOD_VF]] 705; INTERLEAVE-NEXT: br label %[[VECTOR_BODY:.*]] 706; INTERLEAVE: [[VECTOR_BODY]]: 707; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 708; INTERLEAVE-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 1 709; INTERLEAVE-NEXT: [[TMP1:%.*]] = add i32 [[TMP0]], 2 710; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 711; INTERLEAVE-NEXT: [[TMP2:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 712; INTERLEAVE-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 713; INTERLEAVE: [[MIDDLE_BLOCK]]: 714; INTERLEAVE-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[K]], [[N_VEC]] 715; INTERLEAVE-NEXT: br i1 [[CMP_N]], label %[[FOR_END:.*]], label %[[SCALAR_PH]] 716; INTERLEAVE: [[SCALAR_PH]]: 717; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 718; INTERLEAVE-NEXT: br label %[[FOR_BODY:.*]] 719; INTERLEAVE: [[FOR_BODY]]: 720; INTERLEAVE-NEXT: [[INC_PHI:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY]] ] 721; INTERLEAVE-NEXT: [[INC]] = add nsw i32 [[INC_PHI]], 1 722; INTERLEAVE-NEXT: [[INC_2:%.*]] = add i32 [[INC_PHI]], 2 723; INTERLEAVE-NEXT: [[CMP:%.*]] = icmp eq i32 [[INC]], [[K]] 724; INTERLEAVE-NEXT: br i1 [[CMP]], label %[[FOR_END]], label %[[FOR_BODY]], {{!llvm.loop ![0-9]+}} 725; INTERLEAVE: [[FOR_END]]: 726; INTERLEAVE-NEXT: [[INC_2_LCSSA:%.*]] = phi i32 [ [[INC_2]], %[[FOR_BODY]] ], [ [[TMP1]], %[[MIDDLE_BLOCK]] ] 727; INTERLEAVE-NEXT: ret i32 [[INC_2_LCSSA]] 728; 729entry: 730 br label %for.body 731 732for.body: 733 %inc.phi = phi i32 [ 0, %entry ], [ %inc, %for.body ] 734 %inc = add nsw i32 %inc.phi, 1 735 %inc.2 = add i32 %inc.phi, 2 736 %cmp = icmp eq i32 %inc, %k 737 br i1 %cmp, label %for.end, label %for.body 738 739for.end: 740 ret i32 %inc.2 741} 742 743define float @fp_postinc_use_fadd(float %init, ptr noalias nocapture %A, i64 %N, float %fpinc) { 744; VEC-LABEL: define float @fp_postinc_use_fadd( 745; VEC-SAME: float [[INIT:%.*]], ptr noalias captures(none) [[A:%.*]], i64 [[N:%.*]], float [[FPINC:%.*]]) { 746; VEC-NEXT: [[ENTRY:.*]]: 747; VEC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 2 748; VEC-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 749; VEC: [[VECTOR_PH]]: 750; VEC-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 2 751; VEC-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 752; VEC-NEXT: [[DOTCAST:%.*]] = sitofp i64 [[N_VEC]] to float 753; VEC-NEXT: [[TMP0:%.*]] = fmul fast float [[FPINC]], [[DOTCAST]] 754; VEC-NEXT: [[TMP1:%.*]] = fadd fast float [[INIT]], [[TMP0]] 755; VEC-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[INIT]], i64 0 756; VEC-NEXT: [[DOTSPLAT:%.*]] = shufflevector <2 x float> [[DOTSPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer 757; VEC-NEXT: [[DOTSPLATINSERT1:%.*]] = insertelement <2 x float> poison, float [[FPINC]], i64 0 758; VEC-NEXT: [[DOTSPLAT2:%.*]] = shufflevector <2 x float> [[DOTSPLATINSERT1]], <2 x float> poison, <2 x i32> zeroinitializer 759; VEC-NEXT: [[TMP2:%.*]] = fmul fast <2 x float> <float 0.000000e+00, float 1.000000e+00>, [[DOTSPLAT2]] 760; VEC-NEXT: [[INDUCTION:%.*]] = fadd fast <2 x float> [[DOTSPLAT]], [[TMP2]] 761; VEC-NEXT: [[TMP3:%.*]] = fmul fast float [[FPINC]], 2.000000e+00 762; VEC-NEXT: [[DOTSPLATINSERT3:%.*]] = insertelement <2 x float> poison, float [[TMP3]], i64 0 763; VEC-NEXT: [[DOTSPLAT4:%.*]] = shufflevector <2 x float> [[DOTSPLATINSERT3]], <2 x float> poison, <2 x i32> zeroinitializer 764; VEC-NEXT: br label %[[VECTOR_BODY:.*]] 765; VEC: [[VECTOR_BODY]]: 766; VEC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 767; VEC-NEXT: [[VEC_IND:%.*]] = phi <2 x float> [ [[INDUCTION]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] 768; VEC-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 0 769; VEC-NEXT: [[TMP5:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP4]] 770; VEC-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, ptr [[TMP5]], i32 0 771; VEC-NEXT: store <2 x float> [[VEC_IND]], ptr [[TMP6]], align 4 772; VEC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 773; VEC-NEXT: [[VEC_IND_NEXT]] = fadd fast <2 x float> [[VEC_IND]], [[DOTSPLAT4]] 774; VEC-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 775; VEC-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 776; VEC: [[MIDDLE_BLOCK]]: 777; VEC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 778; VEC-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]] 779; VEC: [[SCALAR_PH]]: 780; VEC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 781; VEC-NEXT: [[BC_RESUME_VAL5:%.*]] = phi float [ [[TMP1]], %[[MIDDLE_BLOCK]] ], [ [[INIT]], %[[ENTRY]] ] 782; VEC-NEXT: br label %[[LOOP:.*]] 783; VEC: [[LOOP]]: 784; VEC-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 785; VEC-NEXT: [[FP_IV:%.*]] = phi float [ [[BC_RESUME_VAL5]], %[[SCALAR_PH]] ], [ [[ADD:%.*]], %[[LOOP]] ] 786; VEC-NEXT: [[GEP_A:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]] 787; VEC-NEXT: store float [[FP_IV]], ptr [[GEP_A]], align 4 788; VEC-NEXT: [[ADD]] = fadd fast float [[FP_IV]], [[FPINC]] 789; VEC-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 790; VEC-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] 791; VEC-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], {{!llvm.loop ![0-9]+}} 792; VEC: [[EXIT]]: 793; VEC-NEXT: [[ADD_LCSSA:%.*]] = phi float [ [[ADD]], %[[LOOP]] ], [ [[TMP1]], %[[MIDDLE_BLOCK]] ] 794; VEC-NEXT: ret float [[ADD_LCSSA]] 795; 796; INTERLEAVE-LABEL: define float @fp_postinc_use_fadd( 797; INTERLEAVE-SAME: float [[INIT:%.*]], ptr noalias captures(none) [[A:%.*]], i64 [[N:%.*]], float [[FPINC:%.*]]) { 798; INTERLEAVE-NEXT: [[ENTRY:.*]]: 799; INTERLEAVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 2 800; INTERLEAVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 801; INTERLEAVE: [[VECTOR_PH]]: 802; INTERLEAVE-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 2 803; INTERLEAVE-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 804; INTERLEAVE-NEXT: [[DOTCAST:%.*]] = sitofp i64 [[N_VEC]] to float 805; INTERLEAVE-NEXT: [[TMP0:%.*]] = fmul fast float [[FPINC]], [[DOTCAST]] 806; INTERLEAVE-NEXT: [[TMP1:%.*]] = fadd fast float [[INIT]], [[TMP0]] 807; INTERLEAVE-NEXT: br label %[[VECTOR_BODY:.*]] 808; INTERLEAVE: [[VECTOR_BODY]]: 809; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 810; INTERLEAVE-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 0 811; INTERLEAVE-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 1 812; INTERLEAVE-NEXT: [[DOTCAST1:%.*]] = sitofp i64 [[INDEX]] to float 813; INTERLEAVE-NEXT: [[TMP4:%.*]] = fmul fast float [[FPINC]], [[DOTCAST1]] 814; INTERLEAVE-NEXT: [[OFFSET_IDX:%.*]] = fadd fast float [[INIT]], [[TMP4]] 815; INTERLEAVE-NEXT: [[TMP5:%.*]] = fmul fast float 0.000000e+00, [[FPINC]] 816; INTERLEAVE-NEXT: [[TMP6:%.*]] = fadd fast float [[OFFSET_IDX]], [[TMP5]] 817; INTERLEAVE-NEXT: [[TMP7:%.*]] = fmul fast float 1.000000e+00, [[FPINC]] 818; INTERLEAVE-NEXT: [[TMP8:%.*]] = fadd fast float [[OFFSET_IDX]], [[TMP7]] 819; INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP2]] 820; INTERLEAVE-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP3]] 821; INTERLEAVE-NEXT: store float [[TMP6]], ptr [[TMP9]], align 4 822; INTERLEAVE-NEXT: store float [[TMP8]], ptr [[TMP10]], align 4 823; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 824; INTERLEAVE-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 825; INTERLEAVE-NEXT: br i1 [[TMP11]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 826; INTERLEAVE: [[MIDDLE_BLOCK]]: 827; INTERLEAVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 828; INTERLEAVE-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]] 829; INTERLEAVE: [[SCALAR_PH]]: 830; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 831; INTERLEAVE-NEXT: [[BC_RESUME_VAL2:%.*]] = phi float [ [[TMP1]], %[[MIDDLE_BLOCK]] ], [ [[INIT]], %[[ENTRY]] ] 832; INTERLEAVE-NEXT: br label %[[LOOP:.*]] 833; INTERLEAVE: [[LOOP]]: 834; INTERLEAVE-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 835; INTERLEAVE-NEXT: [[FP_IV:%.*]] = phi float [ [[BC_RESUME_VAL2]], %[[SCALAR_PH]] ], [ [[ADD:%.*]], %[[LOOP]] ] 836; INTERLEAVE-NEXT: [[GEP_A:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]] 837; INTERLEAVE-NEXT: store float [[FP_IV]], ptr [[GEP_A]], align 4 838; INTERLEAVE-NEXT: [[ADD]] = fadd fast float [[FP_IV]], [[FPINC]] 839; INTERLEAVE-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 840; INTERLEAVE-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] 841; INTERLEAVE-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], {{!llvm.loop ![0-9]+}} 842; INTERLEAVE: [[EXIT]]: 843; INTERLEAVE-NEXT: [[ADD_LCSSA:%.*]] = phi float [ [[ADD]], %[[LOOP]] ], [ [[TMP1]], %[[MIDDLE_BLOCK]] ] 844; INTERLEAVE-NEXT: ret float [[ADD_LCSSA]] 845; 846entry: 847 br label %loop 848 849loop: 850 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] 851 %fp.iv = phi float [ %init, %entry ], [ %add, %loop ] 852 %gep.A = getelementptr inbounds float, ptr %A, i64 %iv 853 store float %fp.iv, ptr %gep.A, align 4 854 %add = fadd fast float %fp.iv, %fpinc 855 %iv.next = add nuw nsw i64 %iv, 1 856 %ec = icmp eq i64 %iv.next, %N 857 br i1 %ec, label %exit, label %loop 858 859exit: 860 ret float %add 861} 862 863define float @fp_postinc_use_fadd_ops_swapped(float %init, ptr noalias nocapture %A, i64 %N, float %fpinc) { 864; VEC-LABEL: define float @fp_postinc_use_fadd_ops_swapped( 865; VEC-SAME: float [[INIT:%.*]], ptr noalias captures(none) [[A:%.*]], i64 [[N:%.*]], float [[FPINC:%.*]]) { 866; VEC-NEXT: [[ENTRY:.*]]: 867; VEC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 2 868; VEC-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 869; VEC: [[VECTOR_PH]]: 870; VEC-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 2 871; VEC-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 872; VEC-NEXT: [[DOTCAST:%.*]] = sitofp i64 [[N_VEC]] to float 873; VEC-NEXT: [[TMP0:%.*]] = fmul fast float [[FPINC]], [[DOTCAST]] 874; VEC-NEXT: [[TMP1:%.*]] = fadd fast float [[INIT]], [[TMP0]] 875; VEC-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[INIT]], i64 0 876; VEC-NEXT: [[DOTSPLAT:%.*]] = shufflevector <2 x float> [[DOTSPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer 877; VEC-NEXT: [[DOTSPLATINSERT1:%.*]] = insertelement <2 x float> poison, float [[FPINC]], i64 0 878; VEC-NEXT: [[DOTSPLAT2:%.*]] = shufflevector <2 x float> [[DOTSPLATINSERT1]], <2 x float> poison, <2 x i32> zeroinitializer 879; VEC-NEXT: [[TMP2:%.*]] = fmul fast <2 x float> <float 0.000000e+00, float 1.000000e+00>, [[DOTSPLAT2]] 880; VEC-NEXT: [[INDUCTION:%.*]] = fadd fast <2 x float> [[DOTSPLAT]], [[TMP2]] 881; VEC-NEXT: [[TMP3:%.*]] = fmul fast float [[FPINC]], 2.000000e+00 882; VEC-NEXT: [[DOTSPLATINSERT3:%.*]] = insertelement <2 x float> poison, float [[TMP3]], i64 0 883; VEC-NEXT: [[DOTSPLAT4:%.*]] = shufflevector <2 x float> [[DOTSPLATINSERT3]], <2 x float> poison, <2 x i32> zeroinitializer 884; VEC-NEXT: br label %[[VECTOR_BODY:.*]] 885; VEC: [[VECTOR_BODY]]: 886; VEC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 887; VEC-NEXT: [[VEC_IND:%.*]] = phi <2 x float> [ [[INDUCTION]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] 888; VEC-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 0 889; VEC-NEXT: [[TMP5:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP4]] 890; VEC-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, ptr [[TMP5]], i32 0 891; VEC-NEXT: store <2 x float> [[VEC_IND]], ptr [[TMP6]], align 4 892; VEC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 893; VEC-NEXT: [[VEC_IND_NEXT]] = fadd fast <2 x float> [[VEC_IND]], [[DOTSPLAT4]] 894; VEC-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 895; VEC-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 896; VEC: [[MIDDLE_BLOCK]]: 897; VEC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 898; VEC-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]] 899; VEC: [[SCALAR_PH]]: 900; VEC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 901; VEC-NEXT: [[BC_RESUME_VAL5:%.*]] = phi float [ [[TMP1]], %[[MIDDLE_BLOCK]] ], [ [[INIT]], %[[ENTRY]] ] 902; VEC-NEXT: br label %[[LOOP:.*]] 903; VEC: [[LOOP]]: 904; VEC-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 905; VEC-NEXT: [[FP_IV:%.*]] = phi float [ [[BC_RESUME_VAL5]], %[[SCALAR_PH]] ], [ [[ADD:%.*]], %[[LOOP]] ] 906; VEC-NEXT: [[GEP_A:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]] 907; VEC-NEXT: store float [[FP_IV]], ptr [[GEP_A]], align 4 908; VEC-NEXT: [[ADD]] = fadd fast float [[FPINC]], [[FP_IV]] 909; VEC-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 910; VEC-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] 911; VEC-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], {{!llvm.loop ![0-9]+}} 912; VEC: [[EXIT]]: 913; VEC-NEXT: [[ADD_LCSSA:%.*]] = phi float [ [[ADD]], %[[LOOP]] ], [ [[TMP1]], %[[MIDDLE_BLOCK]] ] 914; VEC-NEXT: ret float [[ADD_LCSSA]] 915; 916; INTERLEAVE-LABEL: define float @fp_postinc_use_fadd_ops_swapped( 917; INTERLEAVE-SAME: float [[INIT:%.*]], ptr noalias captures(none) [[A:%.*]], i64 [[N:%.*]], float [[FPINC:%.*]]) { 918; INTERLEAVE-NEXT: [[ENTRY:.*]]: 919; INTERLEAVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 2 920; INTERLEAVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 921; INTERLEAVE: [[VECTOR_PH]]: 922; INTERLEAVE-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 2 923; INTERLEAVE-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 924; INTERLEAVE-NEXT: [[DOTCAST:%.*]] = sitofp i64 [[N_VEC]] to float 925; INTERLEAVE-NEXT: [[TMP0:%.*]] = fmul fast float [[FPINC]], [[DOTCAST]] 926; INTERLEAVE-NEXT: [[TMP1:%.*]] = fadd fast float [[INIT]], [[TMP0]] 927; INTERLEAVE-NEXT: br label %[[VECTOR_BODY:.*]] 928; INTERLEAVE: [[VECTOR_BODY]]: 929; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 930; INTERLEAVE-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 0 931; INTERLEAVE-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 1 932; INTERLEAVE-NEXT: [[DOTCAST1:%.*]] = sitofp i64 [[INDEX]] to float 933; INTERLEAVE-NEXT: [[TMP4:%.*]] = fmul fast float [[FPINC]], [[DOTCAST1]] 934; INTERLEAVE-NEXT: [[OFFSET_IDX:%.*]] = fadd fast float [[INIT]], [[TMP4]] 935; INTERLEAVE-NEXT: [[TMP5:%.*]] = fmul fast float 0.000000e+00, [[FPINC]] 936; INTERLEAVE-NEXT: [[TMP6:%.*]] = fadd fast float [[OFFSET_IDX]], [[TMP5]] 937; INTERLEAVE-NEXT: [[TMP7:%.*]] = fmul fast float 1.000000e+00, [[FPINC]] 938; INTERLEAVE-NEXT: [[TMP8:%.*]] = fadd fast float [[OFFSET_IDX]], [[TMP7]] 939; INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP2]] 940; INTERLEAVE-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP3]] 941; INTERLEAVE-NEXT: store float [[TMP6]], ptr [[TMP9]], align 4 942; INTERLEAVE-NEXT: store float [[TMP8]], ptr [[TMP10]], align 4 943; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 944; INTERLEAVE-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 945; INTERLEAVE-NEXT: br i1 [[TMP11]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 946; INTERLEAVE: [[MIDDLE_BLOCK]]: 947; INTERLEAVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 948; INTERLEAVE-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]] 949; INTERLEAVE: [[SCALAR_PH]]: 950; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 951; INTERLEAVE-NEXT: [[BC_RESUME_VAL2:%.*]] = phi float [ [[TMP1]], %[[MIDDLE_BLOCK]] ], [ [[INIT]], %[[ENTRY]] ] 952; INTERLEAVE-NEXT: br label %[[LOOP:.*]] 953; INTERLEAVE: [[LOOP]]: 954; INTERLEAVE-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 955; INTERLEAVE-NEXT: [[FP_IV:%.*]] = phi float [ [[BC_RESUME_VAL2]], %[[SCALAR_PH]] ], [ [[ADD:%.*]], %[[LOOP]] ] 956; INTERLEAVE-NEXT: [[GEP_A:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]] 957; INTERLEAVE-NEXT: store float [[FP_IV]], ptr [[GEP_A]], align 4 958; INTERLEAVE-NEXT: [[ADD]] = fadd fast float [[FPINC]], [[FP_IV]] 959; INTERLEAVE-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 960; INTERLEAVE-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] 961; INTERLEAVE-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], {{!llvm.loop ![0-9]+}} 962; INTERLEAVE: [[EXIT]]: 963; INTERLEAVE-NEXT: [[ADD_LCSSA:%.*]] = phi float [ [[ADD]], %[[LOOP]] ], [ [[TMP1]], %[[MIDDLE_BLOCK]] ] 964; INTERLEAVE-NEXT: ret float [[ADD_LCSSA]] 965; 966entry: 967 br label %loop 968 969loop: 970 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] 971 %fp.iv = phi float [ %init, %entry ], [ %add, %loop ] 972 %gep.A = getelementptr inbounds float, ptr %A, i64 %iv 973 store float %fp.iv, ptr %gep.A, align 4 974 %add = fadd fast float %fpinc, %fp.iv 975 %iv.next = add nuw nsw i64 %iv, 1 976 %ec = icmp eq i64 %iv.next, %N 977 br i1 %ec, label %exit, label %loop 978 979exit: 980 ret float %add 981} 982 983define float @fp_postinc_use_fsub(float %init, ptr noalias nocapture %A, i64 %N, float %fpinc) { 984; VEC-LABEL: define float @fp_postinc_use_fsub( 985; VEC-SAME: float [[INIT:%.*]], ptr noalias captures(none) [[A:%.*]], i64 [[N:%.*]], float [[FPINC:%.*]]) { 986; VEC-NEXT: [[ENTRY:.*]]: 987; VEC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 2 988; VEC-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 989; VEC: [[VECTOR_PH]]: 990; VEC-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 2 991; VEC-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 992; VEC-NEXT: [[DOTCAST:%.*]] = sitofp i64 [[N_VEC]] to float 993; VEC-NEXT: [[TMP0:%.*]] = fmul fast float [[FPINC]], [[DOTCAST]] 994; VEC-NEXT: [[TMP1:%.*]] = fsub fast float [[INIT]], [[TMP0]] 995; VEC-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[INIT]], i64 0 996; VEC-NEXT: [[DOTSPLAT:%.*]] = shufflevector <2 x float> [[DOTSPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer 997; VEC-NEXT: [[DOTSPLATINSERT1:%.*]] = insertelement <2 x float> poison, float [[FPINC]], i64 0 998; VEC-NEXT: [[DOTSPLAT2:%.*]] = shufflevector <2 x float> [[DOTSPLATINSERT1]], <2 x float> poison, <2 x i32> zeroinitializer 999; VEC-NEXT: [[TMP2:%.*]] = fmul fast <2 x float> <float 0.000000e+00, float 1.000000e+00>, [[DOTSPLAT2]] 1000; VEC-NEXT: [[INDUCTION:%.*]] = fsub fast <2 x float> [[DOTSPLAT]], [[TMP2]] 1001; VEC-NEXT: [[TMP3:%.*]] = fmul fast float [[FPINC]], 2.000000e+00 1002; VEC-NEXT: [[DOTSPLATINSERT3:%.*]] = insertelement <2 x float> poison, float [[TMP3]], i64 0 1003; VEC-NEXT: [[DOTSPLAT4:%.*]] = shufflevector <2 x float> [[DOTSPLATINSERT3]], <2 x float> poison, <2 x i32> zeroinitializer 1004; VEC-NEXT: br label %[[VECTOR_BODY:.*]] 1005; VEC: [[VECTOR_BODY]]: 1006; VEC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 1007; VEC-NEXT: [[VEC_IND:%.*]] = phi <2 x float> [ [[INDUCTION]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] 1008; VEC-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 0 1009; VEC-NEXT: [[TMP5:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP4]] 1010; VEC-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, ptr [[TMP5]], i32 0 1011; VEC-NEXT: store <2 x float> [[VEC_IND]], ptr [[TMP6]], align 4 1012; VEC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 1013; VEC-NEXT: [[VEC_IND_NEXT]] = fsub fast <2 x float> [[VEC_IND]], [[DOTSPLAT4]] 1014; VEC-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 1015; VEC-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 1016; VEC: [[MIDDLE_BLOCK]]: 1017; VEC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 1018; VEC-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]] 1019; VEC: [[SCALAR_PH]]: 1020; VEC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 1021; VEC-NEXT: [[BC_RESUME_VAL5:%.*]] = phi float [ [[TMP1]], %[[MIDDLE_BLOCK]] ], [ [[INIT]], %[[ENTRY]] ] 1022; VEC-NEXT: br label %[[LOOP:.*]] 1023; VEC: [[LOOP]]: 1024; VEC-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 1025; VEC-NEXT: [[FP_IV:%.*]] = phi float [ [[BC_RESUME_VAL5]], %[[SCALAR_PH]] ], [ [[ADD:%.*]], %[[LOOP]] ] 1026; VEC-NEXT: [[GEP_A:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]] 1027; VEC-NEXT: store float [[FP_IV]], ptr [[GEP_A]], align 4 1028; VEC-NEXT: [[ADD]] = fsub fast float [[FP_IV]], [[FPINC]] 1029; VEC-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 1030; VEC-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] 1031; VEC-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], {{!llvm.loop ![0-9]+}} 1032; VEC: [[EXIT]]: 1033; VEC-NEXT: [[ADD_LCSSA:%.*]] = phi float [ [[ADD]], %[[LOOP]] ], [ [[TMP1]], %[[MIDDLE_BLOCK]] ] 1034; VEC-NEXT: ret float [[ADD_LCSSA]] 1035; 1036; INTERLEAVE-LABEL: define float @fp_postinc_use_fsub( 1037; INTERLEAVE-SAME: float [[INIT:%.*]], ptr noalias captures(none) [[A:%.*]], i64 [[N:%.*]], float [[FPINC:%.*]]) { 1038; INTERLEAVE-NEXT: [[ENTRY:.*]]: 1039; INTERLEAVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 2 1040; INTERLEAVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 1041; INTERLEAVE: [[VECTOR_PH]]: 1042; INTERLEAVE-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 2 1043; INTERLEAVE-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 1044; INTERLEAVE-NEXT: [[DOTCAST:%.*]] = sitofp i64 [[N_VEC]] to float 1045; INTERLEAVE-NEXT: [[TMP0:%.*]] = fmul fast float [[FPINC]], [[DOTCAST]] 1046; INTERLEAVE-NEXT: [[TMP1:%.*]] = fsub fast float [[INIT]], [[TMP0]] 1047; INTERLEAVE-NEXT: br label %[[VECTOR_BODY:.*]] 1048; INTERLEAVE: [[VECTOR_BODY]]: 1049; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 1050; INTERLEAVE-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 0 1051; INTERLEAVE-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 1 1052; INTERLEAVE-NEXT: [[DOTCAST1:%.*]] = sitofp i64 [[INDEX]] to float 1053; INTERLEAVE-NEXT: [[TMP4:%.*]] = fmul fast float [[FPINC]], [[DOTCAST1]] 1054; INTERLEAVE-NEXT: [[OFFSET_IDX:%.*]] = fsub fast float [[INIT]], [[TMP4]] 1055; INTERLEAVE-NEXT: [[TMP5:%.*]] = fmul fast float 0.000000e+00, [[FPINC]] 1056; INTERLEAVE-NEXT: [[TMP6:%.*]] = fsub fast float [[OFFSET_IDX]], [[TMP5]] 1057; INTERLEAVE-NEXT: [[TMP7:%.*]] = fmul fast float 1.000000e+00, [[FPINC]] 1058; INTERLEAVE-NEXT: [[TMP8:%.*]] = fsub fast float [[OFFSET_IDX]], [[TMP7]] 1059; INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP2]] 1060; INTERLEAVE-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP3]] 1061; INTERLEAVE-NEXT: store float [[TMP6]], ptr [[TMP9]], align 4 1062; INTERLEAVE-NEXT: store float [[TMP8]], ptr [[TMP10]], align 4 1063; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 1064; INTERLEAVE-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 1065; INTERLEAVE-NEXT: br i1 [[TMP11]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 1066; INTERLEAVE: [[MIDDLE_BLOCK]]: 1067; INTERLEAVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 1068; INTERLEAVE-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]] 1069; INTERLEAVE: [[SCALAR_PH]]: 1070; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 1071; INTERLEAVE-NEXT: [[BC_RESUME_VAL2:%.*]] = phi float [ [[TMP1]], %[[MIDDLE_BLOCK]] ], [ [[INIT]], %[[ENTRY]] ] 1072; INTERLEAVE-NEXT: br label %[[LOOP:.*]] 1073; INTERLEAVE: [[LOOP]]: 1074; INTERLEAVE-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 1075; INTERLEAVE-NEXT: [[FP_IV:%.*]] = phi float [ [[BC_RESUME_VAL2]], %[[SCALAR_PH]] ], [ [[ADD:%.*]], %[[LOOP]] ] 1076; INTERLEAVE-NEXT: [[GEP_A:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]] 1077; INTERLEAVE-NEXT: store float [[FP_IV]], ptr [[GEP_A]], align 4 1078; INTERLEAVE-NEXT: [[ADD]] = fsub fast float [[FP_IV]], [[FPINC]] 1079; INTERLEAVE-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 1080; INTERLEAVE-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] 1081; INTERLEAVE-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], {{!llvm.loop ![0-9]+}} 1082; INTERLEAVE: [[EXIT]]: 1083; INTERLEAVE-NEXT: [[ADD_LCSSA:%.*]] = phi float [ [[ADD]], %[[LOOP]] ], [ [[TMP1]], %[[MIDDLE_BLOCK]] ] 1084; INTERLEAVE-NEXT: ret float [[ADD_LCSSA]] 1085; 1086entry: 1087 br label %loop 1088 1089loop: 1090 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] 1091 %fp.iv = phi float [ %init, %entry ], [ %add, %loop ] 1092 %gep.A = getelementptr inbounds float, ptr %A, i64 %iv 1093 store float %fp.iv, ptr %gep.A, align 4 1094 %add = fsub fast float %fp.iv, %fpinc 1095 %iv.next = add nuw nsw i64 %iv, 1 1096 %ec = icmp eq i64 %iv.next, %N 1097 br i1 %ec, label %exit, label %loop 1098 1099exit: 1100 ret float %add 1101} 1102 1103; Test case for https://github.com/llvm/llvm-project/issues/121745. 1104define i32 @test_iv_uniform_with_outside_use_scev_simplification(ptr %dst) { 1105; VEC-LABEL: define i32 @test_iv_uniform_with_outside_use_scev_simplification( 1106; VEC-SAME: ptr [[DST:%.*]]) { 1107; VEC-NEXT: [[ENTRY:.*]]: 1108; VEC-NEXT: [[STEP_1:%.*]] = sext i8 0 to i32 1109; VEC-NEXT: [[STEP_2:%.*]] = add nsw i32 [[STEP_1]], 1 1110; VEC-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 1111; VEC: [[VECTOR_PH]]: 1112; VEC-NEXT: br label %[[VECTOR_BODY:.*]] 1113; VEC: [[VECTOR_BODY]]: 1114; VEC-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 1115; VEC-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 0 1116; VEC-NEXT: [[TMP6:%.*]] = add i32 [[INDEX]], 1 1117; VEC-NEXT: [[TMP1:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP0]] 1118; VEC-NEXT: [[TMP2:%.*]] = getelementptr inbounds i16, ptr [[TMP1]], i32 0 1119; VEC-NEXT: store <2 x i16> zeroinitializer, ptr [[TMP2]], align 2 1120; VEC-NEXT: [[TMP4:%.*]] = add i32 [[STEP_2]], [[TMP0]] 1121; VEC-NEXT: [[TMP5:%.*]] = add i32 [[STEP_2]], [[TMP6]] 1122; VEC-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 1123; VEC-NEXT: [[TMP3:%.*]] = icmp eq i32 [[INDEX_NEXT]], 8 1124; VEC-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 1125; VEC: [[MIDDLE_BLOCK]]: 1126; VEC-NEXT: br i1 true, label %[[E_EXIT:.*]], label %[[SCALAR_PH]] 1127; VEC: [[SCALAR_PH]]: 1128; VEC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 8, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 1129; VEC-NEXT: br label %[[LOOP:.*]] 1130; VEC: [[LOOP]]: 1131; VEC-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 1132; VEC-NEXT: [[GEP_DST:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[IV]] 1133; VEC-NEXT: store i16 0, ptr [[GEP_DST]], align 2 1134; VEC-NEXT: [[IV_NEXT]] = add i32 [[STEP_2]], [[IV]] 1135; VEC-NEXT: [[CMP_I:%.*]] = icmp slt i32 [[IV_NEXT]], 8 1136; VEC-NEXT: br i1 [[CMP_I]], label %[[LOOP]], label %[[E_EXIT]], {{!llvm.loop ![0-9]+}} 1137; VEC: [[E_EXIT]]: 1138; VEC-NEXT: [[RES:%.*]] = phi i32 [ [[IV_NEXT]], %[[LOOP]] ], [ [[TMP5]], %[[MIDDLE_BLOCK]] ] 1139; VEC-NEXT: ret i32 [[RES]] 1140; 1141; INTERLEAVE-LABEL: define i32 @test_iv_uniform_with_outside_use_scev_simplification( 1142; INTERLEAVE-SAME: ptr [[DST:%.*]]) { 1143; INTERLEAVE-NEXT: [[ENTRY:.*]]: 1144; INTERLEAVE-NEXT: [[STEP_1:%.*]] = sext i8 0 to i32 1145; INTERLEAVE-NEXT: [[STEP_2:%.*]] = add nsw i32 [[STEP_1]], 1 1146; INTERLEAVE-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 1147; INTERLEAVE: [[VECTOR_PH]]: 1148; INTERLEAVE-NEXT: br label %[[VECTOR_BODY:.*]] 1149; INTERLEAVE: [[VECTOR_BODY]]: 1150; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 1151; INTERLEAVE-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 0 1152; INTERLEAVE-NEXT: [[TMP1:%.*]] = add i32 [[INDEX]], 1 1153; INTERLEAVE-NEXT: [[TMP2:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP0]] 1154; INTERLEAVE-NEXT: [[TMP3:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP1]] 1155; INTERLEAVE-NEXT: store i16 0, ptr [[TMP2]], align 2 1156; INTERLEAVE-NEXT: store i16 0, ptr [[TMP3]], align 2 1157; INTERLEAVE-NEXT: [[TMP5:%.*]] = add i32 [[STEP_2]], [[TMP1]] 1158; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 1159; INTERLEAVE-NEXT: [[TMP4:%.*]] = icmp eq i32 [[INDEX_NEXT]], 8 1160; INTERLEAVE-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 1161; INTERLEAVE: [[MIDDLE_BLOCK]]: 1162; INTERLEAVE-NEXT: br i1 true, label %[[E_EXIT:.*]], label %[[SCALAR_PH]] 1163; INTERLEAVE: [[SCALAR_PH]]: 1164; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 8, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 1165; INTERLEAVE-NEXT: br label %[[LOOP:.*]] 1166; INTERLEAVE: [[LOOP]]: 1167; INTERLEAVE-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 1168; INTERLEAVE-NEXT: [[GEP_DST:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[IV]] 1169; INTERLEAVE-NEXT: store i16 0, ptr [[GEP_DST]], align 2 1170; INTERLEAVE-NEXT: [[IV_NEXT]] = add i32 [[STEP_2]], [[IV]] 1171; INTERLEAVE-NEXT: [[CMP_I:%.*]] = icmp slt i32 [[IV_NEXT]], 8 1172; INTERLEAVE-NEXT: br i1 [[CMP_I]], label %[[LOOP]], label %[[E_EXIT]], {{!llvm.loop ![0-9]+}} 1173; INTERLEAVE: [[E_EXIT]]: 1174; INTERLEAVE-NEXT: [[RES:%.*]] = phi i32 [ [[IV_NEXT]], %[[LOOP]] ], [ [[TMP5]], %[[MIDDLE_BLOCK]] ] 1175; INTERLEAVE-NEXT: ret i32 [[RES]] 1176; 1177entry: 1178 %step.1 = sext i8 0 to i32 1179 %step.2 = add nsw i32 %step.1, 1 1180 br label %loop 1181 1182loop: 1183 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ] 1184 %gep.dst = getelementptr inbounds i16, ptr %dst, i32 %iv 1185 store i16 0, ptr %gep.dst, align 2 1186 %iv.next = add i32 %step.2, %iv 1187 %cmp.i = icmp slt i32 %iv.next, 8 1188 br i1 %cmp.i, label %loop, label %e.exit 1189 1190e.exit: 1191 %res = phi i32 [ %iv.next, %loop ] 1192 ret i32 %res 1193} 1194 1195define i32 @test_iv_uniform_with_outside_use_scev_simplification_2(ptr %dst) { 1196; VEC-LABEL: define i32 @test_iv_uniform_with_outside_use_scev_simplification_2( 1197; VEC-SAME: ptr [[DST:%.*]]) { 1198; VEC-NEXT: [[ENTRY:.*]]: 1199; VEC-NEXT: [[STEP_1:%.*]] = sext i8 0 to i32 1200; VEC-NEXT: [[STEP_2:%.*]] = add nsw i32 [[STEP_1]], 1 1201; VEC-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 1202; VEC: [[VECTOR_PH]]: 1203; VEC-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[STEP_2]], i64 0 1204; VEC-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer 1205; VEC-NEXT: br label %[[VECTOR_BODY:.*]] 1206; VEC: [[VECTOR_BODY]]: 1207; VEC-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 1208; VEC-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 2>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] 1209; VEC-NEXT: [[OFFSET_IDX:%.*]] = mul i32 [[INDEX]], 2 1210; VEC-NEXT: [[TMP0:%.*]] = add i32 [[OFFSET_IDX]], 0 1211; VEC-NEXT: [[TMP1:%.*]] = add i32 [[OFFSET_IDX]], 2 1212; VEC-NEXT: [[TMP2:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP0]] 1213; VEC-NEXT: [[TMP3:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP1]] 1214; VEC-NEXT: store i16 0, ptr [[TMP2]], align 2 1215; VEC-NEXT: store i16 0, ptr [[TMP3]], align 2 1216; VEC-NEXT: [[TMP4:%.*]] = add <2 x i32> [[VEC_IND]], splat (i32 1) 1217; VEC-NEXT: [[TMP5:%.*]] = add <2 x i32> [[BROADCAST_SPLAT]], [[TMP4]] 1218; VEC-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 1219; VEC-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 4) 1220; VEC-NEXT: [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], 4 1221; VEC-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 1222; VEC: [[MIDDLE_BLOCK]]: 1223; VEC-NEXT: [[TMP7:%.*]] = extractelement <2 x i32> [[TMP5]], i32 1 1224; VEC-NEXT: br i1 true, label %[[E_EXIT:.*]], label %[[SCALAR_PH]] 1225; VEC: [[SCALAR_PH]]: 1226; VEC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 8, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 1227; VEC-NEXT: br label %[[LOOP:.*]] 1228; VEC: [[LOOP]]: 1229; VEC-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 1230; VEC-NEXT: [[GEP_DST:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[IV]] 1231; VEC-NEXT: store i16 0, ptr [[GEP_DST]], align 2 1232; VEC-NEXT: [[INC:%.*]] = add i32 [[IV]], 1 1233; VEC-NEXT: [[IV_NEXT]] = add i32 [[STEP_2]], [[INC]] 1234; VEC-NEXT: [[CMP_I:%.*]] = icmp slt i32 [[IV_NEXT]], 8 1235; VEC-NEXT: br i1 [[CMP_I]], label %[[LOOP]], label %[[E_EXIT]], {{!llvm.loop ![0-9]+}} 1236; VEC: [[E_EXIT]]: 1237; VEC-NEXT: [[RES:%.*]] = phi i32 [ [[IV_NEXT]], %[[LOOP]] ], [ [[TMP7]], %[[MIDDLE_BLOCK]] ] 1238; VEC-NEXT: ret i32 [[RES]] 1239; 1240; INTERLEAVE-LABEL: define i32 @test_iv_uniform_with_outside_use_scev_simplification_2( 1241; INTERLEAVE-SAME: ptr [[DST:%.*]]) { 1242; INTERLEAVE-NEXT: [[ENTRY:.*]]: 1243; INTERLEAVE-NEXT: [[STEP_1:%.*]] = sext i8 0 to i32 1244; INTERLEAVE-NEXT: [[STEP_2:%.*]] = add nsw i32 [[STEP_1]], 1 1245; INTERLEAVE-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 1246; INTERLEAVE: [[VECTOR_PH]]: 1247; INTERLEAVE-NEXT: br label %[[VECTOR_BODY:.*]] 1248; INTERLEAVE: [[VECTOR_BODY]]: 1249; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 1250; INTERLEAVE-NEXT: [[OFFSET_IDX:%.*]] = mul i32 [[INDEX]], 2 1251; INTERLEAVE-NEXT: [[TMP0:%.*]] = add i32 [[OFFSET_IDX]], 0 1252; INTERLEAVE-NEXT: [[TMP1:%.*]] = add i32 [[OFFSET_IDX]], 2 1253; INTERLEAVE-NEXT: [[TMP2:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP0]] 1254; INTERLEAVE-NEXT: [[TMP3:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP1]] 1255; INTERLEAVE-NEXT: store i16 0, ptr [[TMP2]], align 2 1256; INTERLEAVE-NEXT: store i16 0, ptr [[TMP3]], align 2 1257; INTERLEAVE-NEXT: [[TMP4:%.*]] = add i32 [[TMP1]], 1 1258; INTERLEAVE-NEXT: [[TMP5:%.*]] = add i32 [[STEP_2]], [[TMP4]] 1259; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 1260; INTERLEAVE-NEXT: [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], 4 1261; INTERLEAVE-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 1262; INTERLEAVE: [[MIDDLE_BLOCK]]: 1263; INTERLEAVE-NEXT: br i1 true, label %[[E_EXIT:.*]], label %[[SCALAR_PH]] 1264; INTERLEAVE: [[SCALAR_PH]]: 1265; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 8, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 1266; INTERLEAVE-NEXT: br label %[[LOOP:.*]] 1267; INTERLEAVE: [[LOOP]]: 1268; INTERLEAVE-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] 1269; INTERLEAVE-NEXT: [[GEP_DST:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[IV]] 1270; INTERLEAVE-NEXT: store i16 0, ptr [[GEP_DST]], align 2 1271; INTERLEAVE-NEXT: [[INC:%.*]] = add i32 [[IV]], 1 1272; INTERLEAVE-NEXT: [[IV_NEXT]] = add i32 [[STEP_2]], [[INC]] 1273; INTERLEAVE-NEXT: [[CMP_I:%.*]] = icmp slt i32 [[IV_NEXT]], 8 1274; INTERLEAVE-NEXT: br i1 [[CMP_I]], label %[[LOOP]], label %[[E_EXIT]], {{!llvm.loop ![0-9]+}} 1275; INTERLEAVE: [[E_EXIT]]: 1276; INTERLEAVE-NEXT: [[RES:%.*]] = phi i32 [ [[IV_NEXT]], %[[LOOP]] ], [ [[TMP5]], %[[MIDDLE_BLOCK]] ] 1277; INTERLEAVE-NEXT: ret i32 [[RES]] 1278; 1279entry: 1280 %step.1 = sext i8 0 to i32 1281 %step.2 = add nsw i32 %step.1, 1 1282 br label %loop 1283 1284loop: 1285 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ] 1286 %gep.dst = getelementptr inbounds i16, ptr %dst, i32 %iv 1287 store i16 0, ptr %gep.dst, align 2 1288 %inc = add i32 %iv, 1 1289 %iv.next = add i32 %step.2, %inc 1290 %cmp.i = icmp slt i32 %iv.next, 8 1291 br i1 %cmp.i, label %loop, label %e.exit 1292 1293e.exit: 1294 %res = phi i32 [ %iv.next, %loop ] 1295 ret i32 %res 1296} 1297 1298; Test case for https://github.com/llvm/llvm-project/issues/122496. 1299define i32 @iv_ext_used_outside( ptr %dst) { 1300; VEC-LABEL: define i32 @iv_ext_used_outside( 1301; VEC-SAME: ptr [[DST:%.*]]) { 1302; VEC-NEXT: [[ENTRY:.*]]: 1303; VEC-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 1304; VEC: [[VECTOR_PH]]: 1305; VEC-NEXT: br label %[[VECTOR_BODY:.*]] 1306; VEC: [[VECTOR_BODY]]: 1307; VEC-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 1308; VEC-NEXT: [[VEC_IND:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] 1309; VEC-NEXT: [[OFFSET_IDX:%.*]] = trunc i32 [[INDEX]] to i16 1310; VEC-NEXT: [[TMP0:%.*]] = add i16 [[OFFSET_IDX]], 0 1311; VEC-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw i32, ptr [[DST]], i16 [[TMP0]] 1312; VEC-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP1]], i32 0 1313; VEC-NEXT: store <2 x i32> zeroinitializer, ptr [[TMP2]], align 4 1314; VEC-NEXT: [[TMP5:%.*]] = add nuw nsw <2 x i16> [[VEC_IND]], splat (i16 1) 1315; VEC-NEXT: [[TMP3:%.*]] = extractelement <2 x i16> [[TMP5]], i32 0 1316; VEC-NEXT: [[TMP4:%.*]] = zext nneg i16 [[TMP3]] to i32 1317; VEC-NEXT: [[TMP8:%.*]] = extractelement <2 x i16> [[TMP5]], i32 1 1318; VEC-NEXT: [[TMP7:%.*]] = zext nneg i16 [[TMP8]] to i32 1319; VEC-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 1320; VEC-NEXT: [[VEC_IND_NEXT]] = add <2 x i16> [[VEC_IND]], splat (i16 2) 1321; VEC-NEXT: [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], 128 1322; VEC-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 1323; VEC: [[MIDDLE_BLOCK]]: 1324; VEC-NEXT: br i1 false, label %[[EXIT:.*]], label %[[SCALAR_PH]] 1325; VEC: [[SCALAR_PH]]: 1326; VEC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ 128, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 1327; VEC-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ 128, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 1328; VEC-NEXT: br label %[[LOOP:.*]] 1329; VEC: [[LOOP]]: 1330; VEC-NEXT: [[IV_1:%.*]] = phi i16 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_1_NEXT:%.*]], %[[LOOP]] ] 1331; VEC-NEXT: [[IV_2:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], %[[SCALAR_PH]] ], [ [[IV_1_EXT:%.*]], %[[LOOP]] ] 1332; VEC-NEXT: [[GEP:%.*]] = getelementptr inbounds nuw i32, ptr [[DST]], i16 [[IV_1]] 1333; VEC-NEXT: store i32 0, ptr [[GEP]], align 4 1334; VEC-NEXT: [[IV_1_NEXT]] = add nuw nsw i16 [[IV_1]], 1 1335; VEC-NEXT: [[IV_1_EXT]] = zext nneg i16 [[IV_1_NEXT]] to i32 1336; VEC-NEXT: [[EC:%.*]] = icmp samesign ult i16 [[IV_1]], 128 1337; VEC-NEXT: br i1 [[EC]], label %[[LOOP]], label %[[EXIT]], {{!llvm.loop ![0-9]+}} 1338; VEC: [[EXIT]]: 1339; VEC-NEXT: [[IV_1_EXT_LCSSA:%.*]] = phi i32 [ [[IV_1_EXT]], %[[LOOP]] ], [ [[TMP7]], %[[MIDDLE_BLOCK]] ] 1340; VEC-NEXT: ret i32 [[IV_1_EXT_LCSSA]] 1341; 1342; INTERLEAVE-LABEL: define i32 @iv_ext_used_outside( 1343; INTERLEAVE-SAME: ptr [[DST:%.*]]) { 1344; INTERLEAVE-NEXT: [[ENTRY:.*]]: 1345; INTERLEAVE-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 1346; INTERLEAVE: [[VECTOR_PH]]: 1347; INTERLEAVE-NEXT: br label %[[VECTOR_BODY:.*]] 1348; INTERLEAVE: [[VECTOR_BODY]]: 1349; INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 1350; INTERLEAVE-NEXT: [[OFFSET_IDX:%.*]] = trunc i32 [[INDEX]] to i16 1351; INTERLEAVE-NEXT: [[TMP0:%.*]] = add i16 [[OFFSET_IDX]], 0 1352; INTERLEAVE-NEXT: [[TMP1:%.*]] = add i16 [[OFFSET_IDX]], 1 1353; INTERLEAVE-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw i32, ptr [[DST]], i16 [[TMP0]] 1354; INTERLEAVE-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw i32, ptr [[DST]], i16 [[TMP1]] 1355; INTERLEAVE-NEXT: store i32 0, ptr [[TMP2]], align 4 1356; INTERLEAVE-NEXT: store i32 0, ptr [[TMP3]], align 4 1357; INTERLEAVE-NEXT: [[TMP4:%.*]] = add nuw nsw i16 [[TMP1]], 1 1358; INTERLEAVE-NEXT: [[TMP5:%.*]] = zext nneg i16 [[TMP4]] to i32 1359; INTERLEAVE-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 1360; INTERLEAVE-NEXT: [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], 128 1361; INTERLEAVE-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}} 1362; INTERLEAVE: [[MIDDLE_BLOCK]]: 1363; INTERLEAVE-NEXT: br i1 false, label %[[EXIT:.*]], label %[[SCALAR_PH]] 1364; INTERLEAVE: [[SCALAR_PH]]: 1365; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ 128, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 1366; INTERLEAVE-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ 128, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 1367; INTERLEAVE-NEXT: br label %[[LOOP:.*]] 1368; INTERLEAVE: [[LOOP]]: 1369; INTERLEAVE-NEXT: [[IV_1:%.*]] = phi i16 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_1_NEXT:%.*]], %[[LOOP]] ] 1370; INTERLEAVE-NEXT: [[IV_2:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], %[[SCALAR_PH]] ], [ [[IV_1_EXT:%.*]], %[[LOOP]] ] 1371; INTERLEAVE-NEXT: [[GEP:%.*]] = getelementptr inbounds nuw i32, ptr [[DST]], i16 [[IV_1]] 1372; INTERLEAVE-NEXT: store i32 0, ptr [[GEP]], align 4 1373; INTERLEAVE-NEXT: [[IV_1_NEXT]] = add nuw nsw i16 [[IV_1]], 1 1374; INTERLEAVE-NEXT: [[IV_1_EXT]] = zext nneg i16 [[IV_1_NEXT]] to i32 1375; INTERLEAVE-NEXT: [[EC:%.*]] = icmp samesign ult i16 [[IV_1]], 128 1376; INTERLEAVE-NEXT: br i1 [[EC]], label %[[LOOP]], label %[[EXIT]], {{!llvm.loop ![0-9]+}} 1377; INTERLEAVE: [[EXIT]]: 1378; INTERLEAVE-NEXT: [[IV_1_EXT_LCSSA:%.*]] = phi i32 [ [[IV_1_EXT]], %[[LOOP]] ], [ [[TMP5]], %[[MIDDLE_BLOCK]] ] 1379; INTERLEAVE-NEXT: ret i32 [[IV_1_EXT_LCSSA]] 1380; 1381entry: 1382 br label %loop 1383 1384loop: 1385 %iv.1 = phi i16 [ 0, %entry ], [ %iv.1.next, %loop ] 1386 %iv.2 = phi i32 [ 0, %entry ], [ %iv.1.ext, %loop ] 1387 %gep = getelementptr inbounds nuw i32, ptr %dst, i16 %iv.1 1388 store i32 0, ptr %gep, align 4 1389 %iv.1.next = add nuw nsw i16 %iv.1, 1 1390 %iv.1.ext = zext nneg i16 %iv.1.next to i32 1391 %ec = icmp samesign ult i16 %iv.1, 128 1392 br i1 %ec, label %loop, label %exit 1393 1394exit: 1395 %iv.1.ext.lcssa = phi i32 [ %iv.1.ext, %loop ] 1396 ret i32 %iv.1.ext.lcssa 1397} 1398 1399; Test case for https://github.com/llvm/llvm-project/issues/122602. 1400define i64 @test_iv_increment_incremented(ptr %dst) { 1401; VEC-LABEL: define i64 @test_iv_increment_incremented( 1402; VEC-SAME: ptr [[DST:%.*]]) { 1403; VEC-NEXT: [[ENTRY:.*]]: 1404; VEC-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 1405; VEC: [[VECTOR_PH]]: 1406; VEC-NEXT: br label %[[VECTOR_BODY:.*]] 1407; VEC: [[VECTOR_BODY]]: 1408; VEC-NEXT: [[TMP0:%.*]] = getelementptr i16, ptr [[DST]], i64 3 1409; VEC-NEXT: [[TMP1:%.*]] = getelementptr i16, ptr [[TMP0]], i32 0 1410; VEC-NEXT: [[TMP2:%.*]] = getelementptr i16, ptr [[TMP1]], i32 -1 1411; VEC-NEXT: store <2 x i16> splat (i16 1), ptr [[TMP2]], align 2 1412; VEC-NEXT: [[TMP3:%.*]] = add i64 2, -1 1413; VEC-NEXT: [[TMP5:%.*]] = add i64 1, -1 1414; VEC-NEXT: [[TMP4:%.*]] = add i64 [[TMP3]], 1 1415; VEC-NEXT: [[TMP6:%.*]] = add i64 [[TMP5]], 1 1416; VEC-NEXT: br label %[[MIDDLE_BLOCK:.*]] 1417; VEC: [[MIDDLE_BLOCK]]: 1418; VEC-NEXT: br i1 true, label %[[EXIT:.*]], label %[[SCALAR_PH]] 1419; VEC: [[SCALAR_PH]]: 1420; VEC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1, %[[MIDDLE_BLOCK]] ], [ 3, %[[ENTRY]] ] 1421; VEC-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i64 [ 0, %[[MIDDLE_BLOCK]] ], [ 2, %[[ENTRY]] ] 1422; VEC-NEXT: br label %[[LOOP:.*]] 1423; VEC: [[LOOP]]: 1424; VEC-NEXT: [[IV_1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_1_NEXT:%.*]], %[[LOOP]] ] 1425; VEC-NEXT: [[IV_2:%.*]] = phi i64 [ [[BC_RESUME_VAL1]], %[[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], %[[LOOP]] ] 1426; VEC-NEXT: [[GEP:%.*]] = getelementptr i16, ptr [[DST]], i64 [[IV_1]] 1427; VEC-NEXT: store i16 1, ptr [[GEP]], align 2 1428; VEC-NEXT: [[IV_2_NEXT]] = add i64 [[IV_2]], -1 1429; VEC-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_2_NEXT]], 0 1430; VEC-NEXT: [[IV_1_NEXT]] = add i64 [[IV_2_NEXT]], 1 1431; VEC-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], {{!llvm.loop ![0-9]+}} 1432; VEC: [[EXIT]]: 1433; VEC-NEXT: [[IV_1_NEXT_LCSSA:%.*]] = phi i64 [ [[IV_1_NEXT]], %[[LOOP]] ], [ [[TMP6]], %[[MIDDLE_BLOCK]] ] 1434; VEC-NEXT: ret i64 [[IV_1_NEXT_LCSSA]] 1435; 1436; INTERLEAVE-LABEL: define i64 @test_iv_increment_incremented( 1437; INTERLEAVE-SAME: ptr [[DST:%.*]]) { 1438; INTERLEAVE-NEXT: [[ENTRY:.*]]: 1439; INTERLEAVE-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 1440; INTERLEAVE: [[VECTOR_PH]]: 1441; INTERLEAVE-NEXT: br label %[[VECTOR_BODY:.*]] 1442; INTERLEAVE: [[VECTOR_BODY]]: 1443; INTERLEAVE-NEXT: [[TMP0:%.*]] = getelementptr i16, ptr [[DST]], i64 3 1444; INTERLEAVE-NEXT: [[TMP1:%.*]] = getelementptr i16, ptr [[DST]], i64 2 1445; INTERLEAVE-NEXT: store i16 1, ptr [[TMP0]], align 2 1446; INTERLEAVE-NEXT: store i16 1, ptr [[TMP1]], align 2 1447; INTERLEAVE-NEXT: [[TMP2:%.*]] = add i64 1, -1 1448; INTERLEAVE-NEXT: [[TMP3:%.*]] = add i64 [[TMP2]], 1 1449; INTERLEAVE-NEXT: br label %[[MIDDLE_BLOCK:.*]] 1450; INTERLEAVE: [[MIDDLE_BLOCK]]: 1451; INTERLEAVE-NEXT: br i1 true, label %[[EXIT:.*]], label %[[SCALAR_PH]] 1452; INTERLEAVE: [[SCALAR_PH]]: 1453; INTERLEAVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1, %[[MIDDLE_BLOCK]] ], [ 3, %[[ENTRY]] ] 1454; INTERLEAVE-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i64 [ 0, %[[MIDDLE_BLOCK]] ], [ 2, %[[ENTRY]] ] 1455; INTERLEAVE-NEXT: br label %[[LOOP:.*]] 1456; INTERLEAVE: [[LOOP]]: 1457; INTERLEAVE-NEXT: [[IV_1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_1_NEXT:%.*]], %[[LOOP]] ] 1458; INTERLEAVE-NEXT: [[IV_2:%.*]] = phi i64 [ [[BC_RESUME_VAL1]], %[[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], %[[LOOP]] ] 1459; INTERLEAVE-NEXT: [[GEP:%.*]] = getelementptr i16, ptr [[DST]], i64 [[IV_1]] 1460; INTERLEAVE-NEXT: store i16 1, ptr [[GEP]], align 2 1461; INTERLEAVE-NEXT: [[IV_2_NEXT]] = add i64 [[IV_2]], -1 1462; INTERLEAVE-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_2_NEXT]], 0 1463; INTERLEAVE-NEXT: [[IV_1_NEXT]] = add i64 [[IV_2_NEXT]], 1 1464; INTERLEAVE-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], {{!llvm.loop ![0-9]+}} 1465; INTERLEAVE: [[EXIT]]: 1466; INTERLEAVE-NEXT: [[IV_1_NEXT_LCSSA:%.*]] = phi i64 [ [[IV_1_NEXT]], %[[LOOP]] ], [ [[TMP3]], %[[MIDDLE_BLOCK]] ] 1467; INTERLEAVE-NEXT: ret i64 [[IV_1_NEXT_LCSSA]] 1468; 1469entry: 1470 br label %loop 1471 1472loop: 1473 %iv.1 = phi i64 [ 3, %entry ], [ %iv.1.next, %loop ] 1474 %iv.2 = phi i64 [ 2, %entry ], [ %iv.2.next, %loop ] 1475 %gep = getelementptr i16, ptr %dst, i64 %iv.1 1476 store i16 1, ptr %gep, align 2 1477 %iv.2.next = add i64 %iv.2, -1 1478 %ec = icmp eq i64 %iv.2.next, 0 1479 %iv.1.next = add i64 %iv.2.next, 1 1480 br i1 %ec, label %exit, label %loop 1481 1482exit: 1483 ret i64 %iv.1.next 1484} 1485