1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; This test verifies that the loop vectorizer will not vectorizes low trip count 3; loops that require runtime checks (Trip count is computed with profile info). 4; REQUIRES: asserts 5; RUN: opt < %s -loop-vectorize -loop-vectorize-with-block-frequency -S | FileCheck %s 6 7target datalayout = "E-m:e-p:32:32-i64:32-f64:32:64-a:0:32-n32-S128" 8 9@tab = common global [32 x i8] zeroinitializer, align 1 10 11define i32 @foo_low_trip_count1(i32 %bound) { 12; Simple loop with low tripcount. Should not be vectorized. 13; CHECK-LABEL: @foo_low_trip_count1( 14; CHECK-NEXT: entry: 15; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[BOUND:%.*]], 1 16; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 17; CHECK: vector.ph: 18; CHECK-NEXT: [[N_RND_UP:%.*]] = add i32 [[TMP0]], 3 19; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[N_RND_UP]], 4 20; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[N_RND_UP]], [[N_MOD_VF]] 21; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i32 [[TMP0]], 1 22; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[TRIP_COUNT_MINUS_1]], i32 0 23; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer 24; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 25; CHECK: vector.body: 26; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE12:%.*]] ] 27; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE12]] ] 28; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[INDEX]], 0 29; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[INDEX]], 1 30; CHECK-NEXT: [[TMP3:%.*]] = add i32 [[INDEX]], 2 31; CHECK-NEXT: [[TMP4:%.*]] = add i32 [[INDEX]], 3 32; CHECK-NEXT: [[TMP5:%.*]] = icmp ule <4 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]] 33; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP1]] 34; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP2]] 35; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP3]] 36; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP4]] 37; CHECK-NEXT: [[TMP10:%.*]] = extractelement <4 x i1> [[TMP5]], i32 0 38; CHECK-NEXT: br i1 [[TMP10]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]] 39; CHECK: pred.load.if: 40; CHECK-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP6]], align 1 41; CHECK-NEXT: [[TMP12:%.*]] = insertelement <4 x i8> poison, i8 [[TMP11]], i32 0 42; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE]] 43; CHECK: pred.load.continue: 44; CHECK-NEXT: [[TMP13:%.*]] = phi <4 x i8> [ poison, [[VECTOR_BODY]] ], [ [[TMP12]], [[PRED_LOAD_IF]] ] 45; CHECK-NEXT: [[TMP14:%.*]] = extractelement <4 x i1> [[TMP5]], i32 1 46; CHECK-NEXT: br i1 [[TMP14]], label [[PRED_LOAD_IF1:%.*]], label [[PRED_LOAD_CONTINUE2:%.*]] 47; CHECK: pred.load.if1: 48; CHECK-NEXT: [[TMP15:%.*]] = load i8, i8* [[TMP7]], align 1 49; CHECK-NEXT: [[TMP16:%.*]] = insertelement <4 x i8> [[TMP13]], i8 [[TMP15]], i32 1 50; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE2]] 51; CHECK: pred.load.continue2: 52; CHECK-NEXT: [[TMP17:%.*]] = phi <4 x i8> [ [[TMP13]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP16]], [[PRED_LOAD_IF1]] ] 53; CHECK-NEXT: [[TMP18:%.*]] = extractelement <4 x i1> [[TMP5]], i32 2 54; CHECK-NEXT: br i1 [[TMP18]], label [[PRED_LOAD_IF3:%.*]], label [[PRED_LOAD_CONTINUE4:%.*]] 55; CHECK: pred.load.if3: 56; CHECK-NEXT: [[TMP19:%.*]] = load i8, i8* [[TMP8]], align 1 57; CHECK-NEXT: [[TMP20:%.*]] = insertelement <4 x i8> [[TMP17]], i8 [[TMP19]], i32 2 58; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE4]] 59; CHECK: pred.load.continue4: 60; CHECK-NEXT: [[TMP21:%.*]] = phi <4 x i8> [ [[TMP17]], [[PRED_LOAD_CONTINUE2]] ], [ [[TMP20]], [[PRED_LOAD_IF3]] ] 61; CHECK-NEXT: [[TMP22:%.*]] = extractelement <4 x i1> [[TMP5]], i32 3 62; CHECK-NEXT: br i1 [[TMP22]], label [[PRED_LOAD_IF5:%.*]], label [[PRED_LOAD_CONTINUE6:%.*]] 63; CHECK: pred.load.if5: 64; CHECK-NEXT: [[TMP23:%.*]] = load i8, i8* [[TMP9]], align 1 65; CHECK-NEXT: [[TMP24:%.*]] = insertelement <4 x i8> [[TMP21]], i8 [[TMP23]], i32 3 66; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE6]] 67; CHECK: pred.load.continue6: 68; CHECK-NEXT: [[TMP25:%.*]] = phi <4 x i8> [ [[TMP21]], [[PRED_LOAD_CONTINUE4]] ], [ [[TMP24]], [[PRED_LOAD_IF5]] ] 69; CHECK-NEXT: [[TMP26:%.*]] = icmp eq <4 x i8> [[TMP25]], zeroinitializer 70; CHECK-NEXT: [[TMP27:%.*]] = select <4 x i1> [[TMP26]], <4 x i8> <i8 2, i8 2, i8 2, i8 2>, <4 x i8> <i8 1, i8 1, i8 1, i8 1> 71; CHECK-NEXT: [[TMP28:%.*]] = extractelement <4 x i1> [[TMP5]], i32 0 72; CHECK-NEXT: br i1 [[TMP28]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]] 73; CHECK: pred.store.if: 74; CHECK-NEXT: [[TMP29:%.*]] = extractelement <4 x i8> [[TMP27]], i32 0 75; CHECK-NEXT: store i8 [[TMP29]], i8* [[TMP6]], align 1 76; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]] 77; CHECK: pred.store.continue: 78; CHECK-NEXT: [[TMP30:%.*]] = extractelement <4 x i1> [[TMP5]], i32 1 79; CHECK-NEXT: br i1 [[TMP30]], label [[PRED_STORE_IF7:%.*]], label [[PRED_STORE_CONTINUE8:%.*]] 80; CHECK: pred.store.if7: 81; CHECK-NEXT: [[TMP31:%.*]] = extractelement <4 x i8> [[TMP27]], i32 1 82; CHECK-NEXT: store i8 [[TMP31]], i8* [[TMP7]], align 1 83; CHECK-NEXT: br label [[PRED_STORE_CONTINUE8]] 84; CHECK: pred.store.continue8: 85; CHECK-NEXT: [[TMP32:%.*]] = extractelement <4 x i1> [[TMP5]], i32 2 86; CHECK-NEXT: br i1 [[TMP32]], label [[PRED_STORE_IF9:%.*]], label [[PRED_STORE_CONTINUE10:%.*]] 87; CHECK: pred.store.if9: 88; CHECK-NEXT: [[TMP33:%.*]] = extractelement <4 x i8> [[TMP27]], i32 2 89; CHECK-NEXT: store i8 [[TMP33]], i8* [[TMP8]], align 1 90; CHECK-NEXT: br label [[PRED_STORE_CONTINUE10]] 91; CHECK: pred.store.continue10: 92; CHECK-NEXT: [[TMP34:%.*]] = extractelement <4 x i1> [[TMP5]], i32 3 93; CHECK-NEXT: br i1 [[TMP34]], label [[PRED_STORE_IF11:%.*]], label [[PRED_STORE_CONTINUE12]] 94; CHECK: pred.store.if11: 95; CHECK-NEXT: [[TMP35:%.*]] = extractelement <4 x i8> [[TMP27]], i32 3 96; CHECK-NEXT: store i8 [[TMP35]], i8* [[TMP9]], align 1 97; CHECK-NEXT: br label [[PRED_STORE_CONTINUE12]] 98; CHECK: pred.store.continue12: 99; CHECK-NEXT: [[INDEX_NEXT]] = add i32 [[INDEX]], 4 100; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4> 101; CHECK-NEXT: [[TMP36:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 102; CHECK-NEXT: br i1 [[TMP36]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !prof [[PROF0:![0-9]+]], !llvm.loop [[LOOP1:![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 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 107; CHECK-NEXT: br label [[FOR_BODY:%.*]] 108; CHECK: for.body: 109; CHECK-NEXT: [[I_08:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] 110; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[I_08]] 111; CHECK-NEXT: [[TMP37:%.*]] = load i8, i8* [[ARRAYIDX]], align 1 112; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP37]], 0 113; CHECK-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1 114; CHECK-NEXT: store i8 [[DOT]], i8* [[ARRAYIDX]], align 1 115; CHECK-NEXT: [[INC]] = add nsw i32 [[I_08]], 1 116; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[I_08]], [[BOUND]] 117; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END]], label [[FOR_BODY]], !prof [[PROF3:![0-9]+]], !llvm.loop [[LOOP4:![0-9]+]] 118; CHECK: for.end: 119; CHECK-NEXT: ret i32 0 120; 121 122entry: 123 br label %for.body 124 125for.body: ; preds = %for.body, %entry 126 %i.08 = phi i32 [ 0, %entry ], [ %inc, %for.body ] 127 %arrayidx = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 %i.08 128 %0 = load i8, i8* %arrayidx, align 1 129 %cmp1 = icmp eq i8 %0, 0 130 %. = select i1 %cmp1, i8 2, i8 1 131 store i8 %., i8* %arrayidx, align 1 132 %inc = add nsw i32 %i.08, 1 133 %exitcond = icmp eq i32 %i.08, %bound 134 br i1 %exitcond, label %for.end, label %for.body, !prof !1 135 136for.end: ; preds = %for.body 137 ret i32 0 138} 139 140define i32 @foo_low_trip_count2(i32 %bound) !prof !0 { 141; The loop has a same invocation count with the function, but has a low 142; trip_count per invocation and not worth to vectorize. 143; CHECK-LABEL: @foo_low_trip_count2( 144; CHECK-NEXT: entry: 145; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[BOUND:%.*]], 1 146; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 147; CHECK: vector.ph: 148; CHECK-NEXT: [[N_RND_UP:%.*]] = add i32 [[TMP0]], 3 149; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[N_RND_UP]], 4 150; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[N_RND_UP]], [[N_MOD_VF]] 151; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i32 [[TMP0]], 1 152; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[TRIP_COUNT_MINUS_1]], i32 0 153; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer 154; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 155; CHECK: vector.body: 156; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE12:%.*]] ] 157; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE12]] ] 158; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[INDEX]], 0 159; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[INDEX]], 1 160; CHECK-NEXT: [[TMP3:%.*]] = add i32 [[INDEX]], 2 161; CHECK-NEXT: [[TMP4:%.*]] = add i32 [[INDEX]], 3 162; CHECK-NEXT: [[TMP5:%.*]] = icmp ule <4 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]] 163; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP1]] 164; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP2]] 165; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP3]] 166; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP4]] 167; CHECK-NEXT: [[TMP10:%.*]] = extractelement <4 x i1> [[TMP5]], i32 0 168; CHECK-NEXT: br i1 [[TMP10]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]] 169; CHECK: pred.load.if: 170; CHECK-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP6]], align 1 171; CHECK-NEXT: [[TMP12:%.*]] = insertelement <4 x i8> poison, i8 [[TMP11]], i32 0 172; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE]] 173; CHECK: pred.load.continue: 174; CHECK-NEXT: [[TMP13:%.*]] = phi <4 x i8> [ poison, [[VECTOR_BODY]] ], [ [[TMP12]], [[PRED_LOAD_IF]] ] 175; CHECK-NEXT: [[TMP14:%.*]] = extractelement <4 x i1> [[TMP5]], i32 1 176; CHECK-NEXT: br i1 [[TMP14]], label [[PRED_LOAD_IF1:%.*]], label [[PRED_LOAD_CONTINUE2:%.*]] 177; CHECK: pred.load.if1: 178; CHECK-NEXT: [[TMP15:%.*]] = load i8, i8* [[TMP7]], align 1 179; CHECK-NEXT: [[TMP16:%.*]] = insertelement <4 x i8> [[TMP13]], i8 [[TMP15]], i32 1 180; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE2]] 181; CHECK: pred.load.continue2: 182; CHECK-NEXT: [[TMP17:%.*]] = phi <4 x i8> [ [[TMP13]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP16]], [[PRED_LOAD_IF1]] ] 183; CHECK-NEXT: [[TMP18:%.*]] = extractelement <4 x i1> [[TMP5]], i32 2 184; CHECK-NEXT: br i1 [[TMP18]], label [[PRED_LOAD_IF3:%.*]], label [[PRED_LOAD_CONTINUE4:%.*]] 185; CHECK: pred.load.if3: 186; CHECK-NEXT: [[TMP19:%.*]] = load i8, i8* [[TMP8]], align 1 187; CHECK-NEXT: [[TMP20:%.*]] = insertelement <4 x i8> [[TMP17]], i8 [[TMP19]], i32 2 188; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE4]] 189; CHECK: pred.load.continue4: 190; CHECK-NEXT: [[TMP21:%.*]] = phi <4 x i8> [ [[TMP17]], [[PRED_LOAD_CONTINUE2]] ], [ [[TMP20]], [[PRED_LOAD_IF3]] ] 191; CHECK-NEXT: [[TMP22:%.*]] = extractelement <4 x i1> [[TMP5]], i32 3 192; CHECK-NEXT: br i1 [[TMP22]], label [[PRED_LOAD_IF5:%.*]], label [[PRED_LOAD_CONTINUE6:%.*]] 193; CHECK: pred.load.if5: 194; CHECK-NEXT: [[TMP23:%.*]] = load i8, i8* [[TMP9]], align 1 195; CHECK-NEXT: [[TMP24:%.*]] = insertelement <4 x i8> [[TMP21]], i8 [[TMP23]], i32 3 196; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE6]] 197; CHECK: pred.load.continue6: 198; CHECK-NEXT: [[TMP25:%.*]] = phi <4 x i8> [ [[TMP21]], [[PRED_LOAD_CONTINUE4]] ], [ [[TMP24]], [[PRED_LOAD_IF5]] ] 199; CHECK-NEXT: [[TMP26:%.*]] = icmp eq <4 x i8> [[TMP25]], zeroinitializer 200; CHECK-NEXT: [[TMP27:%.*]] = select <4 x i1> [[TMP26]], <4 x i8> <i8 2, i8 2, i8 2, i8 2>, <4 x i8> <i8 1, i8 1, i8 1, i8 1> 201; CHECK-NEXT: [[TMP28:%.*]] = extractelement <4 x i1> [[TMP5]], i32 0 202; CHECK-NEXT: br i1 [[TMP28]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]] 203; CHECK: pred.store.if: 204; CHECK-NEXT: [[TMP29:%.*]] = extractelement <4 x i8> [[TMP27]], i32 0 205; CHECK-NEXT: store i8 [[TMP29]], i8* [[TMP6]], align 1 206; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]] 207; CHECK: pred.store.continue: 208; CHECK-NEXT: [[TMP30:%.*]] = extractelement <4 x i1> [[TMP5]], i32 1 209; CHECK-NEXT: br i1 [[TMP30]], label [[PRED_STORE_IF7:%.*]], label [[PRED_STORE_CONTINUE8:%.*]] 210; CHECK: pred.store.if7: 211; CHECK-NEXT: [[TMP31:%.*]] = extractelement <4 x i8> [[TMP27]], i32 1 212; CHECK-NEXT: store i8 [[TMP31]], i8* [[TMP7]], align 1 213; CHECK-NEXT: br label [[PRED_STORE_CONTINUE8]] 214; CHECK: pred.store.continue8: 215; CHECK-NEXT: [[TMP32:%.*]] = extractelement <4 x i1> [[TMP5]], i32 2 216; CHECK-NEXT: br i1 [[TMP32]], label [[PRED_STORE_IF9:%.*]], label [[PRED_STORE_CONTINUE10:%.*]] 217; CHECK: pred.store.if9: 218; CHECK-NEXT: [[TMP33:%.*]] = extractelement <4 x i8> [[TMP27]], i32 2 219; CHECK-NEXT: store i8 [[TMP33]], i8* [[TMP8]], align 1 220; CHECK-NEXT: br label [[PRED_STORE_CONTINUE10]] 221; CHECK: pred.store.continue10: 222; CHECK-NEXT: [[TMP34:%.*]] = extractelement <4 x i1> [[TMP5]], i32 3 223; CHECK-NEXT: br i1 [[TMP34]], label [[PRED_STORE_IF11:%.*]], label [[PRED_STORE_CONTINUE12]] 224; CHECK: pred.store.if11: 225; CHECK-NEXT: [[TMP35:%.*]] = extractelement <4 x i8> [[TMP27]], i32 3 226; CHECK-NEXT: store i8 [[TMP35]], i8* [[TMP9]], align 1 227; CHECK-NEXT: br label [[PRED_STORE_CONTINUE12]] 228; CHECK: pred.store.continue12: 229; CHECK-NEXT: [[INDEX_NEXT]] = add i32 [[INDEX]], 4 230; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4> 231; CHECK-NEXT: [[TMP36:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 232; CHECK-NEXT: br i1 [[TMP36]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !prof [[PROF0]], !llvm.loop [[LOOP7:![0-9]+]] 233; CHECK: middle.block: 234; CHECK-NEXT: br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]] 235; CHECK: scalar.ph: 236; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 237; CHECK-NEXT: br label [[FOR_BODY:%.*]] 238; CHECK: for.body: 239; CHECK-NEXT: [[I_08:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] 240; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[I_08]] 241; CHECK-NEXT: [[TMP37:%.*]] = load i8, i8* [[ARRAYIDX]], align 1 242; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP37]], 0 243; CHECK-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1 244; CHECK-NEXT: store i8 [[DOT]], i8* [[ARRAYIDX]], align 1 245; CHECK-NEXT: [[INC]] = add nsw i32 [[I_08]], 1 246; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[I_08]], [[BOUND]] 247; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END]], label [[FOR_BODY]], !prof [[PROF3]], !llvm.loop [[LOOP8:![0-9]+]] 248; CHECK: for.end: 249; CHECK-NEXT: ret i32 0 250; 251 252entry: 253 br label %for.body 254 255for.body: ; preds = %for.body, %entry 256 %i.08 = phi i32 [ 0, %entry ], [ %inc, %for.body ] 257 %arrayidx = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 %i.08 258 %0 = load i8, i8* %arrayidx, align 1 259 %cmp1 = icmp eq i8 %0, 0 260 %. = select i1 %cmp1, i8 2, i8 1 261 store i8 %., i8* %arrayidx, align 1 262 %inc = add nsw i32 %i.08, 1 263 %exitcond = icmp eq i32 %i.08, %bound 264 br i1 %exitcond, label %for.end, label %for.body, !prof !1 265 266for.end: ; preds = %for.body 267 ret i32 0 268} 269 270define i32 @foo_low_trip_count3(i1 %cond, i32 %bound) !prof !0 { 271; The loop has low invocation count compare to the function invocation count, 272; but has a high trip count per invocation. Vectorize it. 273; CHECK-LABEL: @foo_low_trip_count3( 274; CHECK-NEXT: entry: 275; CHECK-NEXT: br i1 [[COND:%.*]], label [[FOR_PREHEADER:%.*]], label [[FOR_END:%.*]], !prof [[PROF9:![0-9]+]] 276; CHECK: for.preheader: 277; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[BOUND:%.*]], 1 278; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 4 279; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 280; CHECK: vector.ph: 281; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], 4 282; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]] 283; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 284; CHECK: vector.body: 285; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 286; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[INDEX]], 0 287; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP1]] 288; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, i8* [[TMP2]], i32 0 289; CHECK-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to <4 x i8>* 290; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, <4 x i8>* [[TMP4]], align 1 291; CHECK-NEXT: [[TMP5:%.*]] = icmp eq <4 x i8> [[WIDE_LOAD]], zeroinitializer 292; CHECK-NEXT: [[TMP6:%.*]] = select <4 x i1> [[TMP5]], <4 x i8> <i8 2, i8 2, i8 2, i8 2>, <4 x i8> <i8 1, i8 1, i8 1, i8 1> 293; CHECK-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP3]] to <4 x i8>* 294; CHECK-NEXT: store <4 x i8> [[TMP6]], <4 x i8>* [[TMP7]], align 1 295; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 296; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 297; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !prof [[PROF10:![0-9]+]], !llvm.loop [[LOOP11:![0-9]+]] 298; CHECK: middle.block: 299; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]] 300; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[SCALAR_PH]] 301; CHECK: scalar.ph: 302; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_PREHEADER]] ] 303; CHECK-NEXT: br label [[FOR_BODY:%.*]] 304; CHECK: for.body: 305; CHECK-NEXT: [[I_08:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] 306; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[I_08]] 307; CHECK-NEXT: [[TMP9:%.*]] = load i8, i8* [[ARRAYIDX]], align 1 308; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP9]], 0 309; CHECK-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1 310; CHECK-NEXT: store i8 [[DOT]], i8* [[ARRAYIDX]], align 1 311; CHECK-NEXT: [[INC]] = add nsw i32 [[I_08]], 1 312; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[I_08]], [[BOUND]] 313; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY]], !prof [[PROF12:![0-9]+]], !llvm.loop [[LOOP13:![0-9]+]] 314; CHECK: for.end.loopexit: 315; CHECK-NEXT: br label [[FOR_END]] 316; CHECK: for.end: 317; CHECK-NEXT: ret i32 0 318; 319entry: 320 br i1 %cond, label %for.preheader, label %for.end, !prof !2 321 322for.preheader: 323 br label %for.body 324 325for.body: ; preds = %for.body, %entry 326 %i.08 = phi i32 [ 0, %for.preheader ], [ %inc, %for.body ] 327 %arrayidx = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 %i.08 328 %0 = load i8, i8* %arrayidx, align 1 329 %cmp1 = icmp eq i8 %0, 0 330 %. = select i1 %cmp1, i8 2, i8 1 331 store i8 %., i8* %arrayidx, align 1 332 %inc = add nsw i32 %i.08, 1 333 %exitcond = icmp eq i32 %i.08, %bound 334 br i1 %exitcond, label %for.end, label %for.body, !prof !3 335 336for.end: ; preds = %for.body 337 ret i32 0 338} 339 340define i32 @foo_low_trip_count_icmp_sgt(i32 %bound) { 341; Simple loop with low tripcount and inequality test for exit. 342; Should not be vectorized. 343; CHECK-LABEL: @foo_low_trip_count_icmp_sgt( 344; CHECK-NEXT: entry: 345; CHECK-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[BOUND:%.*]], i32 -1) 346; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[SMAX]], 2 347; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 348; CHECK: vector.ph: 349; CHECK-NEXT: [[N_RND_UP:%.*]] = add i32 [[TMP0]], 3 350; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[N_RND_UP]], 4 351; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[N_RND_UP]], [[N_MOD_VF]] 352; CHECK-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i32 [[TMP0]], 1 353; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[TRIP_COUNT_MINUS_1]], i32 0 354; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer 355; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 356; CHECK: vector.body: 357; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE12:%.*]] ] 358; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE12]] ] 359; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[INDEX]], 0 360; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[INDEX]], 1 361; CHECK-NEXT: [[TMP3:%.*]] = add i32 [[INDEX]], 2 362; CHECK-NEXT: [[TMP4:%.*]] = add i32 [[INDEX]], 3 363; CHECK-NEXT: [[TMP5:%.*]] = icmp ule <4 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]] 364; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP1]] 365; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP2]] 366; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP3]] 367; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP4]] 368; CHECK-NEXT: [[TMP10:%.*]] = extractelement <4 x i1> [[TMP5]], i32 0 369; CHECK-NEXT: br i1 [[TMP10]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]] 370; CHECK: pred.load.if: 371; CHECK-NEXT: [[TMP11:%.*]] = load i8, i8* [[TMP6]], align 1 372; CHECK-NEXT: [[TMP12:%.*]] = insertelement <4 x i8> poison, i8 [[TMP11]], i32 0 373; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE]] 374; CHECK: pred.load.continue: 375; CHECK-NEXT: [[TMP13:%.*]] = phi <4 x i8> [ poison, [[VECTOR_BODY]] ], [ [[TMP12]], [[PRED_LOAD_IF]] ] 376; CHECK-NEXT: [[TMP14:%.*]] = extractelement <4 x i1> [[TMP5]], i32 1 377; CHECK-NEXT: br i1 [[TMP14]], label [[PRED_LOAD_IF1:%.*]], label [[PRED_LOAD_CONTINUE2:%.*]] 378; CHECK: pred.load.if1: 379; CHECK-NEXT: [[TMP15:%.*]] = load i8, i8* [[TMP7]], align 1 380; CHECK-NEXT: [[TMP16:%.*]] = insertelement <4 x i8> [[TMP13]], i8 [[TMP15]], i32 1 381; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE2]] 382; CHECK: pred.load.continue2: 383; CHECK-NEXT: [[TMP17:%.*]] = phi <4 x i8> [ [[TMP13]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP16]], [[PRED_LOAD_IF1]] ] 384; CHECK-NEXT: [[TMP18:%.*]] = extractelement <4 x i1> [[TMP5]], i32 2 385; CHECK-NEXT: br i1 [[TMP18]], label [[PRED_LOAD_IF3:%.*]], label [[PRED_LOAD_CONTINUE4:%.*]] 386; CHECK: pred.load.if3: 387; CHECK-NEXT: [[TMP19:%.*]] = load i8, i8* [[TMP8]], align 1 388; CHECK-NEXT: [[TMP20:%.*]] = insertelement <4 x i8> [[TMP17]], i8 [[TMP19]], i32 2 389; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE4]] 390; CHECK: pred.load.continue4: 391; CHECK-NEXT: [[TMP21:%.*]] = phi <4 x i8> [ [[TMP17]], [[PRED_LOAD_CONTINUE2]] ], [ [[TMP20]], [[PRED_LOAD_IF3]] ] 392; CHECK-NEXT: [[TMP22:%.*]] = extractelement <4 x i1> [[TMP5]], i32 3 393; CHECK-NEXT: br i1 [[TMP22]], label [[PRED_LOAD_IF5:%.*]], label [[PRED_LOAD_CONTINUE6:%.*]] 394; CHECK: pred.load.if5: 395; CHECK-NEXT: [[TMP23:%.*]] = load i8, i8* [[TMP9]], align 1 396; CHECK-NEXT: [[TMP24:%.*]] = insertelement <4 x i8> [[TMP21]], i8 [[TMP23]], i32 3 397; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE6]] 398; CHECK: pred.load.continue6: 399; CHECK-NEXT: [[TMP25:%.*]] = phi <4 x i8> [ [[TMP21]], [[PRED_LOAD_CONTINUE4]] ], [ [[TMP24]], [[PRED_LOAD_IF5]] ] 400; CHECK-NEXT: [[TMP26:%.*]] = icmp eq <4 x i8> [[TMP25]], zeroinitializer 401; CHECK-NEXT: [[TMP27:%.*]] = select <4 x i1> [[TMP26]], <4 x i8> <i8 2, i8 2, i8 2, i8 2>, <4 x i8> <i8 1, i8 1, i8 1, i8 1> 402; CHECK-NEXT: [[TMP28:%.*]] = extractelement <4 x i1> [[TMP5]], i32 0 403; CHECK-NEXT: br i1 [[TMP28]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]] 404; CHECK: pred.store.if: 405; CHECK-NEXT: [[TMP29:%.*]] = extractelement <4 x i8> [[TMP27]], i32 0 406; CHECK-NEXT: store i8 [[TMP29]], i8* [[TMP6]], align 1 407; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]] 408; CHECK: pred.store.continue: 409; CHECK-NEXT: [[TMP30:%.*]] = extractelement <4 x i1> [[TMP5]], i32 1 410; CHECK-NEXT: br i1 [[TMP30]], label [[PRED_STORE_IF7:%.*]], label [[PRED_STORE_CONTINUE8:%.*]] 411; CHECK: pred.store.if7: 412; CHECK-NEXT: [[TMP31:%.*]] = extractelement <4 x i8> [[TMP27]], i32 1 413; CHECK-NEXT: store i8 [[TMP31]], i8* [[TMP7]], align 1 414; CHECK-NEXT: br label [[PRED_STORE_CONTINUE8]] 415; CHECK: pred.store.continue8: 416; CHECK-NEXT: [[TMP32:%.*]] = extractelement <4 x i1> [[TMP5]], i32 2 417; CHECK-NEXT: br i1 [[TMP32]], label [[PRED_STORE_IF9:%.*]], label [[PRED_STORE_CONTINUE10:%.*]] 418; CHECK: pred.store.if9: 419; CHECK-NEXT: [[TMP33:%.*]] = extractelement <4 x i8> [[TMP27]], i32 2 420; CHECK-NEXT: store i8 [[TMP33]], i8* [[TMP8]], align 1 421; CHECK-NEXT: br label [[PRED_STORE_CONTINUE10]] 422; CHECK: pred.store.continue10: 423; CHECK-NEXT: [[TMP34:%.*]] = extractelement <4 x i1> [[TMP5]], i32 3 424; CHECK-NEXT: br i1 [[TMP34]], label [[PRED_STORE_IF11:%.*]], label [[PRED_STORE_CONTINUE12]] 425; CHECK: pred.store.if11: 426; CHECK-NEXT: [[TMP35:%.*]] = extractelement <4 x i8> [[TMP27]], i32 3 427; CHECK-NEXT: store i8 [[TMP35]], i8* [[TMP9]], align 1 428; CHECK-NEXT: br label [[PRED_STORE_CONTINUE12]] 429; CHECK: pred.store.continue12: 430; CHECK-NEXT: [[INDEX_NEXT]] = add i32 [[INDEX]], 4 431; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4> 432; CHECK-NEXT: [[TMP36:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 433; CHECK-NEXT: br i1 [[TMP36]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !prof [[PROF0]], !llvm.loop [[LOOP14:![0-9]+]] 434; CHECK: middle.block: 435; CHECK-NEXT: br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]] 436; CHECK: scalar.ph: 437; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 438; CHECK-NEXT: br label [[FOR_BODY:%.*]] 439; CHECK: for.body: 440; CHECK-NEXT: [[I_08:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] 441; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[I_08]] 442; CHECK-NEXT: [[TMP37:%.*]] = load i8, i8* [[ARRAYIDX]], align 1 443; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP37]], 0 444; CHECK-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1 445; CHECK-NEXT: store i8 [[DOT]], i8* [[ARRAYIDX]], align 1 446; CHECK-NEXT: [[INC]] = add nsw i32 [[I_08]], 1 447; CHECK-NEXT: [[EXITCOND:%.*]] = icmp sgt i32 [[I_08]], [[BOUND]] 448; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END]], label [[FOR_BODY]], !prof [[PROF3]], !llvm.loop [[LOOP15:![0-9]+]] 449; CHECK: for.end: 450; CHECK-NEXT: ret i32 0 451; 452 453entry: 454 br label %for.body 455 456for.body: ; preds = %for.body, %entry 457 %i.08 = phi i32 [ 0, %entry ], [ %inc, %for.body ] 458 %arrayidx = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 %i.08 459 %0 = load i8, i8* %arrayidx, align 1 460 %cmp1 = icmp eq i8 %0, 0 461 %. = select i1 %cmp1, i8 2, i8 1 462 store i8 %., i8* %arrayidx, align 1 463 %inc = add nsw i32 %i.08, 1 464 %exitcond = icmp sgt i32 %i.08, %bound 465 br i1 %exitcond, label %for.end, label %for.body, !prof !1 466 467for.end: ; preds = %for.body 468 ret i32 0 469} 470 471define i32 @const_low_trip_count() { 472; Simple loop with constant, small trip count and no profiling info. 473; CHECK-LABEL: @const_low_trip_count( 474; CHECK-NEXT: entry: 475; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 476; CHECK: vector.ph: 477; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 478; CHECK: vector.body: 479; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE12:%.*]] ] 480; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE12]] ] 481; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 0 482; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[INDEX]], 1 483; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[INDEX]], 2 484; CHECK-NEXT: [[TMP3:%.*]] = add i32 [[INDEX]], 3 485; CHECK-NEXT: [[TMP4:%.*]] = icmp ule <4 x i32> [[VEC_IND]], <i32 2, i32 2, i32 2, i32 2> 486; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP0]] 487; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP1]] 488; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP2]] 489; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP3]] 490; CHECK-NEXT: [[TMP9:%.*]] = extractelement <4 x i1> [[TMP4]], i32 0 491; CHECK-NEXT: br i1 [[TMP9]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]] 492; CHECK: pred.load.if: 493; CHECK-NEXT: [[TMP10:%.*]] = load i8, i8* [[TMP5]], align 1 494; CHECK-NEXT: [[TMP11:%.*]] = insertelement <4 x i8> poison, i8 [[TMP10]], i32 0 495; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE]] 496; CHECK: pred.load.continue: 497; CHECK-NEXT: [[TMP12:%.*]] = phi <4 x i8> [ poison, [[VECTOR_BODY]] ], [ [[TMP11]], [[PRED_LOAD_IF]] ] 498; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x i1> [[TMP4]], i32 1 499; CHECK-NEXT: br i1 [[TMP13]], label [[PRED_LOAD_IF1:%.*]], label [[PRED_LOAD_CONTINUE2:%.*]] 500; CHECK: pred.load.if1: 501; CHECK-NEXT: [[TMP14:%.*]] = load i8, i8* [[TMP6]], align 1 502; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x i8> [[TMP12]], i8 [[TMP14]], i32 1 503; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE2]] 504; CHECK: pred.load.continue2: 505; CHECK-NEXT: [[TMP16:%.*]] = phi <4 x i8> [ [[TMP12]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP15]], [[PRED_LOAD_IF1]] ] 506; CHECK-NEXT: [[TMP17:%.*]] = extractelement <4 x i1> [[TMP4]], i32 2 507; CHECK-NEXT: br i1 [[TMP17]], label [[PRED_LOAD_IF3:%.*]], label [[PRED_LOAD_CONTINUE4:%.*]] 508; CHECK: pred.load.if3: 509; CHECK-NEXT: [[TMP18:%.*]] = load i8, i8* [[TMP7]], align 1 510; CHECK-NEXT: [[TMP19:%.*]] = insertelement <4 x i8> [[TMP16]], i8 [[TMP18]], i32 2 511; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE4]] 512; CHECK: pred.load.continue4: 513; CHECK-NEXT: [[TMP20:%.*]] = phi <4 x i8> [ [[TMP16]], [[PRED_LOAD_CONTINUE2]] ], [ [[TMP19]], [[PRED_LOAD_IF3]] ] 514; CHECK-NEXT: [[TMP21:%.*]] = extractelement <4 x i1> [[TMP4]], i32 3 515; CHECK-NEXT: br i1 [[TMP21]], label [[PRED_LOAD_IF5:%.*]], label [[PRED_LOAD_CONTINUE6:%.*]] 516; CHECK: pred.load.if5: 517; CHECK-NEXT: [[TMP22:%.*]] = load i8, i8* [[TMP8]], align 1 518; CHECK-NEXT: [[TMP23:%.*]] = insertelement <4 x i8> [[TMP20]], i8 [[TMP22]], i32 3 519; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE6]] 520; CHECK: pred.load.continue6: 521; CHECK-NEXT: [[TMP24:%.*]] = phi <4 x i8> [ [[TMP20]], [[PRED_LOAD_CONTINUE4]] ], [ [[TMP23]], [[PRED_LOAD_IF5]] ] 522; CHECK-NEXT: [[TMP25:%.*]] = icmp eq <4 x i8> [[TMP24]], zeroinitializer 523; CHECK-NEXT: [[TMP26:%.*]] = select <4 x i1> [[TMP25]], <4 x i8> <i8 2, i8 2, i8 2, i8 2>, <4 x i8> <i8 1, i8 1, i8 1, i8 1> 524; CHECK-NEXT: [[TMP27:%.*]] = extractelement <4 x i1> [[TMP4]], i32 0 525; CHECK-NEXT: br i1 [[TMP27]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]] 526; CHECK: pred.store.if: 527; CHECK-NEXT: [[TMP28:%.*]] = extractelement <4 x i8> [[TMP26]], i32 0 528; CHECK-NEXT: store i8 [[TMP28]], i8* [[TMP5]], align 1 529; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]] 530; CHECK: pred.store.continue: 531; CHECK-NEXT: [[TMP29:%.*]] = extractelement <4 x i1> [[TMP4]], i32 1 532; CHECK-NEXT: br i1 [[TMP29]], label [[PRED_STORE_IF7:%.*]], label [[PRED_STORE_CONTINUE8:%.*]] 533; CHECK: pred.store.if7: 534; CHECK-NEXT: [[TMP30:%.*]] = extractelement <4 x i8> [[TMP26]], i32 1 535; CHECK-NEXT: store i8 [[TMP30]], i8* [[TMP6]], align 1 536; CHECK-NEXT: br label [[PRED_STORE_CONTINUE8]] 537; CHECK: pred.store.continue8: 538; CHECK-NEXT: [[TMP31:%.*]] = extractelement <4 x i1> [[TMP4]], i32 2 539; CHECK-NEXT: br i1 [[TMP31]], label [[PRED_STORE_IF9:%.*]], label [[PRED_STORE_CONTINUE10:%.*]] 540; CHECK: pred.store.if9: 541; CHECK-NEXT: [[TMP32:%.*]] = extractelement <4 x i8> [[TMP26]], i32 2 542; CHECK-NEXT: store i8 [[TMP32]], i8* [[TMP7]], align 1 543; CHECK-NEXT: br label [[PRED_STORE_CONTINUE10]] 544; CHECK: pred.store.continue10: 545; CHECK-NEXT: [[TMP33:%.*]] = extractelement <4 x i1> [[TMP4]], i32 3 546; CHECK-NEXT: br i1 [[TMP33]], label [[PRED_STORE_IF11:%.*]], label [[PRED_STORE_CONTINUE12]] 547; CHECK: pred.store.if11: 548; CHECK-NEXT: [[TMP34:%.*]] = extractelement <4 x i8> [[TMP26]], i32 3 549; CHECK-NEXT: store i8 [[TMP34]], i8* [[TMP8]], align 1 550; CHECK-NEXT: br label [[PRED_STORE_CONTINUE12]] 551; CHECK: pred.store.continue12: 552; CHECK-NEXT: [[INDEX_NEXT]] = add i32 [[INDEX]], 4 553; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4> 554; CHECK-NEXT: [[TMP35:%.*]] = icmp eq i32 [[INDEX_NEXT]], 4 555; CHECK-NEXT: br i1 [[TMP35]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]] 556; CHECK: middle.block: 557; CHECK-NEXT: br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]] 558; CHECK: scalar.ph: 559; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 4, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 560; CHECK-NEXT: br label [[FOR_BODY:%.*]] 561; CHECK: for.body: 562; CHECK-NEXT: [[I_08:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] 563; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[I_08]] 564; CHECK-NEXT: [[TMP36:%.*]] = load i8, i8* [[ARRAYIDX]], align 1 565; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP36]], 0 566; CHECK-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1 567; CHECK-NEXT: store i8 [[DOT]], i8* [[ARRAYIDX]], align 1 568; CHECK-NEXT: [[INC]] = add nsw i32 [[I_08]], 1 569; CHECK-NEXT: [[EXITCOND:%.*]] = icmp slt i32 [[I_08]], 2 570; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP17:![0-9]+]] 571; CHECK: for.end: 572; CHECK-NEXT: ret i32 0 573; 574 575entry: 576 br label %for.body 577 578for.body: ; preds = %for.body, %entry 579 %i.08 = phi i32 [ 0, %entry ], [ %inc, %for.body ] 580 %arrayidx = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 %i.08 581 %0 = load i8, i8* %arrayidx, align 1 582 %cmp1 = icmp eq i8 %0, 0 583 %. = select i1 %cmp1, i8 2, i8 1 584 store i8 %., i8* %arrayidx, align 1 585 %inc = add nsw i32 %i.08, 1 586 %exitcond = icmp slt i32 %i.08, 2 587 br i1 %exitcond, label %for.body, label %for.end 588 589for.end: ; preds = %for.body 590 ret i32 0 591} 592 593define i32 @const_large_trip_count() { 594; Simple loop with constant large trip count and no profiling info. 595; CHECK-LABEL: @const_large_trip_count( 596; CHECK-NEXT: entry: 597; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 598; CHECK: vector.ph: 599; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 600; CHECK: vector.body: 601; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 602; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 0 603; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP0]] 604; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, i8* [[TMP1]], i32 0 605; CHECK-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to <4 x i8>* 606; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, <4 x i8>* [[TMP3]], align 1 607; CHECK-NEXT: [[TMP4:%.*]] = icmp eq <4 x i8> [[WIDE_LOAD]], zeroinitializer 608; CHECK-NEXT: [[TMP5:%.*]] = select <4 x i1> [[TMP4]], <4 x i8> <i8 2, i8 2, i8 2, i8 2>, <4 x i8> <i8 1, i8 1, i8 1, i8 1> 609; CHECK-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP2]] to <4 x i8>* 610; CHECK-NEXT: store <4 x i8> [[TMP5]], <4 x i8>* [[TMP6]], align 1 611; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 612; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000 613; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]] 614; CHECK: middle.block: 615; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 1001, 1000 616; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 617; CHECK: scalar.ph: 618; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1000, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 619; CHECK-NEXT: br label [[FOR_BODY:%.*]] 620; CHECK: for.body: 621; CHECK-NEXT: [[I_08:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] 622; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[I_08]] 623; CHECK-NEXT: [[TMP8:%.*]] = load i8, i8* [[ARRAYIDX]], align 1 624; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP8]], 0 625; CHECK-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1 626; CHECK-NEXT: store i8 [[DOT]], i8* [[ARRAYIDX]], align 1 627; CHECK-NEXT: [[INC]] = add nsw i32 [[I_08]], 1 628; CHECK-NEXT: [[EXITCOND:%.*]] = icmp slt i32 [[I_08]], 1000 629; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP19:![0-9]+]] 630; CHECK: for.end: 631; CHECK-NEXT: ret i32 0 632; 633 634entry: 635 br label %for.body 636 637for.body: ; preds = %for.body, %entry 638 %i.08 = phi i32 [ 0, %entry ], [ %inc, %for.body ] 639 %arrayidx = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 %i.08 640 %0 = load i8, i8* %arrayidx, align 1 641 %cmp1 = icmp eq i8 %0, 0 642 %. = select i1 %cmp1, i8 2, i8 1 643 store i8 %., i8* %arrayidx, align 1 644 %inc = add nsw i32 %i.08, 1 645 %exitcond = icmp slt i32 %i.08, 1000 646 br i1 %exitcond, label %for.body, label %for.end 647 648for.end: ; preds = %for.body 649 ret i32 0 650} 651 652define i32 @const_small_trip_count_step() { 653; Simple loop with static, small trip count and no profiling info. 654; CHECK-LABEL: @const_small_trip_count_step( 655; CHECK-NEXT: entry: 656; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 657; CHECK: vector.ph: 658; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 659; CHECK: vector.body: 660; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE12:%.*]] ] 661; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i32 [[INDEX]], 5 662; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[OFFSET_IDX]], 0 663; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[OFFSET_IDX]], 5 664; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[OFFSET_IDX]], 10 665; CHECK-NEXT: [[TMP3:%.*]] = add i32 [[OFFSET_IDX]], 15 666; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[INDEX]], i32 0 667; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer 668; CHECK-NEXT: [[VEC_IV:%.*]] = add <4 x i32> [[BROADCAST_SPLAT]], <i32 0, i32 1, i32 2, i32 3> 669; CHECK-NEXT: [[TMP4:%.*]] = icmp ule <4 x i32> [[VEC_IV]], <i32 2, i32 2, i32 2, i32 2> 670; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP0]] 671; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP1]] 672; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP2]] 673; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP3]] 674; CHECK-NEXT: [[TMP9:%.*]] = extractelement <4 x i1> [[TMP4]], i32 0 675; CHECK-NEXT: br i1 [[TMP9]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]] 676; CHECK: pred.load.if: 677; CHECK-NEXT: [[TMP10:%.*]] = load i8, i8* [[TMP5]], align 1 678; CHECK-NEXT: [[TMP11:%.*]] = insertelement <4 x i8> poison, i8 [[TMP10]], i32 0 679; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE]] 680; CHECK: pred.load.continue: 681; CHECK-NEXT: [[TMP12:%.*]] = phi <4 x i8> [ poison, [[VECTOR_BODY]] ], [ [[TMP11]], [[PRED_LOAD_IF]] ] 682; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x i1> [[TMP4]], i32 1 683; CHECK-NEXT: br i1 [[TMP13]], label [[PRED_LOAD_IF1:%.*]], label [[PRED_LOAD_CONTINUE2:%.*]] 684; CHECK: pred.load.if1: 685; CHECK-NEXT: [[TMP14:%.*]] = load i8, i8* [[TMP6]], align 1 686; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x i8> [[TMP12]], i8 [[TMP14]], i32 1 687; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE2]] 688; CHECK: pred.load.continue2: 689; CHECK-NEXT: [[TMP16:%.*]] = phi <4 x i8> [ [[TMP12]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP15]], [[PRED_LOAD_IF1]] ] 690; CHECK-NEXT: [[TMP17:%.*]] = extractelement <4 x i1> [[TMP4]], i32 2 691; CHECK-NEXT: br i1 [[TMP17]], label [[PRED_LOAD_IF3:%.*]], label [[PRED_LOAD_CONTINUE4:%.*]] 692; CHECK: pred.load.if3: 693; CHECK-NEXT: [[TMP18:%.*]] = load i8, i8* [[TMP7]], align 1 694; CHECK-NEXT: [[TMP19:%.*]] = insertelement <4 x i8> [[TMP16]], i8 [[TMP18]], i32 2 695; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE4]] 696; CHECK: pred.load.continue4: 697; CHECK-NEXT: [[TMP20:%.*]] = phi <4 x i8> [ [[TMP16]], [[PRED_LOAD_CONTINUE2]] ], [ [[TMP19]], [[PRED_LOAD_IF3]] ] 698; CHECK-NEXT: [[TMP21:%.*]] = extractelement <4 x i1> [[TMP4]], i32 3 699; CHECK-NEXT: br i1 [[TMP21]], label [[PRED_LOAD_IF5:%.*]], label [[PRED_LOAD_CONTINUE6:%.*]] 700; CHECK: pred.load.if5: 701; CHECK-NEXT: [[TMP22:%.*]] = load i8, i8* [[TMP8]], align 1 702; CHECK-NEXT: [[TMP23:%.*]] = insertelement <4 x i8> [[TMP20]], i8 [[TMP22]], i32 3 703; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE6]] 704; CHECK: pred.load.continue6: 705; CHECK-NEXT: [[TMP24:%.*]] = phi <4 x i8> [ [[TMP20]], [[PRED_LOAD_CONTINUE4]] ], [ [[TMP23]], [[PRED_LOAD_IF5]] ] 706; CHECK-NEXT: [[TMP25:%.*]] = icmp eq <4 x i8> [[TMP24]], zeroinitializer 707; CHECK-NEXT: [[TMP26:%.*]] = select <4 x i1> [[TMP25]], <4 x i8> <i8 2, i8 2, i8 2, i8 2>, <4 x i8> <i8 1, i8 1, i8 1, i8 1> 708; CHECK-NEXT: [[TMP27:%.*]] = extractelement <4 x i1> [[TMP4]], i32 0 709; CHECK-NEXT: br i1 [[TMP27]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]] 710; CHECK: pred.store.if: 711; CHECK-NEXT: [[TMP28:%.*]] = extractelement <4 x i8> [[TMP26]], i32 0 712; CHECK-NEXT: store i8 [[TMP28]], i8* [[TMP5]], align 1 713; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]] 714; CHECK: pred.store.continue: 715; CHECK-NEXT: [[TMP29:%.*]] = extractelement <4 x i1> [[TMP4]], i32 1 716; CHECK-NEXT: br i1 [[TMP29]], label [[PRED_STORE_IF7:%.*]], label [[PRED_STORE_CONTINUE8:%.*]] 717; CHECK: pred.store.if7: 718; CHECK-NEXT: [[TMP30:%.*]] = extractelement <4 x i8> [[TMP26]], i32 1 719; CHECK-NEXT: store i8 [[TMP30]], i8* [[TMP6]], align 1 720; CHECK-NEXT: br label [[PRED_STORE_CONTINUE8]] 721; CHECK: pred.store.continue8: 722; CHECK-NEXT: [[TMP31:%.*]] = extractelement <4 x i1> [[TMP4]], i32 2 723; CHECK-NEXT: br i1 [[TMP31]], label [[PRED_STORE_IF9:%.*]], label [[PRED_STORE_CONTINUE10:%.*]] 724; CHECK: pred.store.if9: 725; CHECK-NEXT: [[TMP32:%.*]] = extractelement <4 x i8> [[TMP26]], i32 2 726; CHECK-NEXT: store i8 [[TMP32]], i8* [[TMP7]], align 1 727; CHECK-NEXT: br label [[PRED_STORE_CONTINUE10]] 728; CHECK: pred.store.continue10: 729; CHECK-NEXT: [[TMP33:%.*]] = extractelement <4 x i1> [[TMP4]], i32 3 730; CHECK-NEXT: br i1 [[TMP33]], label [[PRED_STORE_IF11:%.*]], label [[PRED_STORE_CONTINUE12]] 731; CHECK: pred.store.if11: 732; CHECK-NEXT: [[TMP34:%.*]] = extractelement <4 x i8> [[TMP26]], i32 3 733; CHECK-NEXT: store i8 [[TMP34]], i8* [[TMP8]], align 1 734; CHECK-NEXT: br label [[PRED_STORE_CONTINUE12]] 735; CHECK: pred.store.continue12: 736; CHECK-NEXT: [[INDEX_NEXT]] = add i32 [[INDEX]], 4 737; CHECK-NEXT: [[TMP35:%.*]] = icmp eq i32 [[INDEX_NEXT]], 4 738; CHECK-NEXT: br i1 [[TMP35]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]] 739; CHECK: middle.block: 740; CHECK-NEXT: br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]] 741; CHECK: scalar.ph: 742; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 20, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 743; CHECK-NEXT: br label [[FOR_BODY:%.*]] 744; CHECK: for.body: 745; CHECK-NEXT: [[I_08:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] 746; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[I_08]] 747; CHECK-NEXT: [[TMP36:%.*]] = load i8, i8* [[ARRAYIDX]], align 1 748; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP36]], 0 749; CHECK-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1 750; CHECK-NEXT: store i8 [[DOT]], i8* [[ARRAYIDX]], align 1 751; CHECK-NEXT: [[INC]] = add nsw i32 [[I_08]], 5 752; CHECK-NEXT: [[EXITCOND:%.*]] = icmp slt i32 [[I_08]], 10 753; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP21:![0-9]+]] 754; CHECK: for.end: 755; CHECK-NEXT: ret i32 0 756; 757 758entry: 759 br label %for.body 760 761for.body: ; preds = %for.body, %entry 762 %i.08 = phi i32 [ 0, %entry ], [ %inc, %for.body ] 763 %arrayidx = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 %i.08 764 %0 = load i8, i8* %arrayidx, align 1 765 %cmp1 = icmp eq i8 %0, 0 766 %. = select i1 %cmp1, i8 2, i8 1 767 store i8 %., i8* %arrayidx, align 1 768 %inc = add nsw i32 %i.08, 5 769 %exitcond = icmp slt i32 %i.08, 10 770 br i1 %exitcond, label %for.body, label %for.end 771 772for.end: ; preds = %for.body 773 ret i32 0 774} 775 776define i32 @const_trip_over_profile() { 777; constant trip count takes precedence over profile data 778; CHECK-LABEL: @const_trip_over_profile( 779; CHECK-NEXT: entry: 780; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 781; CHECK: vector.ph: 782; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 783; CHECK: vector.body: 784; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 785; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 0 786; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[TMP0]] 787; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, i8* [[TMP1]], i32 0 788; CHECK-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to <4 x i8>* 789; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, <4 x i8>* [[TMP3]], align 1 790; CHECK-NEXT: [[TMP4:%.*]] = icmp eq <4 x i8> [[WIDE_LOAD]], zeroinitializer 791; CHECK-NEXT: [[TMP5:%.*]] = select <4 x i1> [[TMP4]], <4 x i8> <i8 2, i8 2, i8 2, i8 2>, <4 x i8> <i8 1, i8 1, i8 1, i8 1> 792; CHECK-NEXT: [[TMP6:%.*]] = bitcast i8* [[TMP2]] to <4 x i8>* 793; CHECK-NEXT: store <4 x i8> [[TMP5]], <4 x i8>* [[TMP6]], align 1 794; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 795; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000 796; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]] 797; CHECK: middle.block: 798; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 1001, 1000 799; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 800; CHECK: scalar.ph: 801; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1000, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 802; CHECK-NEXT: br label [[FOR_BODY:%.*]] 803; CHECK: for.body: 804; CHECK-NEXT: [[I_08:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] 805; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 [[I_08]] 806; CHECK-NEXT: [[TMP8:%.*]] = load i8, i8* [[ARRAYIDX]], align 1 807; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i8 [[TMP8]], 0 808; CHECK-NEXT: [[DOT:%.*]] = select i1 [[CMP1]], i8 2, i8 1 809; CHECK-NEXT: store i8 [[DOT]], i8* [[ARRAYIDX]], align 1 810; CHECK-NEXT: [[INC]] = add nsw i32 [[I_08]], 1 811; CHECK-NEXT: [[EXITCOND:%.*]] = icmp slt i32 [[I_08]], 1000 812; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END]], !prof [[PROF3]], !llvm.loop [[LOOP23:![0-9]+]] 813; CHECK: for.end: 814; CHECK-NEXT: ret i32 0 815; 816 817entry: 818 br label %for.body 819 820for.body: ; preds = %for.body, %entry 821 %i.08 = phi i32 [ 0, %entry ], [ %inc, %for.body ] 822 %arrayidx = getelementptr inbounds [32 x i8], [32 x i8]* @tab, i32 0, i32 %i.08 823 %0 = load i8, i8* %arrayidx, align 1 824 %cmp1 = icmp eq i8 %0, 0 825 %. = select i1 %cmp1, i8 2, i8 1 826 store i8 %., i8* %arrayidx, align 1 827 %inc = add nsw i32 %i.08, 1 828 %exitcond = icmp slt i32 %i.08, 1000 829 br i1 %exitcond, label %for.body, label %for.end, !prof !1 830 831for.end: ; preds = %for.body 832 ret i32 0 833} 834 835; original loop has latchExitWeight=10 and backedgeTakenWeight=10,000, 836; therefore estimatedBackedgeTakenCount=1,000 and estimatedTripCount=1,001. 837; Vectorizing by 4 produces estimatedTripCounts of 1,001/4=250 and 1,001%4=1 838; for vectorized and remainder loops, respectively, therefore their 839; estimatedBackedgeTakenCounts are 249 and 0, and so the weights recorded with 840; loop invocation weights of 10 are the above {10, 2490} and {10, 0}. 841 842!0 = !{!"function_entry_count", i64 100} 843!1 = !{!"branch_weights", i32 100, i32 0} 844!2 = !{!"branch_weights", i32 10, i32 90} 845!3 = !{!"branch_weights", i32 10, i32 10000} 846