1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2 3; To test epilogue-vectorization we need to make sure that the vectorizer actually vectorizes the loop. 4; Without a target triple this becomes difficult, unless we force vectorization through user hints. 5; Currently user provided vectorization hints prevent epilogue vectorization unless the forced 6; VF is the same as the epilogue vectorization VF. To make these tests target independent we'll use a 7; trick where both VFs are forced to be the same value. Mismatching VFs are tested in target specific tests. 8; RUN: opt -passes='loop-vectorize' -force-vector-width=4 -enable-epilogue-vectorization -epilogue-vectorization-force-VF=4 -S %s | FileCheck %s 9 10; Some simpler cases are found profitable even without triple or user hints. 11; RUN: opt -passes='loop-vectorize' -enable-epilogue-vectorization -epilogue-vectorization-force-VF=2 -S %s | FileCheck --check-prefix=CHECK-PROFITABLE-BY-DEFAULT %s 12 13target datalayout = "e-m:e-i64:64-n32:64-v128:128:128" 14 15define dso_local void @f1(float* noalias %aa, float* noalias %bb, float* noalias %cc, i32 signext %N) { 16; CHECK-LABEL: @f1( 17; CHECK-NEXT: entry: 18; CHECK-NEXT: [[CMP1:%.*]] = icmp sgt i32 [[N:%.*]], 0 19; CHECK-NEXT: br i1 [[CMP1]], label [[ITER_CHECK:%.*]], label [[FOR_END:%.*]] 20; CHECK: iter.check: 21; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i64 22; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 4 23; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]] 24; CHECK: vector.main.loop.iter.check: 25; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 4 26; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]] 27; CHECK: vector.ph: 28; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 4 29; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]] 30; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 31; CHECK: vector.body: 32; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 33; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 34; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds float, float* [[BB:%.*]], i64 [[TMP0]] 35; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, float* [[TMP1]], i32 0 36; CHECK-NEXT: [[TMP3:%.*]] = bitcast float* [[TMP2]] to <4 x float>* 37; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, <4 x float>* [[TMP3]], align 4 38; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, float* [[CC:%.*]], i64 [[TMP0]] 39; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds float, float* [[TMP4]], i32 0 40; CHECK-NEXT: [[TMP6:%.*]] = bitcast float* [[TMP5]] to <4 x float>* 41; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <4 x float>, <4 x float>* [[TMP6]], align 4 42; CHECK-NEXT: [[TMP7:%.*]] = fadd fast <4 x float> [[WIDE_LOAD]], [[WIDE_LOAD2]] 43; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, float* [[AA:%.*]], i64 [[TMP0]] 44; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, float* [[TMP8]], i32 0 45; CHECK-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to <4 x float>* 46; CHECK-NEXT: store <4 x float> [[TMP7]], <4 x float>* [[TMP10]], align 4 47; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 48; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 49; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 50; CHECK: middle.block: 51; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]] 52; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] 53; CHECK: vec.epilog.iter.check: 54; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_VEC]] 55; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 4 56; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] 57; CHECK: vec.epilog.ph: 58; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] 59; CHECK-NEXT: [[N_MOD_VF3:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 4 60; CHECK-NEXT: [[N_VEC4:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF3]] 61; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] 62; CHECK: vec.epilog.vector.body: 63; CHECK-NEXT: [[OFFSET_IDX:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT9:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 64; CHECK-NEXT: [[TMP12:%.*]] = add i64 [[OFFSET_IDX]], 0 65; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds float, float* [[BB]], i64 [[TMP12]] 66; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds float, float* [[TMP13]], i32 0 67; CHECK-NEXT: [[TMP15:%.*]] = bitcast float* [[TMP14]] to <4 x float>* 68; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <4 x float>, <4 x float>* [[TMP15]], align 4 69; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds float, float* [[CC]], i64 [[TMP12]] 70; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, float* [[TMP16]], i32 0 71; CHECK-NEXT: [[TMP18:%.*]] = bitcast float* [[TMP17]] to <4 x float>* 72; CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <4 x float>, <4 x float>* [[TMP18]], align 4 73; CHECK-NEXT: [[TMP19:%.*]] = fadd fast <4 x float> [[WIDE_LOAD7]], [[WIDE_LOAD8]] 74; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds float, float* [[AA]], i64 [[TMP12]] 75; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, float* [[TMP20]], i32 0 76; CHECK-NEXT: [[TMP22:%.*]] = bitcast float* [[TMP21]] to <4 x float>* 77; CHECK-NEXT: store <4 x float> [[TMP19]], <4 x float>* [[TMP22]], align 4 78; CHECK-NEXT: [[INDEX_NEXT9]] = add nuw i64 [[OFFSET_IDX]], 4 79; CHECK-NEXT: [[TMP23:%.*]] = icmp eq i64 [[INDEX_NEXT9]], [[N_VEC4]] 80; CHECK-NEXT: br i1 [[TMP23]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]] 81; CHECK: vec.epilog.middle.block: 82; CHECK-NEXT: [[CMP_N5:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC4]] 83; CHECK-NEXT: br i1 [[CMP_N5]], label [[FOR_END_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH]] 84; CHECK: vec.epilog.scalar.ph: 85; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC4]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK]] ] 86; CHECK-NEXT: br label [[FOR_BODY:%.*]] 87; CHECK: for.body: 88; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] 89; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[BB]], i64 [[INDVARS_IV]] 90; CHECK-NEXT: [[TMP24:%.*]] = load float, float* [[ARRAYIDX]], align 4 91; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[CC]], i64 [[INDVARS_IV]] 92; CHECK-NEXT: [[TMP25:%.*]] = load float, float* [[ARRAYIDX2]], align 4 93; CHECK-NEXT: [[ADD:%.*]] = fadd fast float [[TMP24]], [[TMP25]] 94; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds float, float* [[AA]], i64 [[INDVARS_IV]] 95; CHECK-NEXT: store float [[ADD]], float* [[ARRAYIDX4]], align 4 96; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 97; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]] 98; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]], !llvm.loop [[LOOP4:![0-9]+]] 99; CHECK: for.end.loopexit: 100; CHECK-NEXT: br label [[FOR_END]] 101; CHECK: for.end: 102; CHECK-NEXT: ret void 103; 104; CHECK-PROFITABLE-BY-DEFAULT-LABEL: @f1( 105; CHECK-PROFITABLE-BY-DEFAULT-NEXT: entry: 106; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[CMP1:%.*]] = icmp sgt i32 [[N:%.*]], 0 107; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[CMP1]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]] 108; CHECK-PROFITABLE-BY-DEFAULT: for.body.preheader: 109; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i64 110; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br label [[FOR_BODY:%.*]] 111; CHECK-PROFITABLE-BY-DEFAULT: for.body: 112; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] 113; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[BB:%.*]], i64 [[INDVARS_IV]] 114; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP0:%.*]] = load float, float* [[ARRAYIDX]], align 4 115; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[CC:%.*]], i64 [[INDVARS_IV]] 116; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP1:%.*]] = load float, float* [[ARRAYIDX2]], align 4 117; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[ADD:%.*]] = fadd fast float [[TMP0]], [[TMP1]] 118; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds float, float* [[AA:%.*]], i64 [[INDVARS_IV]] 119; CHECK-PROFITABLE-BY-DEFAULT-NEXT: store float [[ADD]], float* [[ARRAYIDX4]], align 4 120; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 121; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]] 122; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT:%.*]] 123; CHECK-PROFITABLE-BY-DEFAULT: for.end.loopexit: 124; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br label [[FOR_END]] 125; CHECK-PROFITABLE-BY-DEFAULT: for.end: 126; CHECK-PROFITABLE-BY-DEFAULT-NEXT: ret void 127; 128entry: 129 %cmp1 = icmp sgt i32 %N, 0 130 br i1 %cmp1, label %for.body.preheader, label %for.end 131 132for.body.preheader: ; preds = %entry 133 %wide.trip.count = zext i32 %N to i64 134 br label %for.body 135 136for.body: ; preds = %for.body.preheader, %for.body 137 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ] 138 %arrayidx = getelementptr inbounds float, float* %bb, i64 %indvars.iv 139 %0 = load float, float* %arrayidx, align 4 140 %arrayidx2 = getelementptr inbounds float, float* %cc, i64 %indvars.iv 141 %1 = load float, float* %arrayidx2, align 4 142 %add = fadd fast float %0, %1 143 %arrayidx4 = getelementptr inbounds float, float* %aa, i64 %indvars.iv 144 store float %add, float* %arrayidx4, align 4 145 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 146 %exitcond = icmp ne i64 %indvars.iv.next, %wide.trip.count 147 br i1 %exitcond, label %for.body, label %for.end.loopexit 148 149for.end.loopexit: ; preds = %for.body 150 br label %for.end 151 152for.end: ; preds = %for.end.loopexit, %entry 153 ret void 154} 155 156define dso_local signext i32 @f2(float* noalias %A, float* noalias %B, i32 signext %n) { 157; CHECK-LABEL: @f2( 158; CHECK-NEXT: entry: 159; CHECK-NEXT: [[CMP1:%.*]] = icmp sgt i32 [[N:%.*]], 1 160; CHECK-NEXT: br i1 [[CMP1]], label [[ITER_CHECK:%.*]], label [[FOR_END:%.*]] 161; CHECK: iter.check: 162; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[N]], -1 163; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[TMP0]] to i64 164; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 4 165; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]] 166; CHECK: vector.scevcheck: 167; CHECK-NEXT: [[TMP1:%.*]] = add nsw i64 [[WIDE_TRIP_COUNT]], -1 168; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[TMP1]] to i32 169; CHECK-NEXT: [[MUL:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 1, i32 [[TMP2]]) 170; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL]], 0 171; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL]], 1 172; CHECK-NEXT: [[TMP3:%.*]] = sub i32 [[TMP0]], [[MUL_RESULT]] 173; CHECK-NEXT: [[TMP4:%.*]] = icmp sgt i32 [[TMP3]], [[TMP0]] 174; CHECK-NEXT: [[TMP5:%.*]] = or i1 [[TMP4]], [[MUL_OVERFLOW]] 175; CHECK-NEXT: [[TMP6:%.*]] = icmp ugt i64 [[TMP1]], 4294967295 176; CHECK-NEXT: [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]] 177; CHECK-NEXT: br i1 [[TMP7]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]] 178; CHECK: vector.main.loop.iter.check: 179; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 4 180; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]] 181; CHECK: vector.ph: 182; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 4 183; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]] 184; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 185; CHECK: vector.body: 186; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 187; CHECK-NEXT: [[OFFSET_IDX:%.*]] = trunc i64 [[INDEX]] to i32 188; CHECK-NEXT: [[TMP8:%.*]] = add i32 [[OFFSET_IDX]], 0 189; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[INDEX]], 0 190; CHECK-NEXT: [[TMP10:%.*]] = xor i32 [[TMP8]], -1 191; CHECK-NEXT: [[TMP11:%.*]] = add i32 [[TMP10]], [[N]] 192; CHECK-NEXT: [[TMP12:%.*]] = sext i32 [[TMP11]] to i64 193; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 [[TMP12]] 194; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds float, float* [[TMP13]], i32 0 195; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds float, float* [[TMP14]], i32 -3 196; CHECK-NEXT: [[TMP16:%.*]] = bitcast float* [[TMP15]] to <4 x float>* 197; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, <4 x float>* [[TMP16]], align 4 198; CHECK-NEXT: [[REVERSE:%.*]] = shufflevector <4 x float> [[WIDE_LOAD]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 199; CHECK-NEXT: [[TMP17:%.*]] = fadd fast <4 x float> [[REVERSE]], <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00> 200; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[TMP9]] 201; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds float, float* [[TMP18]], i32 0 202; CHECK-NEXT: [[TMP20:%.*]] = bitcast float* [[TMP19]] to <4 x float>* 203; CHECK-NEXT: store <4 x float> [[TMP17]], <4 x float>* [[TMP20]], align 4 204; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 205; CHECK-NEXT: [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 206; CHECK-NEXT: br i1 [[TMP21]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] 207; CHECK: middle.block: 208; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]] 209; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] 210; CHECK: vec.epilog.iter.check: 211; CHECK-NEXT: [[IND_END5:%.*]] = trunc i64 [[N_VEC]] to i32 212; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_VEC]] 213; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 4 214; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] 215; CHECK: vec.epilog.ph: 216; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] 217; CHECK-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 4 218; CHECK-NEXT: [[N_VEC3:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF2]] 219; CHECK-NEXT: [[IND_END:%.*]] = trunc i64 [[N_VEC3]] to i32 220; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] 221; CHECK: vec.epilog.vector.body: 222; CHECK-NEXT: [[OFFSET_IDX9:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT12:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 223; CHECK-NEXT: [[OFFSET_IDX8:%.*]] = trunc i64 [[OFFSET_IDX9]] to i32 224; CHECK-NEXT: [[TMP22:%.*]] = add i32 [[OFFSET_IDX8]], 0 225; CHECK-NEXT: [[TMP23:%.*]] = add i64 [[OFFSET_IDX9]], 0 226; CHECK-NEXT: [[TMP24:%.*]] = xor i32 [[TMP22]], -1 227; CHECK-NEXT: [[TMP25:%.*]] = add i32 [[TMP24]], [[N]] 228; CHECK-NEXT: [[TMP26:%.*]] = sext i32 [[TMP25]] to i64 229; CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP26]] 230; CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds float, float* [[TMP27]], i32 0 231; CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, float* [[TMP28]], i32 -3 232; CHECK-NEXT: [[TMP30:%.*]] = bitcast float* [[TMP29]] to <4 x float>* 233; CHECK-NEXT: [[WIDE_LOAD10:%.*]] = load <4 x float>, <4 x float>* [[TMP30]], align 4 234; CHECK-NEXT: [[REVERSE11:%.*]] = shufflevector <4 x float> [[WIDE_LOAD10]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 235; CHECK-NEXT: [[TMP31:%.*]] = fadd fast <4 x float> [[REVERSE11]], <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00> 236; CHECK-NEXT: [[TMP32:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP23]] 237; CHECK-NEXT: [[TMP33:%.*]] = getelementptr inbounds float, float* [[TMP32]], i32 0 238; CHECK-NEXT: [[TMP34:%.*]] = bitcast float* [[TMP33]] to <4 x float>* 239; CHECK-NEXT: store <4 x float> [[TMP31]], <4 x float>* [[TMP34]], align 4 240; CHECK-NEXT: [[INDEX_NEXT12]] = add nuw i64 [[OFFSET_IDX9]], 4 241; CHECK-NEXT: [[TMP35:%.*]] = icmp eq i64 [[INDEX_NEXT12]], [[N_VEC3]] 242; CHECK-NEXT: br i1 [[TMP35]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] 243; CHECK: vec.epilog.middle.block: 244; CHECK-NEXT: [[CMP_N6:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC3]] 245; CHECK-NEXT: br i1 [[CMP_N6]], label [[FOR_END_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH]] 246; CHECK: vec.epilog.scalar.ph: 247; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC3]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ITER_CHECK]] ] 248; CHECK-NEXT: [[BC_RESUME_VAL4:%.*]] = phi i32 [ [[IND_END]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END5]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ITER_CHECK]] ] 249; CHECK-NEXT: br label [[FOR_BODY:%.*]] 250; CHECK: for.body: 251; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] 252; CHECK-NEXT: [[I_014:%.*]] = phi i32 [ [[BC_RESUME_VAL4]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ] 253; CHECK-NEXT: [[TMP36:%.*]] = xor i32 [[I_014]], -1 254; CHECK-NEXT: [[SUB2:%.*]] = add i32 [[TMP36]], [[N]] 255; CHECK-NEXT: [[IDXPROM:%.*]] = sext i32 [[SUB2]] to i64 256; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[IDXPROM]] 257; CHECK-NEXT: [[TMP37:%.*]] = load float, float* [[ARRAYIDX]], align 4 258; CHECK-NEXT: [[CONV3:%.*]] = fadd fast float [[TMP37]], 1.000000e+00 259; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV]] 260; CHECK-NEXT: store float [[CONV3]], float* [[ARRAYIDX5]], align 4 261; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 262; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[I_014]], 1 263; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]] 264; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]], !llvm.loop [[LOOP7:![0-9]+]] 265; CHECK: for.end.loopexit: 266; CHECK-NEXT: br label [[FOR_END]] 267; CHECK: for.end: 268; CHECK-NEXT: ret i32 0 269; 270; CHECK-PROFITABLE-BY-DEFAULT-LABEL: @f2( 271; CHECK-PROFITABLE-BY-DEFAULT-NEXT: entry: 272; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[CMP1:%.*]] = icmp sgt i32 [[N:%.*]], 1 273; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[CMP1]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]] 274; CHECK-PROFITABLE-BY-DEFAULT: for.body.preheader: 275; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP0:%.*]] = add i32 [[N]], -1 276; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[TMP0]] to i64 277; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br label [[FOR_BODY:%.*]] 278; CHECK-PROFITABLE-BY-DEFAULT: for.body: 279; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] 280; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[I_014:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INC:%.*]], [[FOR_BODY]] ] 281; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP1:%.*]] = xor i32 [[I_014]], -1 282; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[SUB2:%.*]] = add i32 [[TMP1]], [[N]] 283; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[IDXPROM:%.*]] = sext i32 [[SUB2]] to i64 284; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 [[IDXPROM]] 285; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP2:%.*]] = load float, float* [[ARRAYIDX]], align 4 286; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[CONV3:%.*]] = fadd fast float [[TMP2]], 1.000000e+00 287; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[INDVARS_IV]] 288; CHECK-PROFITABLE-BY-DEFAULT-NEXT: store float [[CONV3]], float* [[ARRAYIDX5]], align 4 289; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 290; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INC]] = add nuw nsw i32 [[I_014]], 1 291; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]] 292; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT:%.*]] 293; CHECK-PROFITABLE-BY-DEFAULT: for.end.loopexit: 294; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br label [[FOR_END]] 295; CHECK-PROFITABLE-BY-DEFAULT: for.end: 296; CHECK-PROFITABLE-BY-DEFAULT-NEXT: ret i32 0 297; 298entry: 299 %cmp1 = icmp sgt i32 %n, 1 300 br i1 %cmp1, label %for.body.preheader, label %for.end 301 302for.body.preheader: ; preds = %entry 303 %0 = add i32 %n, -1 304 %wide.trip.count = zext i32 %0 to i64 305 br label %for.body 306 307for.body: ; preds = %for.body.preheader, %for.body 308 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ] 309 %i.014 = phi i32 [ 0, %for.body.preheader ], [ %inc, %for.body ] 310 %1 = xor i32 %i.014, -1 311 %sub2 = add i32 %1, %n 312 %idxprom = sext i32 %sub2 to i64 313 %arrayidx = getelementptr inbounds float, float* %B, i64 %idxprom 314 %2 = load float, float* %arrayidx, align 4 315 %conv3 = fadd fast float %2, 1.000000e+00 316 %arrayidx5 = getelementptr inbounds float, float* %A, i64 %indvars.iv 317 store float %conv3, float* %arrayidx5, align 4 318 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 319 %inc = add nuw nsw i32 %i.014, 1 320 %exitcond = icmp ne i64 %indvars.iv.next, %wide.trip.count 321 br i1 %exitcond, label %for.body, label %for.end.loopexit 322 323for.end.loopexit: ; preds = %for.body 324 br label %for.end 325 326for.end: ; preds = %for.end.loopexit, %entry 327 ret i32 0 328} 329 330define void @f3(i8* noalias %A, i64 %n) { 331; CHECK-LABEL: @f3( 332; CHECK-NEXT: iter.check: 333; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4 334; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]] 335; CHECK: vector.main.loop.iter.check: 336; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[N]], 4 337; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]] 338; CHECK: vector.ph: 339; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 340; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 341; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 342; CHECK: vector.body: 343; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 344; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 345; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, i8* [[A:%.*]], i64 [[TMP0]] 346; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, i8* [[TMP1]], i32 0 347; CHECK-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to <4 x i8>* 348; CHECK-NEXT: store <4 x i8> <i8 1, i8 1, i8 1, i8 1>, <4 x i8>* [[TMP3]], align 1 349; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 350; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 351; CHECK-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] 352; CHECK: middle.block: 353; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 354; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] 355; CHECK: vec.epilog.iter.check: 356; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[N]], [[N_VEC]] 357; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 4 358; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] 359; CHECK: vec.epilog.ph: 360; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] 361; CHECK-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[N]], 4 362; CHECK-NEXT: [[N_VEC3:%.*]] = sub i64 [[N]], [[N_MOD_VF2]] 363; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] 364; CHECK: vec.epilog.vector.body: 365; CHECK-NEXT: [[OFFSET_IDX:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT6:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 366; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 0 367; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, i8* [[A]], i64 [[TMP5]] 368; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[TMP6]], i32 0 369; CHECK-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to <4 x i8>* 370; CHECK-NEXT: store <4 x i8> <i8 1, i8 1, i8 1, i8 1>, <4 x i8>* [[TMP8]], align 1 371; CHECK-NEXT: [[INDEX_NEXT6]] = add nuw i64 [[OFFSET_IDX]], 4 372; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT6]], [[N_VEC3]] 373; CHECK-NEXT: br i1 [[TMP9]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] 374; CHECK: vec.epilog.middle.block: 375; CHECK-NEXT: [[CMP_N4:%.*]] = icmp eq i64 [[N]], [[N_VEC3]] 376; CHECK-NEXT: br i1 [[CMP_N4]], label [[FOR_END_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH]] 377; CHECK: vec.epilog.scalar.ph: 378; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC3]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ] 379; CHECK-NEXT: br label [[FOR_BODY:%.*]] 380; CHECK: for.body: 381; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 382; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, i8* [[A]], i64 [[IV]] 383; CHECK-NEXT: store i8 1, i8* [[ARRAYIDX]], align 1 384; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 385; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[IV_NEXT]], [[N]] 386; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]], !llvm.loop [[LOOP10:![0-9]+]] 387; CHECK: for.end.loopexit: 388; CHECK-NEXT: br label [[FOR_END:%.*]] 389; CHECK: for.end: 390; CHECK-NEXT: ret void 391; 392; CHECK-PROFITABLE-BY-DEFAULT-LABEL: @f3( 393; CHECK-PROFITABLE-BY-DEFAULT-NEXT: iter.check: 394; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 2 395; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]] 396; CHECK-PROFITABLE-BY-DEFAULT: vector.main.loop.iter.check: 397; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[N]], 4 398; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]] 399; CHECK-PROFITABLE-BY-DEFAULT: vector.ph: 400; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 401; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 402; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br label [[VECTOR_BODY:%.*]] 403; CHECK-PROFITABLE-BY-DEFAULT: vector.body: 404; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 405; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 406; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, i8* [[A:%.*]], i64 [[TMP0]] 407; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, i8* [[TMP1]], i32 0 408; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to <4 x i8>* 409; CHECK-PROFITABLE-BY-DEFAULT-NEXT: store <4 x i8> <i8 1, i8 1, i8 1, i8 1>, <4 x i8>* [[TMP3]], align 1 410; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 411; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 412; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 413; CHECK-PROFITABLE-BY-DEFAULT: middle.block: 414; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 415; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] 416; CHECK-PROFITABLE-BY-DEFAULT: vec.epilog.iter.check: 417; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[N]], [[N_VEC]] 418; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 2 419; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] 420; CHECK-PROFITABLE-BY-DEFAULT: vec.epilog.ph: 421; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] 422; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[N]], 2 423; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[N_VEC3:%.*]] = sub i64 [[N]], [[N_MOD_VF2]] 424; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] 425; CHECK-PROFITABLE-BY-DEFAULT: vec.epilog.vector.body: 426; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[OFFSET_IDX:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT6:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 427; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 0 428; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, i8* [[A]], i64 [[TMP5]] 429; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[TMP6]], i32 0 430; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to <2 x i8>* 431; CHECK-PROFITABLE-BY-DEFAULT-NEXT: store <2 x i8> <i8 1, i8 1>, <2 x i8>* [[TMP8]], align 1 432; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDEX_NEXT6]] = add nuw i64 [[OFFSET_IDX]], 2 433; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT6]], [[N_VEC3]] 434; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[TMP9]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]] 435; CHECK-PROFITABLE-BY-DEFAULT: vec.epilog.middle.block: 436; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[CMP_N4:%.*]] = icmp eq i64 [[N]], [[N_VEC3]] 437; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[CMP_N4]], label [[FOR_END_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH]] 438; CHECK-PROFITABLE-BY-DEFAULT: vec.epilog.scalar.ph: 439; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC3]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ] 440; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br label [[FOR_BODY:%.*]] 441; CHECK-PROFITABLE-BY-DEFAULT: for.body: 442; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 443; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, i8* [[A]], i64 [[IV]] 444; CHECK-PROFITABLE-BY-DEFAULT-NEXT: store i8 1, i8* [[ARRAYIDX]], align 1 445; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 446; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[IV_NEXT]], [[N]] 447; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]], !llvm.loop [[LOOP4:![0-9]+]] 448; CHECK-PROFITABLE-BY-DEFAULT: for.end.loopexit: 449; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br label [[FOR_END:%.*]] 450; CHECK-PROFITABLE-BY-DEFAULT: for.end: 451; CHECK-PROFITABLE-BY-DEFAULT-NEXT: ret void 452; 453entry: 454 br label %for.body 455 456for.body: ; preds = %for.body.preheader, %for.body 457 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 458 %arrayidx = getelementptr inbounds i8, i8* %A, i64 %iv 459 store i8 1, i8* %arrayidx, align 1 460 %iv.next = add nuw nsw i64 %iv, 1 461 %exitcond = icmp ne i64 %iv.next, %n 462 br i1 %exitcond, label %for.body, label %for.end.loopexit 463 464for.end.loopexit: ; preds = %for.body 465 br label %for.end 466 467for.end: ; preds = %for.end.loopexit, %entry 468 ret void 469} 470 471; Test case for PR54745. 472define void @induction_resume_value_requires_non_trivial_scev_expansion(i8* %dst) { 473; CHECK-LABEL: @induction_resume_value_requires_non_trivial_scev_expansion( 474; CHECK-NEXT: entry: 475; CHECK-NEXT: br label [[OUTER_HEADER:%.*]] 476; CHECK: outer.header: 477; CHECK-NEXT: [[INDUCTION_IV:%.*]] = phi i8 [ [[INDUCTION_IV_NEXT:%.*]], [[OUTER_LATCH:%.*]] ], [ -56, [[ENTRY:%.*]] ] 478; CHECK-NEXT: [[INDVAR:%.*]] = phi i8 [ [[INDVAR_NEXT:%.*]], [[OUTER_LATCH]] ], [ 0, [[ENTRY]] ] 479; CHECK-NEXT: [[OUTER_IV:%.*]] = phi i64 [ 2, [[ENTRY]] ], [ [[OUTER_IV_NEXT:%.*]], [[OUTER_LATCH]] ] 480; CHECK-NEXT: [[P2:%.*]] = phi i32 [ -202, [[ENTRY]] ], [ [[ADD:%.*]], [[OUTER_LATCH]] ] 481; CHECK-NEXT: [[TMP0:%.*]] = mul i8 [[INDVAR]], -1 482; CHECK-NEXT: [[TMP1:%.*]] = add i8 [[TMP0]], -3 483; CHECK-NEXT: [[TRUNC_IV:%.*]] = trunc i64 [[OUTER_IV]] to i32 484; CHECK-NEXT: [[ADD]] = add i32 [[P2]], [[TRUNC_IV]] 485; CHECK-NEXT: [[TRUNC_ADD:%.*]] = trunc i32 [[ADD]] to i8 486; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 487; CHECK: vector.ph: 488; CHECK-NEXT: [[IND_END:%.*]] = mul i8 84, [[INDUCTION_IV]] 489; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x i8> poison, i8 [[INDUCTION_IV]], i32 0 490; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x i8> [[DOTSPLATINSERT]], <4 x i8> poison, <4 x i32> zeroinitializer 491; CHECK-NEXT: [[TMP2:%.*]] = mul <4 x i8> <i8 0, i8 1, i8 2, i8 3>, [[DOTSPLAT]] 492; CHECK-NEXT: [[INDUCTION:%.*]] = add <4 x i8> zeroinitializer, [[TMP2]] 493; CHECK-NEXT: [[TMP3:%.*]] = mul i8 [[INDUCTION_IV]], 4 494; CHECK-NEXT: [[DOTSPLATINSERT2:%.*]] = insertelement <4 x i8> poison, i8 [[TMP3]], i32 0 495; CHECK-NEXT: [[DOTSPLAT3:%.*]] = shufflevector <4 x i8> [[DOTSPLATINSERT2]], <4 x i8> poison, <4 x i32> zeroinitializer 496; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 497; CHECK: vector.body: 498; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 499; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i8> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] 500; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i64 1, [[INDEX]] 501; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 0 502; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, i8* [[DST:%.*]], i64 [[TMP4]] 503; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, i8* [[TMP5]], i32 0 504; CHECK-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to <4 x i8>* 505; CHECK-NEXT: store <4 x i8> [[VEC_IND]], <4 x i8>* [[TMP7]], align 1 506; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 507; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i8> [[VEC_IND]], [[DOTSPLAT3]] 508; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 84 509; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]] 510; CHECK: middle.block: 511; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 84, 84 512; CHECK-NEXT: br i1 [[CMP_N]], label [[OUTER_LATCH]], label [[SCALAR_PH]] 513; CHECK: scalar.ph: 514; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 85, [[MIDDLE_BLOCK]] ], [ 1, [[OUTER_HEADER]] ] 515; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[OUTER_HEADER]] ] 516; CHECK-NEXT: br label [[INNER:%.*]] 517; CHECK: inner: 518; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[INNER]] ] 519; CHECK-NEXT: [[IV_2:%.*]] = phi i8 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], [[INNER]] ] 520; CHECK-NEXT: [[IV_2_NEXT]] = sub i8 [[IV_2]], [[TRUNC_ADD]] 521; CHECK-NEXT: [[GEP_DST:%.*]] = getelementptr inbounds i8, i8* [[DST]], i64 [[IV]] 522; CHECK-NEXT: store i8 [[IV_2]], i8* [[GEP_DST]], align 1 523; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 524; CHECK-NEXT: [[EC:%.*]] = icmp ugt i64 [[IV]], 83 525; CHECK-NEXT: br i1 [[EC]], label [[OUTER_LATCH]], label [[INNER]], !llvm.loop [[LOOP12:![0-9]+]] 526; CHECK: outer.latch: 527; CHECK-NEXT: [[OUTER_IV_NEXT]] = add nuw nsw i64 [[OUTER_IV]], 1 528; CHECK-NEXT: [[INDVAR_NEXT]] = add i8 [[INDVAR]], 1 529; CHECK-NEXT: [[INDUCTION_IV_NEXT]] = add i8 [[INDUCTION_IV]], [[TMP1]] 530; CHECK-NEXT: br label [[OUTER_HEADER]] 531; 532; CHECK-PROFITABLE-BY-DEFAULT-LABEL: @induction_resume_value_requires_non_trivial_scev_expansion( 533; CHECK-PROFITABLE-BY-DEFAULT-NEXT: entry: 534; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br label [[OUTER_HEADER:%.*]] 535; CHECK-PROFITABLE-BY-DEFAULT: outer.header: 536; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDUCTION_IV:%.*]] = phi i8 [ [[INDUCTION_IV_NEXT:%.*]], [[OUTER_LATCH:%.*]] ], [ -56, [[ENTRY:%.*]] ] 537; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDVAR:%.*]] = phi i8 [ [[INDVAR_NEXT:%.*]], [[OUTER_LATCH]] ], [ 0, [[ENTRY]] ] 538; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[OUTER_IV:%.*]] = phi i64 [ 2, [[ENTRY]] ], [ [[OUTER_IV_NEXT:%.*]], [[OUTER_LATCH]] ] 539; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[P2:%.*]] = phi i32 [ -202, [[ENTRY]] ], [ [[ADD:%.*]], [[OUTER_LATCH]] ] 540; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP0:%.*]] = mul i8 [[INDVAR]], -1 541; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP1:%.*]] = add i8 [[TMP0]], -3 542; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TRUNC_IV:%.*]] = trunc i64 [[OUTER_IV]] to i32 543; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[ADD]] = add i32 [[P2]], [[TRUNC_IV]] 544; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TRUNC_ADD:%.*]] = trunc i32 [[ADD]] to i8 545; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 546; CHECK-PROFITABLE-BY-DEFAULT: vector.ph: 547; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[IND_END:%.*]] = mul i8 84, [[INDUCTION_IV]] 548; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x i8> poison, i8 [[INDUCTION_IV]], i32 0 549; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x i8> [[DOTSPLATINSERT]], <4 x i8> poison, <4 x i32> zeroinitializer 550; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP2:%.*]] = mul <4 x i8> <i8 0, i8 1, i8 2, i8 3>, [[DOTSPLAT]] 551; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDUCTION:%.*]] = add <4 x i8> zeroinitializer, [[TMP2]] 552; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP3:%.*]] = mul i8 [[INDUCTION_IV]], 4 553; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[DOTSPLATINSERT2:%.*]] = insertelement <4 x i8> poison, i8 [[TMP3]], i32 0 554; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[DOTSPLAT3:%.*]] = shufflevector <4 x i8> [[DOTSPLATINSERT2]], <4 x i8> poison, <4 x i32> zeroinitializer 555; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br label [[VECTOR_BODY:%.*]] 556; CHECK-PROFITABLE-BY-DEFAULT: vector.body: 557; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 558; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[VEC_IND:%.*]] = phi <4 x i8> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] 559; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[OFFSET_IDX:%.*]] = add i64 1, [[INDEX]] 560; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 0 561; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, i8* [[DST:%.*]], i64 [[TMP4]] 562; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, i8* [[TMP5]], i32 0 563; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to <4 x i8>* 564; CHECK-PROFITABLE-BY-DEFAULT-NEXT: store <4 x i8> [[VEC_IND]], <4 x i8>* [[TMP7]], align 1 565; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 566; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[VEC_IND_NEXT]] = add <4 x i8> [[VEC_IND]], [[DOTSPLAT3]] 567; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 84 568; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] 569; CHECK-PROFITABLE-BY-DEFAULT: middle.block: 570; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[CMP_N:%.*]] = icmp eq i64 84, 84 571; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[CMP_N]], label [[OUTER_LATCH]], label [[SCALAR_PH]] 572; CHECK-PROFITABLE-BY-DEFAULT: scalar.ph: 573; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 85, [[MIDDLE_BLOCK]] ], [ 1, [[OUTER_HEADER]] ] 574; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[OUTER_HEADER]] ] 575; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br label [[INNER:%.*]] 576; CHECK-PROFITABLE-BY-DEFAULT: inner: 577; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[INNER]] ] 578; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[IV_2:%.*]] = phi i8 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], [[INNER]] ] 579; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[IV_2_NEXT]] = sub i8 [[IV_2]], [[TRUNC_ADD]] 580; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[GEP_DST:%.*]] = getelementptr inbounds i8, i8* [[DST]], i64 [[IV]] 581; CHECK-PROFITABLE-BY-DEFAULT-NEXT: store i8 [[IV_2]], i8* [[GEP_DST]], align 1 582; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 583; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[EC:%.*]] = icmp ugt i64 [[IV]], 83 584; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br i1 [[EC]], label [[OUTER_LATCH]], label [[INNER]], !llvm.loop [[LOOP6:![0-9]+]] 585; CHECK-PROFITABLE-BY-DEFAULT: outer.latch: 586; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[OUTER_IV_NEXT]] = add nuw nsw i64 [[OUTER_IV]], 1 587; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDVAR_NEXT]] = add i8 [[INDVAR]], 1 588; CHECK-PROFITABLE-BY-DEFAULT-NEXT: [[INDUCTION_IV_NEXT]] = add i8 [[INDUCTION_IV]], [[TMP1]] 589; CHECK-PROFITABLE-BY-DEFAULT-NEXT: br label [[OUTER_HEADER]] 590; 591entry: 592 br label %outer.header 593 594outer.header: 595 %outer.iv = phi i64 [ 2, %entry ], [ %outer.iv.next, %outer.latch ] 596 %p2 = phi i32 [ -202, %entry ], [ %add, %outer.latch ] 597 %trunc.iv = trunc i64 %outer.iv to i32 598 %add = add i32 %p2, %trunc.iv 599 %trunc.add = trunc i32 %add to i8 600 br label %inner 601 602inner: 603 %iv = phi i64 [ 1, %outer.header ], [ %iv.next, %inner ] 604 %iv.2 = phi i8 [ 0, %outer.header ], [ %iv.2.next, %inner ] 605 %iv.2.next = sub i8 %iv.2, %trunc.add 606 %gep.dst = getelementptr inbounds i8, i8* %dst, i64 %iv 607 store i8 %iv.2, i8* %gep.dst 608 %iv.next = add nuw nsw i64 %iv, 1 609 %ec = icmp ugt i64 %iv, 83 610 br i1 %ec, label %outer.latch, label %inner 611 612outer.latch: 613 %outer.iv.next = add nuw nsw i64 %outer.iv, 1 614 br label %outer.header 615} 616