1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5 2; REQUIRES: asserts 3; RUN: opt < %s -passes='function(loop-vectorize,instcombine)' -force-vector-width=2 -force-vector-interleave=1 -debug-only=loop-vectorize -disable-output -print-after=instcombine 2>&1 | FileCheck %s 4 5target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128" 6 7; CHECK-LABEL: phi_two_incoming_values 8; CHECK: LV: Found an estimated cost of 1 for VF 2 For instruction: %i = phi i64 [ %i.next, %if.end ], [ 0, %entry ] 9; CHECK: LV: Found an estimated cost of 1 for VF 2 For instruction: %tmp5 = phi i32 [ %tmp1, %for.body ], [ %tmp4, %if.then ] 10; 11define void @phi_two_incoming_values(ptr noalias %a, ptr noalias %b, i64 %n) { 12; CHECK-LABEL: define void @phi_two_incoming_values( 13; CHECK-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) { 14; CHECK-NEXT: [[ENTRY:.*]]: 15; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N]], 1 16; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 2 17; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 18; CHECK: [[VECTOR_PH]]: 19; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[TMP0]], -2 20; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] 21; CHECK: [[VECTOR_BODY]]: 22; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 23; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDEX]] 24; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP2]], align 4 25; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDEX]] 26; CHECK-NEXT: [[TMP4:%.*]] = icmp sgt <2 x i32> [[WIDE_LOAD]], zeroinitializer 27; CHECK-NEXT: [[TMP5:%.*]] = zext <2 x i1> [[TMP4]] to <2 x i32> 28; CHECK-NEXT: [[PREDPHI:%.*]] = add <2 x i32> [[WIDE_LOAD]], [[TMP5]] 29; CHECK-NEXT: store <2 x i32> [[PREDPHI]], ptr [[TMP3]], align 4 30; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 31; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 32; CHECK-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 33; CHECK: [[MIDDLE_BLOCK]]: 34; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]] 35; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_END:.*]], label %[[SCALAR_PH]] 36; CHECK: [[SCALAR_PH]]: 37; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 38; CHECK-NEXT: br label %[[FOR_BODY:.*]] 39; CHECK: [[FOR_BODY]]: 40; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], %[[IF_END:.*]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ] 41; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[I]] 42; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP0]], align 4 43; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[I]] 44; CHECK-NEXT: [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], 0 45; CHECK-NEXT: br i1 [[TMP4]], label %[[IF_THEN:.*]], label %[[IF_END]] 46; CHECK: [[IF_THEN]]: 47; CHECK-NEXT: [[TMP5:%.*]] = add nuw i32 [[TMP2]], 1 48; CHECK-NEXT: br label %[[IF_END]] 49; CHECK: [[IF_END]]: 50; CHECK-NEXT: [[TMP6:%.*]] = phi i32 [ [[TMP2]], %[[FOR_BODY]] ], [ [[TMP5]], %[[IF_THEN]] ] 51; CHECK-NEXT: store i32 [[TMP6]], ptr [[TMP3]], align 4 52; CHECK-NEXT: [[I_NEXT]] = add i64 [[I]], 1 53; CHECK-NEXT: [[COND:%.*]] = icmp eq i64 [[I]], [[N]] 54; CHECK-NEXT: br i1 [[COND]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 55; CHECK: [[FOR_END]]: 56; CHECK-NEXT: ret void 57; 58entry: 59 br label %for.body 60 61for.body: 62 %i = phi i64 [ %i.next, %if.end ], [ 0, %entry ] 63 %tmp0 = getelementptr inbounds i32, ptr %a, i64 %i 64 %tmp1 = load i32, ptr %tmp0, align 4 65 %tmp2 = getelementptr inbounds i32, ptr %b, i64 %i 66 %tmp3 = icmp sgt i32 %tmp1, 0 67 br i1 %tmp3, label %if.then, label %if.end 68 69if.then: 70 %tmp4 = add i32 %tmp1, 1 71 br label %if.end 72 73if.end: 74 %tmp5 = phi i32 [ %tmp1, %for.body ], [ %tmp4, %if.then ] 75 store i32 %tmp5, ptr %tmp2, align 4 76 %i.next = add i64 %i, 1 77 %cond = icmp eq i64 %i, %n 78 br i1 %cond, label %for.end, label %for.body 79 80for.end: 81 ret void 82} 83 84; CHECK-LABEL: phi_three_incoming_values 85; CHECK: LV: Found an estimated cost of 1 for VF 2 For instruction: %i = phi i64 [ %i.next, %if.end ], [ 0, %entry ] 86; CHECK: LV: Found an estimated cost of 2 for VF 2 For instruction: %tmp8 = phi i32 [ 9, %for.body ], [ 3, %if.then ], [ %tmp7, %if.else ] 87; 88define void @phi_three_incoming_values(ptr noalias %a, ptr noalias %b, i64 %n) { 89; CHECK-LABEL: define void @phi_three_incoming_values( 90; CHECK-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) { 91; CHECK-NEXT: [[ENTRY:.*]]: 92; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N]], 1 93; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 2 94; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 95; CHECK: [[VECTOR_PH]]: 96; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[TMP0]], -2 97; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] 98; CHECK: [[VECTOR_BODY]]: 99; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] 100; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDEX]] 101; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP3]], align 4 102; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDEX]] 103; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <2 x i32>, ptr [[TMP4]], align 4 104; CHECK-NEXT: [[TMP5:%.*]] = icmp sgt <2 x i32> [[WIDE_LOAD]], [[WIDE_LOAD2]] 105; CHECK-NEXT: [[TMP6:%.*]] = icmp slt <2 x i32> [[WIDE_LOAD]], splat (i32 20) 106; CHECK-NEXT: [[TMP7:%.*]] = icmp slt <2 x i32> [[WIDE_LOAD2]], splat (i32 4) 107; CHECK-NEXT: [[TMP8:%.*]] = select <2 x i1> [[TMP7]], <2 x i32> splat (i32 4), <2 x i32> splat (i32 5) 108; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP6]], <2 x i32> [[TMP8]], <2 x i32> splat (i32 3) 109; CHECK-NEXT: [[PREDPHI3:%.*]] = select <2 x i1> [[TMP5]], <2 x i32> [[PREDPHI]], <2 x i32> splat (i32 9) 110; CHECK-NEXT: store <2 x i32> [[PREDPHI3]], ptr [[TMP3]], align 4 111; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 112; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 113; CHECK-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] 114; CHECK: [[MIDDLE_BLOCK]]: 115; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]] 116; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_END:.*]], label %[[SCALAR_PH]] 117; CHECK: [[SCALAR_PH]]: 118; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ] 119; CHECK-NEXT: br label %[[FOR_BODY:.*]] 120; CHECK: [[FOR_BODY]]: 121; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], %[[IF_END:.*]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ] 122; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[I]] 123; CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr [[TMP0]], align 4 124; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[I]] 125; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[TMP4]], align 4 126; CHECK-NEXT: [[TMP6:%.*]] = icmp sgt i32 [[TMP3]], [[TMP5]] 127; CHECK-NEXT: br i1 [[TMP6]], label %[[IF_THEN:.*]], label %[[IF_END]] 128; CHECK: [[IF_THEN]]: 129; CHECK-NEXT: [[TMP7:%.*]] = icmp sgt i32 [[TMP3]], 19 130; CHECK-NEXT: br i1 [[TMP7]], label %[[IF_END]], label %[[IF_ELSE:.*]] 131; CHECK: [[IF_ELSE]]: 132; CHECK-NEXT: [[TMP8:%.*]] = icmp slt i32 [[TMP5]], 4 133; CHECK-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 4, i32 5 134; CHECK-NEXT: br label %[[IF_END]] 135; CHECK: [[IF_END]]: 136; CHECK-NEXT: [[TMP10:%.*]] = phi i32 [ 9, %[[FOR_BODY]] ], [ 3, %[[IF_THEN]] ], [ [[TMP9]], %[[IF_ELSE]] ] 137; CHECK-NEXT: store i32 [[TMP10]], ptr [[TMP0]], align 4 138; CHECK-NEXT: [[I_NEXT]] = add i64 [[I]], 1 139; CHECK-NEXT: [[COND:%.*]] = icmp eq i64 [[I]], [[N]] 140; CHECK-NEXT: br i1 [[COND]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] 141; CHECK: [[FOR_END]]: 142; CHECK-NEXT: ret void 143; 144entry: 145 br label %for.body 146 147for.body: 148 %i = phi i64 [ %i.next, %if.end ], [ 0, %entry ] 149 %tmp0 = getelementptr inbounds i32, ptr %a, i64 %i 150 %tmp1 = load i32, ptr %tmp0, align 4 151 %tmp2 = getelementptr inbounds i32, ptr %b, i64 %i 152 %tmp3 = load i32, ptr %tmp2, align 4 153 %tmp4 = icmp sgt i32 %tmp1, %tmp3 154 br i1 %tmp4, label %if.then, label %if.end 155 156if.then: 157 %tmp5 = icmp sgt i32 %tmp1, 19 158 br i1 %tmp5, label %if.end, label %if.else 159 160if.else: 161 %tmp6 = icmp slt i32 %tmp3, 4 162 %tmp7 = select i1 %tmp6, i32 4, i32 5 163 br label %if.end 164 165if.end: 166 %tmp8 = phi i32 [ 9, %for.body ], [ 3, %if.then ], [ %tmp7, %if.else ] 167 store i32 %tmp8, ptr %tmp0, align 4 168 %i.next = add i64 %i, 1 169 %cond = icmp eq i64 %i, %n 170 br i1 %cond, label %for.end, label %for.body 171 172for.end: 173 ret void 174} 175 176; Test case for https://github.com/llvm/llvm-project/issues/113794. 177define i32 @red_phi_0(i32 %start, ptr %src) { 178; CHECK-LABEL: define i32 @red_phi_0( 179; CHECK-SAME: i32 [[START:%.*]], ptr [[SRC:%.*]]) { 180; CHECK-NEXT: [[ENTRY:.*:]] 181; CHECK-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] 182; CHECK: [[VECTOR_PH]]: 183; CHECK-NEXT: [[TMP0:%.*]] = insertelement <2 x i32> <i32 poison, i32 0>, i32 [[START]], i64 0 184; 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: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 188; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100 189; CHECK-NEXT: br i1 [[TMP1]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] 190; CHECK: [[MIDDLE_BLOCK]]: 191; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[TMP0]]) 192; CHECK-NEXT: br i1 true, label %[[EXIT:.*]], label %[[SCALAR_PH]] 193; CHECK: [[SCALAR_PH]]: 194; CHECK-NEXT: br label %[[LOOP:.*]] 195; CHECK: [[LOOP]]: 196; CHECK-NEXT: br i1 poison, label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP7:![0-9]+]] 197; CHECK: [[EXIT]]: 198; CHECK-NEXT: [[RES:%.*]] = phi i32 [ poison, %[[LOOP]] ], [ [[TMP2]], %[[MIDDLE_BLOCK]] ] 199; CHECK-NEXT: ret i32 [[RES]] 200; 201entry: 202 br label %loop 203 204loop: 205 %red = phi i32 [ %start, %entry ], [ %red.next, %loop ] 206 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] 207 %red.next = add i32 0, %red 208 %iv.next = add i64 %iv, 1 209 %exitcond.not = icmp eq i64 %iv.next, 100 210 br i1 %exitcond.not, label %exit, label %loop 211 212exit: 213 %res = phi i32 [ %red.next, %loop ] 214 ret i32 %res 215} 216