233 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 52; RUN: opt -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -S < %s | FileCheck %s --check-prefix=CHECK3 4define i64 @select_icmp_nuw_nsw(ptr %a, ptr %b, i64 %ii, i64 %n) {5; CHECK-LABEL: define i64 @select_icmp_nuw_nsw(6; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]], i64 [[II:%.*]], i64 [[N:%.*]]) {7; CHECK-NEXT: [[ENTRY:.*]]:8; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 49; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]10; CHECK: [[VECTOR_PH]]:11; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 412; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]13; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]14; CHECK: [[VECTOR_BODY]]:15; CHECK-NEXT: [[TMP9:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]16; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]17; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]18; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP9]]19; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP10]], align 820; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[TMP9]]21; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <4 x i64>, ptr [[TMP3]], align 822; CHECK-NEXT: [[TMP5:%.*]] = icmp sgt <4 x i64> [[WIDE_LOAD]], [[WIDE_LOAD1]]23; CHECK-NEXT: [[TMP6]] = select <4 x i1> [[TMP5]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]24; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[TMP9]], 425; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[VEC_IND]], splat (i64 4)26; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]27; CHECK-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]28; CHECK: [[MIDDLE_BLOCK]]:29; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vector.reduce.smax.v4i64(<4 x i64> [[TMP6]])30; CHECK-NEXT: [[RDX_SELECT_CMP:%.*]] = icmp ne i64 [[TMP8]], -922337203685477580831; CHECK-NEXT: [[RDX_SELECT:%.*]] = select i1 [[RDX_SELECT_CMP]], i64 [[TMP8]], i64 [[II]]32; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]33; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]34; CHECK: [[SCALAR_PH]]:35; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]36; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[RDX_SELECT]], %[[MIDDLE_BLOCK]] ], [ [[II]], %[[ENTRY]] ]37; CHECK-NEXT: br label %[[FOR_BODY:.*]]38; CHECK: [[FOR_BODY]]:39; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[INC:%.*]], %[[FOR_BODY]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]40; CHECK-NEXT: [[RDX:%.*]] = phi i64 [ [[COND:%.*]], %[[FOR_BODY]] ], [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ]41; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]]42; CHECK-NEXT: [[TMP0:%.*]] = load i64, ptr [[ARRAYIDX]], align 843; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[IV]]44; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr [[ARRAYIDX1]], align 845; CHECK-NEXT: [[CMP2:%.*]] = icmp sgt i64 [[TMP0]], [[TMP1]]46; CHECK-NEXT: [[COND]] = select i1 [[CMP2]], i64 [[IV]], i64 [[RDX]]47; CHECK-NEXT: [[INC]] = add nuw nsw i64 [[IV]], 148; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[N]]49; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[EXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]50; CHECK: [[EXIT]]:51; CHECK-NEXT: [[COND_LCSSA:%.*]] = phi i64 [ [[COND]], %[[FOR_BODY]] ], [ [[RDX_SELECT]], %[[MIDDLE_BLOCK]] ]52; CHECK-NEXT: ret i64 [[COND_LCSSA]]53;54entry:55 br label %for.body56 57for.body: ; preds = %entry, %for.body58 %iv = phi i64 [ %inc, %for.body ], [ 0, %entry ]59 %rdx = phi i64 [ %cond, %for.body ], [ %ii, %entry ]60 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv61 %0 = load i64, ptr %arrayidx, align 862 %arrayidx1 = getelementptr inbounds i64, ptr %b, i64 %iv63 %1 = load i64, ptr %arrayidx1, align 864 %cmp2 = icmp sgt i64 %0, %165 %cond = select i1 %cmp2, i64 %iv, i64 %rdx66 %inc = add nuw nsw i64 %iv, 167 %exitcond.not = icmp eq i64 %inc, %n68 br i1 %exitcond.not, label %exit, label %for.body69 70exit: ; preds = %for.body71 ret i64 %cond72}73 74define i64 @select_icmp_nsw(ptr %a, ptr %b, i64 %ii, i64 %n) {75; CHECK-LABEL: define i64 @select_icmp_nsw(76; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]], i64 [[II:%.*]], i64 [[N:%.*]]) {77; CHECK-NEXT: [[ENTRY:.*]]:78; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 479; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]80; CHECK: [[VECTOR_PH]]:81; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 482; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]83; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]84; CHECK: [[VECTOR_BODY]]:85; CHECK-NEXT: [[TMP9:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]86; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]87; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ splat (i64 -9223372036854775808), %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]88; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP9]]89; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP10]], align 890; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[TMP9]]91; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <4 x i64>, ptr [[TMP3]], align 892; CHECK-NEXT: [[TMP5:%.*]] = icmp sgt <4 x i64> [[WIDE_LOAD]], [[WIDE_LOAD1]]93; CHECK-NEXT: [[TMP6]] = select <4 x i1> [[TMP5]], <4 x i64> [[VEC_IND]], <4 x i64> [[VEC_PHI]]94; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[TMP9]], 495; CHECK-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)96; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]97; CHECK-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]98; CHECK: [[MIDDLE_BLOCK]]:99; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vector.reduce.smax.v4i64(<4 x i64> [[TMP6]])100; CHECK-NEXT: [[RDX_SELECT_CMP:%.*]] = icmp ne i64 [[TMP8]], -9223372036854775808101; CHECK-NEXT: [[RDX_SELECT:%.*]] = select i1 [[RDX_SELECT_CMP]], i64 [[TMP8]], i64 [[II]]102; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]103; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]104; CHECK: [[SCALAR_PH]]:105; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]106; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[RDX_SELECT]], %[[MIDDLE_BLOCK]] ], [ [[II]], %[[ENTRY]] ]107; CHECK-NEXT: br label %[[FOR_BODY:.*]]108; CHECK: [[FOR_BODY]]:109; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[INC:%.*]], %[[FOR_BODY]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]110; CHECK-NEXT: [[RDX:%.*]] = phi i64 [ [[COND:%.*]], %[[FOR_BODY]] ], [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ]111; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]]112; CHECK-NEXT: [[TMP0:%.*]] = load i64, ptr [[ARRAYIDX]], align 8113; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[IV]]114; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr [[ARRAYIDX1]], align 8115; CHECK-NEXT: [[CMP2:%.*]] = icmp sgt i64 [[TMP0]], [[TMP1]]116; CHECK-NEXT: [[COND]] = select i1 [[CMP2]], i64 [[IV]], i64 [[RDX]]117; CHECK-NEXT: [[INC]] = add nsw i64 [[IV]], 1118; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[N]]119; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[EXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]120; CHECK: [[EXIT]]:121; CHECK-NEXT: [[COND_LCSSA:%.*]] = phi i64 [ [[COND]], %[[FOR_BODY]] ], [ [[RDX_SELECT]], %[[MIDDLE_BLOCK]] ]122; CHECK-NEXT: ret i64 [[COND_LCSSA]]123;124entry:125 br label %for.body126 127for.body: ; preds = %entry, %for.body128 %iv = phi i64 [ %inc, %for.body ], [ 0, %entry ]129 %rdx = phi i64 [ %cond, %for.body ], [ %ii, %entry ]130 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv131 %0 = load i64, ptr %arrayidx, align 8132 %arrayidx1 = getelementptr inbounds i64, ptr %b, i64 %iv133 %1 = load i64, ptr %arrayidx1, align 8134 %cmp2 = icmp sgt i64 %0, %1135 %cond = select i1 %cmp2, i64 %iv, i64 %rdx136 %inc = add nsw i64 %iv, 1137 %exitcond.not = icmp eq i64 %inc, %n138 br i1 %exitcond.not, label %exit, label %for.body139 140exit: ; preds = %for.body141 ret i64 %cond142}143 144define i64 @select_icmp_nuw(ptr %a, ptr %b, i64 %ii, i64 %n) {145; CHECK-LABEL: define i64 @select_icmp_nuw(146; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]], i64 [[II:%.*]], i64 [[N:%.*]]) {147; CHECK-NEXT: [[ENTRY:.*]]:148; CHECK-NEXT: br label %[[FOR_BODY:.*]]149; CHECK: [[FOR_BODY]]:150; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 0, %[[ENTRY]] ]151; CHECK-NEXT: [[RDX:%.*]] = phi i64 [ [[COND:%.*]], %[[FOR_BODY]] ], [ [[II]], %[[ENTRY]] ]152; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]]153; CHECK-NEXT: [[TMP0:%.*]] = load i64, ptr [[ARRAYIDX]], align 8154; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[IV]]155; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr [[ARRAYIDX1]], align 8156; CHECK-NEXT: [[CMP2:%.*]] = icmp sgt i64 [[TMP0]], [[TMP1]]157; CHECK-NEXT: [[COND]] = select i1 [[CMP2]], i64 [[IV]], i64 [[RDX]]158; CHECK-NEXT: [[INC]] = add nuw i64 [[IV]], 1159; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[N]]160; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[EXIT:.*]], label %[[FOR_BODY]]161; CHECK: [[EXIT]]:162; CHECK-NEXT: [[COND_LCSSA:%.*]] = phi i64 [ [[COND]], %[[FOR_BODY]] ]163; CHECK-NEXT: ret i64 [[COND_LCSSA]]164;165entry:166 br label %for.body167 168for.body: ; preds = %entry, %for.body169 %iv = phi i64 [ %inc, %for.body ], [ 0, %entry ]170 %rdx = phi i64 [ %cond, %for.body ], [ %ii, %entry ]171 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv172 %0 = load i64, ptr %arrayidx, align 8173 %arrayidx1 = getelementptr inbounds i64, ptr %b, i64 %iv174 %1 = load i64, ptr %arrayidx1, align 8175 %cmp2 = icmp sgt i64 %0, %1176 %cond = select i1 %cmp2, i64 %iv, i64 %rdx177 %inc = add nuw i64 %iv, 1178 %exitcond.not = icmp eq i64 %inc, %n179 br i1 %exitcond.not, label %exit, label %for.body180 181exit: ; preds = %for.body182 ret i64 %cond183}184 185define i64 @select_icmp_noflag(ptr %a, ptr %b, i64 %ii, i64 %n) {186; CHECK-LABEL: define i64 @select_icmp_noflag(187; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]], i64 [[II:%.*]], i64 [[N:%.*]]) {188; CHECK-NEXT: [[ENTRY:.*]]:189; CHECK-NEXT: br label %[[FOR_BODY:.*]]190; CHECK: [[FOR_BODY]]:191; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 0, %[[ENTRY]] ]192; CHECK-NEXT: [[RDX:%.*]] = phi i64 [ [[COND:%.*]], %[[FOR_BODY]] ], [ [[II]], %[[ENTRY]] ]193; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]]194; CHECK-NEXT: [[TMP0:%.*]] = load i64, ptr [[ARRAYIDX]], align 8195; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[IV]]196; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr [[ARRAYIDX1]], align 8197; CHECK-NEXT: [[CMP2:%.*]] = icmp sgt i64 [[TMP0]], [[TMP1]]198; CHECK-NEXT: [[COND]] = select i1 [[CMP2]], i64 [[IV]], i64 [[RDX]]199; CHECK-NEXT: [[INC]] = add i64 [[IV]], 1200; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[N]]201; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[EXIT:.*]], label %[[FOR_BODY]]202; CHECK: [[EXIT]]:203; CHECK-NEXT: [[COND_LCSSA:%.*]] = phi i64 [ [[COND]], %[[FOR_BODY]] ]204; CHECK-NEXT: ret i64 [[COND_LCSSA]]205;206entry:207 br label %for.body208 209for.body: ; preds = %entry, %for.body210 %iv = phi i64 [ %inc, %for.body ], [ 0, %entry ]211 %rdx = phi i64 [ %cond, %for.body ], [ %ii, %entry ]212 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv213 %0 = load i64, ptr %arrayidx, align 8214 %arrayidx1 = getelementptr inbounds i64, ptr %b, i64 %iv215 %1 = load i64, ptr %arrayidx1, align 8216 %cmp2 = icmp sgt i64 %0, %1217 %cond = select i1 %cmp2, i64 %iv, i64 %rdx218 %inc = add i64 %iv, 1219 %exitcond.not = icmp eq i64 %inc, %n220 br i1 %exitcond.not, label %exit, label %for.body221 222exit: ; preds = %for.body223 ret i64 %cond224}225;.226; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}227; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}228; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}229; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}230; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}231; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META2]], [[META1]]}232;.233