165 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=loop-unroll -unroll-runtime=true -unroll-runtime-epilog=true -unroll-runtime-other-exit-predictable=true -verify-loop-lcssa -verify-dom-info -verify-loop-info -S | FileCheck %s --check-prefix=ENABLED3; RUN: opt < %s -passes=loop-unroll -unroll-runtime=true -unroll-runtime-epilog=true -unroll-runtime-other-exit-predictable=false -verify-loop-lcssa -verify-dom-info -verify-loop-info -S | FileCheck %s --check-prefix=DISABLED4 5define i32 @test(ptr nocapture %a, i64 %n) {6; ENABLED-LABEL: @test(7; ENABLED-NEXT: entry:8; ENABLED-NEXT: [[TMP0:%.*]] = freeze i64 [[N:%.*]]9; ENABLED-NEXT: [[TMP1:%.*]] = add i64 [[TMP0]], -110; ENABLED-NEXT: [[XTRAITER:%.*]] = and i64 [[TMP0]], 711; ENABLED-NEXT: [[TMP2:%.*]] = icmp ult i64 [[TMP1]], 712; ENABLED-NEXT: br i1 [[TMP2]], label [[HEADER_EPIL_PREHEADER:%.*]], label [[ENTRY_NEW:%.*]]13; ENABLED: entry.new:14; ENABLED-NEXT: [[UNROLL_ITER:%.*]] = sub i64 [[TMP0]], [[XTRAITER]]15; ENABLED-NEXT: br label [[HEADER:%.*]]16; ENABLED: header:17; ENABLED-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[INDVARS_IV_NEXT_7:%.*]], [[FOR_BODY_7:%.*]] ]18; ENABLED-NEXT: [[SUM_02:%.*]] = phi i32 [ 0, [[ENTRY_NEW]] ], [ [[ADD_7:%.*]], [[FOR_BODY_7]] ]19; ENABLED-NEXT: [[NITER:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[NITER_NEXT_7:%.*]], [[FOR_BODY_7]] ]20; ENABLED-NEXT: [[CMP:%.*]] = icmp eq i64 [[N]], 4221; ENABLED-NEXT: br i1 [[CMP]], label [[FOR_EXIT2_LOOPEXIT:%.*]], label [[FOR_BODY:%.*]]22; ENABLED: for.body:23; ENABLED-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]24; ENABLED-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 425; ENABLED-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP3]], [[SUM_02]]26; ENABLED-NEXT: [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 127; ENABLED-NEXT: br i1 false, label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_1:%.*]]28; ENABLED: for.body.1:29; ENABLED-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT]]30; ENABLED-NEXT: [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 431; ENABLED-NEXT: [[ADD_1:%.*]] = add nsw i32 [[TMP4]], [[ADD]]32; ENABLED-NEXT: [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 233; ENABLED-NEXT: [[CMP_2:%.*]] = icmp eq i64 [[N]], 4234; ENABLED-NEXT: br i1 [[CMP_2]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_2:%.*]]35; ENABLED: for.body.2:36; ENABLED-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_1]]37; ENABLED-NEXT: [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX_2]], align 438; ENABLED-NEXT: [[ADD_2:%.*]] = add nsw i32 [[TMP5]], [[ADD_1]]39; ENABLED-NEXT: [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 340; ENABLED-NEXT: br i1 false, label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_3:%.*]]41; ENABLED: for.body.3:42; ENABLED-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_2]]43; ENABLED-NEXT: [[TMP6:%.*]] = load i32, ptr [[ARRAYIDX_3]], align 444; ENABLED-NEXT: [[ADD_3:%.*]] = add nsw i32 [[TMP6]], [[ADD_2]]45; ENABLED-NEXT: [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 446; ENABLED-NEXT: [[CMP_4:%.*]] = icmp eq i64 [[N]], 4247; ENABLED-NEXT: br i1 [[CMP_4]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_4:%.*]]48; ENABLED: for.body.4:49; ENABLED-NEXT: [[ARRAYIDX_4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_3]]50; ENABLED-NEXT: [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX_4]], align 451; ENABLED-NEXT: [[ADD_4:%.*]] = add nsw i32 [[TMP7]], [[ADD_3]]52; ENABLED-NEXT: [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 553; ENABLED-NEXT: br i1 false, label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_5:%.*]]54; ENABLED: for.body.5:55; ENABLED-NEXT: [[ARRAYIDX_5:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_4]]56; ENABLED-NEXT: [[TMP8:%.*]] = load i32, ptr [[ARRAYIDX_5]], align 457; ENABLED-NEXT: [[ADD_5:%.*]] = add nsw i32 [[TMP8]], [[ADD_4]]58; ENABLED-NEXT: [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 659; ENABLED-NEXT: [[CMP_6:%.*]] = icmp eq i64 [[N]], 4260; ENABLED-NEXT: br i1 [[CMP_6]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_6:%.*]]61; ENABLED: for.body.6:62; ENABLED-NEXT: [[ARRAYIDX_6:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_5]]63; ENABLED-NEXT: [[TMP9:%.*]] = load i32, ptr [[ARRAYIDX_6]], align 464; ENABLED-NEXT: [[ADD_6:%.*]] = add nsw i32 [[TMP9]], [[ADD_5]]65; ENABLED-NEXT: [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 766; ENABLED-NEXT: br i1 false, label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_7]]67; ENABLED: for.body.7:68; ENABLED-NEXT: [[ARRAYIDX_7:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_6]]69; ENABLED-NEXT: [[TMP10:%.*]] = load i32, ptr [[ARRAYIDX_7]], align 470; ENABLED-NEXT: [[ADD_7]] = add nsw i32 [[TMP10]], [[ADD_6]]71; ENABLED-NEXT: [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV]], 872; ENABLED-NEXT: [[NITER_NEXT_7]] = add i64 [[NITER]], 873; ENABLED-NEXT: [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NEXT_7]], [[UNROLL_ITER]]74; ENABLED-NEXT: br i1 [[NITER_NCMP_7]], label [[FOR_END_UNR_LCSSA:%.*]], label [[HEADER]]75; ENABLED: for.end.unr-lcssa:76; ENABLED-NEXT: [[SUM_0_LCSSA_PH_PH:%.*]] = phi i32 [ [[ADD_7]], [[FOR_BODY_7]] ]77; ENABLED-NEXT: [[INDVARS_IV_UNR_PH:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_7]], [[FOR_BODY_7]] ]78; ENABLED-NEXT: [[SUM_02_UNR_PH:%.*]] = phi i32 [ [[ADD_7]], [[FOR_BODY_7]] ]79; ENABLED-NEXT: [[LCMP_MOD:%.*]] = icmp ne i64 [[XTRAITER]], 080; ENABLED-NEXT: br i1 [[LCMP_MOD]], label [[HEADER_EPIL_PREHEADER]], label [[FOR_END:%.*]]81; ENABLED: header.epil.preheader:82; ENABLED-NEXT: [[INDVARS_IV_EPIL_INIT:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_UNR_PH]], [[FOR_END_UNR_LCSSA]] ]83; ENABLED-NEXT: [[SUM_02_EPIL_INIT:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[SUM_02_UNR_PH]], [[FOR_END_UNR_LCSSA]] ]84; ENABLED-NEXT: [[LCMP_MOD2:%.*]] = icmp ne i64 [[XTRAITER]], 085; ENABLED-NEXT: call void @llvm.assume(i1 [[LCMP_MOD2]])86; ENABLED-NEXT: br label [[HEADER_EPIL:%.*]]87; ENABLED: header.epil:88; ENABLED-NEXT: [[INDVARS_IV_EPIL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL:%.*]] ], [ [[INDVARS_IV_EPIL_INIT]], [[HEADER_EPIL_PREHEADER]] ]89; ENABLED-NEXT: [[SUM_02_EPIL:%.*]] = phi i32 [ [[ADD_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[SUM_02_EPIL_INIT]], [[HEADER_EPIL_PREHEADER]] ]90; ENABLED-NEXT: [[EPIL_ITER:%.*]] = phi i64 [ 0, [[HEADER_EPIL_PREHEADER]] ], [ [[EPIL_ITER_NEXT:%.*]], [[FOR_BODY_EPIL]] ]91; ENABLED-NEXT: [[CMP_EPIL:%.*]] = icmp eq i64 [[N]], 4292; ENABLED-NEXT: br i1 [[CMP_EPIL]], label [[FOR_EXIT2_LOOPEXIT2:%.*]], label [[FOR_BODY_EPIL]]93; ENABLED: for.body.epil:94; ENABLED-NEXT: [[ARRAYIDX_EPIL:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_EPIL]]95; ENABLED-NEXT: [[TMP11:%.*]] = load i32, ptr [[ARRAYIDX_EPIL]], align 496; ENABLED-NEXT: [[ADD_EPIL]] = add nsw i32 [[TMP11]], [[SUM_02_EPIL]]97; ENABLED-NEXT: [[INDVARS_IV_NEXT_EPIL]] = add i64 [[INDVARS_IV_EPIL]], 198; ENABLED-NEXT: [[EXITCOND_EPIL:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT_EPIL]], [[N]]99; ENABLED-NEXT: [[EPIL_ITER_NEXT]] = add i64 [[EPIL_ITER]], 1100; ENABLED-NEXT: [[EPIL_ITER_CMP:%.*]] = icmp ne i64 [[EPIL_ITER_NEXT]], [[XTRAITER]]101; ENABLED-NEXT: br i1 [[EPIL_ITER_CMP]], label [[HEADER_EPIL]], label [[FOR_END_EPILOG_LCSSA:%.*]], !llvm.loop [[LOOP0:![0-9]+]]102; ENABLED: for.end.epilog-lcssa:103; ENABLED-NEXT: [[SUM_0_LCSSA_PH1:%.*]] = phi i32 [ [[ADD_EPIL]], [[FOR_BODY_EPIL]] ]104; ENABLED-NEXT: br label [[FOR_END]]105; ENABLED: for.end:106; ENABLED-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ [[SUM_0_LCSSA_PH_PH]], [[FOR_END_UNR_LCSSA]] ], [ [[SUM_0_LCSSA_PH1]], [[FOR_END_EPILOG_LCSSA]] ]107; ENABLED-NEXT: ret i32 [[SUM_0_LCSSA]]108; ENABLED: for.exit2.loopexit:109; ENABLED-NEXT: [[RETVAL_PH:%.*]] = phi i32 [ [[SUM_02]], [[HEADER]] ], [ [[ADD]], [[FOR_BODY]] ], [ [[ADD_1]], [[FOR_BODY_1]] ], [ [[ADD_2]], [[FOR_BODY_2]] ], [ [[ADD_3]], [[FOR_BODY_3]] ], [ [[ADD_4]], [[FOR_BODY_4]] ], [ [[ADD_5]], [[FOR_BODY_5]] ], [ [[ADD_6]], [[FOR_BODY_6]] ]110; ENABLED-NEXT: br label [[FOR_EXIT2:%.*]]111; ENABLED: for.exit2.loopexit3:112; ENABLED-NEXT: [[RETVAL_PH3:%.*]] = phi i32 [ [[SUM_02_EPIL]], [[HEADER_EPIL]] ]113; ENABLED-NEXT: br label [[FOR_EXIT2]]114; ENABLED: for.exit2:115; ENABLED-NEXT: [[RETVAL:%.*]] = phi i32 [ [[RETVAL_PH]], [[FOR_EXIT2_LOOPEXIT]] ], [ [[RETVAL_PH3]], [[FOR_EXIT2_LOOPEXIT2]] ]116; ENABLED-NEXT: ret i32 [[RETVAL]]117;118; DISABLED-LABEL: @test(119; DISABLED-NEXT: entry:120; DISABLED-NEXT: br label [[HEADER:%.*]]121; DISABLED: header:122; DISABLED-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY:%.*]] ], [ 0, [[ENTRY:%.*]] ]123; DISABLED-NEXT: [[SUM_02:%.*]] = phi i32 [ [[ADD:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY]] ]124; DISABLED-NEXT: [[CMP:%.*]] = icmp eq i64 [[N:%.*]], 42125; DISABLED-NEXT: br i1 [[CMP]], label [[FOR_EXIT2:%.*]], label [[FOR_BODY]]126; DISABLED: for.body:127; DISABLED-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]128; DISABLED-NEXT: [[TMP0:%.*]] = load i32, ptr [[ARRAYIDX]], align 4129; DISABLED-NEXT: [[ADD]] = add nsw i32 [[TMP0]], [[SUM_02]]130; DISABLED-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1131; DISABLED-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[N]]132; DISABLED-NEXT: br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[HEADER]]133; DISABLED: for.end:134; DISABLED-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ [[ADD]], [[FOR_BODY]] ]135; DISABLED-NEXT: ret i32 [[SUM_0_LCSSA]]136; DISABLED: for.exit2:137; DISABLED-NEXT: [[RETVAL:%.*]] = phi i32 [ [[SUM_02]], [[HEADER]] ]138; DISABLED-NEXT: ret i32 [[RETVAL]]139;140entry:141 br label %header142 143header:144 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]145 %sum.02 = phi i32 [ %add, %for.body ], [ 0, %entry ]146 %cmp = icmp eq i64 %n, 42147 br i1 %cmp, label %for.exit2, label %for.body148 149for.body:150 %arrayidx = getelementptr inbounds i32, ptr %a, i64 %indvars.iv151 %0 = load i32, ptr %arrayidx, align 4152 %add = add nsw i32 %0, %sum.02153 %indvars.iv.next = add i64 %indvars.iv, 1154 %exitcond = icmp eq i64 %indvars.iv.next, %n155 br i1 %exitcond, label %for.end, label %header156 157for.end:158 %sum.0.lcssa = phi i32 [ %add, %for.body ]159 ret i32 %sum.0.lcssa160 161for.exit2:162 %retval = phi i32 [ %sum.02, %header ]163 ret i32 %retval164}165