1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -S -passes=loop-unroll -unroll-runtime=true -unroll-count=16 | FileCheck --check-prefix=UNROLL-16 %s 3; RUN: opt < %s -S -passes=loop-unroll -unroll-runtime=true -unroll-count=4 | FileCheck --check-prefix=UNROLL-4 %s 4 5; RUN: opt < %s -S -passes='require<opt-remark-emit>,loop-unroll' -unroll-runtime=true -unroll-count=16 | FileCheck --check-prefix=UNROLL-16 %s 6; RUN: opt < %s -S -passes='require<opt-remark-emit>,loop-unroll' -unroll-runtime=true -unroll-count=4 | FileCheck --check-prefix=UNROLL-4 %s 7 8; Given that the trip-count of this loop is a 3-bit value, we cannot 9; safely unroll it with a count of anything more than 8. 10 11define i3 @test(ptr %a, i3 %n) { 12; UNROLL-16-LABEL: @test( 13; UNROLL-16-NEXT: entry: 14; UNROLL-16-NEXT: [[CMP1:%.*]] = icmp eq i3 [[N:%.*]], 0 15; UNROLL-16-NEXT: br i1 [[CMP1]], label [[FOR_END:%.*]], label [[FOR_BODY_PREHEADER:%.*]] 16; UNROLL-16: for.body.preheader: 17; UNROLL-16-NEXT: br label [[FOR_BODY:%.*]] 18; UNROLL-16: for.body: 19; UNROLL-16-NEXT: [[TMP0:%.*]] = load i3, ptr [[A:%.*]], align 1 20; UNROLL-16-NEXT: [[EXITCOND:%.*]] = icmp eq i3 1, [[N]] 21; UNROLL-16-NEXT: br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT:%.*]], label [[FOR_BODY_1:%.*]] 22; UNROLL-16: for.body.1: 23; UNROLL-16-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i3, ptr [[A]], i64 1 24; UNROLL-16-NEXT: [[TMP1:%.*]] = load i3, ptr [[ARRAYIDX_1]], align 1 25; UNROLL-16-NEXT: [[ADD_1:%.*]] = add nsw i3 [[TMP1]], [[TMP0]] 26; UNROLL-16-NEXT: [[EXITCOND_1:%.*]] = icmp eq i3 2, [[N]] 27; UNROLL-16-NEXT: br i1 [[EXITCOND_1]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY_2:%.*]] 28; UNROLL-16: for.body.2: 29; UNROLL-16-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i3, ptr [[A]], i64 2 30; UNROLL-16-NEXT: [[TMP2:%.*]] = load i3, ptr [[ARRAYIDX_2]], align 1 31; UNROLL-16-NEXT: [[ADD_2:%.*]] = add nsw i3 [[TMP2]], [[ADD_1]] 32; UNROLL-16-NEXT: [[EXITCOND_2:%.*]] = icmp eq i3 3, [[N]] 33; UNROLL-16-NEXT: br i1 [[EXITCOND_2]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY_3:%.*]] 34; UNROLL-16: for.body.3: 35; UNROLL-16-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i3, ptr [[A]], i64 3 36; UNROLL-16-NEXT: [[TMP3:%.*]] = load i3, ptr [[ARRAYIDX_3]], align 1 37; UNROLL-16-NEXT: [[ADD_3:%.*]] = add nsw i3 [[TMP3]], [[ADD_2]] 38; UNROLL-16-NEXT: [[EXITCOND_3:%.*]] = icmp eq i3 -4, [[N]] 39; UNROLL-16-NEXT: br i1 [[EXITCOND_3]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY_4:%.*]] 40; UNROLL-16: for.body.4: 41; UNROLL-16-NEXT: [[ARRAYIDX_4:%.*]] = getelementptr inbounds i3, ptr [[A]], i64 4 42; UNROLL-16-NEXT: [[TMP4:%.*]] = load i3, ptr [[ARRAYIDX_4]], align 1 43; UNROLL-16-NEXT: [[ADD_4:%.*]] = add nsw i3 [[TMP4]], [[ADD_3]] 44; UNROLL-16-NEXT: [[EXITCOND_4:%.*]] = icmp eq i3 -3, [[N]] 45; UNROLL-16-NEXT: br i1 [[EXITCOND_4]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY_5:%.*]] 46; UNROLL-16: for.body.5: 47; UNROLL-16-NEXT: [[ARRAYIDX_5:%.*]] = getelementptr inbounds i3, ptr [[A]], i64 5 48; UNROLL-16-NEXT: [[TMP5:%.*]] = load i3, ptr [[ARRAYIDX_5]], align 1 49; UNROLL-16-NEXT: [[ADD_5:%.*]] = add nsw i3 [[TMP5]], [[ADD_4]] 50; UNROLL-16-NEXT: [[EXITCOND_5:%.*]] = icmp eq i3 -2, [[N]] 51; UNROLL-16-NEXT: br i1 [[EXITCOND_5]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY_6:%.*]] 52; UNROLL-16: for.body.6: 53; UNROLL-16-NEXT: [[ARRAYIDX_6:%.*]] = getelementptr inbounds i3, ptr [[A]], i64 6 54; UNROLL-16-NEXT: [[TMP6:%.*]] = load i3, ptr [[ARRAYIDX_6]], align 1 55; UNROLL-16-NEXT: [[ADD_6:%.*]] = add nsw i3 [[TMP6]], [[ADD_5]] 56; UNROLL-16-NEXT: br label [[FOR_END_LOOPEXIT]] 57; UNROLL-16: for.end.loopexit: 58; UNROLL-16-NEXT: [[ADD_LCSSA:%.*]] = phi i3 [ [[TMP0]], [[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]] ] 59; UNROLL-16-NEXT: br label [[FOR_END]] 60; UNROLL-16: for.end: 61; UNROLL-16-NEXT: [[SUM_0_LCSSA:%.*]] = phi i3 [ 0, [[ENTRY:%.*]] ], [ [[ADD_LCSSA]], [[FOR_END_LOOPEXIT]] ] 62; UNROLL-16-NEXT: ret i3 [[SUM_0_LCSSA]] 63; 64; UNROLL-4-LABEL: @test( 65; UNROLL-4-NEXT: entry: 66; UNROLL-4-NEXT: [[CMP1:%.*]] = icmp eq i3 [[N:%.*]], 0 67; UNROLL-4-NEXT: br i1 [[CMP1]], label [[FOR_END:%.*]], label [[FOR_BODY_PREHEADER:%.*]] 68; UNROLL-4: for.body.preheader: 69; UNROLL-4-NEXT: [[TMP0:%.*]] = add i3 [[N]], -1 70; UNROLL-4-NEXT: [[XTRAITER:%.*]] = and i3 [[N]], 3 71; UNROLL-4-NEXT: [[TMP1:%.*]] = icmp ult i3 [[TMP0]], 3 72; UNROLL-4-NEXT: br i1 [[TMP1]], label [[FOR_END_LOOPEXIT_UNR_LCSSA:%.*]], label [[FOR_BODY_PREHEADER_NEW:%.*]] 73; UNROLL-4: for.body.preheader.new: 74; UNROLL-4-NEXT: [[UNROLL_ITER:%.*]] = sub i3 [[N]], [[XTRAITER]] 75; UNROLL-4-NEXT: br label [[FOR_BODY:%.*]] 76; UNROLL-4: for.body: 77; UNROLL-4-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[INDVARS_IV_NEXT_3:%.*]], [[FOR_BODY]] ] 78; UNROLL-4-NEXT: [[SUM_02:%.*]] = phi i3 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[ADD_3:%.*]], [[FOR_BODY]] ] 79; UNROLL-4-NEXT: [[NITER:%.*]] = phi i3 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_BODY]] ] 80; UNROLL-4-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i3, ptr [[A:%.*]], i64 [[INDVARS_IV]] 81; UNROLL-4-NEXT: [[TMP2:%.*]] = load i3, ptr [[ARRAYIDX]], align 1 82; UNROLL-4-NEXT: [[ADD:%.*]] = add nsw i3 [[TMP2]], [[SUM_02]] 83; UNROLL-4-NEXT: [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1 84; UNROLL-4-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i3, ptr [[A]], i64 [[INDVARS_IV_NEXT]] 85; UNROLL-4-NEXT: [[TMP3:%.*]] = load i3, ptr [[ARRAYIDX_1]], align 1 86; UNROLL-4-NEXT: [[ADD_1:%.*]] = add nsw i3 [[TMP3]], [[ADD]] 87; UNROLL-4-NEXT: [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 2 88; UNROLL-4-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i3, ptr [[A]], i64 [[INDVARS_IV_NEXT_1]] 89; UNROLL-4-NEXT: [[TMP4:%.*]] = load i3, ptr [[ARRAYIDX_2]], align 1 90; UNROLL-4-NEXT: [[ADD_2:%.*]] = add nsw i3 [[TMP4]], [[ADD_1]] 91; UNROLL-4-NEXT: [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 3 92; UNROLL-4-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i3, ptr [[A]], i64 [[INDVARS_IV_NEXT_2]] 93; UNROLL-4-NEXT: [[TMP5:%.*]] = load i3, ptr [[ARRAYIDX_3]], align 1 94; UNROLL-4-NEXT: [[ADD_3]] = add nsw i3 [[TMP5]], [[ADD_2]] 95; UNROLL-4-NEXT: [[INDVARS_IV_NEXT_3]] = add nuw nsw i64 [[INDVARS_IV]], 4 96; UNROLL-4-NEXT: [[NITER_NEXT_3]] = add i3 [[NITER]], -4 97; UNROLL-4-NEXT: [[NITER_NCMP_3:%.*]] = icmp eq i3 [[NITER_NEXT_3]], [[UNROLL_ITER]] 98; UNROLL-4-NEXT: br i1 [[NITER_NCMP_3]], label [[FOR_END_LOOPEXIT_UNR_LCSSA_LOOPEXIT:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 99; UNROLL-4: for.end.loopexit.unr-lcssa.loopexit: 100; UNROLL-4-NEXT: [[ADD_LCSSA_PH_PH:%.*]] = phi i3 [ [[ADD_3]], [[FOR_BODY]] ] 101; UNROLL-4-NEXT: [[INDVARS_IV_UNR_PH:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_3]], [[FOR_BODY]] ] 102; UNROLL-4-NEXT: [[SUM_02_UNR_PH:%.*]] = phi i3 [ [[ADD_3]], [[FOR_BODY]] ] 103; UNROLL-4-NEXT: br label [[FOR_END_LOOPEXIT_UNR_LCSSA]] 104; UNROLL-4: for.end.loopexit.unr-lcssa: 105; UNROLL-4-NEXT: [[ADD_LCSSA_PH:%.*]] = phi i3 [ poison, [[FOR_BODY_PREHEADER]] ], [ [[ADD_LCSSA_PH_PH]], [[FOR_END_LOOPEXIT_UNR_LCSSA_LOOPEXIT]] ] 106; UNROLL-4-NEXT: [[INDVARS_IV_UNR:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_UNR_PH]], [[FOR_END_LOOPEXIT_UNR_LCSSA_LOOPEXIT]] ] 107; UNROLL-4-NEXT: [[SUM_02_UNR:%.*]] = phi i3 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[SUM_02_UNR_PH]], [[FOR_END_LOOPEXIT_UNR_LCSSA_LOOPEXIT]] ] 108; UNROLL-4-NEXT: [[LCMP_MOD:%.*]] = icmp ne i3 [[XTRAITER]], 0 109; UNROLL-4-NEXT: br i1 [[LCMP_MOD]], label [[FOR_BODY_EPIL_PREHEADER:%.*]], label [[FOR_END_LOOPEXIT:%.*]] 110; UNROLL-4: for.body.epil.preheader: 111; UNROLL-4-NEXT: br label [[FOR_BODY_EPIL:%.*]] 112; UNROLL-4: for.body.epil: 113; UNROLL-4-NEXT: [[INDVARS_IV_EPIL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[INDVARS_IV_UNR]], [[FOR_BODY_EPIL_PREHEADER]] ] 114; UNROLL-4-NEXT: [[SUM_02_EPIL:%.*]] = phi i3 [ [[ADD_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[SUM_02_UNR]], [[FOR_BODY_EPIL_PREHEADER]] ] 115; UNROLL-4-NEXT: [[EPIL_ITER:%.*]] = phi i3 [ 0, [[FOR_BODY_EPIL_PREHEADER]] ], [ [[EPIL_ITER_NEXT:%.*]], [[FOR_BODY_EPIL]] ] 116; UNROLL-4-NEXT: [[ARRAYIDX_EPIL:%.*]] = getelementptr inbounds i3, ptr [[A]], i64 [[INDVARS_IV_EPIL]] 117; UNROLL-4-NEXT: [[TMP6:%.*]] = load i3, ptr [[ARRAYIDX_EPIL]], align 1 118; UNROLL-4-NEXT: [[ADD_EPIL]] = add nsw i3 [[TMP6]], [[SUM_02_EPIL]] 119; UNROLL-4-NEXT: [[INDVARS_IV_NEXT_EPIL]] = add i64 [[INDVARS_IV_EPIL]], 1 120; UNROLL-4-NEXT: [[LFTR_WIDEIV_EPIL:%.*]] = trunc i64 [[INDVARS_IV_NEXT_EPIL]] to i3 121; UNROLL-4-NEXT: [[EXITCOND_EPIL:%.*]] = icmp eq i3 [[LFTR_WIDEIV_EPIL]], [[N]] 122; UNROLL-4-NEXT: [[EPIL_ITER_NEXT]] = add i3 [[EPIL_ITER]], 1 123; UNROLL-4-NEXT: [[EPIL_ITER_CMP:%.*]] = icmp ne i3 [[EPIL_ITER_NEXT]], [[XTRAITER]] 124; UNROLL-4-NEXT: br i1 [[EPIL_ITER_CMP]], label [[FOR_BODY_EPIL]], label [[FOR_END_LOOPEXIT_EPILOG_LCSSA:%.*]], !llvm.loop [[LOOP2:![0-9]+]] 125; UNROLL-4: for.end.loopexit.epilog-lcssa: 126; UNROLL-4-NEXT: [[ADD_LCSSA_PH1:%.*]] = phi i3 [ [[ADD_EPIL]], [[FOR_BODY_EPIL]] ] 127; UNROLL-4-NEXT: br label [[FOR_END_LOOPEXIT]] 128; UNROLL-4: for.end.loopexit: 129; UNROLL-4-NEXT: [[ADD_LCSSA:%.*]] = phi i3 [ [[ADD_LCSSA_PH]], [[FOR_END_LOOPEXIT_UNR_LCSSA]] ], [ [[ADD_LCSSA_PH1]], [[FOR_END_LOOPEXIT_EPILOG_LCSSA]] ] 130; UNROLL-4-NEXT: br label [[FOR_END]] 131; UNROLL-4: for.end: 132; UNROLL-4-NEXT: [[SUM_0_LCSSA:%.*]] = phi i3 [ 0, [[ENTRY:%.*]] ], [ [[ADD_LCSSA]], [[FOR_END_LOOPEXIT]] ] 133; UNROLL-4-NEXT: ret i3 [[SUM_0_LCSSA]] 134; 135entry: 136 %cmp1 = icmp eq i3 %n, 0 137 br i1 %cmp1, label %for.end, label %for.body 138 139for.body: ; preds = %for.body, %entry 140 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ] 141 %sum.02 = phi i3 [ %add, %for.body ], [ 0, %entry ] 142 %arrayidx = getelementptr inbounds i3, ptr %a, i64 %indvars.iv 143 %0 = load i3, ptr %arrayidx 144 %add = add nsw i3 %0, %sum.02 145 %indvars.iv.next = add i64 %indvars.iv, 1 146 %lftr.wideiv = trunc i64 %indvars.iv.next to i3 147 %exitcond = icmp eq i3 %lftr.wideiv, %n 148 br i1 %exitcond, label %for.end, label %for.body 149 150for.end: ; preds = %for.body, %entry 151 %sum.0.lcssa = phi i3 [ 0, %entry ], [ %add, %for.body ] 152 ret i3 %sum.0.lcssa 153} 154