1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -S -loop-unroll -unroll-runtime=true -unroll-count=4 -unroll-remainder -instcombine | FileCheck %s 3 4define i32 @unroll(i32* nocapture readonly %a, i32* nocapture readonly %b, i32 %N) local_unnamed_addr #0 { 5; 6; CHECK-LABEL: @unroll( 7; CHECK-NEXT: entry: 8; CHECK-NEXT: [[CMP9:%.*]] = icmp eq i32 [[N:%.*]], 0 9; CHECK-NEXT: br i1 [[CMP9]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY_LR_PH:%.*]] 10; CHECK: for.body.lr.ph: 11; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i64 12; CHECK-NEXT: [[TMP0:%.*]] = add nsw i64 [[WIDE_TRIP_COUNT]], -1 13; CHECK-NEXT: [[XTRAITER:%.*]] = and i64 [[WIDE_TRIP_COUNT]], 3 14; CHECK-NEXT: [[TMP1:%.*]] = icmp ult i64 [[TMP0]], 3 15; CHECK-NEXT: br i1 [[TMP1]], label [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA:%.*]], label [[FOR_BODY_LR_PH_NEW:%.*]] 16; CHECK: for.body.lr.ph.new: 17; CHECK-NEXT: [[UNROLL_ITER:%.*]] = and i64 [[WIDE_TRIP_COUNT]], 4294967292 18; CHECK-NEXT: br label [[FOR_BODY:%.*]] 19; CHECK: for.cond.cleanup.loopexit.unr-lcssa.loopexit: 20; CHECK-NEXT: br label [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA]] 21; CHECK: for.cond.cleanup.loopexit.unr-lcssa: 22; CHECK-NEXT: [[ADD_LCSSA_PH:%.*]] = phi i32 [ undef, [[FOR_BODY_LR_PH]] ], [ [[ADD_3:%.*]], [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA_LOOPEXIT:%.*]] ] 23; CHECK-NEXT: [[INDVARS_IV_UNR:%.*]] = phi i64 [ 0, [[FOR_BODY_LR_PH]] ], [ [[INDVARS_IV_NEXT_3:%.*]], [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA_LOOPEXIT]] ] 24; CHECK-NEXT: [[C_010_UNR:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[ADD_3]], [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA_LOOPEXIT]] ] 25; CHECK-NEXT: [[LCMP_MOD_NOT:%.*]] = icmp eq i64 [[XTRAITER]], 0 26; CHECK-NEXT: br i1 [[LCMP_MOD_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[FOR_BODY_EPIL_PREHEADER:%.*]] 27; CHECK: for.body.epil.preheader: 28; CHECK-NEXT: br label [[FOR_BODY_EPIL:%.*]] 29; CHECK: for.body.epil: 30; CHECK-NEXT: [[ARRAYIDX_EPIL:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[INDVARS_IV_UNR]] 31; CHECK-NEXT: [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX_EPIL]], align 4 32; CHECK-NEXT: [[ARRAYIDX2_EPIL:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[INDVARS_IV_UNR]] 33; CHECK-NEXT: [[TMP3:%.*]] = load i32, i32* [[ARRAYIDX2_EPIL]], align 4 34; CHECK-NEXT: [[MUL_EPIL:%.*]] = mul nsw i32 [[TMP3]], [[TMP2]] 35; CHECK-NEXT: [[ADD_EPIL:%.*]] = add nsw i32 [[MUL_EPIL]], [[C_010_UNR]] 36; CHECK-NEXT: [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[XTRAITER]], 1 37; CHECK-NEXT: br i1 [[EPIL_ITER_CMP_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT_EPILOG_LCSSA:%.*]], label [[FOR_BODY_EPIL_1:%.*]] 38; CHECK: for.body.epil.1: 39; CHECK-NEXT: [[INDVARS_IV_NEXT_EPIL:%.*]] = add nuw nsw i64 [[INDVARS_IV_UNR]], 1 40; CHECK-NEXT: [[ARRAYIDX_EPIL_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV_NEXT_EPIL]] 41; CHECK-NEXT: [[TMP4:%.*]] = load i32, i32* [[ARRAYIDX_EPIL_1]], align 4 42; CHECK-NEXT: [[ARRAYIDX2_EPIL_1:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDVARS_IV_NEXT_EPIL]] 43; CHECK-NEXT: [[TMP5:%.*]] = load i32, i32* [[ARRAYIDX2_EPIL_1]], align 4 44; CHECK-NEXT: [[MUL_EPIL_1:%.*]] = mul nsw i32 [[TMP5]], [[TMP4]] 45; CHECK-NEXT: [[ADD_EPIL_1:%.*]] = add nsw i32 [[MUL_EPIL_1]], [[ADD_EPIL]] 46; CHECK-NEXT: [[EPIL_ITER_CMP_1_NOT:%.*]] = icmp eq i64 [[XTRAITER]], 2 47; CHECK-NEXT: br i1 [[EPIL_ITER_CMP_1_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT_EPILOG_LCSSA]], label [[FOR_BODY_EPIL_2:%.*]] 48; CHECK: for.body.epil.2: 49; CHECK-NEXT: [[INDVARS_IV_NEXT_EPIL_1:%.*]] = add nuw nsw i64 [[INDVARS_IV_UNR]], 2 50; CHECK-NEXT: [[ARRAYIDX_EPIL_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV_NEXT_EPIL_1]] 51; CHECK-NEXT: [[TMP6:%.*]] = load i32, i32* [[ARRAYIDX_EPIL_2]], align 4 52; CHECK-NEXT: [[ARRAYIDX2_EPIL_2:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDVARS_IV_NEXT_EPIL_1]] 53; CHECK-NEXT: [[TMP7:%.*]] = load i32, i32* [[ARRAYIDX2_EPIL_2]], align 4 54; CHECK-NEXT: [[MUL_EPIL_2:%.*]] = mul nsw i32 [[TMP7]], [[TMP6]] 55; CHECK-NEXT: [[ADD_EPIL_2:%.*]] = add nsw i32 [[MUL_EPIL_2]], [[ADD_EPIL_1]] 56; CHECK-NEXT: br label [[FOR_COND_CLEANUP_LOOPEXIT_EPILOG_LCSSA]] 57; CHECK: for.cond.cleanup.loopexit.epilog-lcssa: 58; CHECK-NEXT: [[ADD_LCSSA_PH1:%.*]] = phi i32 [ [[ADD_EPIL]], [[FOR_BODY_EPIL]] ], [ [[ADD_EPIL_1]], [[FOR_BODY_EPIL_1]] ], [ [[ADD_EPIL_2]], [[FOR_BODY_EPIL_2]] ] 59; CHECK-NEXT: br label [[FOR_COND_CLEANUP_LOOPEXIT]] 60; CHECK: for.cond.cleanup.loopexit: 61; CHECK-NEXT: [[ADD_LCSSA:%.*]] = phi i32 [ [[ADD_LCSSA_PH]], [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA]] ], [ [[ADD_LCSSA_PH1]], [[FOR_COND_CLEANUP_LOOPEXIT_EPILOG_LCSSA]] ] 62; CHECK-NEXT: br label [[FOR_COND_CLEANUP]] 63; CHECK: for.cond.cleanup: 64; CHECK-NEXT: [[C_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[ADD_LCSSA]], [[FOR_COND_CLEANUP_LOOPEXIT]] ] 65; CHECK-NEXT: ret i32 [[C_0_LCSSA]] 66; CHECK: for.body: 67; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_LR_PH_NEW]] ], [ [[INDVARS_IV_NEXT_3]], [[FOR_BODY]] ] 68; CHECK-NEXT: [[C_010:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH_NEW]] ], [ [[ADD_3]], [[FOR_BODY]] ] 69; CHECK-NEXT: [[NITER:%.*]] = phi i64 [ 0, [[FOR_BODY_LR_PH_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_BODY]] ] 70; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV]] 71; CHECK-NEXT: [[TMP8:%.*]] = load i32, i32* [[ARRAYIDX]], align 4 72; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDVARS_IV]] 73; CHECK-NEXT: [[TMP9:%.*]] = load i32, i32* [[ARRAYIDX2]], align 4 74; CHECK-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP9]], [[TMP8]] 75; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[MUL]], [[C_010]] 76; CHECK-NEXT: [[INDVARS_IV_NEXT:%.*]] = or i64 [[INDVARS_IV]], 1 77; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV_NEXT]] 78; CHECK-NEXT: [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX_1]], align 4 79; CHECK-NEXT: [[ARRAYIDX2_1:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDVARS_IV_NEXT]] 80; CHECK-NEXT: [[TMP11:%.*]] = load i32, i32* [[ARRAYIDX2_1]], align 4 81; CHECK-NEXT: [[MUL_1:%.*]] = mul nsw i32 [[TMP11]], [[TMP10]] 82; CHECK-NEXT: [[ADD_1:%.*]] = add nsw i32 [[MUL_1]], [[ADD]] 83; CHECK-NEXT: [[INDVARS_IV_NEXT_1:%.*]] = or i64 [[INDVARS_IV]], 2 84; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV_NEXT_1]] 85; CHECK-NEXT: [[TMP12:%.*]] = load i32, i32* [[ARRAYIDX_2]], align 4 86; CHECK-NEXT: [[ARRAYIDX2_2:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDVARS_IV_NEXT_1]] 87; CHECK-NEXT: [[TMP13:%.*]] = load i32, i32* [[ARRAYIDX2_2]], align 4 88; CHECK-NEXT: [[MUL_2:%.*]] = mul nsw i32 [[TMP13]], [[TMP12]] 89; CHECK-NEXT: [[ADD_2:%.*]] = add nsw i32 [[MUL_2]], [[ADD_1]] 90; CHECK-NEXT: [[INDVARS_IV_NEXT_2:%.*]] = or i64 [[INDVARS_IV]], 3 91; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV_NEXT_2]] 92; CHECK-NEXT: [[TMP14:%.*]] = load i32, i32* [[ARRAYIDX_3]], align 4 93; CHECK-NEXT: [[ARRAYIDX2_3:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDVARS_IV_NEXT_2]] 94; CHECK-NEXT: [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX2_3]], align 4 95; CHECK-NEXT: [[MUL_3:%.*]] = mul nsw i32 [[TMP15]], [[TMP14]] 96; CHECK-NEXT: [[ADD_3]] = add nsw i32 [[MUL_3]], [[ADD_2]] 97; CHECK-NEXT: [[INDVARS_IV_NEXT_3]] = add nuw nsw i64 [[INDVARS_IV]], 4 98; CHECK-NEXT: [[NITER_NEXT_3]] = add i64 [[NITER]], 4 99; CHECK-NEXT: [[NITER_NCMP_3:%.*]] = icmp eq i64 [[NITER_NEXT_3]], [[UNROLL_ITER]] 100; CHECK-NEXT: br i1 [[NITER_NCMP_3]], label [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 101; 102entry: 103 %cmp9 = icmp eq i32 %N, 0 104 br i1 %cmp9, label %for.cond.cleanup, label %for.body.lr.ph 105 106for.body.lr.ph: 107 %wide.trip.count = zext i32 %N to i64 108 br label %for.body 109 110for.cond.cleanup: 111 %c.0.lcssa = phi i32 [ 0, %entry ], [ %add, %for.body ] 112 ret i32 %c.0.lcssa 113 114 115 116 117 118 119 120 121 122 123 124for.body: 125 %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ] 126 %c.010 = phi i32 [ 0, %for.body.lr.ph ], [ %add, %for.body ] 127 %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv 128 %0 = load i32, i32* %arrayidx, align 4 129 %arrayidx2 = getelementptr inbounds i32, i32* %b, i64 %indvars.iv 130 %1 = load i32, i32* %arrayidx2, align 4 131 %mul = mul nsw i32 %1, %0 132 %add = add nsw i32 %mul, %c.010 133 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 134 %exitcond = icmp eq i64 %indvars.iv.next, %wide.trip.count 135 br i1 %exitcond, label %for.cond.cleanup, label %for.body 136} 137