151 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=loop-simplify,loop-rotate,instcombine,indvars -S -verify-loop-info -verify-dom-info | FileCheck %s3 4; Loopsimplify should be able to merge the two loop exits5; into one, so that loop rotate can rotate the loop, so6; that indvars can promote the induction variable to i647; without needing casts.8 9target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n32:64"10 11define float @test1(ptr %pTmp1, ptr %peakWeight, i32 %bandEdgeIndex) nounwind {12; CHECK-LABEL: @test1(13; CHECK-NEXT: entry:14; CHECK-NEXT: [[T0:%.*]] = load float, ptr [[PEAKWEIGHT:%.*]], align 415; CHECK-NEXT: [[T11:%.*]] = add i32 [[BANDEDGEINDEX:%.*]], -116; CHECK-NEXT: [[T121:%.*]] = icmp sgt i32 [[T11]], 017; CHECK-NEXT: br i1 [[T121]], label [[BB_LR_PH:%.*]], label [[BB3:%.*]]18; CHECK: bb.lr.ph:19; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[T11]] to i6420; CHECK-NEXT: br label [[BB:%.*]]21; CHECK: bb:22; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[BB_LR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[BB]] ]23; CHECK-NEXT: [[DISTERBHI_04:%.*]] = phi float [ 0.000000e+00, [[BB_LR_PH]] ], [ [[T4:%.*]], [[BB]] ]24; CHECK-NEXT: [[PEAKCOUNT_02:%.*]] = phi float [ [[T0]], [[BB_LR_PH]] ], [ [[T9:%.*]], [[BB]] ]25; CHECK-NEXT: [[T2:%.*]] = getelementptr float, ptr [[PTMP1:%.*]], i64 [[INDVARS_IV]]26; CHECK-NEXT: [[T3:%.*]] = load float, ptr [[T2]], align 427; CHECK-NEXT: [[T4]] = fadd float [[T3]], [[DISTERBHI_04]]28; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 129; CHECK-NEXT: [[T7:%.*]] = getelementptr float, ptr [[PEAKWEIGHT]], i64 [[INDVARS_IV_NEXT]]30; CHECK-NEXT: [[T8:%.*]] = load float, ptr [[T7]], align 431; CHECK-NEXT: [[T9]] = fadd float [[T8]], [[PEAKCOUNT_02]]32; CHECK-NEXT: [[T10:%.*]] = fcmp olt float [[T4]], 2.500000e+0033; CHECK-NEXT: [[T12:%.*]] = icmp sgt i64 [[TMP0]], [[INDVARS_IV_NEXT]]34; CHECK-NEXT: [[OR_COND:%.*]] = and i1 [[T10]], [[T12]]35; CHECK-NEXT: br i1 [[OR_COND]], label [[BB]], label [[BB1_BB3_CRIT_EDGE:%.*]]36; CHECK: bb1.bb3_crit_edge:37; CHECK-NEXT: [[T4_LCSSA:%.*]] = phi float [ [[T4]], [[BB]] ]38; CHECK-NEXT: [[T9_LCSSA:%.*]] = phi float [ [[T9]], [[BB]] ]39; CHECK-NEXT: br label [[BB3]]40; CHECK: bb3:41; CHECK-NEXT: [[PEAKCOUNT_0_LCSSA:%.*]] = phi float [ [[T9_LCSSA]], [[BB1_BB3_CRIT_EDGE]] ], [ [[T0]], [[ENTRY:%.*]] ]42; CHECK-NEXT: [[DISTERBHI_0_LCSSA:%.*]] = phi float [ [[T4_LCSSA]], [[BB1_BB3_CRIT_EDGE]] ], [ 0.000000e+00, [[ENTRY]] ]43; CHECK-NEXT: [[T13:%.*]] = fdiv float [[PEAKCOUNT_0_LCSSA]], [[DISTERBHI_0_LCSSA]]44; CHECK-NEXT: ret float [[T13]]45;46entry:47 %t0 = load float, ptr %peakWeight, align 448 br label %bb149 50bb: ; preds = %bb251 %t1 = sext i32 %hiPart.0 to i6452 %t2 = getelementptr float, ptr %pTmp1, i64 %t153 %t3 = load float, ptr %t2, align 454 %t4 = fadd float %t3, %distERBhi.055 %t5 = add i32 %hiPart.0, 156 %t6 = sext i32 %t5 to i6457 %t7 = getelementptr float, ptr %peakWeight, i64 %t658 %t8 = load float, ptr %t7, align 459 %t9 = fadd float %t8, %peakCount.060 br label %bb161 62bb1: ; preds = %bb, %entry63 %peakCount.0 = phi float [ %t0, %entry ], [ %t9, %bb ]64 %hiPart.0 = phi i32 [ 0, %entry ], [ %t5, %bb ]65 %distERBhi.0 = phi float [ 0.000000e+00, %entry ], [ %t4, %bb ]66 %t10 = fcmp uge float %distERBhi.0, 2.500000e+0067 br i1 %t10, label %bb3, label %bb268 69bb2: ; preds = %bb170 %t11 = add i32 %bandEdgeIndex, -171 %t12 = icmp sgt i32 %t11, %hiPart.072 br i1 %t12, label %bb, label %bb373 74bb3: ; preds = %bb2, %bb175 %t13 = fdiv float %peakCount.0, %distERBhi.076 ret float %t1377}78 79; Same test as above.80; This would crash because we assumed TTI was available to process the metadata.81 82define float @merge_branches_profile_metadata(ptr %pTmp1, ptr %peakWeight, i32 %bandEdgeIndex) nounwind {83; CHECK-LABEL: @merge_branches_profile_metadata(84; CHECK-NEXT: entry:85; CHECK-NEXT: [[T0:%.*]] = load float, ptr [[PEAKWEIGHT:%.*]], align 486; CHECK-NEXT: [[T11:%.*]] = add i32 [[BANDEDGEINDEX:%.*]], -187; CHECK-NEXT: [[T121:%.*]] = icmp sgt i32 [[T11]], 088; CHECK-NEXT: br i1 [[T121]], label [[BB_LR_PH:%.*]], label [[BB3:%.*]], !prof [[PROF0:![0-9]+]]89; CHECK: bb.lr.ph:90; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[T11]] to i6491; CHECK-NEXT: br label [[BB:%.*]]92; CHECK: bb:93; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[BB_LR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[BB]] ]94; CHECK-NEXT: [[DISTERBHI_04:%.*]] = phi float [ 0.000000e+00, [[BB_LR_PH]] ], [ [[T4:%.*]], [[BB]] ]95; CHECK-NEXT: [[PEAKCOUNT_02:%.*]] = phi float [ [[T0]], [[BB_LR_PH]] ], [ [[T9:%.*]], [[BB]] ]96; CHECK-NEXT: [[T2:%.*]] = getelementptr float, ptr [[PTMP1:%.*]], i64 [[INDVARS_IV]]97; CHECK-NEXT: [[T3:%.*]] = load float, ptr [[T2]], align 498; CHECK-NEXT: [[T4]] = fadd float [[T3]], [[DISTERBHI_04]]99; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1100; CHECK-NEXT: [[T7:%.*]] = getelementptr float, ptr [[PEAKWEIGHT]], i64 [[INDVARS_IV_NEXT]]101; CHECK-NEXT: [[T8:%.*]] = load float, ptr [[T7]], align 4102; CHECK-NEXT: [[T9]] = fadd float [[T8]], [[PEAKCOUNT_02]]103; CHECK-NEXT: [[T10:%.*]] = fcmp olt float [[T4]], 2.500000e+00104; CHECK-NEXT: [[T12:%.*]] = icmp sgt i64 [[TMP0]], [[INDVARS_IV_NEXT]]105; CHECK-NEXT: [[OR_COND:%.*]] = and i1 [[T10]], [[T12]]106; CHECK-NEXT: br i1 [[OR_COND]], label [[BB]], label [[BB1_BB3_CRIT_EDGE:%.*]], !prof [[PROF1:![0-9]+]]107; CHECK: bb1.bb3_crit_edge:108; CHECK-NEXT: [[T4_LCSSA:%.*]] = phi float [ [[T4]], [[BB]] ]109; CHECK-NEXT: [[T9_LCSSA:%.*]] = phi float [ [[T9]], [[BB]] ]110; CHECK-NEXT: br label [[BB3]]111; CHECK: bb3:112; CHECK-NEXT: [[PEAKCOUNT_0_LCSSA:%.*]] = phi float [ [[T9_LCSSA]], [[BB1_BB3_CRIT_EDGE]] ], [ [[T0]], [[ENTRY:%.*]] ]113; CHECK-NEXT: [[DISTERBHI_0_LCSSA:%.*]] = phi float [ [[T4_LCSSA]], [[BB1_BB3_CRIT_EDGE]] ], [ 0.000000e+00, [[ENTRY]] ]114; CHECK-NEXT: [[T13:%.*]] = fdiv float [[PEAKCOUNT_0_LCSSA]], [[DISTERBHI_0_LCSSA]]115; CHECK-NEXT: ret float [[T13]]116;117entry:118 %t0 = load float, ptr %peakWeight, align 4119 br label %bb1120 121bb: ; preds = %bb2122 %t1 = sext i32 %hiPart.0 to i64123 %t2 = getelementptr float, ptr %pTmp1, i64 %t1124 %t3 = load float, ptr %t2, align 4125 %t4 = fadd float %t3, %distERBhi.0126 %t5 = add i32 %hiPart.0, 1127 %t6 = sext i32 %t5 to i64128 %t7 = getelementptr float, ptr %peakWeight, i64 %t6129 %t8 = load float, ptr %t7, align 4130 %t9 = fadd float %t8, %peakCount.0131 br label %bb1132 133bb1: ; preds = %bb, %entry134 %peakCount.0 = phi float [ %t0, %entry ], [ %t9, %bb ]135 %hiPart.0 = phi i32 [ 0, %entry ], [ %t5, %bb ]136 %distERBhi.0 = phi float [ 0.000000e+00, %entry ], [ %t4, %bb ]137 %t10 = fcmp uge float %distERBhi.0, 2.500000e+00138 br i1 %t10, label %bb3, label %bb2, !prof !0139 140bb2: ; preds = %bb1141 %t11 = add i32 %bandEdgeIndex, -1142 %t12 = icmp sgt i32 %t11, %hiPart.0143 br i1 %t12, label %bb, label %bb3144 145bb3: ; preds = %bb2, %bb1146 %t13 = fdiv float %peakCount.0, %distERBhi.0147 ret float %t13148}149 150!0 = !{!"branch_weights", i32 2000, i32 1}151