1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -loop-vectorize -force-vector-width=2 -force-vector-interleave=1 -S %s | FileCheck %s 3 4; Make sure LV does not crash when creating runtime checks involving values from 5; other loops. 6define i16 @test(i16** %arg, i64 %N) { 7; CHECK-LABEL: @test( 8; CHECK-NEXT: entry: 9; CHECK-NEXT: br label [[OUTER:%.*]] 10; CHECK: outer: 11; CHECK-NEXT: [[L_1:%.*]] = load i16*, i16** [[ARG:%.*]], align 8 12; CHECK-NEXT: [[L_2:%.*]] = load i16*, i16** [[ARG]], align 8 13; CHECK-NEXT: [[C_1:%.*]] = call i1 @cond() 14; CHECK-NEXT: br i1 [[C_1]], label [[OUTER_BACKEDGE:%.*]], label [[INNER_PREHEADER:%.*]] 15; CHECK: outer.backedge: 16; CHECK-NEXT: br label [[OUTER]] 17; CHECK: inner.preheader: 18; CHECK-NEXT: br label [[INNER:%.*]] 19; CHECK: inner: 20; CHECK-NEXT: [[C_2:%.*]] = call i1 @cond() 21; CHECK-NEXT: br i1 [[C_2]], label [[OUTER_LATCH:%.*]], label [[INNER_BB:%.*]] 22; CHECK: inner.bb: 23; CHECK-NEXT: [[C_3:%.*]] = call i1 @cond() 24; CHECK-NEXT: br i1 [[C_3]], label [[LOOP3_PREHEADER:%.*]], label [[INNER_LATCH:%.*]] 25; CHECK: loop3.preheader: 26; CHECK-NEXT: [[L_1_LCSSA11:%.*]] = phi i16* [ [[L_1]], [[INNER_BB]] ] 27; CHECK-NEXT: [[L_1_LCSSA:%.*]] = phi i16* [ [[L_1]], [[INNER_BB]] ] 28; CHECK-NEXT: [[L_2_LCSSA:%.*]] = phi i16* [ [[L_2]], [[INNER_BB]] ] 29; CHECK-NEXT: [[L_2_LCSSA4:%.*]] = bitcast i16* [[L_2_LCSSA]] to i8* 30; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N:%.*]], 1 31; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 2 32; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] 33; CHECK: vector.memcheck: 34; CHECK-NEXT: [[UGLYGEP:%.*]] = getelementptr i8, i8* [[L_2_LCSSA4]], i64 1 35; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i16, i16* [[L_1_LCSSA]], i64 1 36; CHECK-NEXT: [[SCEVGEP9:%.*]] = bitcast i16* [[SCEVGEP]] to i8* 37; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[N]], 2 38; CHECK-NEXT: [[SCEVGEP10:%.*]] = getelementptr i16, i16* [[L_1_LCSSA11]], i64 [[TMP1]] 39; CHECK-NEXT: [[SCEVGEP1013:%.*]] = bitcast i16* [[SCEVGEP10]] to i8* 40; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult i8* [[L_2_LCSSA4]], [[SCEVGEP1013]] 41; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult i8* [[SCEVGEP9]], [[UGLYGEP]] 42; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] 43; CHECK-NEXT: [[MEMCHECK_CONFLICT:%.*]] = and i1 [[FOUND_CONFLICT]], true 44; CHECK-NEXT: br i1 [[MEMCHECK_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] 45; CHECK: vector.ph: 46; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 2 47; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]] 48; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 49; CHECK: vector.body: 50; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 51; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 0 52; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 1 53; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i16, i16* [[L_1]], i64 [[TMP3]] 54; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i16, i16* [[TMP4]], i32 0 55; CHECK-NEXT: [[TMP6:%.*]] = bitcast i16* [[TMP5]] to <2 x i16>* 56; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i16>, <2 x i16>* [[TMP6]], align 2, !alias.scope !0 57; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i16, i16* [[L_2]], i64 0 58; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x i16> [[WIDE_LOAD]], i32 0 59; CHECK-NEXT: store i16 [[TMP8]], i16* [[TMP7]], align 2, !alias.scope !3, !noalias !0 60; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x i16> [[WIDE_LOAD]], i32 1 61; CHECK-NEXT: store i16 [[TMP9]], i16* [[TMP7]], align 2, !alias.scope !3, !noalias !0 62; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2 63; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 64; CHECK-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP5:!llvm.loop !.*]] 65; CHECK: middle.block: 66; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]] 67; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]] 68; CHECK: scalar.ph: 69; CHECK-NEXT: [[L_18:%.*]] = phi i16* [ [[L_1_LCSSA]], [[VECTOR_MEMCHECK]] ], [ [[L_1_LCSSA]], [[LOOP3_PREHEADER]] ], [ [[L_1_LCSSA]], [[MIDDLE_BLOCK]] ] 70; CHECK-NEXT: [[L_23:%.*]] = phi i16* [ [[L_2_LCSSA]], [[VECTOR_MEMCHECK]] ], [ [[L_2_LCSSA]], [[LOOP3_PREHEADER]] ], [ [[L_2_LCSSA]], [[MIDDLE_BLOCK]] ] 71; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP3_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ] 72; CHECK-NEXT: br label [[LOOP3:%.*]] 73; CHECK: inner.latch: 74; CHECK-NEXT: [[C_4:%.*]] = call i1 @cond() 75; CHECK-NEXT: br i1 [[C_4]], label [[EXIT_LOOPEXIT1:%.*]], label [[INNER]] 76; CHECK: outer.latch: 77; CHECK-NEXT: br label [[OUTER_BACKEDGE]] 78; CHECK: loop3: 79; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP3]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 80; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 81; CHECK-NEXT: [[C_5:%.*]] = icmp ult i64 [[IV]], [[N]] 82; CHECK-NEXT: [[GEP_1:%.*]] = getelementptr inbounds i16, i16* [[L_18]], i64 [[IV_NEXT]] 83; CHECK-NEXT: [[LOOP_L_1:%.*]] = load i16, i16* [[GEP_1]], align 2 84; CHECK-NEXT: [[GEP_2:%.*]] = getelementptr inbounds i16, i16* [[L_23]], i64 0 85; CHECK-NEXT: store i16 [[LOOP_L_1]], i16* [[GEP_2]], align 2 86; CHECK-NEXT: br i1 [[C_5]], label [[LOOP3]], label [[EXIT_LOOPEXIT]], [[LOOP7:!llvm.loop !.*]] 87; CHECK: exit.loopexit: 88; CHECK-NEXT: [[L_17:%.*]] = phi i16* [ [[L_1_LCSSA]], [[MIDDLE_BLOCK]] ], [ [[L_18]], [[LOOP3]] ] 89; CHECK-NEXT: br label [[EXIT:%.*]] 90; CHECK: exit.loopexit1: 91; CHECK-NEXT: [[L_1_LCSSA5:%.*]] = phi i16* [ [[L_1]], [[INNER_LATCH]] ] 92; CHECK-NEXT: br label [[EXIT]] 93; CHECK: exit: 94; CHECK-NEXT: [[L_16:%.*]] = phi i16* [ [[L_1_LCSSA5]], [[EXIT_LOOPEXIT1]] ], [ [[L_17]], [[EXIT_LOOPEXIT]] ] 95; CHECK-NEXT: [[L_3:%.*]] = load i16, i16* [[L_16]], align 2 96; CHECK-NEXT: ret i16 [[L_3]] 97; 98entry: 99 br label %outer 100 101outer: 102 %l.1 = load i16*, i16** %arg, align 8 103 %l.2 = load i16*, i16** %arg, align 8 104 %c.1 = call i1 @cond() 105 br i1 %c.1, label %outer, label %inner 106 107inner: ; preds = %bb15, %bb1 108 %c.2 = call i1 @cond() 109 br i1 %c.2, label %outer.latch, label %inner.bb 110 111inner.bb: ; preds = %bb3 112 %c.3 = call i1 @cond() 113 br i1 %c.3, label %loop3, label %inner.latch 114 115inner.latch: ; preds = %bb4 116 %c.4 = call i1 @cond() 117 br i1 %c.4, label %exit, label %inner 118 119outer.latch: ; preds = %bb3 120 br label %outer 121 122loop3: ; preds = %bb9, %bb4 123 %iv = phi i64 [ %iv.next, %loop3 ], [ 0, %inner.bb ] 124 %iv.next = add nsw nuw i64 %iv, 1 125 %c.5 = icmp ult i64 %iv, %N 126 %gep.1 = getelementptr inbounds i16, i16* %l.1, i64 %iv.next 127 %loop.l.1 = load i16, i16* %gep.1, align 2 128 %gep.2 = getelementptr inbounds i16, i16* %l.2, i64 0 129 store i16 %loop.l.1, i16* %gep.2 , align 2 130 br i1 %c.5, label %loop3, label %exit 131 132exit: ; preds = %bb15, %bb5 133 %l.3 = load i16, i16* %l.1, align 2 134 ret i16 %l.3 135} 136 137declare i1 @cond() 138