1; REQUIRES: asserts 2; RUN: opt < %s -loop-vectorize -force-vector-width=2 -force-vector-interleave=1 -instcombine -debug-only=loop-vectorize -disable-output -print-after=instcombine 2>&1 | FileCheck %s 3 4target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128" 5 6; CHECK-LABEL: vector_gep 7; CHECK-NOT: LV: Found scalar instruction: %tmp0 = getelementptr inbounds i32, i32* %b, i64 %i 8; CHECK: vector.body: 9; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ] 10; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, %vector.ph ], [ [[VEC_IND_NEXT:%.*]], %vector.body ] 11; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, i32* %b, <2 x i64> [[VEC_IND]] 12; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32*, i32** %a, i64 [[INDEX]] 13; CHECK-NEXT: [[TMP3:%.*]] = bitcast i32** [[TMP2]] to <2 x i32*>* 14; CHECK-NEXT: store <2 x i32*> [[TMP1]], <2 x i32*>* [[TMP3]], align 8 15; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2 16; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2> 17; CHECK: br i1 {{.*}}, label %middle.block, label %vector.body 18; 19define void @vector_gep(i32** %a, i32 *%b, i64 %n) { 20entry: 21 br label %for.body 22 23for.body: 24 %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ] 25 %tmp0 = getelementptr inbounds i32, i32* %b, i64 %i 26 %tmp1 = getelementptr inbounds i32*, i32** %a, i64 %i 27 store i32* %tmp0, i32** %tmp1, align 8 28 %i.next = add nuw nsw i64 %i, 1 29 %cond = icmp slt i64 %i.next, %n 30 br i1 %cond, label %for.body, label %for.end 31 32for.end: 33 ret void 34} 35 36; CHECK-LABEL: scalar_store 37; CHECK: LV: Found scalar instruction: %tmp1 = getelementptr inbounds i32*, i32** %a, i64 %i 38; CHECK-NEXT: LV: Found scalar instruction: %tmp0 = getelementptr inbounds i32, i32* %b, i64 %i 39; CHECK-NEXT: LV: Found scalar instruction: %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ] 40; CHECK-NEXT: LV: Found scalar instruction: %i.next = add nuw nsw i64 %i, 2 41; CHECK: vector.body: 42; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ] 43; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 1 44; CHECK-NEXT: [[TMP4:%.*]] = or i64 [[OFFSET_IDX]], 2 45; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32* %b, i64 [[OFFSET_IDX]] 46; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, i32* %b, i64 [[TMP4]] 47; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32*, i32** %a, i64 [[OFFSET_IDX]] 48; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32*, i32** %a, i64 [[TMP4]] 49; CHECK-NEXT: store i32* [[TMP5]], i32** [[TMP7]], align 8 50; CHECK-NEXT: store i32* [[TMP6]], i32** [[TMP8]], align 8 51; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2 52; CHECK: br i1 {{.*}}, label %middle.block, label %vector.body 53; 54define void @scalar_store(i32** %a, i32 *%b, i64 %n) { 55entry: 56 br label %for.body 57 58for.body: 59 %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ] 60 %tmp0 = getelementptr inbounds i32, i32* %b, i64 %i 61 %tmp1 = getelementptr inbounds i32*, i32** %a, i64 %i 62 store i32* %tmp0, i32** %tmp1, align 8 63 %i.next = add nuw nsw i64 %i, 2 64 %cond = icmp slt i64 %i.next, %n 65 br i1 %cond, label %for.body, label %for.end 66 67for.end: 68 ret void 69} 70 71; CHECK-LABEL: expansion 72; CHECK: LV: Found scalar instruction: %tmp3 = getelementptr inbounds i32*, i32** %tmp2, i64 %i 73; CHECK-NEXT: LV: Found scalar instruction: %tmp1 = bitcast i64* %tmp0 to i32* 74; CHECK-NEXT: LV: Found scalar instruction: %tmp2 = getelementptr inbounds i32*, i32** %a, i64 0 75; CHECK-NEXT: LV: Found scalar instruction: %tmp0 = getelementptr inbounds i64, i64* %b, i64 %i 76; CHECK-NEXT: LV: Found scalar instruction: %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ] 77; CHECK-NEXT: LV: Found scalar instruction: %i.next = add nuw nsw i64 %i, 2 78; CHECK: vector.body: 79; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ] 80; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 1 81; CHECK-NEXT: [[TMP4:%.*]] = or i64 [[OFFSET_IDX]], 2 82; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, i64* %b, i64 [[OFFSET_IDX]] 83; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i64, i64* %b, i64 [[TMP4]] 84; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32*, i32** %a, i64 [[OFFSET_IDX]] 85; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32*, i32** %a, i64 [[TMP4]] 86; CHECK-NEXT: [[TMP9:%.*]] = bitcast i32** [[TMP7]] to i64** 87; CHECK-NEXT: store i64* [[TMP5]], i64** [[TMP9]], align 8 88; CHECK-NEXT: [[TMP10:%.*]] = bitcast i32** [[TMP8]] to i64** 89; CHECK-NEXT: store i64* [[TMP6]], i64** [[TMP10]], align 8 90; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2 91; CHECK: br i1 {{.*}}, label %middle.block, label %vector.body 92; 93define void @expansion(i32** %a, i64 *%b, i64 %n) { 94entry: 95 br label %for.body 96 97for.body: 98 %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ] 99 %tmp0 = getelementptr inbounds i64, i64* %b, i64 %i 100 %tmp1 = bitcast i64* %tmp0 to i32* 101 %tmp2 = getelementptr inbounds i32*, i32** %a, i64 0 102 %tmp3 = getelementptr inbounds i32*, i32** %tmp2, i64 %i 103 store i32* %tmp1, i32** %tmp3, align 8 104 %i.next = add nuw nsw i64 %i, 2 105 %cond = icmp slt i64 %i.next, %n 106 br i1 %cond, label %for.body, label %for.end 107 108for.end: 109 ret void 110} 111 112; CHECK-LABEL: no_gep_or_bitcast 113; CHECK-NOT: LV: Found scalar instruction: %tmp1 = load i32*, i32** %tmp0, align 8 114; CHECK: LV: Found scalar instruction: %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ] 115; CHECK-NEXT: LV: Found scalar instruction: %i.next = add nuw nsw i64 %i, 1 116; CHECK: vector.body: 117; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ] 118; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32*, i32** %a, i64 [[INDEX]] 119; CHECK-NEXT: [[TMP2:%.*]] = bitcast i32** [[TMP1]] to <2 x i32*>* 120; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32*>, <2 x i32*>* [[TMP2]], align 8 121; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i32*> [[WIDE_LOAD]], i32 0 122; CHECK-NEXT: store i32 0, i32* [[TMP3]], align 8 123; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i32*> [[WIDE_LOAD]], i32 1 124; CHECK-NEXT: store i32 0, i32* [[TMP4]], align 8 125; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2 126; CHECK: br i1 {{.*}}, label %middle.block, label %vector.body 127; 128define void @no_gep_or_bitcast(i32** noalias %a, i64 %n) { 129entry: 130 br label %for.body 131 132for.body: 133 %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ] 134 %tmp0 = getelementptr inbounds i32*, i32** %a, i64 %i 135 %tmp1 = load i32*, i32** %tmp0, align 8 136 store i32 0, i32* %tmp1, align 8 137 %i.next = add nuw nsw i64 %i, 1 138 %cond = icmp slt i64 %i.next, %n 139 br i1 %cond, label %for.body, label %for.end 140 141for.end: 142 ret void 143} 144