1; RUN: opt < %s -loop-vectorize -force-vector-interleave=2 -force-vector-width=4 -S | FileCheck %s 2 3target 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-n8:16:32:64-S128" 4 5; Make sure consecutive vector generates correct negative indices. 6; PR15882 7 8; CHECK: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ] 9; CHECK: %offset.idx = sub i64 %startval, %index 10; CHECK: %[[a0:.+]] = add i64 %offset.idx, 0 11; CHECK: %[[a1:.+]] = add i64 %offset.idx, -1 12; CHECK: %[[a2:.+]] = add i64 %offset.idx, -2 13; CHECK: %[[a3:.+]] = add i64 %offset.idx, -3 14; CHECK: %[[a4:.+]] = add i64 %offset.idx, -4 15; CHECK: %[[a5:.+]] = add i64 %offset.idx, -5 16; CHECK: %[[a6:.+]] = add i64 %offset.idx, -6 17; CHECK: %[[a7:.+]] = add i64 %offset.idx, -7 18 19define i32 @reverse_induction_i64(i64 %startval, i32 * %ptr) { 20entry: 21 br label %for.body 22 23for.body: 24 %add.i7 = phi i64 [ %startval, %entry ], [ %add.i, %for.body ] 25 %i.06 = phi i32 [ 0, %entry ], [ %inc4, %for.body ] 26 %redux5 = phi i32 [ 0, %entry ], [ %inc.redux, %for.body ] 27 %add.i = add i64 %add.i7, -1 28 %kind_.i = getelementptr inbounds i32, i32* %ptr, i64 %add.i 29 %tmp.i1 = load i32, i32* %kind_.i, align 4 30 %inc.redux = add i32 %tmp.i1, %redux5 31 %inc4 = add i32 %i.06, 1 32 %exitcond = icmp ne i32 %inc4, 1024 33 br i1 %exitcond, label %for.body, label %loopend 34 35loopend: 36 ret i32 %inc.redux 37} 38 39; CHECK-LABEL: @reverse_induction_i128( 40; CHECK: %index = phi i128 [ 0, %vector.ph ], [ %index.next, %vector.body ] 41; CHECK: %offset.idx = sub i128 %startval, %index 42; CHECK: %[[a0:.+]] = add i128 %offset.idx, 0 43; CHECK: %[[a1:.+]] = add i128 %offset.idx, -1 44; CHECK: %[[a2:.+]] = add i128 %offset.idx, -2 45; CHECK: %[[a3:.+]] = add i128 %offset.idx, -3 46; CHECK: %[[a4:.+]] = add i128 %offset.idx, -4 47; CHECK: %[[a5:.+]] = add i128 %offset.idx, -5 48; CHECK: %[[a6:.+]] = add i128 %offset.idx, -6 49; CHECK: %[[a7:.+]] = add i128 %offset.idx, -7 50 51define i32 @reverse_induction_i128(i128 %startval, i32 * %ptr) { 52entry: 53 br label %for.body 54 55for.body: 56 %add.i7 = phi i128 [ %startval, %entry ], [ %add.i, %for.body ] 57 %i.06 = phi i32 [ 0, %entry ], [ %inc4, %for.body ] 58 %redux5 = phi i32 [ 0, %entry ], [ %inc.redux, %for.body ] 59 %add.i = add i128 %add.i7, -1 60 %kind_.i = getelementptr inbounds i32, i32* %ptr, i128 %add.i 61 %tmp.i1 = load i32, i32* %kind_.i, align 4 62 %inc.redux = add i32 %tmp.i1, %redux5 63 %inc4 = add i32 %i.06, 1 64 %exitcond = icmp ne i32 %inc4, 1024 65 br i1 %exitcond, label %for.body, label %loopend 66 67loopend: 68 ret i32 %inc.redux 69} 70 71; CHECK-LABEL: @reverse_induction_i16( 72; CHECK: %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ] 73; CHECK: %offset.idx = sub i16 %startval, {{.*}} 74; CHECK: %[[a0:.+]] = add i16 %offset.idx, 0 75; CHECK: %[[a1:.+]] = add i16 %offset.idx, -1 76; CHECK: %[[a2:.+]] = add i16 %offset.idx, -2 77; CHECK: %[[a3:.+]] = add i16 %offset.idx, -3 78; CHECK: %[[a4:.+]] = add i16 %offset.idx, -4 79; CHECK: %[[a5:.+]] = add i16 %offset.idx, -5 80; CHECK: %[[a6:.+]] = add i16 %offset.idx, -6 81; CHECK: %[[a7:.+]] = add i16 %offset.idx, -7 82 83define i32 @reverse_induction_i16(i16 %startval, i32 * %ptr) { 84entry: 85 br label %for.body 86 87for.body: 88 %add.i7 = phi i16 [ %startval, %entry ], [ %add.i, %for.body ] 89 %i.06 = phi i32 [ 0, %entry ], [ %inc4, %for.body ] 90 %redux5 = phi i32 [ 0, %entry ], [ %inc.redux, %for.body ] 91 %add.i = add i16 %add.i7, -1 92 %kind_.i = getelementptr inbounds i32, i32* %ptr, i16 %add.i 93 %tmp.i1 = load i32, i32* %kind_.i, align 4 94 %inc.redux = add i32 %tmp.i1, %redux5 95 %inc4 = add i32 %i.06, 1 96 %exitcond = icmp ne i32 %inc4, 1024 97 br i1 %exitcond, label %for.body, label %loopend 98 99loopend: 100 ret i32 %inc.redux 101} 102 103 104@a = common global [1024 x i32] zeroinitializer, align 16 105 106; We incorrectly transformed this loop into an empty one because we left the 107; induction variable in i8 type and truncated the exit value 1024 to 0. 108; int a[1024]; 109; 110; void fail() { 111; int reverse_induction = 1023; 112; unsigned char forward_induction = 0; 113; while ((reverse_induction) >= 0) { 114; forward_induction++; 115; a[reverse_induction] = forward_induction; 116; --reverse_induction; 117; } 118; } 119 120; CHECK-LABEL: @reverse_forward_induction_i64_i8( 121; CHECK: vector.body 122; CHECK: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ] 123; CHECK: %vec.ind = phi <4 x i64> [ <i64 1023, i64 1022, i64 1021, i64 1020>, %vector.ph ] 124; CHECK: %step.add = add <4 x i64> %vec.ind, <i64 -4, i64 -4, i64 -4, i64 -4> 125; CHECK: trunc i64 %index to i8 126 127define void @reverse_forward_induction_i64_i8() { 128entry: 129 br label %while.body 130 131while.body: 132 %indvars.iv = phi i64 [ 1023, %entry ], [ %indvars.iv.next, %while.body ] 133 %forward_induction.05 = phi i8 [ 0, %entry ], [ %inc, %while.body ] 134 %inc = add i8 %forward_induction.05, 1 135 %conv = zext i8 %inc to i32 136 %arrayidx = getelementptr inbounds [1024 x i32], [1024 x i32]* @a, i64 0, i64 %indvars.iv 137 store i32 %conv, i32* %arrayidx, align 4 138 %indvars.iv.next = add i64 %indvars.iv, -1 139 %0 = trunc i64 %indvars.iv to i32 140 %cmp = icmp sgt i32 %0, 0 141 br i1 %cmp, label %while.body, label %while.end 142 143while.end: 144 ret void 145} 146 147; CHECK-LABEL: @reverse_forward_induction_i64_i8_signed( 148; CHECK: vector.body: 149; CHECK: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ] 150; CHECK: %vec.ind = phi <4 x i64> [ <i64 1023, i64 1022, i64 1021, i64 1020>, %vector.ph ] 151; CHECK: %step.add = add <4 x i64> %vec.ind, <i64 -4, i64 -4, i64 -4, i64 -4> 152 153define void @reverse_forward_induction_i64_i8_signed() { 154entry: 155 br label %while.body 156 157while.body: 158 %indvars.iv = phi i64 [ 1023, %entry ], [ %indvars.iv.next, %while.body ] 159 %forward_induction.05 = phi i8 [ -127, %entry ], [ %inc, %while.body ] 160 %inc = add i8 %forward_induction.05, 1 161 %conv = sext i8 %inc to i32 162 %arrayidx = getelementptr inbounds [1024 x i32], [1024 x i32]* @a, i64 0, i64 %indvars.iv 163 store i32 %conv, i32* %arrayidx, align 4 164 %indvars.iv.next = add i64 %indvars.iv, -1 165 %0 = trunc i64 %indvars.iv to i32 166 %cmp = icmp sgt i32 %0, 0 167 br i1 %cmp, label %while.body, label %while.end 168 169while.end: 170 ret void 171} 172