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