Home | History | Annotate | Download | only in AArch64
      1 ; RUN: llc -O3 -mtriple arm64-apple-ios3 -aarch64-gep-opt=false %s -o - | FileCheck %s
      2 ; <rdar://problem/13621857>
      3 
      4 @block = common global i8* null, align 8
      5 
      6 define i32 @fct(i32 %i1, i32 %i2) {
      7 ; CHECK: @fct
      8 ; Sign extension is used more than once, thus it should not be folded.
      9 ; CodeGenPrepare is not sharing sext across uses, thus this is folded because
     10 ; of that.
     11 ; _CHECK-NOT_: , sxtw]
     12 entry:
     13   %idxprom = sext i32 %i1 to i64
     14   %0 = load i8*, i8** @block, align 8
     15   %arrayidx = getelementptr inbounds i8, i8* %0, i64 %idxprom
     16   %1 = load i8, i8* %arrayidx, align 1
     17   %idxprom1 = sext i32 %i2 to i64
     18   %arrayidx2 = getelementptr inbounds i8, i8* %0, i64 %idxprom1
     19   %2 = load i8, i8* %arrayidx2, align 1
     20   %cmp = icmp eq i8 %1, %2
     21   br i1 %cmp, label %if.end, label %if.then
     22 
     23 if.then:                                          ; preds = %entry
     24   %cmp7 = icmp ugt i8 %1, %2
     25   %conv8 = zext i1 %cmp7 to i32
     26   br label %return
     27 
     28 if.end:                                           ; preds = %entry
     29   %inc = add nsw i32 %i1, 1
     30   %inc9 = add nsw i32 %i2, 1
     31   %idxprom10 = sext i32 %inc to i64
     32   %arrayidx11 = getelementptr inbounds i8, i8* %0, i64 %idxprom10
     33   %3 = load i8, i8* %arrayidx11, align 1
     34   %idxprom12 = sext i32 %inc9 to i64
     35   %arrayidx13 = getelementptr inbounds i8, i8* %0, i64 %idxprom12
     36   %4 = load i8, i8* %arrayidx13, align 1
     37   %cmp16 = icmp eq i8 %3, %4
     38   br i1 %cmp16, label %if.end23, label %if.then18
     39 
     40 if.then18:                                        ; preds = %if.end
     41   %cmp21 = icmp ugt i8 %3, %4
     42   %conv22 = zext i1 %cmp21 to i32
     43   br label %return
     44 
     45 if.end23:                                         ; preds = %if.end
     46   %inc24 = add nsw i32 %i1, 2
     47   %inc25 = add nsw i32 %i2, 2
     48   %idxprom26 = sext i32 %inc24 to i64
     49   %arrayidx27 = getelementptr inbounds i8, i8* %0, i64 %idxprom26
     50   %5 = load i8, i8* %arrayidx27, align 1
     51   %idxprom28 = sext i32 %inc25 to i64
     52   %arrayidx29 = getelementptr inbounds i8, i8* %0, i64 %idxprom28
     53   %6 = load i8, i8* %arrayidx29, align 1
     54   %cmp32 = icmp eq i8 %5, %6
     55   br i1 %cmp32, label %return, label %if.then34
     56 
     57 if.then34:                                        ; preds = %if.end23
     58   %cmp37 = icmp ugt i8 %5, %6
     59   %conv38 = zext i1 %cmp37 to i32
     60   br label %return
     61 
     62 return:                                           ; preds = %if.end23, %if.then34, %if.then18, %if.then
     63   %retval.0 = phi i32 [ %conv8, %if.then ], [ %conv22, %if.then18 ], [ %conv38, %if.then34 ], [ 1, %if.end23 ]
     64   ret i32 %retval.0
     65 }
     66 
     67 define i32 @fct1(i32 %i1, i32 %i2) optsize {
     68 ; CHECK: @fct1
     69 ; Addressing are folded when optimizing for code size.
     70 ; CHECK: , sxtw]
     71 ; CHECK: , sxtw]
     72 entry:
     73   %idxprom = sext i32 %i1 to i64
     74   %0 = load i8*, i8** @block, align 8
     75   %arrayidx = getelementptr inbounds i8, i8* %0, i64 %idxprom
     76   %1 = load i8, i8* %arrayidx, align 1
     77   %idxprom1 = sext i32 %i2 to i64
     78   %arrayidx2 = getelementptr inbounds i8, i8* %0, i64 %idxprom1
     79   %2 = load i8, i8* %arrayidx2, align 1
     80   %cmp = icmp eq i8 %1, %2
     81   br i1 %cmp, label %if.end, label %if.then
     82 
     83 if.then:                                          ; preds = %entry
     84   %cmp7 = icmp ugt i8 %1, %2
     85   %conv8 = zext i1 %cmp7 to i32
     86   br label %return
     87 
     88 if.end:                                           ; preds = %entry
     89   %inc = add nsw i32 %i1, 1
     90   %inc9 = add nsw i32 %i2, 1
     91   %idxprom10 = sext i32 %inc to i64
     92   %arrayidx11 = getelementptr inbounds i8, i8* %0, i64 %idxprom10
     93   %3 = load i8, i8* %arrayidx11, align 1
     94   %idxprom12 = sext i32 %inc9 to i64
     95   %arrayidx13 = getelementptr inbounds i8, i8* %0, i64 %idxprom12
     96   %4 = load i8, i8* %arrayidx13, align 1
     97   %cmp16 = icmp eq i8 %3, %4
     98   br i1 %cmp16, label %if.end23, label %if.then18
     99 
    100 if.then18:                                        ; preds = %if.end
    101   %cmp21 = icmp ugt i8 %3, %4
    102   %conv22 = zext i1 %cmp21 to i32
    103   br label %return
    104 
    105 if.end23:                                         ; preds = %if.end
    106   %inc24 = add nsw i32 %i1, 2
    107   %inc25 = add nsw i32 %i2, 2
    108   %idxprom26 = sext i32 %inc24 to i64
    109   %arrayidx27 = getelementptr inbounds i8, i8* %0, i64 %idxprom26
    110   %5 = load i8, i8* %arrayidx27, align 1
    111   %idxprom28 = sext i32 %inc25 to i64
    112   %arrayidx29 = getelementptr inbounds i8, i8* %0, i64 %idxprom28
    113   %6 = load i8, i8* %arrayidx29, align 1
    114   %cmp32 = icmp eq i8 %5, %6
    115   br i1 %cmp32, label %return, label %if.then34
    116 
    117 if.then34:                                        ; preds = %if.end23
    118   %cmp37 = icmp ugt i8 %5, %6
    119   %conv38 = zext i1 %cmp37 to i32
    120   br label %return
    121 
    122 return:                                           ; preds = %if.end23, %if.then34, %if.then18, %if.then
    123   %retval.0 = phi i32 [ %conv8, %if.then ], [ %conv22, %if.then18 ], [ %conv38, %if.then34 ], [ 1, %if.end23 ]
    124   ret i32 %retval.0
    125 }
    126 
    127 ; CHECK: @test
    128 ; CHECK-NOT: , uxtw #2]
    129 define i32 @test(i32* %array, i8 zeroext %c, i32 %arg) {
    130 entry:
    131   %conv = zext i8 %c to i32
    132   %add = sub i32 0, %arg
    133   %tobool = icmp eq i32 %conv, %add
    134   br i1 %tobool, label %if.end, label %if.then
    135 
    136 if.then:                                          ; preds = %entry
    137   %idxprom = zext i8 %c to i64
    138   %arrayidx = getelementptr inbounds i32, i32* %array, i64 %idxprom
    139   %0 = load volatile i32, i32* %arrayidx, align 4
    140   %1 = load volatile i32, i32* %arrayidx, align 4
    141   %add3 = add nsw i32 %1, %0
    142   br label %if.end
    143 
    144 if.end:                                           ; preds = %entry, %if.then
    145   %res.0 = phi i32 [ %add3, %if.then ], [ 0, %entry ]
    146   ret i32 %res.0
    147 }
    148 
    149 
    150 ; CHECK: @test2
    151 ; CHECK: , uxtw #2]
    152 ; CHECK: , uxtw #2]
    153 define i32 @test2(i32* %array, i8 zeroext %c, i32 %arg) optsize {
    154 entry:
    155   %conv = zext i8 %c to i32
    156   %add = sub i32 0, %arg
    157   %tobool = icmp eq i32 %conv, %add
    158   br i1 %tobool, label %if.end, label %if.then
    159 
    160 if.then:                                          ; preds = %entry
    161   %idxprom = zext i8 %c to i64
    162   %arrayidx = getelementptr inbounds i32, i32* %array, i64 %idxprom
    163   %0 = load volatile i32, i32* %arrayidx, align 4
    164   %1 = load volatile i32, i32* %arrayidx, align 4
    165   %add3 = add nsw i32 %1, %0
    166   br label %if.end
    167 
    168 if.end:                                           ; preds = %entry, %if.then
    169   %res.0 = phi i32 [ %add3, %if.then ], [ 0, %entry ]
    170   ret i32 %res.0
    171 }
    172