Home | History | Annotate | Download | only in IndVarSimplify
      1 ; RUN: opt < %s -indvars -S | FileCheck %s
      2 target triple = "aarch64--linux-gnu"
      3 
      4 ; Provide legal integer types.
      5 target datalayout = "n8:16:32:64"
      6 
      7 
      8 ; Check the loop exit i32 compare instruction and operand are widened to i64
      9 ; instead of truncating IV before its use in the i32 compare instruction.
     10 
     11 @idx = common global i32 0, align 4
     12 @e = common global i32 0, align 4
     13 @ptr = common global i32* null, align 8
     14 
     15 ; CHECK-LABEL: @test1
     16 ; CHECK: for.body.lr.ph:
     17 ; CHECK: sext i32
     18 ; CHECK: for.cond:
     19 ; CHECK: icmp slt i64
     20 ; CHECK: for.body:
     21 ; CHECK: phi i64
     22 
     23 define i32 @test1() {
     24 entry:
     25   store i32 -1, i32* @idx, align 4
     26   %0 = load i32, i32* @e, align 4
     27   %cmp4 = icmp slt i32 %0, 0
     28   br i1 %cmp4, label %for.end.loopexit, label %for.body.lr.ph
     29 
     30 for.body.lr.ph:
     31   %1 = load i32*, i32** @ptr, align 8
     32   %2 = load i32, i32* @e, align 4
     33   br label %for.body
     34 
     35 for.cond:
     36   %inc = add nsw i32 %i.05, 1
     37   %cmp = icmp slt i32 %i.05, %2
     38   br i1 %cmp, label %for.body, label %for.cond.for.end.loopexit_crit_edge
     39 
     40 for.body:
     41   %i.05 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.cond ]
     42   %idxprom = sext i32 %i.05 to i64
     43   %arrayidx = getelementptr inbounds i32, i32* %1, i64 %idxprom
     44   %3 = load i32, i32* %arrayidx, align 4
     45   %tobool = icmp eq i32 %3, 0
     46   br i1 %tobool, label %if.then, label %for.cond
     47 
     48 if.then:
     49   %i.05.lcssa = phi i32 [ %i.05, %for.body ]
     50   store i32 %i.05.lcssa, i32* @idx, align 4
     51   br label %for.end
     52 
     53 for.cond.for.end.loopexit_crit_edge:
     54   br label %for.end.loopexit
     55 
     56 for.end.loopexit:
     57   br label %for.end
     58 
     59 for.end:
     60   %4 = load i32, i32* @idx, align 4
     61   ret i32 %4
     62 }
     63 
     64 ; CHECK-LABEL: @test2
     65 ; CHECK: for.body4.us
     66 ; CHECK: %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
     67 ; CHECK: %cmp2.us = icmp slt i64
     68 ; CHECK-NOT: %2 = trunc i64 %indvars.iv.next to i32
     69 ; CHECK-NOT: %cmp2.us = icmp slt i32
     70 
     71 define void @test2([8 x i8]* %a, i8* %b, i8 %limit) {
     72 entry:
     73   %conv = zext i8 %limit to i32
     74   br i1 undef, label %for.cond1.preheader, label %for.cond1.preheader.us
     75 
     76 for.cond1.preheader.us:
     77   %storemerge5.us = phi i32 [ 0, %entry ], [ %inc14.us, %for.inc13.us ]
     78   br i1 true, label %for.body4.lr.ph.us, label %for.inc13.us
     79 
     80 for.inc13.us:
     81   %inc14.us = add nsw i32 %storemerge5.us, 1
     82   %cmp.us = icmp slt i32 %inc14.us, 4
     83   br i1 %cmp.us, label %for.cond1.preheader.us, label %for.end
     84 
     85 for.body4.us:
     86   %storemerge14.us = phi i32 [ 0, %for.body4.lr.ph.us ], [ %inc.us, %for.body4.us ]
     87   %idxprom.us = sext i32 %storemerge14.us to i64
     88   %arrayidx6.us = getelementptr inbounds [8 x i8], [8 x i8]* %a, i64 %idxprom5.us, i64 %idxprom.us
     89   %0 = load i8, i8* %arrayidx6.us, align 1
     90   %idxprom7.us = zext i8 %0 to i64
     91   %arrayidx8.us = getelementptr inbounds i8, i8* %b, i64 %idxprom7.us
     92   %1 = load i8, i8* %arrayidx8.us, align 1
     93   store i8 %1, i8* %arrayidx6.us, align 1
     94   %inc.us = add nsw i32 %storemerge14.us, 1
     95   %cmp2.us = icmp slt i32 %inc.us, %conv
     96   br i1 %cmp2.us, label %for.body4.us, label %for.inc13.us
     97 
     98 for.body4.lr.ph.us:
     99   %idxprom5.us = sext i32 %storemerge5.us to i64
    100   br label %for.body4.us
    101 
    102 for.cond1.preheader:
    103   %storemerge5 = phi i32 [ 0, %entry ], [ %inc14, %for.inc13 ]
    104   br i1 false, label %for.inc13, label %for.inc13
    105 
    106 for.inc13:
    107   %inc14 = add nsw i32 %storemerge5, 1
    108   %cmp = icmp slt i32 %inc14, 4
    109   br i1 %cmp, label %for.cond1.preheader, label %for.end
    110 
    111 for.end:
    112   ret void
    113 }
    114 
    115 ; CHECK-LABEL: @test3
    116 ; CHECK: sext i32 %b
    117 ; CHECK: for.cond:
    118 ; CHECK: phi i64
    119 ; CHECK: icmp slt i64
    120 
    121 define i32 @test3(i32* %a, i32 %b) {
    122 entry:
    123   br label %for.cond
    124 
    125 for.cond:
    126   %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.body ]
    127   %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
    128   %cmp = icmp slt i32 %i.0, %b
    129   br i1 %cmp, label %for.body, label %for.end
    130 
    131 for.body:
    132   %idxprom = sext i32 %i.0 to i64
    133   %arrayidx = getelementptr inbounds i32, i32* %a, i64 %idxprom
    134   %0 = load i32, i32* %arrayidx, align 4
    135   %add = add nsw i32 %sum.0, %0
    136   %inc = add nsw i32 %i.0, 1
    137   br label %for.cond
    138 
    139 for.end:
    140   ret i32 %sum.0
    141 }
    142 
    143 declare i32 @fn1(i8 signext)
    144 
    145 ; PR21030
    146 ; CHECK-LABEL: @test4
    147 ; CHECK: for.body:
    148 ; CHECK: phi i32
    149 ; CHECK: icmp sgt i8
    150 
    151 define i32 @test4(i32 %a) {
    152 entry:
    153   br label %for.body
    154 
    155 for.body:
    156   %c.07 = phi i8 [ -3, %entry ], [ %dec, %for.body ]
    157   %conv6 = zext i8 %c.07 to i32
    158   %or = or i32 %a, %conv6
    159   %conv3 = trunc i32 %or to i8
    160   %call = call i32 @fn1(i8 signext %conv3)
    161   %dec = add i8 %c.07, -1
    162   %cmp = icmp sgt i8 %dec, -14
    163   br i1 %cmp, label %for.body, label %for.end
    164 
    165 for.end:
    166   ret i32 0
    167 }
    168 
    169 ; CHECK-LABEL: @test5
    170 ; CHECK: zext i32 %b
    171 ; CHECK: for.cond:
    172 ; CHECK: phi i64
    173 ; CHECK: icmp ule i64
    174 
    175 define i32 @test5(i32* %a, i32 %b) {
    176 entry:
    177   br label %for.cond
    178 
    179 for.cond:
    180   %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.body ]
    181   %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
    182   %cmp = icmp ule i32 %i.0, %b
    183   br i1 %cmp, label %for.body, label %for.end
    184 
    185 for.body:
    186   %idxprom = zext i32 %i.0 to i64
    187   %arrayidx = getelementptr inbounds i32, i32* %a, i64 %idxprom
    188   %0 = load i32, i32* %arrayidx, align 4
    189   %add = add nsw i32 %sum.0, %0
    190   %inc = add nsw i32 %i.0, 1
    191   br label %for.cond
    192 
    193 for.end:
    194   ret i32 %sum.0
    195 }
    196 
    197 define i32 @test6(i32* %a, i32 %b) {
    198 ; CHECK-LABEL: @test6(
    199 ; CHECK: [[B_SEXT:%[a-z0-9]+]] = sext i32 %b to i64
    200 ; CHECK: for.cond:
    201 ; CHECK: icmp sle i64 %indvars.iv, [[B_SEXT]]
    202 
    203 entry:
    204   br label %for.cond
    205 
    206 for.cond:
    207   %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.body ]
    208   %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
    209   %cmp = icmp sle i32 %i.0, %b
    210   br i1 %cmp, label %for.body, label %for.end
    211 
    212 for.body:
    213   %idxprom = zext i32 %i.0 to i64
    214   %arrayidx = getelementptr inbounds i32, i32* %a, i64 %idxprom
    215   %0 = load i32, i32* %arrayidx, align 4
    216   %add = add nsw i32 %sum.0, %0
    217   %inc = add nsw i32 %i.0, 1
    218   br label %for.cond
    219 
    220 for.end:
    221   ret i32 %sum.0
    222 }
    223 
    224 define i32 @test7(i32* %a, i32 %b) {
    225 ; CHECK-LABEL: @test7(
    226 ; CHECK: [[B_ZEXT:%[a-z0-9]+]] = zext i32 %b to i64
    227 ; CHECK: [[B_SEXT:%[a-z0-9]+]] = sext i32 %b to i64
    228 ; CHECK: for.cond:
    229 ; CHECK: icmp ule i64 %indvars.iv, [[B_ZEXT]]
    230 ; CHECK: for.body:
    231 ; CHECK: icmp sle i64 %indvars.iv, [[B_SEXT]]
    232 
    233 entry:
    234   br label %for.cond
    235 
    236 for.cond:
    237   %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.body ]
    238   %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
    239   %cmp = icmp ule i32 %i.0, %b
    240   br i1 %cmp, label %for.body, label %for.end
    241 
    242 for.body:
    243   %idxprom = sext i32 %i.0 to i64
    244   %arrayidx = getelementptr inbounds i32, i32* %a, i64 %idxprom
    245   %0 = load i32, i32* %arrayidx, align 4
    246   %add = add nsw i32 %sum.0, %0
    247   %inc = add nsw i32 %i.0, 1
    248   %cmp2 = icmp sle i32 %i.0, %b
    249   br i1 %cmp2, label %for.cond, label %for.end
    250 
    251 for.end:
    252   ret i32 %sum.0
    253 }
    254 
    255 define i32 @test8(i32* %a, i32 %b, i32 %init) {
    256 ; CHECK-LABEL: @test8(
    257 ; CHECK: [[INIT_SEXT:%[a-z0-9]+]] = sext i32 %init to i64
    258 ; CHECK: [[B_ZEXT:%[a-z0-9]+]] = zext i32 %b to i64
    259 ; CHECK: for.cond:
    260 ;     Note: %indvars.iv is the sign extension of %i.0
    261 ; CHECK: %indvars.iv = phi i64 [ [[INIT_SEXT]], %for.cond.preheader ], [ %indvars.iv.next, %for.body ]
    262 ; CHECK: icmp ule i64 %indvars.iv, [[B_ZEXT]]
    263 
    264 entry:
    265   %e = icmp sgt i32 %init, 0
    266   br i1 %e, label %for.cond, label %leave
    267 
    268 for.cond:
    269   %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.body ]
    270   %i.0 = phi i32 [ %init, %entry ], [ %inc, %for.body ]
    271   %cmp = icmp ule i32 %i.0, %b
    272   br i1 %cmp, label %for.body, label %for.end
    273 
    274 for.body:
    275   %idxprom = sext i32 %i.0 to i64
    276   %arrayidx = getelementptr inbounds i32, i32* %a, i64 %idxprom
    277   %0 = load i32, i32* %arrayidx, align 4
    278   %add = add nsw i32 %sum.0, %0
    279   %inc = add nsw i32 %i.0, 1
    280   %cmp2 = icmp slt i32 0, %inc
    281   br i1 %cmp2, label %for.cond, label %for.end
    282 
    283 for.end:
    284   ret i32 %sum.0
    285 
    286 leave:
    287   ret i32 0
    288 }
    289 
    290 define i32 @test9(i32* %a, i32 %b, i32 %init) {
    291 ; CHECK-LABEL: @test9(
    292 ; CHECK: [[INIT_ZEXT:%[a-z0-9]+]] = zext i32 %init to i64
    293 ; CHECK: [[B_SEXT:%[a-z0-9]+]] = sext i32 %b to i64
    294 ; CHECK: for.cond:
    295 ;     Note: %indvars.iv is the zero extension of %i.0
    296 ; CHECK: %indvars.iv = phi i64 [ [[INIT_ZEXT]], %for.cond.preheader ], [ %indvars.iv.next, %for.body ]
    297 ; CHECK: icmp slt i64 %indvars.iv, [[B_SEXT]]
    298 
    299 entry:
    300   %e = icmp sgt i32 %init, 0
    301   br i1 %e, label %for.cond, label %leave
    302 
    303 for.cond:
    304   %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.body ]
    305   %i.0 = phi i32 [ %init, %entry ], [ %inc, %for.body ]
    306   %cmp = icmp slt i32 %i.0, %b
    307   br i1 %cmp, label %for.body, label %for.end
    308 
    309 for.body:
    310   %idxprom = zext i32 %i.0 to i64
    311   %arrayidx = getelementptr inbounds i32, i32* %a, i64 %idxprom
    312   %0 = load i32, i32* %arrayidx, align 4
    313   %add = add nsw i32 %sum.0, %0
    314   %inc = add nsw i32 %i.0, 1
    315   %cmp2 = icmp slt i32 0, %inc
    316   br i1 %cmp2, label %for.cond, label %for.end
    317 
    318 for.end:
    319   ret i32 %sum.0
    320 
    321 leave:
    322   ret i32 0
    323 }
    324 
    325 declare void @consume.i64(i64)
    326 declare void @consume.i1(i1)
    327 
    328 define i32 @test10(i32 %v) {
    329 ; CHECK-LABEL: @test10(
    330  entry:
    331 ; CHECK-NOT: zext
    332   br label %loop
    333 
    334  loop:
    335 ; CHECK: loop:
    336 ; CHECK: %indvars.iv = phi i64 [ %indvars.iv.next, %loop ], [ 0, %entry ]
    337 ; CHECK: %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
    338 ; CHECK: [[MUL:%[a-z0-9]+]] = mul nsw i64 %indvars.iv, -1
    339 ; CHECK: [[MUL_TRUNC:%[a-z0-9]+]] = trunc i64 [[MUL]] to i32
    340 ; CHECK: [[CMP:%[a-z0-9]+]] = icmp eq i32 [[MUL_TRUNC]], %v
    341 ; CHECK: call void @consume.i1(i1 [[CMP]])
    342 
    343   %i = phi i32 [ 0, %entry ], [ %i.inc, %loop ]
    344   %i.inc = add i32 %i, 1
    345   %iv = mul i32 %i, -1
    346   %cmp = icmp eq i32 %iv, %v
    347   call void @consume.i1(i1 %cmp)
    348   %be.cond = icmp slt i32 %i.inc, 11
    349   %ext = sext i32 %iv to i64
    350   call void @consume.i64(i64 %ext)
    351   br i1 %be.cond, label %loop, label %leave
    352 
    353  leave:
    354   ret i32 22
    355 }
    356