Home | History | Annotate | Download | only in X86 
      1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
      2 ; RUN: llc < %s -mtriple=i686-unknown | FileCheck %s --check-prefix=X32
      3 ; RUN: llc < %s -mtriple=x86_64-unknown | FileCheck %s --check-prefix=X64
      4 ; RUN: llc < %s -mtriple=i686-unknown -mattr=+popcnt | FileCheck %s --check-prefix=X32-POPCNT
      5 ; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+popcnt | FileCheck %s --check-prefix=X64-POPCNT
      6 
      7 define i8 @cnt8(i8 %x) nounwind readnone {
      8 ; X32-LABEL: cnt8:
      9 ; X32:       # %bb.0:
     10 ; X32-NEXT:    movb {{[0-9]+}}(%esp), %cl
     11 ; X32-NEXT:    movl %ecx, %eax
     12 ; X32-NEXT:    shrb %al
     13 ; X32-NEXT:    andb $85, %al
     14 ; X32-NEXT:    subb %al, %cl
     15 ; X32-NEXT:    movl %ecx, %eax
     16 ; X32-NEXT:    andb $51, %al
     17 ; X32-NEXT:    shrb $2, %cl
     18 ; X32-NEXT:    andb $51, %cl
     19 ; X32-NEXT:    addb %al, %cl
     20 ; X32-NEXT:    movl %ecx, %eax
     21 ; X32-NEXT:    shrb $4, %al
     22 ; X32-NEXT:    addb %cl, %al
     23 ; X32-NEXT:    andb $15, %al
     24 ; X32-NEXT:    retl
     25 ;
     26 ; X64-LABEL: cnt8:
     27 ; X64:       # %bb.0:
     28 ; X64-NEXT:    movl %edi, %eax
     29 ; X64-NEXT:    shrb %al
     30 ; X64-NEXT:    andb $85, %al
     31 ; X64-NEXT:    subb %al, %dil
     32 ; X64-NEXT:    movl %edi, %eax
     33 ; X64-NEXT:    andb $51, %al
     34 ; X64-NEXT:    shrb $2, %dil
     35 ; X64-NEXT:    andb $51, %dil
     36 ; X64-NEXT:    addb %al, %dil
     37 ; X64-NEXT:    movl %edi, %eax
     38 ; X64-NEXT:    shrb $4, %al
     39 ; X64-NEXT:    addb %dil, %al
     40 ; X64-NEXT:    andb $15, %al
     41 ; X64-NEXT:    retq
     42 ;
     43 ; X32-POPCNT-LABEL: cnt8:
     44 ; X32-POPCNT:       # %bb.0:
     45 ; X32-POPCNT-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
     46 ; X32-POPCNT-NEXT:    popcntl %eax, %eax
     47 ; X32-POPCNT-NEXT:    # kill: def $al killed $al killed $eax
     48 ; X32-POPCNT-NEXT:    retl
     49 ;
     50 ; X64-POPCNT-LABEL: cnt8:
     51 ; X64-POPCNT:       # %bb.0:
     52 ; X64-POPCNT-NEXT:    movzbl %dil, %eax
     53 ; X64-POPCNT-NEXT:    popcntl %eax, %eax
     54 ; X64-POPCNT-NEXT:    # kill: def $al killed $al killed $eax
     55 ; X64-POPCNT-NEXT:    retq
     56   %cnt = tail call i8 @llvm.ctpop.i8(i8 %x)
     57   ret i8 %cnt
     58 }
     59 
     60 define i16 @cnt16(i16 %x) nounwind readnone {
     61 ; X32-LABEL: cnt16:
     62 ; X32:       # %bb.0:
     63 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
     64 ; X32-NEXT:    movl %eax, %ecx
     65 ; X32-NEXT:    shrl %ecx
     66 ; X32-NEXT:    andl $21845, %ecx # imm = 0x5555
     67 ; X32-NEXT:    subl %ecx, %eax
     68 ; X32-NEXT:    movl %eax, %ecx
     69 ; X32-NEXT:    andl $13107, %ecx # imm = 0x3333
     70 ; X32-NEXT:    shrl $2, %eax
     71 ; X32-NEXT:    andl $13107, %eax # imm = 0x3333
     72 ; X32-NEXT:    addl %ecx, %eax
     73 ; X32-NEXT:    movl %eax, %ecx
     74 ; X32-NEXT:    shrl $4, %ecx
     75 ; X32-NEXT:    addl %eax, %ecx
     76 ; X32-NEXT:    andl $3855, %ecx # imm = 0xF0F
     77 ; X32-NEXT:    movl %ecx, %eax
     78 ; X32-NEXT:    shll $8, %eax
     79 ; X32-NEXT:    addl %ecx, %eax
     80 ; X32-NEXT:    movzbl %ah, %eax
     81 ; X32-NEXT:    # kill: def $ax killed $ax killed $eax
     82 ; X32-NEXT:    retl
     83 ;
     84 ; X64-LABEL: cnt16:
     85 ; X64:       # %bb.0:
     86 ; X64-NEXT:    movl %edi, %eax
     87 ; X64-NEXT:    shrl %eax
     88 ; X64-NEXT:    andl $21845, %eax # imm = 0x5555
     89 ; X64-NEXT:    subl %eax, %edi
     90 ; X64-NEXT:    movl %edi, %eax
     91 ; X64-NEXT:    andl $13107, %eax # imm = 0x3333
     92 ; X64-NEXT:    shrl $2, %edi
     93 ; X64-NEXT:    andl $13107, %edi # imm = 0x3333
     94 ; X64-NEXT:    addl %eax, %edi
     95 ; X64-NEXT:    movl %edi, %eax
     96 ; X64-NEXT:    shrl $4, %eax
     97 ; X64-NEXT:    addl %edi, %eax
     98 ; X64-NEXT:    andl $3855, %eax # imm = 0xF0F
     99 ; X64-NEXT:    movl %eax, %ecx
    100 ; X64-NEXT:    shll $8, %ecx
    101 ; X64-NEXT:    addl %eax, %ecx
    102 ; X64-NEXT:    movzbl %ch, %eax
    103 ; X64-NEXT:    # kill: def $ax killed $ax killed $eax
    104 ; X64-NEXT:    retq
    105 ;
    106 ; X32-POPCNT-LABEL: cnt16:
    107 ; X32-POPCNT:       # %bb.0:
    108 ; X32-POPCNT-NEXT:    popcntw {{[0-9]+}}(%esp), %ax
    109 ; X32-POPCNT-NEXT:    retl
    110 ;
    111 ; X64-POPCNT-LABEL: cnt16:
    112 ; X64-POPCNT:       # %bb.0:
    113 ; X64-POPCNT-NEXT:    popcntw %di, %ax
    114 ; X64-POPCNT-NEXT:    retq
    115   %cnt = tail call i16 @llvm.ctpop.i16(i16 %x)
    116   ret i16 %cnt
    117 }
    118 
    119 define i32 @cnt32(i32 %x) nounwind readnone {
    120 ; X32-LABEL: cnt32:
    121 ; X32:       # %bb.0:
    122 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
    123 ; X32-NEXT:    movl %eax, %ecx
    124 ; X32-NEXT:    shrl %ecx
    125 ; X32-NEXT:    andl $1431655765, %ecx # imm = 0x55555555
    126 ; X32-NEXT:    subl %ecx, %eax
    127 ; X32-NEXT:    movl %eax, %ecx
    128 ; X32-NEXT:    andl $858993459, %ecx # imm = 0x33333333
    129 ; X32-NEXT:    shrl $2, %eax
    130 ; X32-NEXT:    andl $858993459, %eax # imm = 0x33333333
    131 ; X32-NEXT:    addl %ecx, %eax
    132 ; X32-NEXT:    movl %eax, %ecx
    133 ; X32-NEXT:    shrl $4, %ecx
    134 ; X32-NEXT:    addl %eax, %ecx
    135 ; X32-NEXT:    andl $252645135, %ecx # imm = 0xF0F0F0F
    136 ; X32-NEXT:    imull $16843009, %ecx, %eax # imm = 0x1010101
    137 ; X32-NEXT:    shrl $24, %eax
    138 ; X32-NEXT:    retl
    139 ;
    140 ; X64-LABEL: cnt32:
    141 ; X64:       # %bb.0:
    142 ; X64-NEXT:    movl %edi, %eax
    143 ; X64-NEXT:    shrl %eax
    144 ; X64-NEXT:    andl $1431655765, %eax # imm = 0x55555555
    145 ; X64-NEXT:    subl %eax, %edi
    146 ; X64-NEXT:    movl %edi, %eax
    147 ; X64-NEXT:    andl $858993459, %eax # imm = 0x33333333
    148 ; X64-NEXT:    shrl $2, %edi
    149 ; X64-NEXT:    andl $858993459, %edi # imm = 0x33333333
    150 ; X64-NEXT:    addl %eax, %edi
    151 ; X64-NEXT:    movl %edi, %eax
    152 ; X64-NEXT:    shrl $4, %eax
    153 ; X64-NEXT:    addl %edi, %eax
    154 ; X64-NEXT:    andl $252645135, %eax # imm = 0xF0F0F0F
    155 ; X64-NEXT:    imull $16843009, %eax, %eax # imm = 0x1010101
    156 ; X64-NEXT:    shrl $24, %eax
    157 ; X64-NEXT:    retq
    158 ;
    159 ; X32-POPCNT-LABEL: cnt32:
    160 ; X32-POPCNT:       # %bb.0:
    161 ; X32-POPCNT-NEXT:    popcntl {{[0-9]+}}(%esp), %eax
    162 ; X32-POPCNT-NEXT:    retl
    163 ;
    164 ; X64-POPCNT-LABEL: cnt32:
    165 ; X64-POPCNT:       # %bb.0:
    166 ; X64-POPCNT-NEXT:    popcntl %edi, %eax
    167 ; X64-POPCNT-NEXT:    retq
    168   %cnt = tail call i32 @llvm.ctpop.i32(i32 %x)
    169   ret i32 %cnt
    170 }
    171 
    172 define i64 @cnt64(i64 %x) nounwind readnone {
    173 ; X32-LABEL: cnt64:
    174 ; X32:       # %bb.0:
    175 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
    176 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
    177 ; X32-NEXT:    movl %ecx, %edx
    178 ; X32-NEXT:    shrl %edx
    179 ; X32-NEXT:    andl $1431655765, %edx # imm = 0x55555555
    180 ; X32-NEXT:    subl %edx, %ecx
    181 ; X32-NEXT:    movl %ecx, %edx
    182 ; X32-NEXT:    andl $858993459, %edx # imm = 0x33333333
    183 ; X32-NEXT:    shrl $2, %ecx
    184 ; X32-NEXT:    andl $858993459, %ecx # imm = 0x33333333
    185 ; X32-NEXT:    addl %edx, %ecx
    186 ; X32-NEXT:    movl %ecx, %edx
    187 ; X32-NEXT:    shrl $4, %edx
    188 ; X32-NEXT:    addl %ecx, %edx
    189 ; X32-NEXT:    andl $252645135, %edx # imm = 0xF0F0F0F
    190 ; X32-NEXT:    imull $16843009, %edx, %ecx # imm = 0x1010101
    191 ; X32-NEXT:    shrl $24, %ecx
    192 ; X32-NEXT:    movl %eax, %edx
    193 ; X32-NEXT:    shrl %edx
    194 ; X32-NEXT:    andl $1431655765, %edx # imm = 0x55555555
    195 ; X32-NEXT:    subl %edx, %eax
    196 ; X32-NEXT:    movl %eax, %edx
    197 ; X32-NEXT:    andl $858993459, %edx # imm = 0x33333333
    198 ; X32-NEXT:    shrl $2, %eax
    199 ; X32-NEXT:    andl $858993459, %eax # imm = 0x33333333
    200 ; X32-NEXT:    addl %edx, %eax
    201 ; X32-NEXT:    movl %eax, %edx
    202 ; X32-NEXT:    shrl $4, %edx
    203 ; X32-NEXT:    addl %eax, %edx
    204 ; X32-NEXT:    andl $252645135, %edx # imm = 0xF0F0F0F
    205 ; X32-NEXT:    imull $16843009, %edx, %eax # imm = 0x1010101
    206 ; X32-NEXT:    shrl $24, %eax
    207 ; X32-NEXT:    addl %ecx, %eax
    208 ; X32-NEXT:    xorl %edx, %edx
    209 ; X32-NEXT:    retl
    210 ;
    211 ; X64-LABEL: cnt64:
    212 ; X64:       # %bb.0:
    213 ; X64-NEXT:    movq %rdi, %rax
    214 ; X64-NEXT:    shrq %rax
    215 ; X64-NEXT:    movabsq $6148914691236517205, %rcx # imm = 0x5555555555555555
    216 ; X64-NEXT:    andq %rax, %rcx
    217 ; X64-NEXT:    subq %rcx, %rdi
    218 ; X64-NEXT:    movabsq $3689348814741910323, %rax # imm = 0x3333333333333333
    219 ; X64-NEXT:    movq %rdi, %rcx
    220 ; X64-NEXT:    andq %rax, %rcx
    221 ; X64-NEXT:    shrq $2, %rdi
    222 ; X64-NEXT:    andq %rax, %rdi
    223 ; X64-NEXT:    addq %rcx, %rdi
    224 ; X64-NEXT:    movq %rdi, %rax
    225 ; X64-NEXT:    shrq $4, %rax
    226 ; X64-NEXT:    leaq (%rax,%rdi), %rax
    227 ; X64-NEXT:    movabsq $1085102592571150095, %rcx # imm = 0xF0F0F0F0F0F0F0F
    228 ; X64-NEXT:    andq %rax, %rcx
    229 ; X64-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
    230 ; X64-NEXT:    imulq %rcx, %rax
    231 ; X64-NEXT:    shrq $56, %rax
    232 ; X64-NEXT:    retq
    233 ;
    234 ; X32-POPCNT-LABEL: cnt64:
    235 ; X32-POPCNT:       # %bb.0:
    236 ; X32-POPCNT-NEXT:    popcntl {{[0-9]+}}(%esp), %ecx
    237 ; X32-POPCNT-NEXT:    popcntl {{[0-9]+}}(%esp), %eax
    238 ; X32-POPCNT-NEXT:    addl %ecx, %eax
    239 ; X32-POPCNT-NEXT:    xorl %edx, %edx
    240 ; X32-POPCNT-NEXT:    retl
    241 ;
    242 ; X64-POPCNT-LABEL: cnt64:
    243 ; X64-POPCNT:       # %bb.0:
    244 ; X64-POPCNT-NEXT:    popcntq %rdi, %rax
    245 ; X64-POPCNT-NEXT:    retq
    246   %cnt = tail call i64 @llvm.ctpop.i64(i64 %x)
    247   ret i64 %cnt
    248 }
    249 
    250 declare i8 @llvm.ctpop.i8(i8) nounwind readnone
    251 declare i16 @llvm.ctpop.i16(i16) nounwind readnone
    252 declare i32 @llvm.ctpop.i32(i32) nounwind readnone
    253 declare i64 @llvm.ctpop.i64(i64) nounwind readnone
    254