sysdeps/i386/i586/memcpy.S

   1 /* Highly optimized version for i586.
   2    Copyright (C) 1997-2021 Free Software Foundation, Inc.
   3    This file is part of the GNU C Library.
   4
   5    The GNU C Library is free software; you can redistribute it and/or
   6    modify it under the terms of the GNU Lesser General Public
   7    License as published by the Free Software Foundation; either
   8    version 2.1 of the License, or (at your option) any later version.
   9
  10    The GNU C Library is distributed in the hope that it will be useful,
  11    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13    Lesser General Public License for more details.
  14
  15    You should have received a copy of the GNU Lesser General Public
  16    License along with the GNU C Library; if not, see
  17    <https://www.gnu.org/licenses/>.  */
  18
  19 #include <sysdep.h>
  20 #include "asm-syntax.h"
  21
  22 #define PARMS   4+8     /* space for 2 saved regs */
  23 #define RTN     PARMS
  24 #define DEST    RTN
  25 #define SRC     DEST+4
  26 #define LEN     SRC+4
  27
  28         .text
  29 #if defined PIC && IS_IN (libc)
  30 ENTRY (__memcpy_chk)
  31         movl    12(%esp), %eax
  32         cmpl    %eax, 16(%esp)
  33         jb      HIDDEN_JUMPTARGET (__chk_fail)
  34 END (__memcpy_chk)
  35 #endif
  36 ENTRY (memcpy)
  37
  38         pushl   %edi
  39         cfi_adjust_cfa_offset (4)
  40         pushl   %esi
  41         cfi_adjust_cfa_offset (4)
  42
  43         movl    DEST(%esp), %edi
  44         cfi_rel_offset (edi, 4)
  45         movl    SRC(%esp), %esi
  46         cfi_rel_offset (esi, 0)
  47         movl    LEN(%esp), %ecx
  48         movl    %edi, %eax
  49
  50         /* We need this in any case.  */
  51         cld
  52
  53         /* Cutoff for the big loop is a size of 32 bytes since otherwise
  54            the loop will never be entered.  */
  55         cmpl    $32, %ecx
  56         jbe     L(1)
  57
  58         negl    %eax
  59         andl    $3, %eax
  60         subl    %eax, %ecx
  61         xchgl   %eax, %ecx
  62
  63         rep; movsb
  64
  65         movl    %eax, %ecx
  66         subl    $32, %ecx
  67         js      L(2)
  68
  69         /* Read ahead to make sure we write in the cache since the stupid
  70            i586 designers haven't implemented read-on-write-miss.  */
  71         movl    (%edi), %eax
  72 L(3):   movl    28(%edi), %edx
  73
  74         /* Now correct the loop counter.  Please note that in the following
  75            code the flags are not changed anymore.  */
  76         subl    $32, %ecx
  77
  78         movl    (%esi), %eax
  79         movl    4(%esi), %edx
  80         movl    %eax, (%edi)
  81         movl    %edx, 4(%edi)
  82         movl    8(%esi), %eax
  83         movl    12(%esi), %edx
  84         movl    %eax, 8(%edi)
  85         movl    %edx, 12(%edi)
  86         movl    16(%esi), %eax
  87         movl    20(%esi), %edx
  88         movl    %eax, 16(%edi)
  89         movl    %edx, 20(%edi)
  90         movl    24(%esi), %eax
  91         movl    28(%esi), %edx
  92         movl    %eax, 24(%edi)
  93         movl    %edx, 28(%edi)
  94
  95         leal    32(%esi), %esi
  96         leal    32(%edi), %edi
  97
  98         jns     L(3)
  99
 100         /* Correct extra loop counter modification.  */
 101 L(2):   addl    $32, %ecx
 102 #ifndef USE_AS_MEMPCPY
 103         movl    DEST(%esp), %eax
 104 #endif
 105
 106 L(1):   rep; movsb
 107
 108 #ifdef USE_AS_MEMPCPY
 109         movl    %edi, %eax
 110 #endif
 111
 112         popl    %esi
 113         cfi_adjust_cfa_offset (-4)
 114         cfi_restore (esi)
 115         popl    %edi
 116         cfi_adjust_cfa_offset (-4)
 117         cfi_restore (edi)
 118
 119         ret
 120 END (memcpy)
 121 #ifndef USE_AS_MEMPCPY
 122 libc_hidden_builtin_def (memcpy)
 123 #endif