sysdeps/i386/stpncpy.S

   1 /* stpncpy -- copy no more then N bytes from SRC to DEST, returning the
   2               address of the terminating '\0' in DEST.
   3 For Intel 80x86, x>=3.
   4 Copyright (C) 1994, 1995 Free Software Foundation, Inc.
   5 Contributed by Ulrich Drepper <drepper@gnu.ai.mit.edu>
   6 Some bug fixes by Alan Modra <Alan@SPRI.Levels.UniSA.Edu.Au>
   7   - original wrote n+1 chars in some cases.
   8   - stpncpy() ought to behave like strncpy() ie. not null-terminate
   9     if limited by n.  glibc-1.09 stpncpy() does this.
  10 This file is part of the GNU C Library.
  11
  12 The GNU C Library is free software; you can redistribute it and/or
  13 modify it under the terms of the GNU Library General Public License as
  14 published by the Free Software Foundation; either version 2 of the
  15 License, or (at your option) any later version.
  16
  17 The GNU C Library is distributed in the hope that it will be useful,
  18 but WITHOUT ANY WARRANTY; without even the implied warranty of
  19 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  20 Library General Public License for more details.
  21
  22 You should have received a copy of the GNU Library General Public
  23 License along with the GNU C Library; see the file COPYING.LIB.  If
  24 not, write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  25 Boston, MA 02111-1307, USA.  */
  26
  27 #include <sysdep.h>
  28 #include "asm-syntax.h"
  29
  30 /*
  31    INPUT PARAMETERS:
  32    dest         (sp + 4)
  33    src          (sp + 8)
  34    maxlen       (sp + 12)
  35 */
  36
  37         .text
  38 ENTRY (__stpncpy)
  39
  40         pushl %esi
  41
  42         movl 8(%esp), %eax      /* load destination pointer */
  43         movl 12(%esp), %esi     /* load source pointer */
  44         movl 16(%esp), %ecx     /* load maximal length */
  45
  46         subl %eax, %esi         /* magic: reduce number of loop variants
  47                                    to one using addressing mode */
  48         jmp L1                  /* jump to loop "head" */
  49
  50         ALIGN(4)
  51
  52         /* Four times unfolded loop with two loop counters.  We get the
  53            the third value (the source address) by using the index+base
  54            adressing mode.  */
  55 L2:     movb (%eax,%esi), %dl   /* load current char */
  56         movb %dl, (%eax)        /* and store it */
  57         testb %dl, %dl          /* was it NUL? */
  58         jz L7                   /* yes, then exit */
  59
  60         movb 1(%eax,%esi), %dl  /* load current char */
  61         movb %dl, 1(%eax)       /* and store it */
  62         testb %dl, %dl          /* was it NUL? */
  63         jz L6                   /* yes, then exit */
  64
  65         movb 2(%eax,%esi), %dl  /* load current char */
  66         movb %dl, 2(%eax)       /* and store it */
  67         testb %dl, %dl          /* was it NUL? */
  68         jz L5                   /* yes, then exit */
  69
  70         movb 3(%eax,%esi), %dl  /* load current char */
  71         movb %dl, 3(%eax)       /* and store it */
  72         testb %dl, %dl          /* was it NUL? */
  73         jz L4                   /* yes, then exit */
  74
  75         addl $4, %eax           /* increment loop counter for full round */
  76
  77 L1:     subl $4, %ecx           /* still more than 4 bytes allowed? */
  78         jae L2                  /* yes, then go to start of loop */
  79
  80         /* The maximal remaining 15 bytes are not processed in a loop.  */
  81
  82         addl $4, %ecx           /* correct above subtraction */
  83         jz L9                   /* maximal allowed char reached => go to end */
  84
  85         movb (%eax,%esi), %dl   /* load current char */
  86         movb %dl, (%eax)        /* and store it */
  87         testb %dl, %dl          /* was it NUL? */
  88         jz L3                   /* yes, then exit */
  89
  90         incl %eax               /* increment pointer */
  91         decl %ecx               /* decrement length counter */
  92         jz L9                   /* no more allowed => exit */
  93
  94         movb (%eax,%esi), %dl   /* load current char */
  95         movb %dl, (%eax)        /* and store it */
  96         testb %dl, %dl          /* was it NUL? */
  97         jz L3                   /* yes, then exit */
  98
  99         incl %eax               /* increment pointer */
 100         decl %ecx               /* decrement length counter */
 101         jz L9                   /* no more allowed => exit */
 102
 103         movb (%eax,%esi), %dl   /* load current char */
 104         movb %dl, (%eax)        /* and store it */
 105         testb %dl, %dl          /* was it NUL? */
 106         jz L3                   /* yes, then exit */
 107
 108         incl %eax               /* increment pointer */
 109         jmp L9                  /* we don't have to test for counter underflow
 110                                    because we know we had a most 3 bytes
 111                                    remaining => exit */
 112
 113         /* When coming from the main loop we have to adjust the pointer.  */
 114 L4:     decl %ecx               /* decrement counter */
 115         incl %eax               /* increment pointer */
 116
 117 L5:     decl %ecx               /* increment pointer */
 118         incl %eax               /* increment pointer */
 119
 120 L6:     decl %ecx               /* increment pointer */
 121         incl %eax               /* increment pointer */
 122 L7:
 123
 124         addl $3, %ecx           /* correct pre-decrementation of counter
 125                                    at the beginning of the loop; but why 3
 126                                    and not 4?  Very simple, we have to count
 127                                    the NUL char we already wrote.  */
 128         jz L9                   /* counter is also 0 => exit */
 129
 130         /* We now have to fill the rest of the buffer with NUL.  This
 131            is done in a tricky way.  Please note that the adressing mode
 132            used below is not the same we used above.  Here we use the
 133            %ecx register.  */
 134 L8:
 135         movb $0, (%ecx,%eax)    /* store NUL char */
 136 L3:     decl %ecx               /* all bytes written? */
 137         jnz L8                  /* no, then again */
 138
 139 L9:     popl %esi               /* restore saved register content */
 140
 141         ret
 142
 143 weak_alias (__stpncpy, stpncpy)