sysdeps/i386/add_n.S

   1 /* Add two limb vectors of the same length > 0 and store sum in a third
   2    limb vector.
   3    Copyright (C) 1992-2013 Free Software Foundation, Inc.
   4    This file is part of the GNU MP Library.
   5
   6    The GNU MP Library is free software; you can redistribute it and/or modify
   7    it under the terms of the GNU Lesser General Public License as published by
   8    the Free Software Foundation; either version 2.1 of the License, or (at your
   9    option) any later version.
  10
  11    The GNU MP Library is distributed in the hope that it will be useful, but
  12    WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  13    or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
  14    License for more details.
  15
  16    You should have received a copy of the GNU Lesser General Public License
  17    along with the GNU MP Library; see the file COPYING.LIB.  If not,
  18    see <http://www.gnu.org/licenses/>.  */
  19
  20 #include "sysdep.h"
  21 #include "asm-syntax.h"
  22
  23 #define PARMS   4+8     /* space for 2 saved regs */
  24 #define RES     PARMS
  25 #define S1      RES+4
  26 #define S2      S1+4
  27 #define SIZE    S2+4
  28
  29         .text
  30 ENTRY (__mpn_add_n)
  31
  32         pushl %edi
  33         cfi_adjust_cfa_offset (4)
  34         pushl %esi
  35         cfi_adjust_cfa_offset (4)
  36
  37         movl RES(%esp),%edi
  38         cfi_rel_offset (edi, 4)
  39         movl S1(%esp),%esi
  40         cfi_rel_offset (esi, 0)
  41         movl S2(%esp),%edx
  42         movl SIZE(%esp),%ecx
  43         movl    %ecx,%eax
  44         shrl    $3,%ecx                 /* compute count for unrolled loop */
  45         negl    %eax
  46         andl    $7,%eax                 /* get index where to start loop */
  47         jz      L(oop)                  /* necessary special case for 0 */
  48         incl    %ecx                    /* adjust loop count */
  49         shll    $2,%eax                 /* adjustment for pointers... */
  50         subl    %eax,%edi               /* ... since they are offset ... */
  51         subl    %eax,%esi               /* ... by a constant when we ... */
  52         subl    %eax,%edx               /* ... enter the loop */
  53         shrl    $2,%eax                 /* restore previous value */
  54 #ifdef PIC
  55 /* Calculate start address in loop for PIC.  Due to limitations in some
  56    assemblers, Loop-L0-3 cannot be put into the leal */
  57         call    L(0)
  58         cfi_adjust_cfa_offset (4)
  59 L(0):   leal    (%eax,%eax,8),%eax
  60         addl    (%esp),%eax
  61         addl    $(L(oop)-L(0)-3),%eax
  62         addl    $4,%esp
  63         cfi_adjust_cfa_offset (-4)
  64 #else
  65 /* Calculate start address in loop for non-PIC.  */
  66         leal    (L(oop) - 3)(%eax,%eax,8),%eax
  67 #endif
  68         jmp     *%eax                   /* jump into loop */
  69         ALIGN (3)
  70 L(oop): movl    (%esi),%eax
  71         adcl    (%edx),%eax
  72         movl    %eax,(%edi)
  73         movl    4(%esi),%eax
  74         adcl    4(%edx),%eax
  75         movl    %eax,4(%edi)
  76         movl    8(%esi),%eax
  77         adcl    8(%edx),%eax
  78         movl    %eax,8(%edi)
  79         movl    12(%esi),%eax
  80         adcl    12(%edx),%eax
  81         movl    %eax,12(%edi)
  82         movl    16(%esi),%eax
  83         adcl    16(%edx),%eax
  84         movl    %eax,16(%edi)
  85         movl    20(%esi),%eax
  86         adcl    20(%edx),%eax
  87         movl    %eax,20(%edi)
  88         movl    24(%esi),%eax
  89         adcl    24(%edx),%eax
  90         movl    %eax,24(%edi)
  91         movl    28(%esi),%eax
  92         adcl    28(%edx),%eax
  93         movl    %eax,28(%edi)
  94         leal    32(%edi),%edi
  95         leal    32(%esi),%esi
  96         leal    32(%edx),%edx
  97         decl    %ecx
  98         jnz     L(oop)
  99
 100         sbbl    %eax,%eax
 101         negl    %eax
 102
 103         popl %esi
 104         cfi_adjust_cfa_offset (-4)
 105         cfi_restore (esi)
 106         popl %edi
 107         cfi_adjust_cfa_offset (-4)
 108         cfi_restore (edi)
 109
 110         ret
 111 END (__mpn_add_n)