source/libs/gmp/gmp-src/mpn/x86/aors_n.asm

   1 dnl  x86 mpn_add_n/mpn_sub_n -- mpn addition and subtraction.
   2
   3 dnl  Copyright 1992, 1994-1996, 1999-2002 Free Software Foundation, Inc.
   4
   5 dnl  This file is part of the GNU MP Library.
   6 dnl
   7 dnl  The GNU MP Library is free software; you can redistribute it and/or modify
   8 dnl  it under the terms of either:
   9 dnl
  10 dnl    * the GNU Lesser General Public License as published by the Free
  11 dnl      Software Foundation; either version 3 of the License, or (at your
  12 dnl      option) any later version.
  13 dnl
  14 dnl  or
  15 dnl
  16 dnl    * the GNU General Public License as published by the Free Software
  17 dnl      Foundation; either version 2 of the License, or (at your option) any
  18 dnl      later version.
  19 dnl
  20 dnl  or both in parallel, as here.
  21 dnl
  22 dnl  The GNU MP Library is distributed in the hope that it will be useful, but
  23 dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  24 dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  25 dnl  for more details.
  26 dnl
  27 dnl  You should have received copies of the GNU General Public License and the
  28 dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
  29 dnl  see https://www.gnu.org/licenses/.
  30
  31 include(`../config.m4')
  32
  33
  34 C     cycles/limb
  35 C P5    3.375
  36 C P6    3.125
  37 C K6    3.5
  38 C K7    2.25
  39 C P4    8.75
  40
  41
  42 ifdef(`OPERATION_add_n',`
  43         define(M4_inst,        adcl)
  44         define(M4_function_n,  mpn_add_n)
  45         define(M4_function_nc, mpn_add_nc)
  46
  47 ',`ifdef(`OPERATION_sub_n',`
  48         define(M4_inst,        sbbl)
  49         define(M4_function_n,  mpn_sub_n)
  50         define(M4_function_nc, mpn_sub_nc)
  51
  52 ',`m4_error(`Need OPERATION_add_n or OPERATION_sub_n
  53 ')')')
  54
  55 MULFUNC_PROLOGUE(mpn_add_n mpn_add_nc mpn_sub_n mpn_sub_nc)
  56
  57
  58 C mp_limb_t M4_function_n (mp_ptr dst, mp_srcptr src1, mp_srcptr src2,
  59 C                          mp_size_t size);
  60 C mp_limb_t M4_function_nc (mp_ptr dst, mp_srcptr src1, mp_srcptr src2,
  61 C                           mp_size_t size, mp_limb_t carry);
  62
  63 defframe(PARAM_CARRY,20)
  64 defframe(PARAM_SIZE, 16)
  65 defframe(PARAM_SRC2, 12)
  66 defframe(PARAM_SRC1, 8)
  67 defframe(PARAM_DST,  4)
  68
  69         TEXT
  70         ALIGN(8)
  71
  72 PROLOGUE(M4_function_nc)
  73 deflit(`FRAME',0)
  74
  75         pushl   %edi            FRAME_pushl()
  76         pushl   %esi            FRAME_pushl()
  77
  78         movl    PARAM_DST,%edi
  79         movl    PARAM_SRC1,%esi
  80         movl    PARAM_SRC2,%edx
  81         movl    PARAM_SIZE,%ecx
  82
  83         movl    %ecx,%eax
  84         shrl    $3,%ecx                 C compute count for unrolled loop
  85         negl    %eax
  86         andl    $7,%eax                 C get index where to start loop
  87         jz      L(oopgo)                C necessary special case for 0
  88         incl    %ecx                    C adjust loop count
  89         shll    $2,%eax                 C adjustment for pointers...
  90         subl    %eax,%edi               C ... since they are offset ...
  91         subl    %eax,%esi               C ... by a constant when we ...
  92         subl    %eax,%edx               C ... enter the loop
  93         shrl    $2,%eax                 C restore previous value
  94
  95 ifdef(`PIC',`
  96         C Calculate start address in loop for PIC.  Due to limitations in
  97         C old gas, LF(M4_function_n,oop)-L(0a)-3 cannot be put into the leal
  98         call    L(0a)
  99 L(0a):  leal    (%eax,%eax,8),%eax
 100         addl    (%esp),%eax
 101         addl    $L(oop)-L(0a)-3,%eax
 102         addl    $4,%esp
 103 ',`
 104         C Calculate start address in loop for non-PIC.
 105         leal    L(oop)-3(%eax,%eax,8),%eax
 106 ')
 107
 108         C These lines initialize carry from the 5th parameter.  Should be
 109         C possible to simplify.
 110         pushl   %ebp            FRAME_pushl()
 111         movl    PARAM_CARRY,%ebp
 112         shrl    %ebp                    C shift bit 0 into carry
 113         popl    %ebp            FRAME_popl()
 114
 115         jmp     *%eax                   C jump into loop
 116
 117 EPILOGUE()
 118
 119
 120         ALIGN(16)
 121 PROLOGUE(M4_function_n)
 122 deflit(`FRAME',0)
 123
 124         pushl   %edi            FRAME_pushl()
 125         pushl   %esi            FRAME_pushl()
 126
 127         movl    PARAM_DST,%edi
 128         movl    PARAM_SRC1,%esi
 129         movl    PARAM_SRC2,%edx
 130         movl    PARAM_SIZE,%ecx
 131
 132         movl    %ecx,%eax
 133         shrl    $3,%ecx                 C compute count for unrolled loop
 134         negl    %eax
 135         andl    $7,%eax                 C get index where to start loop
 136         jz      L(oop)                  C necessary special case for 0
 137         incl    %ecx                    C adjust loop count
 138         shll    $2,%eax                 C adjustment for pointers...
 139         subl    %eax,%edi               C ... since they are offset ...
 140         subl    %eax,%esi               C ... by a constant when we ...
 141         subl    %eax,%edx               C ... enter the loop
 142         shrl    $2,%eax                 C restore previous value
 143
 144 ifdef(`PIC',`
 145         C Calculate start address in loop for PIC.  Due to limitations in
 146         C some assemblers, L(oop)-L(0b)-3 cannot be put into the leal
 147         call    L(0b)
 148 L(0b):  leal    (%eax,%eax,8),%eax
 149         addl    (%esp),%eax
 150         addl    $L(oop)-L(0b)-3,%eax
 151         addl    $4,%esp
 152 ',`
 153         C Calculate start address in loop for non-PIC.
 154         leal    L(oop)-3(%eax,%eax,8),%eax
 155 ')
 156         jmp     *%eax                   C jump into loop
 157
 158 L(oopgo):
 159         pushl   %ebp            FRAME_pushl()
 160         movl    PARAM_CARRY,%ebp
 161         shrl    %ebp                    C shift bit 0 into carry
 162         popl    %ebp            FRAME_popl()
 163
 164         ALIGN(16)
 165 L(oop): movl    (%esi),%eax
 166         M4_inst (%edx),%eax
 167         movl    %eax,(%edi)
 168         movl    4(%esi),%eax
 169         M4_inst 4(%edx),%eax
 170         movl    %eax,4(%edi)
 171         movl    8(%esi),%eax
 172         M4_inst 8(%edx),%eax
 173         movl    %eax,8(%edi)
 174         movl    12(%esi),%eax
 175         M4_inst 12(%edx),%eax
 176         movl    %eax,12(%edi)
 177         movl    16(%esi),%eax
 178         M4_inst 16(%edx),%eax
 179         movl    %eax,16(%edi)
 180         movl    20(%esi),%eax
 181         M4_inst 20(%edx),%eax
 182         movl    %eax,20(%edi)
 183         movl    24(%esi),%eax
 184         M4_inst 24(%edx),%eax
 185         movl    %eax,24(%edi)
 186         movl    28(%esi),%eax
 187         M4_inst 28(%edx),%eax
 188         movl    %eax,28(%edi)
 189         leal    32(%edi),%edi
 190         leal    32(%esi),%esi
 191         leal    32(%edx),%edx
 192         decl    %ecx
 193         jnz     L(oop)
 194
 195         sbbl    %eax,%eax
 196         negl    %eax
 197
 198         popl    %esi
 199         popl    %edi
 200         ret
 201
 202 EPILOGUE()