source/libs/gmp/gmp-src/mpn/x86/pentium/mod_34lsub1.asm

   1 dnl  Intel P5 mpn_mod_34lsub1 -- mpn remainder modulo 2**24-1.
   2
   3 dnl  Copyright 2000-2002 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 P5: 1.66 cycles/limb
  35
  36
  37 C mp_limb_t mpn_mod_34lsub1 (mp_srcptr src, mp_size_t size)
  38 C
  39
  40 defframe(PARAM_SIZE, 8)
  41 defframe(PARAM_SRC,  4)
  42
  43         TEXT
  44         ALIGN(16)
  45 PROLOGUE(mpn_mod_34lsub1)
  46 deflit(`FRAME',0)
  47
  48         movl    PARAM_SIZE, %ecx
  49         movl    PARAM_SRC, %edx
  50
  51         subl    $2, %ecx
  52         ja      L(three_or_more)
  53
  54         movl    (%edx), %eax
  55         jne     L(one)
  56
  57
  58         movl    4(%edx), %ecx
  59         movl    %eax, %edx
  60
  61         shrl    $24, %edx
  62         andl    $0xFFFFFF, %eax
  63
  64         addl    %edx, %eax
  65         movl    %ecx, %edx
  66
  67         shrl    $16, %ecx
  68         andl    $0xFFFF, %edx
  69
  70         shll    $8, %edx
  71         addl    %ecx, %eax
  72
  73         addl    %edx, %eax
  74
  75 L(one):
  76         ret
  77
  78
  79 L(three_or_more):
  80         C eax
  81         C ebx
  82         C ecx   size-2
  83         C edx   src
  84         C esi
  85         C edi
  86         C ebp
  87
  88         pushl   %ebx    FRAME_pushl()
  89         pushl   %esi    FRAME_pushl()
  90
  91         pushl   %edi    FRAME_pushl()
  92         pushl   %ebp    FRAME_pushl()
  93
  94         xorl    %esi, %esi              C 0mod3
  95         xorl    %edi, %edi              C 1mod3
  96
  97         xorl    %ebp, %ebp              C 2mod3, and clear carry
  98
  99 L(top):
 100         C eax   scratch
 101         C ebx   scratch
 102         C ecx   counter, limbs
 103         C edx   src
 104         C esi   0mod3
 105         C edi   1mod3
 106         C ebp   2mod3
 107
 108         movl    (%edx), %eax
 109         movl    4(%edx), %ebx
 110
 111         adcl    %eax, %esi
 112         movl    8(%edx), %eax
 113
 114         adcl    %ebx, %edi
 115         leal    12(%edx), %edx
 116
 117         adcl    %eax, %ebp
 118         leal    -2(%ecx), %ecx
 119
 120         decl    %ecx
 121         jg      L(top)
 122
 123
 124         C ecx is -2, -1 or 0, representing 0, 1 or 2 more limbs, respectively
 125
 126         movl    $0xFFFFFFFF, %ebx       C mask
 127         incl    %ecx
 128
 129         js      L(combine)              C 0 more
 130
 131         movl    (%edx), %eax
 132         movl    $0xFFFFFF00, %ebx
 133
 134         adcl    %eax, %esi
 135         decl    %ecx
 136
 137         js      L(combine)              C 1 more
 138
 139         movl    4(%edx), %eax
 140         movl    $0xFFFF0000, %ebx
 141
 142         adcl    %eax, %edi
 143
 144
 145
 146 L(combine):
 147         C eax
 148         C ebx   mask
 149         C ecx
 150         C edx
 151         C esi   0mod3
 152         C edi   1mod3
 153         C ebp   2mod3
 154
 155         sbbl    %ecx, %ecx              C carry
 156         movl    %esi, %eax              C 0mod3
 157
 158         andl    %ebx, %ecx              C masked for position
 159         andl    $0xFFFFFF, %eax         C 0mod3 low
 160
 161         shrl    $24, %esi               C 0mod3 high
 162         subl    %ecx, %eax              C apply carry
 163
 164         addl    %esi, %eax              C apply 0mod3
 165         movl    %edi, %ebx              C 1mod3
 166
 167         shrl    $16, %edi               C 1mod3 high
 168         andl    $0x0000FFFF, %ebx
 169
 170         shll    $8, %ebx                C 1mod3 low
 171         addl    %edi, %eax              C apply 1mod3 high
 172
 173         addl    %ebx, %eax              C apply 1mod3 low
 174         movl    %ebp, %ebx              C 2mod3
 175
 176         shrl    $8, %ebp                C 2mod3 high
 177         andl    $0xFF, %ebx
 178
 179         shll    $16, %ebx               C 2mod3 low
 180         addl    %ebp, %eax              C apply 2mod3 high
 181
 182         addl    %ebx, %eax              C apply 2mod3 low
 183
 184         popl    %ebp
 185         popl    %edi
 186
 187         popl    %esi
 188         popl    %ebx
 189
 190         ret
 191
 192 EPILOGUE()