source/libs/gmp/gmp-src/mpn/x86/p6/aors_n.asm

   1 dnl  Intel P6 mpn_add_n/mpn_sub_n -- mpn add or subtract.
   2
   3 dnl  Copyright 2006 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 C TODO:
  34 C  * Avoid indexed addressing, it makes us stall on the two-ported register
  35 C    file.
  36
  37 C                           cycles/limb
  38 C P6 model 0-8,10-12            3.17
  39 C P6 model 9   (Banias)         2.15
  40 C P6 model 13  (Dothan)         2.25
  41
  42
  43 define(`rp',    `%edi')
  44 define(`up',    `%esi')
  45 define(`vp',    `%ebx')
  46 define(`n',     `%ecx')
  47
  48 ifdef(`OPERATION_add_n', `
  49         define(ADCSBB,        adc)
  50         define(func,          mpn_add_n)
  51         define(func_nc,       mpn_add_nc)')
  52 ifdef(`OPERATION_sub_n', `
  53         define(ADCSBB,        sbb)
  54         define(func,          mpn_sub_n)
  55         define(func_nc,       mpn_sub_nc)')
  56
  57 MULFUNC_PROLOGUE(mpn_add_n mpn_add_nc mpn_sub_n mpn_sub_nc)
  58
  59 ASM_START()
  60
  61         TEXT
  62         ALIGN(16)
  63
  64 PROLOGUE(func)
  65         xor     %edx, %edx
  66 L(start):
  67         push    %edi
  68         push    %esi
  69         push    %ebx
  70
  71         mov     16(%esp), rp
  72         mov     20(%esp), up
  73         mov     24(%esp), vp
  74         mov     28(%esp), n
  75
  76         lea     (up,n,4), up
  77         lea     (vp,n,4), vp
  78         lea     (rp,n,4), rp
  79
  80         neg     n
  81         mov     n, %eax
  82         and     $-8, n
  83         and     $7, %eax
  84         shl     $2, %eax                        C 4x
  85 ifdef(`PIC',`
  86         call    L(pic_calc)
  87 L(here):
  88 ',`
  89         lea     L(ent) (%eax,%eax,2), %eax      C 12x
  90 ')
  91
  92         shr     %edx                            C set cy flag
  93         jmp     *%eax
  94
  95 ifdef(`PIC',`
  96 L(pic_calc):
  97         C See mpn/x86/README about old gas bugs
  98         lea     (%eax,%eax,2), %eax
  99         add     $L(ent)-L(here), %eax
 100         add     (%esp), %eax
 101         ret_internal
 102 ')
 103
 104 L(end):
 105         sbb     %eax, %eax
 106         neg     %eax
 107         pop     %ebx
 108         pop     %esi
 109         pop     %edi
 110         ret
 111
 112         ALIGN(16)
 113 L(top):
 114         jecxz   L(end)
 115 L(ent):
 116 Zdisp(  mov,    0,(up,n,4), %eax)
 117 Zdisp(  ADCSBB, 0,(vp,n,4), %eax)
 118 Zdisp(  mov,    %eax, 0,(rp,n,4))
 119
 120         mov     4(up,n,4), %edx
 121         ADCSBB  4(vp,n,4), %edx
 122         mov     %edx, 4(rp,n,4)
 123
 124         mov     8(up,n,4), %eax
 125         ADCSBB  8(vp,n,4), %eax
 126         mov     %eax, 8(rp,n,4)
 127
 128         mov     12(up,n,4), %edx
 129         ADCSBB  12(vp,n,4), %edx
 130         mov     %edx, 12(rp,n,4)
 131
 132         mov     16(up,n,4), %eax
 133         ADCSBB  16(vp,n,4), %eax
 134         mov     %eax, 16(rp,n,4)
 135
 136         mov     20(up,n,4), %edx
 137         ADCSBB  20(vp,n,4), %edx
 138         mov     %edx, 20(rp,n,4)
 139
 140         mov     24(up,n,4), %eax
 141         ADCSBB  24(vp,n,4), %eax
 142         mov     %eax, 24(rp,n,4)
 143
 144         mov     28(up,n,4), %edx
 145         ADCSBB  28(vp,n,4), %edx
 146         mov     %edx, 28(rp,n,4)
 147
 148         lea     8(n), n
 149         jmp     L(top)
 150
 151 EPILOGUE()
 152
 153 PROLOGUE(func_nc)
 154         movl    20(%esp), %edx
 155         jmp     L(start)
 156 EPILOGUE()