release/src/router/nettle/x86_64/ecc-521-modp.asm

   1 C nettle, low-level cryptographics library
   2 C
   3 C Copyright (C) 2013 Niels Möller
   4 C
   5 C The nettle library is free software; you can redistribute it and/or modify
   6 C it under the terms of the GNU Lesser General Public License as published by
   7 C the Free Software Foundation; either version 2.1 of the License, or (at your
   8 C option) any later version.
   9 C
  10 C The nettle library is distributed in the hope that it will be useful, but
  11 C WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  12 C or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
  13 C License for more details.
  14 C
  15 C You should have received a copy of the GNU Lesser General Public License
  16 C along with the nettle library; see the file COPYING.LIB.  If not, write to
  17 C the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston,
  18 C MA 02111-1301, USA.
  19
  20         .file "ecc-521-modp.asm"
  21
  22 define(<RP>, <%rsi>)
  23 define(<U0>, <%rax>)
  24 define(<U1>, <%rbx>)
  25 define(<U2>, <%rcx>)
  26 define(<U3>, <%rdx>)
  27 define(<U4>, <%rbp>)
  28 define(<U5>, <%rdi>)
  29 define(<U6>, <%r8>)
  30 define(<U7>, <%r9>)
  31 define(<U8>, <%r10>)
  32 define(<U9>, <%r11>)
  33 define(<T0>, <%r12>)
  34 define(<T1>, <%r13>)
  35
  36 PROLOGUE(nettle_ecc_521_modp)
  37         W64_ENTRY(2, 0)
  38         push    %rbx
  39         push    %rbp
  40         push    %r12
  41         push    %r13
  42
  43         C Read top 17 limbs, shift left 55 bits
  44         mov     72(RP), U1
  45         mov     U1, U0
  46         shl     $55, U0
  47         shr     $9, U1
  48
  49         mov     80(RP), U2
  50         mov     U2, T0
  51         shr     $9, U2
  52         shl     $55, T0
  53         or      T0, U1
  54
  55         mov     88(RP), U3
  56         mov     U3, T0
  57         shr     $9, U3
  58         shl     $55, T0
  59         or      T0, U2
  60
  61         mov     96(RP), U4
  62         mov     U4, T0
  63         shr     $9, U4
  64         shl     $55, T0
  65         or      T0, U3
  66
  67         mov     104(RP), U5
  68         mov     U5, T0
  69         shr     $9, U5
  70         shl     $55, T0
  71         or      T0, U4
  72
  73         mov     112(RP), U6
  74         mov     U6, T0
  75         shr     $9, U6
  76         shl     $55, T0
  77         or      T0, U5
  78
  79         mov     120(RP), U7
  80         mov     U7, T0
  81         shr     $9, U7
  82         shl     $55, T0
  83         or      T0, U6
  84
  85         mov     128(RP), U8
  86         mov     U8, T0
  87         shr     $9, U8
  88         shl     $55, T0
  89         or      T0, U7
  90
  91         mov     136(RP), U9
  92         mov     U9, T0
  93         shr     $9, U9
  94         shl     $55, T0
  95         or      T0, U8
  96
  97         add       (RP), U0
  98         adc      8(RP), U1
  99         adc     16(RP), U2
 100         adc     24(RP), U3
 101         adc     32(RP), U4
 102         adc     40(RP), U5
 103         adc     48(RP), U6
 104         adc     56(RP), U7
 105         adc     64(RP), U8
 106         adc     $0, U9
 107
 108         C Top limbs are <U9, U8>. Keep low 9 bits of 8, and fold the
 109         C top bits (at most 65 bits).
 110         mov     U8, T0
 111         shr     $9, T0
 112         and     $0x1ff, U8
 113         mov     U9, T1
 114         shl     $55, U9
 115         shr     $9, T1
 116         or      U9, T0
 117
 118         add     T0, U0
 119         mov     U0, (RP)
 120         adc     T1, U1
 121         mov     U1, 8(RP)
 122         adc     $0, U2
 123         mov     U2, 16(RP)
 124         adc     $0, U3
 125         mov     U3, 24(RP)
 126         adc     $0, U4
 127         mov     U4, 32(RP)
 128         adc     $0, U5
 129         mov     U5, 40(RP)
 130         adc     $0, U6
 131         mov     U6, 48(RP)
 132         adc     $0, U7
 133         mov     U7, 56(RP)
 134         adc     $0, U8
 135         mov     U8, 64(RP)
 136
 137         pop     %r13
 138         pop     %r12
 139         pop     %rbp
 140         pop     %rbx
 141         W64_EXIT(2, 0)
 142         ret
 143 EPILOGUE(nettle_ecc_521_modp)