release/src/router/nettle/x86_64/sha256-compress.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 "sha256-compress.asm"
  21 define(<STATE>, <%rdi>)
  22 define(<INPUT>, <%rsi>)
  23 define(<K>, <%rdx>)
  24 define(<SA>, <%eax>)
  25 define(<SB>, <%ebx>)
  26 define(<SC>, <%ecx>)
  27 define(<SD>, <%r8d>)
  28 define(<SE>, <%r9d>)
  29 define(<SF>, <%r10d>)
  30 define(<SG>, <%r11d>)
  31 define(<SH>, <%r12d>)
  32 define(<T0>, <%r13d>)
  33 define(<T1>, <%edi>)    C Overlap STATE
  34 define(<COUNT>, <%r14>)
  35 define(<W>, <%r15d>)
  36
  37 define(<EXPN>, <
  38         movl    OFFSET($1)(%rsp), W
  39         movl    OFFSET(eval(($1 + 14) % 16))(%rsp), T0
  40         movl    T0, T1
  41         shrl    <$>10, T0
  42         roll    <$>13, T1
  43         xorl    T1, T0
  44         roll    <$>2, T1
  45         xorl    T1, T0
  46         addl    T0, W
  47         movl    OFFSET(eval(($1 + 1) % 16))(%rsp), T0
  48         movl    T0, T1
  49         shrl    <$>3, T0
  50         roll    <$>14, T1
  51         xorl    T1, T0
  52         roll    <$>11, T1
  53         xorl    T1, T0
  54         addl    T0, W
  55         addl    OFFSET(eval(($1 + 9) % 16))(%rsp), W
  56         movl    W, OFFSET($1)(%rsp)
  57 >)
  58
  59 C ROUND(A,B,C,D,E,F,G,H,K)
  60 C
  61 C H += S1(E) + Choice(E,F,G) + K + W
  62 C D += H
  63 C H += S0(A) + Majority(A,B,C)
  64 C
  65 C Where
  66 C
  67 C S1(E) = E<<<26 ^ E<<<21 ^ E<<<7
  68 C S0(A) = A<<<30 ^ A<<<19 ^ A<<<10
  69 C Choice (E, F, G) = G^(E&(F^G))
  70 C Majority (A,B,C) = (A&B) + (C&(A^B))
  71
  72 define(<ROUND>, <
  73         movl    $5, T0
  74         movl    $5, T1
  75         roll    <$>7, T0
  76         roll    <$>21, T1
  77         xorl    T0, T1
  78         roll    <$>19, T0
  79         xorl    T0, T1
  80         addl    W, $8
  81         addl    T1, $8
  82         movl    $7, T0
  83         xorl    $6, T0
  84         andl    $5, T0
  85         xorl    $7, T0
  86         addl    OFFSET($9)(K,COUNT,4), $8
  87         addl    T0, $8
  88         addl    $8, $4
  89
  90         movl    $1, T0
  91         movl    $1, T1
  92         roll    <$>10, T0
  93         roll    <$>19, T1
  94         xorl    T0, T1
  95         roll    <$>20, T0
  96         xorl    T0, T1
  97         addl    T1, $8
  98         movl    $1, T0
  99         movl    $1, T1
 100         andl    $2, T0
 101         xorl    $2, T1
 102         addl    T0, $8
 103         andl    $3, T1
 104         addl    T1, $8
 105 >)
 106
 107 define(<NOEXPN>, <
 108         movl    OFFSET($1)(INPUT, COUNT, 4), W
 109         bswapl  W
 110         movl    W, OFFSET($1)(%rsp, COUNT, 4)
 111 >)
 112
 113         C void
 114         C _nettle_sha256_compress(uint32_t *state, const uint8_t *input, const uint32_t *k)
 115
 116         .text
 117         ALIGN(16)
 118
 119 PROLOGUE(_nettle_sha256_compress)
 120         W64_ENTRY(3, 0)
 121
 122         sub     $120, %rsp
 123         mov     %rbx, 64(%rsp)
 124         mov     STATE, 72(%rsp) C Save state, to free a register
 125         mov     %rbp, 80(%rsp)
 126         mov     %r12, 88(%rsp)
 127         mov     %r13, 96(%rsp)
 128         mov     %r14, 104(%rsp)
 129         mov     %r15, 112(%rsp)
 130
 131         movl    (STATE),   SA
 132         movl    4(STATE),  SB
 133         movl    8(STATE),  SC
 134         movl    12(STATE), SD
 135         movl    16(STATE), SE
 136         movl    20(STATE), SF
 137         movl    24(STATE), SG
 138         movl    28(STATE), SH
 139         xor     COUNT, COUNT
 140         ALIGN(16)
 141
 142 .Loop1:
 143         NOEXPN(0) ROUND(SA,SB,SC,SD,SE,SF,SG,SH,0)
 144         NOEXPN(1) ROUND(SH,SA,SB,SC,SD,SE,SF,SG,1)
 145         NOEXPN(2) ROUND(SG,SH,SA,SB,SC,SD,SE,SF,2)
 146         NOEXPN(3) ROUND(SF,SG,SH,SA,SB,SC,SD,SE,3)
 147         NOEXPN(4) ROUND(SE,SF,SG,SH,SA,SB,SC,SD,4)
 148         NOEXPN(5) ROUND(SD,SE,SF,SG,SH,SA,SB,SC,5)
 149         NOEXPN(6) ROUND(SC,SD,SE,SF,SG,SH,SA,SB,6)
 150         NOEXPN(7) ROUND(SB,SC,SD,SE,SF,SG,SH,SA,7)
 151         add     $8, COUNT
 152         cmp     $16, COUNT
 153         jne     .Loop1
 154
 155 .Loop2:
 156         EXPN( 0) ROUND(SA,SB,SC,SD,SE,SF,SG,SH,0)
 157         EXPN( 1) ROUND(SH,SA,SB,SC,SD,SE,SF,SG,1)
 158         EXPN( 2) ROUND(SG,SH,SA,SB,SC,SD,SE,SF,2)
 159         EXPN( 3) ROUND(SF,SG,SH,SA,SB,SC,SD,SE,3)
 160         EXPN( 4) ROUND(SE,SF,SG,SH,SA,SB,SC,SD,4)
 161         EXPN( 5) ROUND(SD,SE,SF,SG,SH,SA,SB,SC,5)
 162         EXPN( 6) ROUND(SC,SD,SE,SF,SG,SH,SA,SB,6)
 163         EXPN( 7) ROUND(SB,SC,SD,SE,SF,SG,SH,SA,7)
 164         EXPN( 8) ROUND(SA,SB,SC,SD,SE,SF,SG,SH,8)
 165         EXPN( 9) ROUND(SH,SA,SB,SC,SD,SE,SF,SG,9)
 166         EXPN(10) ROUND(SG,SH,SA,SB,SC,SD,SE,SF,10)
 167         EXPN(11) ROUND(SF,SG,SH,SA,SB,SC,SD,SE,11)
 168         EXPN(12) ROUND(SE,SF,SG,SH,SA,SB,SC,SD,12)
 169         EXPN(13) ROUND(SD,SE,SF,SG,SH,SA,SB,SC,13)
 170         EXPN(14) ROUND(SC,SD,SE,SF,SG,SH,SA,SB,14)
 171         EXPN(15) ROUND(SB,SC,SD,SE,SF,SG,SH,SA,15)
 172         add     $16, COUNT
 173         cmp     $64, COUNT
 174         jne     .Loop2
 175
 176         mov     72(%rsp), STATE
 177
 178         addl    SA, (STATE)
 179         addl    SB, 4(STATE)
 180         addl    SC, 8(STATE)
 181         addl    SD, 12(STATE)
 182         addl    SE, 16(STATE)
 183         addl    SF, 20(STATE)
 184         addl    SG, 24(STATE)
 185         addl    SH, 28(STATE)
 186
 187         mov     64(%rsp), %rbx
 188         mov     80(%rsp), %rbp
 189         mov     88(%rsp), %r12
 190         mov     96(%rsp), %r13
 191         mov     104(%rsp),%r14
 192         mov     112(%rsp),%r15
 193
 194         add     $120, %rsp
 195         W64_EXIT(3, 0)
 196         ret
 197 EPILOGUE(_nettle_sha256_compress)