release/src/router/nettle/sparc32/aes-encrypt-internal.asm

   1 C -*- mode: asm; asm-comment-char: ?C; -*-
   2 C nettle, low-level cryptographics library
   3 C
   4 C Copyright (C) 2002, 2005 Niels Möller
   5 C
   6 C The nettle library is free software; you can redistribute it and/or modify
   7 C it under the terms of the GNU Lesser General Public License as published by
   8 C the Free Software Foundation; either version 2.1 of the License, or (at your
   9 C option) any later version.
  10 C
  11 C The nettle library is distributed in the hope that it will be useful, but
  12 C WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  13 C or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
  14 C License for more details.
  15 C
  16 C You should have received a copy of the GNU Lesser General Public License
  17 C along with the nettle library; see the file COPYING.LIB.  If not, write to
  18 C the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston,
  19 C MA 02111-1301, USA.
  20
  21 include_src(<sparc32/aes.m4>)
  22
  23 C       Arguments
  24 define(<CTX>,   <%i0>)
  25 define(<T>,     <%i1>)
  26 define(<LENGTH>,<%i2>)
  27 define(<DST>,   <%i3>)
  28 define(<SRC>,   <%i4>)
  29
  30 C       AES state, two copies for unrolling
  31
  32 define(<W0>,    <%l0>)
  33 define(<W1>,    <%l1>)
  34 define(<W2>,    <%l2>)
  35 define(<W3>,    <%l3>)
  36
  37 define(<X0>,    <%l4>)
  38 define(<X1>,    <%l5>)
  39 define(<X2>,    <%l6>)
  40 define(<X3>,    <%l7>)
  41
  42 C       %o0-%03 are used for loop invariants T0-T3
  43 define(<KEY>,   <%o4>)
  44 define(<ROUND>, <%o5>)
  45
  46 C %g1, %g2, %g3 are TMP1, TMP2 and TMP3
  47
  48 C I'm still slightly confused by the frame layout, specified in
  49 C "SYSTEM V APPLICATION BINARY INTERFACE SPARC Processor Supplement".
  50 C However, Sun's cc generates a 104 byte stack frame for a function
  51 C with no local variables, so that should be good enough for us too.
  52
  53 C The sparc32 stack frame looks like
  54 C
  55 C %fp -   4: OS-dependent link field
  56 C %fp -   8: OS-dependent link field
  57 C %fp - 104: OS register save area
  58 define(<FRAME_SIZE>, 104)
  59
  60         .file "aes-encrypt-internal.asm"
  61
  62         C _aes_encrypt(struct aes_context *ctx,
  63         C              const struct aes_table *T,
  64         C              unsigned length, uint8_t *dst,
  65         C              uint8_t *src)
  66
  67         .section        ".text"
  68         .align 16
  69         .proc   020
  70
  71 PROLOGUE(_nettle_aes_encrypt)
  72
  73         save    %sp, -FRAME_SIZE, %sp
  74         cmp     LENGTH, 0
  75         be      .Lend
  76
  77         C       Loop invariants
  78         add     T, AES_TABLE0, T0
  79         add     T, AES_TABLE1, T1
  80         add     T, AES_TABLE2, T2
  81         add     T, AES_TABLE3, T3
  82
  83 .Lblock_loop:
  84         C  Read src, and add initial subkey
  85         add     CTX, AES_KEYS, KEY
  86         AES_LOAD(0, SRC, KEY, W0)
  87         AES_LOAD(1, SRC, KEY, W1)
  88         AES_LOAD(2, SRC, KEY, W2)
  89         AES_LOAD(3, SRC, KEY, W3)
  90
  91         C       Must be even, and includes the final round
  92         ld      [AES_NROUNDS + CTX], ROUND
  93         add     SRC, 16, SRC
  94         add     KEY, 16, KEY
  95
  96         srl     ROUND, 1, ROUND
  97         C       Last two rounds handled specially
  98         sub     ROUND, 1, ROUND
  99 .Lround_loop:
 100         C The AES_ROUND macro uses T0,... T3
 101         C       Transform W -> X
 102         AES_ROUND(0, W0, W1, W2, W3, KEY, X0)
 103         AES_ROUND(1, W1, W2, W3, W0, KEY, X1)
 104         AES_ROUND(2, W2, W3, W0, W1, KEY, X2)
 105         AES_ROUND(3, W3, W0, W1, W2, KEY, X3)
 106
 107         C       Transform X -> W
 108         AES_ROUND(4, X0, X1, X2, X3, KEY, W0)
 109         AES_ROUND(5, X1, X2, X3, X0, KEY, W1)
 110         AES_ROUND(6, X2, X3, X0, X1, KEY, W2)
 111         AES_ROUND(7, X3, X0, X1, X2, KEY, W3)
 112
 113         subcc   ROUND, 1, ROUND
 114         bne     .Lround_loop
 115         add     KEY, 32, KEY
 116
 117         C       Penultimate round
 118         AES_ROUND(0, W0, W1, W2, W3, KEY, X0)
 119         AES_ROUND(1, W1, W2, W3, W0, KEY, X1)
 120         AES_ROUND(2, W2, W3, W0, W1, KEY, X2)
 121         AES_ROUND(3, W3, W0, W1, W2, KEY, X3)
 122
 123         add     KEY, 16, KEY
 124         C       Final round
 125         AES_FINAL_ROUND(0, T, X0, X1, X2, X3, KEY, DST)
 126         AES_FINAL_ROUND(1, T, X1, X2, X3, X0, KEY, DST)
 127         AES_FINAL_ROUND(2, T, X2, X3, X0, X1, KEY, DST)
 128         AES_FINAL_ROUND(3, T, X3, X0, X1, X2, KEY, DST)
 129
 130         subcc   LENGTH, 16, LENGTH
 131         bne     .Lblock_loop
 132         add     DST, 16, DST
 133
 134 .Lend:
 135         ret
 136         restore
 137 EPILOGUE(_nettle_aes_encrypt)
 138
 139 C Some stats from adriana.lysator.liu.se (SS1000$, 85 MHz), for AES 128
 140
 141 C 1:    nettle-1.13 C-code
 142 C 2:    nettle-1.13 assembler
 143 C 3:    New C-code
 144 C 4:    New assembler, first correct version
 145 C 5:    New assembler, with basic scheduling of AES_ROUND.
 146 C 6:    New assembpler, with loop invariants T0-T3.
 147 C 7:    New assembler, with basic scheduling also of AES_FINAL_ROUND.
 148
 149 C       MB/s    cycles/block    Code size (bytes)
 150 C 1     1.2     1107            592
 151 C 2     2.3     572             1032
 152 C 3     2.1     627
 153 C 4     1.8     722
 154 C 5     2.6     496
 155 C 6     3.0     437
 156 C 7     3.1     415             1448