_ Git - BearSSL/blob - src/symcipher/aes_x86ni_cbcdec.c

   1 /*
   2  * Copyright (c) 2017 Thomas Pornin <pornin@bolet.org>
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining
   5  * a copy of this software and associated documentation files (the
   6  * "Software"), to deal in the Software without restriction, including
   7  * without limitation the rights to use, copy, modify, merge, publish,
   8  * distribute, sublicense, and/or sell copies of the Software, and to
   9  * permit persons to whom the Software is furnished to do so, subject to
  10  * the following conditions:
  11  *
  12  * The above copyright notice and this permission notice shall be
  13  * included in all copies or substantial portions of the Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  16  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  17  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  18  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  19  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  20  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  21  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  */
  24
  25 #include "inner.h"
  26
  27 #if BR_AES_X86NI
  28
  29 #if BR_AES_X86NI_GCC
  30 #if BR_AES_X86NI_GCC_OLD
  31 #pragma GCC target("sse2,sse4.1,aes,pclmul")
  32 #endif
  33 #include <wmmintrin.h>
  34 #endif
  35
  36 #if BR_AES_X86NI_MSC
  37 #include <intrin.h>
  38 #endif
  39
  40 /* see bearssl_block.h */
  41 void
  42 br_aes_x86ni_cbcdec_init(br_aes_x86ni_cbcdec_keys *ctx,
  43         const void *key, size_t len)
  44 {
  45         ctx->vtable = &br_aes_x86ni_cbcdec_vtable;
  46         ctx->num_rounds = br_aes_x86ni_keysched_dec(ctx->skey.skni, key, len);
  47 }
  48
  49 /* see bearssl_block.h */
  50 BR_TARGET("sse2,aes")
  51 void
  52 br_aes_x86ni_cbcdec_run(const br_aes_x86ni_cbcdec_keys *ctx,
  53         void *iv, void *data, size_t len)
  54 {
  55         unsigned char *buf;
  56         unsigned num_rounds;
  57         __m128i sk[15], ivx;
  58         unsigned u;
  59
  60         buf = data;
  61         ivx = _mm_loadu_si128(iv);
  62         num_rounds = ctx->num_rounds;
  63         for (u = 0; u <= num_rounds; u ++) {
  64                 sk[u] = _mm_loadu_si128((void *)(ctx->skey.skni + (u << 4)));
  65         }
  66         while (len > 0) {
  67                 __m128i x0, x1, x2, x3, e0, e1, e2, e3;
  68
  69                 x0 = _mm_loadu_si128((void *)(buf +  0));
  70                 if (len >= 64) {
  71                         x1 = _mm_loadu_si128((void *)(buf + 16));
  72                         x2 = _mm_loadu_si128((void *)(buf + 32));
  73                         x3 = _mm_loadu_si128((void *)(buf + 48));
  74                 } else {
  75                         x0 = _mm_loadu_si128((void *)(buf +  0));
  76                         if (len >= 32) {
  77                                 x1 = _mm_loadu_si128((void *)(buf + 16));
  78                                 if (len >= 48) {
  79                                         x2 = _mm_loadu_si128(
  80                                                 (void *)(buf + 32));
  81                                         x3 = x2;
  82                                 } else {
  83                                         x2 = x0;
  84                                         x3 = x1;
  85                                 }
  86                         } else {
  87                                 x1 = x0;
  88                                 x2 = x0;
  89                                 x3 = x0;
  90                         }
  91                 }
  92                 e0 = x0;
  93                 e1 = x1;
  94                 e2 = x2;
  95                 e3 = x3;
  96                 x0 = _mm_xor_si128(x0, sk[0]);
  97                 x1 = _mm_xor_si128(x1, sk[0]);
  98                 x2 = _mm_xor_si128(x2, sk[0]);
  99                 x3 = _mm_xor_si128(x3, sk[0]);
 100                 x0 = _mm_aesdec_si128(x0, sk[1]);
 101                 x1 = _mm_aesdec_si128(x1, sk[1]);
 102                 x2 = _mm_aesdec_si128(x2, sk[1]);
 103                 x3 = _mm_aesdec_si128(x3, sk[1]);
 104                 x0 = _mm_aesdec_si128(x0, sk[2]);
 105                 x1 = _mm_aesdec_si128(x1, sk[2]);
 106                 x2 = _mm_aesdec_si128(x2, sk[2]);
 107                 x3 = _mm_aesdec_si128(x3, sk[2]);
 108                 x0 = _mm_aesdec_si128(x0, sk[3]);
 109                 x1 = _mm_aesdec_si128(x1, sk[3]);
 110                 x2 = _mm_aesdec_si128(x2, sk[3]);
 111                 x3 = _mm_aesdec_si128(x3, sk[3]);
 112                 x0 = _mm_aesdec_si128(x0, sk[4]);
 113                 x1 = _mm_aesdec_si128(x1, sk[4]);
 114                 x2 = _mm_aesdec_si128(x2, sk[4]);
 115                 x3 = _mm_aesdec_si128(x3, sk[4]);
 116                 x0 = _mm_aesdec_si128(x0, sk[5]);
 117                 x1 = _mm_aesdec_si128(x1, sk[5]);
 118                 x2 = _mm_aesdec_si128(x2, sk[5]);
 119                 x3 = _mm_aesdec_si128(x3, sk[5]);
 120                 x0 = _mm_aesdec_si128(x0, sk[6]);
 121                 x1 = _mm_aesdec_si128(x1, sk[6]);
 122                 x2 = _mm_aesdec_si128(x2, sk[6]);
 123                 x3 = _mm_aesdec_si128(x3, sk[6]);
 124                 x0 = _mm_aesdec_si128(x0, sk[7]);
 125                 x1 = _mm_aesdec_si128(x1, sk[7]);
 126                 x2 = _mm_aesdec_si128(x2, sk[7]);
 127                 x3 = _mm_aesdec_si128(x3, sk[7]);
 128                 x0 = _mm_aesdec_si128(x0, sk[8]);
 129                 x1 = _mm_aesdec_si128(x1, sk[8]);
 130                 x2 = _mm_aesdec_si128(x2, sk[8]);
 131                 x3 = _mm_aesdec_si128(x3, sk[8]);
 132                 x0 = _mm_aesdec_si128(x0, sk[9]);
 133                 x1 = _mm_aesdec_si128(x1, sk[9]);
 134                 x2 = _mm_aesdec_si128(x2, sk[9]);
 135                 x3 = _mm_aesdec_si128(x3, sk[9]);
 136                 if (num_rounds == 10) {
 137                         x0 = _mm_aesdeclast_si128(x0, sk[10]);
 138                         x1 = _mm_aesdeclast_si128(x1, sk[10]);
 139                         x2 = _mm_aesdeclast_si128(x2, sk[10]);
 140                         x3 = _mm_aesdeclast_si128(x3, sk[10]);
 141                 } else if (num_rounds == 12) {
 142                         x0 = _mm_aesdec_si128(x0, sk[10]);
 143                         x1 = _mm_aesdec_si128(x1, sk[10]);
 144                         x2 = _mm_aesdec_si128(x2, sk[10]);
 145                         x3 = _mm_aesdec_si128(x3, sk[10]);
 146                         x0 = _mm_aesdec_si128(x0, sk[11]);
 147                         x1 = _mm_aesdec_si128(x1, sk[11]);
 148                         x2 = _mm_aesdec_si128(x2, sk[11]);
 149                         x3 = _mm_aesdec_si128(x3, sk[11]);
 150                         x0 = _mm_aesdeclast_si128(x0, sk[12]);
 151                         x1 = _mm_aesdeclast_si128(x1, sk[12]);
 152                         x2 = _mm_aesdeclast_si128(x2, sk[12]);
 153                         x3 = _mm_aesdeclast_si128(x3, sk[12]);
 154                 } else {
 155                         x0 = _mm_aesdec_si128(x0, sk[10]);
 156                         x1 = _mm_aesdec_si128(x1, sk[10]);
 157                         x2 = _mm_aesdec_si128(x2, sk[10]);
 158                         x3 = _mm_aesdec_si128(x3, sk[10]);
 159                         x0 = _mm_aesdec_si128(x0, sk[11]);
 160                         x1 = _mm_aesdec_si128(x1, sk[11]);
 161                         x2 = _mm_aesdec_si128(x2, sk[11]);
 162                         x3 = _mm_aesdec_si128(x3, sk[11]);
 163                         x0 = _mm_aesdec_si128(x0, sk[12]);
 164                         x1 = _mm_aesdec_si128(x1, sk[12]);
 165                         x2 = _mm_aesdec_si128(x2, sk[12]);
 166                         x3 = _mm_aesdec_si128(x3, sk[12]);
 167                         x0 = _mm_aesdec_si128(x0, sk[13]);
 168                         x1 = _mm_aesdec_si128(x1, sk[13]);
 169                         x2 = _mm_aesdec_si128(x2, sk[13]);
 170                         x3 = _mm_aesdec_si128(x3, sk[13]);
 171                         x0 = _mm_aesdeclast_si128(x0, sk[14]);
 172                         x1 = _mm_aesdeclast_si128(x1, sk[14]);
 173                         x2 = _mm_aesdeclast_si128(x2, sk[14]);
 174                         x3 = _mm_aesdeclast_si128(x3, sk[14]);
 175                 }
 176                 x0 = _mm_xor_si128(x0, ivx);
 177                 x1 = _mm_xor_si128(x1, e0);
 178                 x2 = _mm_xor_si128(x2, e1);
 179                 x3 = _mm_xor_si128(x3, e2);
 180                 ivx = e3;
 181                 _mm_storeu_si128((void *)(buf +  0), x0);
 182                 if (len >= 64) {
 183                         _mm_storeu_si128((void *)(buf + 16), x1);
 184                         _mm_storeu_si128((void *)(buf + 32), x2);
 185                         _mm_storeu_si128((void *)(buf + 48), x3);
 186                         buf += 64;
 187                         len -= 64;
 188                 } else {
 189                         if (len >= 32) {
 190                                 _mm_storeu_si128((void *)(buf + 16), x1);
 191                                 if (len >= 48) {
 192                                         _mm_storeu_si128(
 193                                                 (void *)(buf + 32), x2);
 194                                 }
 195                         }
 196                         break;
 197                 }
 198         }
 199         _mm_storeu_si128(iv, ivx);
 200 }
 201
 202 /* see bearssl_block.h */
 203 const br_block_cbcdec_class br_aes_x86ni_cbcdec_vtable = {
 204         sizeof(br_aes_x86ni_cbcdec_keys),
 205         16,
 206         4,
 207         (void (*)(const br_block_cbcdec_class **, const void *, size_t))
 208                 &br_aes_x86ni_cbcdec_init,
 209         (void (*)(const br_block_cbcdec_class *const *, void *, void *, size_t))
 210                 &br_aes_x86ni_cbcdec_run
 211 };
 212
 213 /* see bearssl_block.h */
 214 const br_block_cbcdec_class *
 215 br_aes_x86ni_cbcdec_get_vtable(void)
 216 {
 217         return br_aes_x86ni_supported() ? &br_aes_x86ni_cbcdec_vtable : NULL;
 218 }
 219
 220 #else
 221
 222 /* see bearssl_block.h */
 223 const br_block_cbcdec_class *
 224 br_aes_x86ni_cbcdec_get_vtable(void)
 225 {
 226         return NULL;
 227 }
 228
 229 #endif