_ Git - BearSSL/blob - src/symcipher/aes_x86ni_ctr.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 <smmintrin.h>
  34 #include <wmmintrin.h>
  35 #define bswap32   __builtin_bswap32
  36 #endif
  37
  38 #if BR_AES_X86NI_MSC
  39 #include <stdlib.h>
  40 #include <intrin.h>
  41 #define bswap32   _byteswap_ulong
  42 #endif
  43
  44 /* see bearssl_block.h */
  45 void
  46 br_aes_x86ni_ctr_init(br_aes_x86ni_ctr_keys *ctx,
  47         const void *key, size_t len)
  48 {
  49         ctx->vtable = &br_aes_x86ni_ctr_vtable;
  50         ctx->num_rounds = br_aes_x86ni_keysched_enc(ctx->skey.skni, key, len);
  51 }
  52
  53 /* see bearssl_block.h */
  54 BR_TARGET("sse2,sse4.1,aes")
  55 uint32_t
  56 br_aes_x86ni_ctr_run(const br_aes_x86ni_ctr_keys *ctx,
  57         const void *iv, uint32_t cc, void *data, size_t len)
  58 {
  59         unsigned char *buf;
  60         unsigned char ivbuf[16];
  61         unsigned num_rounds;
  62         __m128i sk[15];
  63         __m128i ivx;
  64         unsigned u;
  65
  66         buf = data;
  67         memcpy(ivbuf, iv, 12);
  68         num_rounds = ctx->num_rounds;
  69         for (u = 0; u <= num_rounds; u ++) {
  70                 sk[u] = _mm_loadu_si128((void *)(ctx->skey.skni + (u << 4)));
  71         }
  72         ivx = _mm_loadu_si128((void *)ivbuf);
  73         while (len > 0) {
  74                 __m128i x0, x1, x2, x3;
  75
  76                 x0 = _mm_insert_epi32(ivx, bswap32(cc + 0), 3);
  77                 x1 = _mm_insert_epi32(ivx, bswap32(cc + 1), 3);
  78                 x2 = _mm_insert_epi32(ivx, bswap32(cc + 2), 3);
  79                 x3 = _mm_insert_epi32(ivx, bswap32(cc + 3), 3);
  80                 x0 = _mm_xor_si128(x0, sk[0]);
  81                 x1 = _mm_xor_si128(x1, sk[0]);
  82                 x2 = _mm_xor_si128(x2, sk[0]);
  83                 x3 = _mm_xor_si128(x3, sk[0]);
  84                 x0 = _mm_aesenc_si128(x0, sk[1]);
  85                 x1 = _mm_aesenc_si128(x1, sk[1]);
  86                 x2 = _mm_aesenc_si128(x2, sk[1]);
  87                 x3 = _mm_aesenc_si128(x3, sk[1]);
  88                 x0 = _mm_aesenc_si128(x0, sk[2]);
  89                 x1 = _mm_aesenc_si128(x1, sk[2]);
  90                 x2 = _mm_aesenc_si128(x2, sk[2]);
  91                 x3 = _mm_aesenc_si128(x3, sk[2]);
  92                 x0 = _mm_aesenc_si128(x0, sk[3]);
  93                 x1 = _mm_aesenc_si128(x1, sk[3]);
  94                 x2 = _mm_aesenc_si128(x2, sk[3]);
  95                 x3 = _mm_aesenc_si128(x3, sk[3]);
  96                 x0 = _mm_aesenc_si128(x0, sk[4]);
  97                 x1 = _mm_aesenc_si128(x1, sk[4]);
  98                 x2 = _mm_aesenc_si128(x2, sk[4]);
  99                 x3 = _mm_aesenc_si128(x3, sk[4]);
 100                 x0 = _mm_aesenc_si128(x0, sk[5]);
 101                 x1 = _mm_aesenc_si128(x1, sk[5]);
 102                 x2 = _mm_aesenc_si128(x2, sk[5]);
 103                 x3 = _mm_aesenc_si128(x3, sk[5]);
 104                 x0 = _mm_aesenc_si128(x0, sk[6]);
 105                 x1 = _mm_aesenc_si128(x1, sk[6]);
 106                 x2 = _mm_aesenc_si128(x2, sk[6]);
 107                 x3 = _mm_aesenc_si128(x3, sk[6]);
 108                 x0 = _mm_aesenc_si128(x0, sk[7]);
 109                 x1 = _mm_aesenc_si128(x1, sk[7]);
 110                 x2 = _mm_aesenc_si128(x2, sk[7]);
 111                 x3 = _mm_aesenc_si128(x3, sk[7]);
 112                 x0 = _mm_aesenc_si128(x0, sk[8]);
 113                 x1 = _mm_aesenc_si128(x1, sk[8]);
 114                 x2 = _mm_aesenc_si128(x2, sk[8]);
 115                 x3 = _mm_aesenc_si128(x3, sk[8]);
 116                 x0 = _mm_aesenc_si128(x0, sk[9]);
 117                 x1 = _mm_aesenc_si128(x1, sk[9]);
 118                 x2 = _mm_aesenc_si128(x2, sk[9]);
 119                 x3 = _mm_aesenc_si128(x3, sk[9]);
 120                 if (num_rounds == 10) {
 121                         x0 = _mm_aesenclast_si128(x0, sk[10]);
 122                         x1 = _mm_aesenclast_si128(x1, sk[10]);
 123                         x2 = _mm_aesenclast_si128(x2, sk[10]);
 124                         x3 = _mm_aesenclast_si128(x3, sk[10]);
 125                 } else if (num_rounds == 12) {
 126                         x0 = _mm_aesenc_si128(x0, sk[10]);
 127                         x1 = _mm_aesenc_si128(x1, sk[10]);
 128                         x2 = _mm_aesenc_si128(x2, sk[10]);
 129                         x3 = _mm_aesenc_si128(x3, sk[10]);
 130                         x0 = _mm_aesenc_si128(x0, sk[11]);
 131                         x1 = _mm_aesenc_si128(x1, sk[11]);
 132                         x2 = _mm_aesenc_si128(x2, sk[11]);
 133                         x3 = _mm_aesenc_si128(x3, sk[11]);
 134                         x0 = _mm_aesenclast_si128(x0, sk[12]);
 135                         x1 = _mm_aesenclast_si128(x1, sk[12]);
 136                         x2 = _mm_aesenclast_si128(x2, sk[12]);
 137                         x3 = _mm_aesenclast_si128(x3, sk[12]);
 138                 } else {
 139                         x0 = _mm_aesenc_si128(x0, sk[10]);
 140                         x1 = _mm_aesenc_si128(x1, sk[10]);
 141                         x2 = _mm_aesenc_si128(x2, sk[10]);
 142                         x3 = _mm_aesenc_si128(x3, sk[10]);
 143                         x0 = _mm_aesenc_si128(x0, sk[11]);
 144                         x1 = _mm_aesenc_si128(x1, sk[11]);
 145                         x2 = _mm_aesenc_si128(x2, sk[11]);
 146                         x3 = _mm_aesenc_si128(x3, sk[11]);
 147                         x0 = _mm_aesenc_si128(x0, sk[12]);
 148                         x1 = _mm_aesenc_si128(x1, sk[12]);
 149                         x2 = _mm_aesenc_si128(x2, sk[12]);
 150                         x3 = _mm_aesenc_si128(x3, sk[12]);
 151                         x0 = _mm_aesenc_si128(x0, sk[13]);
 152                         x1 = _mm_aesenc_si128(x1, sk[13]);
 153                         x2 = _mm_aesenc_si128(x2, sk[13]);
 154                         x3 = _mm_aesenc_si128(x3, sk[13]);
 155                         x0 = _mm_aesenclast_si128(x0, sk[14]);
 156                         x1 = _mm_aesenclast_si128(x1, sk[14]);
 157                         x2 = _mm_aesenclast_si128(x2, sk[14]);
 158                         x3 = _mm_aesenclast_si128(x3, sk[14]);
 159                 }
 160                 if (len >= 64) {
 161                         x0 = _mm_xor_si128(x0,
 162                                 _mm_loadu_si128((void *)(buf +  0)));
 163                         x1 = _mm_xor_si128(x1,
 164                                 _mm_loadu_si128((void *)(buf + 16)));
 165                         x2 = _mm_xor_si128(x2,
 166                                 _mm_loadu_si128((void *)(buf + 32)));
 167                         x3 = _mm_xor_si128(x3,
 168                                 _mm_loadu_si128((void *)(buf + 48)));
 169                         _mm_storeu_si128((void *)(buf +  0), x0);
 170                         _mm_storeu_si128((void *)(buf + 16), x1);
 171                         _mm_storeu_si128((void *)(buf + 32), x2);
 172                         _mm_storeu_si128((void *)(buf + 48), x3);
 173                         buf += 64;
 174                         len -= 64;
 175                         cc += 4;
 176                 } else {
 177                         unsigned char tmp[64];
 178
 179                         _mm_storeu_si128((void *)(tmp +  0), x0);
 180                         _mm_storeu_si128((void *)(tmp + 16), x1);
 181                         _mm_storeu_si128((void *)(tmp + 32), x2);
 182                         _mm_storeu_si128((void *)(tmp + 48), x3);
 183                         for (u = 0; u < len; u ++) {
 184                                 buf[u] ^= tmp[u];
 185                         }
 186                         cc += (uint32_t)len >> 4;
 187                         break;
 188                 }
 189         }
 190         return cc;
 191 }
 192
 193 /* see bearssl_block.h */
 194 const br_block_ctr_class br_aes_x86ni_ctr_vtable = {
 195         sizeof(br_aes_x86ni_ctr_keys),
 196         16,
 197         4,
 198         (void (*)(const br_block_ctr_class **, const void *, size_t))
 199                 &br_aes_x86ni_ctr_init,
 200         (uint32_t (*)(const br_block_ctr_class *const *,
 201                 const void *, uint32_t, void *, size_t))
 202                 &br_aes_x86ni_ctr_run
 203 };
 204
 205 /* see bearssl_block.h */
 206 const br_block_ctr_class *
 207 br_aes_x86ni_ctr_get_vtable(void)
 208 {
 209         return br_aes_x86ni_supported() ? &br_aes_x86ni_ctr_vtable : NULL;
 210 }
 211
 212 #else
 213
 214 /* see bearssl_block.h */
 215 const br_block_ctr_class *
 216 br_aes_x86ni_ctr_get_vtable(void)
 217 {
 218         return NULL;
 219 }
 220
 221 #endif