1 /******************************************************************************* 2 * Copyright (c) 2013, Intel Corporation 3 * 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions are 8 * met: 9 * 10 * * Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the 16 * distribution. 17 * 18 * * Neither the name of the Intel Corporation nor the names of its 19 * contributors may be used to endorse or promote products derived from 20 * this software without specific prior written permission. 21 * 22 * 23 * THIS SOFTWARE IS PROVIDED BY INTEL CORPORATION ""AS IS"" AND ANY 24 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 26 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL CORPORATION OR 27 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 28 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 29 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 30 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 34 ******************************************************************************** 35 * 36 * Intel SHA Extensions optimized implementation of a SHA-1 update function 37 * 38 * The function takes a pointer to the current hash values, a pointer to the 39 * input data, and a number of 64 byte blocks to process. Once all blocks have 40 * been processed, the digest pointer is updated with the resulting hash value. 41 * The function only processes complete blocks, there is no functionality to 42 * store partial blocks. All message padding and hash value initialization must 43 * be done outside the update function. 44 * 45 * The indented lines in the loop are instructions related to rounds processing. 46 * The non-indented lines are instructions related to the message schedule. 47 * 48 * Author: Sean Gulley <sean.m.gulley@intel.com> 49 * Date: July 2013 50 * 51 ******************************************************************************** 52 * 53 * Example complier command line: 54 * icc intel_sha_extensions_sha1_intrinsic.c 55 * gcc -msha -msse4 intel_sha_extensions_sha1_intrinsic.c 56 * 57 *******************************************************************************/ 58 #include <sys/cdefs.h> 59 #include <sys/types.h> 60 #include <crypto/aesni/aesni_os.h> 61 #include <crypto/aesni/sha_sse.h> 62 63 #include <immintrin.h> 64 65 void intel_sha1_step(uint32_t *digest, const char *data, uint32_t num_blks) { 66 __m128i abcd, e0, e1; 67 __m128i abcd_save, e_save; 68 __m128i msg0, msg1, msg2, msg3; 69 __m128i shuf_mask, e_mask; 70 71 #if 0 72 e_mask = _mm_set_epi64x(0xFFFFFFFF00000000ull, 0x0000000000000000ull); 73 #else 74 (void)e_mask; 75 e0 = _mm_set_epi64x(0, 0); 76 #endif 77 shuf_mask = _mm_set_epi64x(0x0001020304050607ull, 0x08090a0b0c0d0e0full); 78 79 // Load initial hash values 80 abcd = _mm_loadu_si128((__m128i*) digest); 81 e0 = _mm_insert_epi32(e0, *(digest+4), 3); 82 abcd = _mm_shuffle_epi32(abcd, 0x1B); 83 #if 0 84 e0 = _mm_and_si128(e0, e_mask); 85 #endif 86 87 while (num_blks > 0) { 88 // Save hash values for addition after rounds 89 abcd_save = abcd; 90 e_save = e0; 91 92 // Rounds 0-3 93 msg0 = _mm_loadu_si128((const __m128i*) data); 94 msg0 = _mm_shuffle_epi8(msg0, shuf_mask); 95 e0 = _mm_add_epi32(e0, msg0); 96 e1 = abcd; 97 abcd = _mm_sha1rnds4_epu32(abcd, e0, 0); 98 99 // Rounds 4-7 100 msg1 = _mm_loadu_si128((const __m128i*) (data+16)); 101 msg1 = _mm_shuffle_epi8(msg1, shuf_mask); 102 e1 = _mm_sha1nexte_epu32(e1, msg1); 103 e0 = abcd; 104 abcd = _mm_sha1rnds4_epu32(abcd, e1, 0); 105 msg0 = _mm_sha1msg1_epu32(msg0, msg1); 106 107 // Rounds 8-11 108 msg2 = _mm_loadu_si128((const __m128i*) (data+32)); 109 msg2 = _mm_shuffle_epi8(msg2, shuf_mask); 110 e0 = _mm_sha1nexte_epu32(e0, msg2); 111 e1 = abcd; 112 abcd = _mm_sha1rnds4_epu32(abcd, e0, 0); 113 msg1 = _mm_sha1msg1_epu32(msg1, msg2); 114 msg0 = _mm_xor_si128(msg0, msg2); 115 116 // Rounds 12-15 117 msg3 = _mm_loadu_si128((const __m128i*) (data+48)); 118 msg3 = _mm_shuffle_epi8(msg3, shuf_mask); 119 e1 = _mm_sha1nexte_epu32(e1, msg3); 120 e0 = abcd; 121 msg0 = _mm_sha1msg2_epu32(msg0, msg3); 122 abcd = _mm_sha1rnds4_epu32(abcd, e1, 0); 123 msg2 = _mm_sha1msg1_epu32(msg2, msg3); 124 msg1 = _mm_xor_si128(msg1, msg3); 125 126 // Rounds 16-19 127 e0 = _mm_sha1nexte_epu32(e0, msg0); 128 e1 = abcd; 129 msg1 = _mm_sha1msg2_epu32(msg1, msg0); 130 abcd = _mm_sha1rnds4_epu32(abcd, e0, 0); 131 msg3 = _mm_sha1msg1_epu32(msg3, msg0); 132 msg2 = _mm_xor_si128(msg2, msg0); 133 134 // Rounds 20-23 135 e1 = _mm_sha1nexte_epu32(e1, msg1); 136 e0 = abcd; 137 msg2 = _mm_sha1msg2_epu32(msg2, msg1); 138 abcd = _mm_sha1rnds4_epu32(abcd, e1, 1); 139 msg0 = _mm_sha1msg1_epu32(msg0, msg1); 140 msg3 = _mm_xor_si128(msg3, msg1); 141 142 // Rounds 24-27 143 e0 = _mm_sha1nexte_epu32(e0, msg2); 144 e1 = abcd; 145 msg3 = _mm_sha1msg2_epu32(msg3, msg2); 146 abcd = _mm_sha1rnds4_epu32(abcd, e0, 1); 147 msg1 = _mm_sha1msg1_epu32(msg1, msg2); 148 msg0 = _mm_xor_si128(msg0, msg2); 149 150 // Rounds 28-31 151 e1 = _mm_sha1nexte_epu32(e1, msg3); 152 e0 = abcd; 153 msg0 = _mm_sha1msg2_epu32(msg0, msg3); 154 abcd = _mm_sha1rnds4_epu32(abcd, e1, 1); 155 msg2 = _mm_sha1msg1_epu32(msg2, msg3); 156 msg1 = _mm_xor_si128(msg1, msg3); 157 158 // Rounds 32-35 159 e0 = _mm_sha1nexte_epu32(e0, msg0); 160 e1 = abcd; 161 msg1 = _mm_sha1msg2_epu32(msg1, msg0); 162 abcd = _mm_sha1rnds4_epu32(abcd, e0, 1); 163 msg3 = _mm_sha1msg1_epu32(msg3, msg0); 164 msg2 = _mm_xor_si128(msg2, msg0); 165 166 // Rounds 36-39 167 e1 = _mm_sha1nexte_epu32(e1, msg1); 168 e0 = abcd; 169 msg2 = _mm_sha1msg2_epu32(msg2, msg1); 170 abcd = _mm_sha1rnds4_epu32(abcd, e1, 1); 171 msg0 = _mm_sha1msg1_epu32(msg0, msg1); 172 msg3 = _mm_xor_si128(msg3, msg1); 173 174 // Rounds 40-43 175 e0 = _mm_sha1nexte_epu32(e0, msg2); 176 e1 = abcd; 177 msg3 = _mm_sha1msg2_epu32(msg3, msg2); 178 abcd = _mm_sha1rnds4_epu32(abcd, e0, 2); 179 msg1 = _mm_sha1msg1_epu32(msg1, msg2); 180 msg0 = _mm_xor_si128(msg0, msg2); 181 182 // Rounds 44-47 183 e1 = _mm_sha1nexte_epu32(e1, msg3); 184 e0 = abcd; 185 msg0 = _mm_sha1msg2_epu32(msg0, msg3); 186 abcd = _mm_sha1rnds4_epu32(abcd, e1, 2); 187 msg2 = _mm_sha1msg1_epu32(msg2, msg3); 188 msg1 = _mm_xor_si128(msg1, msg3); 189 190 // Rounds 48-51 191 e0 = _mm_sha1nexte_epu32(e0, msg0); 192 e1 = abcd; 193 msg1 = _mm_sha1msg2_epu32(msg1, msg0); 194 abcd = _mm_sha1rnds4_epu32(abcd, e0, 2); 195 msg3 = _mm_sha1msg1_epu32(msg3, msg0); 196 msg2 = _mm_xor_si128(msg2, msg0); 197 198 // Rounds 52-55 199 e1 = _mm_sha1nexte_epu32(e1, msg1); 200 e0 = abcd; 201 msg2 = _mm_sha1msg2_epu32(msg2, msg1); 202 abcd = _mm_sha1rnds4_epu32(abcd, e1, 2); 203 msg0 = _mm_sha1msg1_epu32(msg0, msg1); 204 msg3 = _mm_xor_si128(msg3, msg1); 205 206 // Rounds 56-59 207 e0 = _mm_sha1nexte_epu32(e0, msg2); 208 e1 = abcd; 209 msg3 = _mm_sha1msg2_epu32(msg3, msg2); 210 abcd = _mm_sha1rnds4_epu32(abcd, e0, 2); 211 msg1 = _mm_sha1msg1_epu32(msg1, msg2); 212 msg0 = _mm_xor_si128(msg0, msg2); 213 214 // Rounds 60-63 215 e1 = _mm_sha1nexte_epu32(e1, msg3); 216 e0 = abcd; 217 msg0 = _mm_sha1msg2_epu32(msg0, msg3); 218 abcd = _mm_sha1rnds4_epu32(abcd, e1, 3); 219 msg2 = _mm_sha1msg1_epu32(msg2, msg3); 220 msg1 = _mm_xor_si128(msg1, msg3); 221 222 // Rounds 64-67 223 e0 = _mm_sha1nexte_epu32(e0, msg0); 224 e1 = abcd; 225 msg1 = _mm_sha1msg2_epu32(msg1, msg0); 226 abcd = _mm_sha1rnds4_epu32(abcd, e0, 3); 227 msg3 = _mm_sha1msg1_epu32(msg3, msg0); 228 msg2 = _mm_xor_si128(msg2, msg0); 229 230 // Rounds 68-71 231 e1 = _mm_sha1nexte_epu32(e1, msg1); 232 e0 = abcd; 233 msg2 = _mm_sha1msg2_epu32(msg2, msg1); 234 abcd = _mm_sha1rnds4_epu32(abcd, e1, 3); 235 msg3 = _mm_xor_si128(msg3, msg1); 236 237 // Rounds 72-75 238 e0 = _mm_sha1nexte_epu32(e0, msg2); 239 e1 = abcd; 240 msg3 = _mm_sha1msg2_epu32(msg3, msg2); 241 abcd = _mm_sha1rnds4_epu32(abcd, e0, 3); 242 243 // Rounds 76-79 244 e1 = _mm_sha1nexte_epu32(e1, msg3); 245 e0 = abcd; 246 abcd = _mm_sha1rnds4_epu32(abcd, e1, 3); 247 248 // Add current hash values with previously saved 249 e0 = _mm_sha1nexte_epu32(e0, e_save); 250 abcd = _mm_add_epi32(abcd, abcd_save); 251 252 data += 64; 253 num_blks--; 254 } 255 256 abcd = _mm_shuffle_epi32(abcd, 0x1B); 257 _mm_store_si128((__m128i*) digest, abcd); 258 *(digest+4) = _mm_extract_epi32(e0, 3); 259 } 260 261