Mercurial > libavutil.hg
annotate aes.c @ 212:d31d4880105e libavutil
use 2 state arrays so that fewer temporary variables are needed
author | michael |
---|---|
date | Tue, 16 Jan 2007 17:17:05 +0000 |
parents | 3ef3f2ad70a4 |
children | e50daf73d26c |
rev | line source |
---|---|
164 | 1 /* |
194
3d5efe2de0ae
seems i didnt copy anything from aes128.c so this should be more correct ?
michael
parents:
193
diff
changeset
|
2 * copyright (c) 2007 Michael Niedermayer <michaelni@gmx.at> |
164 | 3 * |
4 * This file is part of FFmpeg. | |
5 * | |
6 * FFmpeg is free software; you can redistribute it and/or | |
7 * modify it under the terms of the GNU Lesser General Public | |
8 * License as published by the Free Software Foundation; either | |
9 * version 2.1 of the License, or (at your option) any later version. | |
10 * | |
11 * FFmpeg is distributed in the hope that it will be useful, | |
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 * Lesser General Public License for more details. | |
15 * | |
16 * You should have received a copy of the GNU Lesser General Public | |
17 * License along with FFmpeg; if not, write to the Free Software | |
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
194
3d5efe2de0ae
seems i didnt copy anything from aes128.c so this should be more correct ?
michael
parents:
193
diff
changeset
|
19 * |
3d5efe2de0ae
seems i didnt copy anything from aes128.c so this should be more correct ?
michael
parents:
193
diff
changeset
|
20 * some optimization ideas from aes128.c by Reimar Doeffinger |
164 | 21 */ |
22 | |
23 #include "common.h" | |
24 #include "aes.h" | |
25 | |
26 typedef struct AVAES{ | |
189 | 27 uint8_t round_key[15][4][4]; |
212
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
28 uint8_t state[2][4][4]; |
164 | 29 int rounds; |
30 }AVAES; | |
31 | |
206 | 32 const int av_aes_size= sizeof(AVAES); |
33 | |
192 | 34 static const uint8_t rcon[10] = { |
35 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1b, 0x36 | |
164 | 36 }; |
37 | |
38 static uint8_t sbox[256]; | |
39 static uint8_t inv_sbox[256]; | |
169 | 40 #ifdef CONFIG_SMALL |
41 static uint32_t enc_multbl[1][256]; | |
42 static uint32_t dec_multbl[1][256]; | |
43 #else | |
44 static uint32_t enc_multbl[4][256]; | |
45 static uint32_t dec_multbl[4][256]; | |
46 #endif | |
164 | 47 |
168 | 48 static inline void addkey(uint64_t state[2], uint64_t round_key[2]){ |
212
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
49 state[2] = state[0] ^ round_key[0]; |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
50 state[3] = state[1] ^ round_key[1]; |
164 | 51 } |
52 | |
212
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
53 static void subshift(uint8_t s0[2][16], uint8_t s1[2][16], uint8_t s3[2][16], uint8_t *box){ |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
54 s0[0][0]=box[s0[1][ 0]]; s0[0][ 4]=box[s0[1][ 4]]; s0[0][ 8]=box[s0[1][ 8]]; s0[0][12]=box[s0[1][12]]; |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
55 s1[0][0]=box[s1[1][ 4]]; s1[0][ 4]=box[s1[1][ 8]]; s1[0][ 8]=box[s1[1][12]]; s1[0][12]=box[s1[1][ 0]]; |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
56 s0[0][2]=box[s0[1][10]]; s0[0][10]=box[s0[1][ 2]]; s0[0][ 6]=box[s0[1][14]]; s0[0][14]=box[s0[1][ 6]]; |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
57 s3[0][0]=box[s3[1][12]]; s3[0][12]=box[s3[1][ 8]]; s3[0][ 8]=box[s3[1][ 4]]; s3[0][ 4]=box[s3[1][ 0]]; |
209 | 58 } |
183 | 59 |
181 | 60 #define ROT(x,s) ((x<<s)|(x>>(32-s))) |
209 | 61 #if 0 |
169 | 62 static inline void mix(uint8_t state[4][4], uint32_t multbl[4][256]){ |
63 int i; | |
64 for(i=0; i<4; i++) | |
65 #ifdef CONFIG_SMALL | |
66 ((uint32_t *)(state))[i] = multbl[0][state[i][0]] ^ ROT(multbl[0][state[i][1]], 8) | |
67 ^ROT(multbl[0][state[i][2]],16) ^ ROT(multbl[0][state[i][3]],24); | |
68 #else | |
69 ((uint32_t *)(state))[i] = multbl[0][state[i][0]] ^ multbl[1][state[i][1]] | |
70 ^multbl[2][state[i][2]] ^ multbl[3][state[i][3]]; | |
71 #endif | |
164 | 72 } |
209 | 73 #endif |
74 | |
212
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
75 static inline void mix(uint8_t state[2][4][4], uint32_t multbl[4][256], int s1, int s3){ |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
76 ((uint32_t *)(state))[0] = multbl[0][state[1][0][0]] ^ multbl[1][state[1][s1 ][1]] |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
77 ^multbl[2][state[1][2][2]] ^ multbl[3][state[1][s3 ][3]]; |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
78 ((uint32_t *)(state))[1] = multbl[0][state[1][1][0]] ^ multbl[1][state[1][s3-1][1]] |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
79 ^multbl[2][state[1][3][2]] ^ multbl[3][state[1][s1-1][3]]; |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
80 ((uint32_t *)(state))[2] = multbl[0][state[1][2][0]] ^ multbl[1][state[1][s3 ][1]] |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
81 ^multbl[2][state[1][0][2]] ^ multbl[3][state[1][s1 ][3]]; |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
82 ((uint32_t *)(state))[3] = multbl[0][state[1][3][0]] ^ multbl[1][state[1][s1-1][1]] |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
83 ^multbl[2][state[1][1][2]] ^ multbl[3][state[1][s3-1][3]]; |
209 | 84 } |
164 | 85 |
191
a2a3c80706e5
merge encrypt and decrypt so the source is simpler and the compiler can choose with inlining if it wants speed or small size
michael
parents:
190
diff
changeset
|
86 static inline void crypt(AVAES *a, int s, uint8_t *sbox, uint32_t *multbl){ |
210 | 87 int r; |
164 | 88 |
188 | 89 for(r=a->rounds; r>1; r--){ |
189 | 90 addkey(a->state, a->round_key[r]); |
212
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
91 mix(a->state, multbl, 3-s, 1+s); |
164 | 92 } |
189 | 93 addkey(a->state, a->round_key[1]); |
212
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
94 subshift(a->state[0][0], a->state[0][0]+3-s, a->state[0][0]+1+s, sbox); |
189 | 95 addkey(a->state, a->round_key[0]); |
164 | 96 } |
97 | |
208
d846d66f99e9
make aes_en/decrypt() static until we decided on the public API
michael
parents:
207
diff
changeset
|
98 static void aes_decrypt(AVAES *a){ |
191
a2a3c80706e5
merge encrypt and decrypt so the source is simpler and the compiler can choose with inlining if it wants speed or small size
michael
parents:
190
diff
changeset
|
99 crypt(a, 0, inv_sbox, dec_multbl); |
a2a3c80706e5
merge encrypt and decrypt so the source is simpler and the compiler can choose with inlining if it wants speed or small size
michael
parents:
190
diff
changeset
|
100 } |
a2a3c80706e5
merge encrypt and decrypt so the source is simpler and the compiler can choose with inlining if it wants speed or small size
michael
parents:
190
diff
changeset
|
101 |
208
d846d66f99e9
make aes_en/decrypt() static until we decided on the public API
michael
parents:
207
diff
changeset
|
102 static void aes_encrypt(AVAES *a){ |
191
a2a3c80706e5
merge encrypt and decrypt so the source is simpler and the compiler can choose with inlining if it wants speed or small size
michael
parents:
190
diff
changeset
|
103 crypt(a, 2, sbox, enc_multbl); |
178 | 104 } |
105 | |
195 | 106 static void init_multbl2(uint8_t tbl[1024], int c[4], uint8_t *log8, uint8_t *alog8, uint8_t *sbox){ |
200 | 107 int i, j; |
183 | 108 for(i=0; i<1024; i++){ |
204
b7ac7733b2f4
replace / by >> this makes the .o ~150 bytes smaller
michael
parents:
203
diff
changeset
|
109 int x= sbox[i>>2]; |
183 | 110 if(x) tbl[i]= alog8[ log8[x] + log8[c[i&3]] ]; |
111 } | |
199 | 112 #ifndef CONFIG_SMALL |
200 | 113 for(j=256; j<1024; j++) |
114 for(i=0; i<4; i++) | |
115 tbl[4*j+i]= tbl[4*j + ((i-1)&3) - 1024]; | |
199 | 116 #endif |
183 | 117 } |
118 | |
164 | 119 // this is based on the reference AES code by Paulo Barreto and Vincent Rijmen |
206 | 120 int av_aes_init(AVAES *a, uint8_t *key, int key_bits, int decrypt) { |
164 | 121 int i, j, t, rconpointer = 0; |
122 uint8_t tk[8][4]; | |
204
b7ac7733b2f4
replace / by >> this makes the .o ~150 bytes smaller
michael
parents:
203
diff
changeset
|
123 int KC= key_bits>>5; |
167 | 124 int rounds= KC + 6; |
169 | 125 uint8_t log8[256]; |
126 uint8_t alog8[512]; | |
164 | 127 |
205
0723fdb8d7a8
check the last? entry written for detecting already initalized tables
michael
parents:
204
diff
changeset
|
128 if(!enc_multbl[4][1023]){ |
164 | 129 j=1; |
130 for(i=0; i<255; i++){ | |
131 alog8[i]= | |
132 alog8[i+255]= j; | |
133 log8[j]= i; | |
134 j^= j+j; | |
135 if(j>255) j^= 0x11B; | |
136 } | |
137 for(i=0; i<256; i++){ | |
138 j= i ? alog8[255-log8[i]] : 0; | |
165 | 139 j ^= (j<<1) ^ (j<<2) ^ (j<<3) ^ (j<<4); |
140 j = (j ^ (j>>8) ^ 99) & 255; | |
164 | 141 inv_sbox[j]= i; |
142 sbox [i]= j; | |
143 } | |
184 | 144 init_multbl2(dec_multbl[0], (int[4]){0xe, 0x9, 0xd, 0xb}, log8, alog8, inv_sbox); |
183 | 145 init_multbl2(enc_multbl[0], (int[4]){0x2, 0x1, 0x1, 0x3}, log8, alog8, sbox); |
164 | 146 } |
147 | |
167 | 148 if(key_bits!=128 && key_bits!=192 && key_bits!=256) |
206 | 149 return -1; |
164 | 150 |
167 | 151 a->rounds= rounds; |
166 | 152 |
164 | 153 memcpy(tk, key, KC*4); |
154 | |
174
263bbdc10c1e
simplify round_key generation by writing over the end but ensuring that theres some irrelevant stuff afterwards
michael
parents:
173
diff
changeset
|
155 for(t= 0; t < (rounds+1)*4;) { |
197 | 156 memcpy(a->round_key[0][t], tk, KC*4); |
174
263bbdc10c1e
simplify round_key generation by writing over the end but ensuring that theres some irrelevant stuff afterwards
michael
parents:
173
diff
changeset
|
157 t+= KC; |
164 | 158 |
159 for(i = 0; i < 4; i++) | |
173 | 160 tk[0][i] ^= sbox[tk[KC-1][(i+1)&3]]; |
164 | 161 tk[0][0] ^= rcon[rconpointer++]; |
162 | |
163 for(j = 1; j < KC; j++){ | |
204
b7ac7733b2f4
replace / by >> this makes the .o ~150 bytes smaller
michael
parents:
203
diff
changeset
|
164 if(KC != 8 || j != KC>>1) |
175 | 165 for(i = 0; i < 4; i++) tk[j][i] ^= tk[j-1][i]; |
164 | 166 else |
175 | 167 for(i = 0; i < 4; i++) tk[j][i] ^= sbox[tk[j-1][i]]; |
164 | 168 } |
169 } | |
184 | 170 |
189 | 171 if(decrypt){ |
190 | 172 for(i=1; i<rounds; i++){ |
212
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
173 uint8_t tmp[2][16]; |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
174 memcpy(tmp[1], a->round_key[i][0], 16); |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
175 subshift(tmp[0], tmp[0]+3, tmp[0]+1, sbox); |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
176 memcpy(tmp[1], tmp[0], 16); |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
177 mix(tmp, dec_multbl, 1, 3); |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
178 memcpy(a->round_key[i][0], tmp[0], 16); |
190 | 179 } |
197 | 180 }else{ |
204
b7ac7733b2f4
replace / by >> this makes the .o ~150 bytes smaller
michael
parents:
203
diff
changeset
|
181 for(i=0; i<(rounds+1)>>1; i++){ |
197 | 182 for(j=0; j<16; j++) |
183 FFSWAP(int, a->round_key[i][0][j], a->round_key[rounds-i][0][j]); | |
184 } | |
187 | 185 } |
184 | 186 |
206 | 187 return 0; |
164 | 188 } |
189 | |
190 #ifdef TEST | |
207
b5ff46799eea
move #include log.h into the selftesting code where it belongs
michael
parents:
206
diff
changeset
|
191 #include "log.h" |
164 | 192 |
193 int main(){ | |
195 | 194 int i,j; |
206 | 195 AVAES ae, ad, b; |
198 | 196 uint8_t rkey[2][16]= { |
197 {0}, | |
198 {0x10, 0xa5, 0x88, 0x69, 0xd7, 0x4b, 0xe5, 0xa3, 0x74, 0xcf, 0x86, 0x7c, 0xfb, 0x47, 0x38, 0x59}}; | |
199 uint8_t pt[16], rpt[2][16]= { | |
200 {0x6a, 0x84, 0x86, 0x7c, 0xd7, 0x7e, 0x12, 0xad, 0x07, 0xea, 0x1b, 0xe8, 0x95, 0xc5, 0x3f, 0xa3}, | |
201 {0}}; | |
202 uint8_t rct[2][16]= { | |
203 {0x73, 0x22, 0x81, 0xc0, 0xa0, 0xaa, 0xb8, 0xf7, 0xa5, 0x4a, 0x0c, 0x67, 0xa0, 0xc4, 0x5e, 0xcf}, | |
204 {0x6d, 0x25, 0x1e, 0x69, 0x44, 0xb0, 0x51, 0xe0, 0x4e, 0xaa, 0x6f, 0xb4, 0xdb, 0xf7, 0x84, 0x65}}; | |
164 | 205 |
206 | 206 av_aes_init(&ae, "PI=3.141592654..", 128, 0); |
207 av_aes_init(&ad, "PI=3.141592654..", 128, 1); | |
179 | 208 av_log_level= AV_LOG_DEBUG; |
209 | |
198 | 210 for(i=0; i<2; i++){ |
206 | 211 av_aes_init(&b, rkey[i], 128, 1); |
212 memcpy(b.state, rct[i], 16); | |
208
d846d66f99e9
make aes_en/decrypt() static until we decided on the public API
michael
parents:
207
diff
changeset
|
213 aes_decrypt(&b); |
198 | 214 for(j=0; j<16; j++) |
212
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
215 if(rpt[i][j] != b.state[1][0][j]) |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
216 av_log(NULL, AV_LOG_ERROR, "%d %02X %02X\n", j, rpt[i][j], b.state[1][0][j]); |
198 | 217 } |
183 | 218 |
164 | 219 for(i=0; i<10000; i++){ |
178 | 220 for(j=0; j<16; j++){ |
221 pt[j]= random(); | |
222 } | |
206 | 223 memcpy(ae.state, pt, 16); |
179 | 224 {START_TIMER |
208
d846d66f99e9
make aes_en/decrypt() static until we decided on the public API
michael
parents:
207
diff
changeset
|
225 aes_encrypt(&ae); |
178 | 226 if(!(i&(i-1))) |
212
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
227 av_log(NULL, AV_LOG_ERROR, "%02X %02X %02X %02X\n", ae.state[1][0][0], ae.state[1][1][1], ae.state[1][2][2], ae.state[1][3][3]); |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
228 memcpy(ad.state[0], ae.state[1], 16); |
208
d846d66f99e9
make aes_en/decrypt() static until we decided on the public API
michael
parents:
207
diff
changeset
|
229 aes_decrypt(&ad); |
179 | 230 STOP_TIMER("aes")} |
178 | 231 for(j=0; j<16; j++){ |
212
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
232 if(pt[j] != ad.state[1][0][j]){ |
d31d4880105e
use 2 state arrays so that fewer temporary variables are needed
michael
parents:
211
diff
changeset
|
233 av_log(NULL, AV_LOG_ERROR, "%d %d %02X %02X\n", i,j, pt[j], ad.state[1][0][j]); |
178 | 234 } |
235 } | |
164 | 236 } |
237 return 0; | |
238 } | |
239 #endif |