2 ---------------------------------------------------------------------------
3 Copyright (c) 1998-2013, Brian Gladman, Worcester, UK. All rights reserved.
5 The redistribution and use of this software (with or without changes)
6 is allowed without the payment of fees or royalties provided that:
8 source code distributions include the above copyright notice, this
9 list of conditions and the following disclaimer;
11 binary distributions include the above copyright notice, this list
12 of conditions and the following disclaimer in their documentation.
14 This software is provided 'as is' with no explicit or implied warranties
15 in respect of its operation, including, but not limited to, correctness
16 and fitness for purpose.
17 ---------------------------------------------------------------------------
18 Issue Date: 20/12/2007
24 #if defined( USE_INTEL_AES_IF_PRESENT )
27 /* map names here to provide the external API ('name' -> 'aes_name') */
28 # define aes_xi(x) aes_ ## x
31 #if defined(__cplusplus)
36 #define si(y,x,k,c) (s(y,c) = word_in(x, c) ^ (k)[c])
37 #define so(y,x,c) word_out(y, c, s(x,c))
40 #define locals(y,x) x[4],y[4]
42 #define locals(y,x) x##0,x##1,x##2,x##3,y##0,y##1,y##2,y##3
45 #define l_copy(y, x) s(y,0) = s(x,0); s(y,1) = s(x,1); \
46 s(y,2) = s(x,2); s(y,3) = s(x,3);
47 #define state_in(y,x,k) si(y,x,k,0); si(y,x,k,1); si(y,x,k,2); si(y,x,k,3)
48 #define state_out(y,x) so(y,x,0); so(y,x,1); so(y,x,2); so(y,x,3)
49 #define round(rm,y,x,k) rm(y,x,k,0); rm(y,x,k,1); rm(y,x,k,2); rm(y,x,k,3)
51 #if ( FUNCS_IN_C & ENCRYPTION_IN_C )
53 /* Visual C++ .Net v7.1 provides the fastest encryption code when using
54 Pentium optimiation with small code but this is poor for decryption
55 so we need to control this with the following VC++ pragmas
58 #if defined( _MSC_VER ) && !defined( _WIN64 )
59 #pragma optimize( "s", on )
62 /* Given the column (c) of the output state variable, the following
63 macros give the input state variables which are needed in its
64 computation for each row (r) of the state. All the alternative
65 macros give the same end values but expand into different ways
66 of calculating these values. In particular the complex macro
67 used for dynamically variable block sizes is designed to expand
68 to a compile time constant whenever possible but will expand to
69 conditional clauses on some branches (I am grateful to Frank
70 Yellin for this construction)
73 #define fwd_var(x,r,c)\
74 ( r == 0 ? ( c == 0 ? s(x,0) : c == 1 ? s(x,1) : c == 2 ? s(x,2) : s(x,3))\
75 : r == 1 ? ( c == 0 ? s(x,1) : c == 1 ? s(x,2) : c == 2 ? s(x,3) : s(x,0))\
76 : r == 2 ? ( c == 0 ? s(x,2) : c == 1 ? s(x,3) : c == 2 ? s(x,0) : s(x,1))\
77 : ( c == 0 ? s(x,3) : c == 1 ? s(x,0) : c == 2 ? s(x,1) : s(x,2)))
81 #define fwd_rnd(y,x,k,c) (s(y,c) = (k)[c] ^ four_tables(x,t_use(f,n),fwd_var,rf1,c))
82 #elif defined(FT1_SET)
84 #define fwd_rnd(y,x,k,c) (s(y,c) = (k)[c] ^ one_table(x,upr,t_use(f,n),fwd_var,rf1,c))
86 #define fwd_rnd(y,x,k,c) (s(y,c) = (k)[c] ^ fwd_mcol(no_table(x,t_use(s,box),fwd_var,rf1,c)))
90 #define fwd_lrnd(y,x,k,c) (s(y,c) = (k)[c] ^ four_tables(x,t_use(f,l),fwd_var,rf1,c))
91 #elif defined(FL1_SET)
92 #define fwd_lrnd(y,x,k,c) (s(y,c) = (k)[c] ^ one_table(x,ups,t_use(f,l),fwd_var,rf1,c))
94 #define fwd_lrnd(y,x,k,c) (s(y,c) = (k)[c] ^ no_table(x,t_use(s,box),fwd_var,rf1,c))
97 AES_RETURN aes_xi(encrypt)(const unsigned char *in, unsigned char *out, const aes_encrypt_ctx cx[1])
98 { uint32_t locals(b0, b1);
100 #if defined( dec_fmvars )
101 dec_fmvars; /* declare variables for fwd_mcol() if needed */
104 if(cx->inf.b[0] != 10 * 16 && cx->inf.b[0] != 12 * 16 && cx->inf.b[0] != 14 * 16)
108 state_in(b0, in, kp);
110 #if (ENC_UNROLL == FULL)
115 round(fwd_rnd, b1, b0, kp + 1 * N_COLS);
116 round(fwd_rnd, b0, b1, kp + 2 * N_COLS);
119 round(fwd_rnd, b1, b0, kp + 1 * N_COLS);
120 round(fwd_rnd, b0, b1, kp + 2 * N_COLS);
123 round(fwd_rnd, b1, b0, kp + 1 * N_COLS);
124 round(fwd_rnd, b0, b1, kp + 2 * N_COLS);
125 round(fwd_rnd, b1, b0, kp + 3 * N_COLS);
126 round(fwd_rnd, b0, b1, kp + 4 * N_COLS);
127 round(fwd_rnd, b1, b0, kp + 5 * N_COLS);
128 round(fwd_rnd, b0, b1, kp + 6 * N_COLS);
129 round(fwd_rnd, b1, b0, kp + 7 * N_COLS);
130 round(fwd_rnd, b0, b1, kp + 8 * N_COLS);
131 round(fwd_rnd, b1, b0, kp + 9 * N_COLS);
132 round(fwd_lrnd, b0, b1, kp +10 * N_COLS);
137 #if (ENC_UNROLL == PARTIAL)
139 for(rnd = 0; rnd < (cx->inf.b[0] >> 5) - 1; ++rnd)
142 round(fwd_rnd, b1, b0, kp);
144 round(fwd_rnd, b0, b1, kp);
147 round(fwd_rnd, b1, b0, kp);
150 for(rnd = 0; rnd < (cx->inf.b[0] >> 4) - 1; ++rnd)
153 round(fwd_rnd, b1, b0, kp);
158 round(fwd_lrnd, b0, b1, kp);
168 #if ( FUNCS_IN_C & DECRYPTION_IN_C)
170 /* Visual C++ .Net v7.1 provides the fastest encryption code when using
171 Pentium optimiation with small code but this is poor for decryption
172 so we need to control this with the following VC++ pragmas
175 #if defined( _MSC_VER ) && !defined( _WIN64 )
176 #pragma optimize( "t", on )
179 /* Given the column (c) of the output state variable, the following
180 macros give the input state variables which are needed in its
181 computation for each row (r) of the state. All the alternative
182 macros give the same end values but expand into different ways
183 of calculating these values. In particular the complex macro
184 used for dynamically variable block sizes is designed to expand
185 to a compile time constant whenever possible but will expand to
186 conditional clauses on some branches (I am grateful to Frank
187 Yellin for this construction)
190 #define inv_var(x,r,c)\
191 ( r == 0 ? ( c == 0 ? s(x,0) : c == 1 ? s(x,1) : c == 2 ? s(x,2) : s(x,3))\
192 : r == 1 ? ( c == 0 ? s(x,3) : c == 1 ? s(x,0) : c == 2 ? s(x,1) : s(x,2))\
193 : r == 2 ? ( c == 0 ? s(x,2) : c == 1 ? s(x,3) : c == 2 ? s(x,0) : s(x,1))\
194 : ( c == 0 ? s(x,1) : c == 1 ? s(x,2) : c == 2 ? s(x,3) : s(x,0)))
198 #define inv_rnd(y,x,k,c) (s(y,c) = (k)[c] ^ four_tables(x,t_use(i,n),inv_var,rf1,c))
199 #elif defined(IT1_SET)
201 #define inv_rnd(y,x,k,c) (s(y,c) = (k)[c] ^ one_table(x,upr,t_use(i,n),inv_var,rf1,c))
203 #define inv_rnd(y,x,k,c) (s(y,c) = inv_mcol((k)[c] ^ no_table(x,t_use(i,box),inv_var,rf1,c)))
207 #define inv_lrnd(y,x,k,c) (s(y,c) = (k)[c] ^ four_tables(x,t_use(i,l),inv_var,rf1,c))
208 #elif defined(IL1_SET)
209 #define inv_lrnd(y,x,k,c) (s(y,c) = (k)[c] ^ one_table(x,ups,t_use(i,l),inv_var,rf1,c))
211 #define inv_lrnd(y,x,k,c) (s(y,c) = (k)[c] ^ no_table(x,t_use(i,box),inv_var,rf1,c))
214 /* This code can work with the decryption key schedule in the */
215 /* order that is used for encrytpion (where the 1st decryption */
216 /* round key is at the high end ot the schedule) or with a key */
217 /* schedule that has been reversed to put the 1st decryption */
218 /* round key at the low end of the schedule in memory (when */
219 /* AES_REV_DKS is defined) */
223 #define rnd_key(n) (kp + n * N_COLS)
226 #define rnd_key(n) (kp - n * N_COLS)
229 AES_RETURN aes_xi(decrypt)(const unsigned char *in, unsigned char *out, const aes_decrypt_ctx cx[1])
230 { uint32_t locals(b0, b1);
231 #if defined( dec_imvars )
232 dec_imvars; /* declare variables for inv_mcol() if needed */
236 if(cx->inf.b[0] != 10 * 16 && cx->inf.b[0] != 12 * 16 && cx->inf.b[0] != 14 * 16)
239 kp = cx->ks + (key_ofs ? (cx->inf.b[0] >> 2) : 0);
240 state_in(b0, in, kp);
242 #if (DEC_UNROLL == FULL)
244 kp = cx->ks + (key_ofs ? 0 : (cx->inf.b[0] >> 2));
248 round(inv_rnd, b1, b0, rnd_key(-13));
249 round(inv_rnd, b0, b1, rnd_key(-12));
251 round(inv_rnd, b1, b0, rnd_key(-11));
252 round(inv_rnd, b0, b1, rnd_key(-10));
254 round(inv_rnd, b1, b0, rnd_key(-9));
255 round(inv_rnd, b0, b1, rnd_key(-8));
256 round(inv_rnd, b1, b0, rnd_key(-7));
257 round(inv_rnd, b0, b1, rnd_key(-6));
258 round(inv_rnd, b1, b0, rnd_key(-5));
259 round(inv_rnd, b0, b1, rnd_key(-4));
260 round(inv_rnd, b1, b0, rnd_key(-3));
261 round(inv_rnd, b0, b1, rnd_key(-2));
262 round(inv_rnd, b1, b0, rnd_key(-1));
263 round(inv_lrnd, b0, b1, rnd_key( 0));
268 #if (DEC_UNROLL == PARTIAL)
270 for(rnd = 0; rnd < (cx->inf.b[0] >> 5) - 1; ++rnd)
273 round(inv_rnd, b1, b0, kp);
275 round(inv_rnd, b0, b1, kp);
278 round(inv_rnd, b1, b0, kp);
281 for(rnd = 0; rnd < (cx->inf.b[0] >> 4) - 1; ++rnd)
284 round(inv_rnd, b1, b0, kp);
289 round(inv_lrnd, b0, b1, kp);
299 #if defined(__cplusplus)