aes.asm 10.6 KB
Newer Older
1
! -*- mode: asm; asm-comment-char: ?!; -*-  
Niels Möller's avatar
Niels Möller committed
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
! nettle, low-level cryptographics library
! 
! Copyright (C) 2002 Niels Mller
!  
! The nettle library is free software; you can redistribute it and/or modify
! it under the terms of the GNU Lesser General Public License as published by
! the Free Software Foundation; either version 2.1 of the License, or (at your
! option) any later version.
! 
! The nettle library is distributed in the hope that it will be useful, but
! WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
! or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
! License for more details.
! 
! You should have received a copy of the GNU Lesser General Public License
! along with the nettle library; see the file COPYING.LIB.  If not, write to
! the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
! MA 02111-1307, USA.

21 22
! NOTE: Some of the %g registers are reserved for operating system etc
! (see gcc/config/sparc.h). We should use only %g1-%g3 to be safe.
Niels Möller's avatar
Niels Möller committed
23
	
24
	! Used registers:	%l0,1,2,3,4,5,6,7
25 26
	!			%i0,1,2,3,4 (%i6=%fp, %i7 = return)
	!			%o0,1,2,3,4,7 (%o6=%sp)
27
	!			%g1,2,3,5
Niels Möller's avatar
Niels Möller committed
28
	
29
	.file	"aes.asm"
Niels Möller's avatar
Niels Möller committed
30
	
31
! Arguments
32
define(ctx, %i0)
33
define(T, %i1)
34
define(length, %i2)
35
define(dst, %i3)
36
define(src, %i4)
37

38
! Loop invariants
39
define(wtxt, %l0)
40
define(tmp, %l1)
41
define(diff, %l2)
42
define(nrounds, %l3)
43 44

! Loop variables
45
define(round, %l4)
46
define(i, %l5)
47
define(key, %o4)
48

49 50 51
! Further loop invariants
define(T0, %l6)
define(T1, %l7)
52
define(T2, %g5)
53
define(T3, %o7)
54 55

C IDX1 cointains the permutation values * 4 + 2
56
define(IDX1, <T + AES_SIDX1 >)
57
C IDX3 cointains the permutation values * 4
58
define(IDX3, <T + AES_SIDX3 >)
59

60
! Teporaries
61
define(t0, %o0)
62
define(t1, %o1)
63
define(t2, %o2)
64 65
define(t3, %o3)

66 67 68
C AES_LOAD(i)
C Get one word of input, XOR with first subkey, store in wtxt
define(<AES_LOAD>, <
69 70 71 72 73
	ldub	[src+3], t3
	ldub	[src+2], t2
	sll	t3, 24, t3
	ldub	[src+1], t1
	
74
	sll	t2, 16, t2
75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
	or	t3, t2, t3
	ldub	[src], t0
	sll	t1, 8, t1
	
	! Get subkey
	ld	[ctx + 0], t2
	or	t3, t1, t3
	or	t3, t0, t3
	xor	t3, t2, t3
	
	st	t3, [wtxt+0]
	add	src, 4, src
	
	C ldub	[src + $1], t0
	C ldub	[src + $1 + 1], t1
	C sll	t1, 8, t1
	C or	t0, t1, t0
	C ldub	[src + $1 + 2], t2
	C sll	t2, 16, t2
	C or	t0, t2, t0
	C ldub	[src + $1 + 3], t1
	C sll	t1, 24, t1
	C or	t0, t1, t0
	C ld	[key + $1], t2
	C xor	t0, t2, t0
	C st	t0, [wtxt + $1]
101 102
	>)dnl

Niels Möller's avatar
Niels Möller committed
103 104 105 106 107 108 109 110 111
C AES_ROUND(i)
C Compute one word in the round function. 
C Input in wtxt, output stored in tmp + i.
C
C The comments mark which j in T->table[j][ Bj(wtxt[IDXi(i)]) ]
C the instruction is a part of. 
C
C The code uses the register %o[j], aka tj, as the primary 
C register for that sub-expression. True for j==1,3.
112 113
define(<AES_ROUND>, <
	ld	[IDX1+$1], t1		! 1
114
	ldub	[wtxt+$1+3], t0		! 0
115
	ldub	[wtxt+t1], t1		! 1
116
	sll	t0, 2, t0		! 0
117
	
118
	ld	[T0+t0], t0		! 0
119
	sll	t1, 2, t1		! 1
120
	ld	[T1+t1], t1		! 1
121
	ld	[IDX3+$1], t2		! 3
122 123
	
	xor	t0, t1, t0		! 0, 1
124
	! IDX2(j) = j XOR 2
125 126 127
	ldub	[wtxt+eval($1 ^ 8)+1], t1	! 2
	ldub	[wtxt+t2], t2		! 3
	sll	t1, 2, t1		! 2
128
	
129 130 131 132
	ld	[T2+t1], t1		! 2
	sll	t2, 2, t2		! 3
	ld	[T3+t2], t2		! 3
	xor	t0, t1, t0		! 0, 1, 2
133
	
134 135
	! Fetch roundkey
	ld	[key + $1], t1
136
	xor	t0, t2, t0		! 0, 1, 2, 3
137
	xor	t0, t1, t0
138
	st	t0, [tmp + $1]>)dnl
Niels Möller's avatar
Niels Möller committed
139 140 141

C AES_FINAL_ROUND(i)
C Compute one word in the final round function. 
142
C Input in wtxt, output converted to an octet string and stored at dst. 
Niels Möller's avatar
Niels Möller committed
143 144 145 146 147
C
C The comments mark which j in T->table[j][ Bj(wtxt[IDXi(i)]) ]
C the instruction is a part of. 
define(<AES_FINAL_ROUND>, <
	ld	[IDX1+$1], t1		! 1
148
	ldub	[wtxt+$1+3], t0		! 0
Niels Möller's avatar
Niels Möller committed
149
	ldub	[wtxt+t1], t1		! 1
150
	ldub	[T+t0], t0		! 0
151
	
152
	ldub	[T+t1], t1		! 1
153
	ld	[IDX3 + $1], t2		! 3
154 155 156
	sll	t1, 8, t1		! 1
	or	t0, t1, t0		! 0, 1
	
Niels Möller's avatar
Niels Möller committed
157
	! IDX2(j) = j XOR 2
158
	ldub	[wtxt+eval($1 ^ 8)+1], t1	! 2
159
	ldub	[wtxt+t2], t2		! 3
160
	ldub	[T+t1], t1		! 2
161
	ldub	[T+t2], t2		! 3
Niels Möller's avatar
Niels Möller committed
162
	
163 164 165 166
	sll	t1, 16, t1		! 2
	or	t0, t1, t0		! 0, 1, 2
	sll	t2, 24, t2		! 3
	ld	[key + $1], t1
Niels Möller's avatar
Niels Möller committed
167
	
168 169 170 171
	or	t0, t2, t0		! 0, 1, 2, 3
	xor	t0, t1, t0
	srl	t0, 24, t1
	stb	t1, [dst+$1+3]
Niels Möller's avatar
Niels Möller committed
172
	
173 174
	srl	t0, 16, t1
	stb	t1, [dst+$1+2]
Niels Möller's avatar
Niels Möller committed
175
	srl	t0, 8, t1
176
	stb	t1, [dst+$1+1]
Niels Möller's avatar
Niels Möller committed
177
	
178
	stb	t0, [dst+$1]>)dnl
179
	
180 181 182 183 184 185 186 187
C The stack frame looks like
C
C %fp -   4: OS-dependent link field
C %fp -   8: OS-dependent link field
C %fp -  24: tmp, uint32_t[4]
C %fp -  40: wtxt, uint32_t[4]
C %fp - 136: OS register save area. 
define(<FRAME_SIZE>, 136)
188

189 190 191 192 193 194
	.section	".text"
	.align 16
	.global _aes_crypt
	.type	_aes_crypt,#function
	.proc	020
	
195
_aes_crypt:
196
	save	%sp, -FRAME_SIZE, %sp
197
	cmp	length, 0
198
	be	.Lend
199 200 201 202
	! wtxt
	add	%fp, -24, wtxt
	
	add	%fp, -40, tmp
203

204
	ld	[ctx + AES_NROUNDS], nrounds
205 206
	! Compute xor, so that we can swap efficiently.
	xor	wtxt, tmp, diff
207
	! The loop variable will be multiplied by 16.
208 209
	! More loop invariants
	add	T, AES_TABLE0, T0
Niels Möller's avatar
Niels Möller committed
210
	
211
	add	T, AES_TABLE1, T1
212 213
	add	T, AES_TABLE2, T2
	add	T, AES_TABLE3, T3
Niels Möller's avatar
Niels Möller committed
214
	
215
	! Read src, and add initial subkey
216
	! Difference between ctx and src.
Niels Möller's avatar
Niels Möller committed
217
	! NOTE: These instructions are duplicated in the delay slot,
218
	! and the instruction before the branch.
219 220 221
	sub	ctx, src, %g2
	! Difference between wtxt and src
	sub	wtxt, src, %g3
222
.Lblock_loop:
223 224
	! For stop condition. Note that src is incremented in the
	! delay slot
225
	add	src, 8, %g1
226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246

	C AES_LOAD(0)	! i = 0
	C AES_LOAD(4)	! i = 1
	C AES_LOAD(8)	! i = 2
	C AES_LOAD(12)	! i = 3
	C add	src, 16, src
			
C .Lsource_loop:
	C Begin loop
	C i = 0
	ldub	[src+3], t3
	ldub	[src+2], t2
	sll	t3, 24, t3
	ldub	[src+1], t1
	
	sll	t2, 16, t2
	or	t3, t2, t3
	ldub	[src], t0
	sll	t1, 8, t1
	
	! Get subkey
247
	ld	[ctx + 0], t2
248 249 250 251
	or	t3, t1, t3
	or	t3, t0, t3
	xor	t3, t2, t3
	
252
	st	t3, [wtxt+0]
253 254 255 256 257 258 259 260 261 262 263 264 265 266
	add	src, 4, src

	C i = 1
	ldub	[src+3], t3
	ldub	[src+2], t2
	sll	t3, 24, t3
	ldub	[src+1], t1
	
	sll	t2, 16, t2
	or	t3, t2, t3
	ldub	[src], t0
	sll	t1, 8, t1
	
	! Get subkey
267
	ld	[ctx + 4], t2
268 269 270 271 272
	or	t3, t1, t3
	or	t3, t0, t3
	xor	t3, t2, t3
	
	C cmp	src, %g1
273
	st	t3, [wtxt + 4]
274 275 276 277 278 279 280 281 282 283 284 285
	C bleu	.Lsource_loop
	add	src, 4, src
	C i = 2
	ldub	[src+3], t3
	ldub	[src+2], t2
	sll	t3, 24, t3
	ldub	[src+1], t1
	
	sll	t2, 16, t2
	or	t3, t2, t3
	ldub	[src], t0
	sll	t1, 8, t1
286
	
287 288 289 290 291 292 293 294 295 296 297
	! Get subkey
	ld	[src+%g2], t2
	or	t3, t1, t3
	or	t3, t0, t3
	xor	t3, t2, t3
	
	C cmp	src, %g1
	st	t3, [src+%g3]
	C bleu	.Lsource_loop
	add	src, 4, src
	C i = 3
298 299 300 301
	ldub	[src+3], t3
	ldub	[src+2], t2
	sll	t3, 24, t3
	ldub	[src+1], t1
302
	
303 304 305 306
	sll	t2, 16, t2
	or	t3, t2, t3
	ldub	[src], t0
	sll	t1, 8, t1
307
	
308
	! Get subkey
309
	ld	[src+%g2], t2
310 311
	or	t3, t1, t3
	or	t3, t0, t3
312
	xor	t3, t2, t3
313
	
314
	C cmp	src, %g1
315
	st	t3, [src+%g3]
316
	C bleu	.Lsource_loop
317
	add	src, 4, src
318 319

	C End loop
320
	
321
	sub	nrounds, 1, round
322
	add	ctx, 16, key
Niels Möller's avatar
Niels Möller committed
323

324
.Lround_loop:
325

Niels Möller's avatar
Niels Möller committed
326 327 328 329
	AES_ROUND(0)	! i = 0
	AES_ROUND(4)	! i = 1
	AES_ROUND(8)	! i = 2
	AES_ROUND(12)	! i = 3
330
			
331
	! switch roles for tmp and wtxt
332
	xor	wtxt, diff, wtxt
333
	xor	tmp, diff, tmp
334

335
	subcc	round, 1, round
336
	bne	.Lround_loop
Niels Möller's avatar
Niels Möller committed
337
	add	key, 16, key
338

339
	C Final round, and storage of the output
340

Niels Möller's avatar
Niels Möller committed
341
	AES_FINAL_ROUND(0)	! i = 0
Niels Möller's avatar
Niels Möller committed
342 343 344
	AES_FINAL_ROUND(4)	! i = 1
	AES_FINAL_ROUND(8)	! i = 2
	AES_FINAL_ROUND(12)	! i = 3
345 346
	add	dst, 16, dst
		
347
	addcc	length, -16, length
348
	sub	ctx, src, %g2
349
	
350
	bne	.Lblock_loop
351
	sub	wtxt, src, %g3
352

353
.Lend:
354 355
	ret
	restore
356 357 358 359
.LLFE1:
.LLfe1:
	.size	_aes_crypt,.LLfe1-_aes_crypt

360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412
	! Benchmarks on my slow sparcstation:	
	! Original C code	
	! aes128 (ECB encrypt): 14.36s, 0.696MB/s
	! aes128 (ECB decrypt): 17.19s, 0.582MB/s
	! aes128 (CBC encrypt): 16.08s, 0.622MB/s
	! aes128 ((CBC decrypt)): 18.79s, 0.532MB/s
	! 
	! aes192 (ECB encrypt): 16.85s, 0.593MB/s
	! aes192 (ECB decrypt): 19.64s, 0.509MB/s
	! aes192 (CBC encrypt): 18.43s, 0.543MB/s
	! aes192 (CBC decrypt): 20.76s, 0.482MB/s
	! 
	! aes256 (ECB encrypt): 19.12s, 0.523MB/s
	! aes256 (ECB decrypt): 22.57s, 0.443MB/s
	! aes256 (CBC encrypt): 20.92s, 0.478MB/s
	! aes256 (CBC decrypt): 23.22s, 0.431MB/s

	! After unrolling key_addition32, and getting rid of
	! some sll x, 2, x, encryption speed is 0.760 MB/s.

	! Next, the C code was optimized to use larger tables and
	! no rotates. New timings:
	! aes128 (ECB encrypt): 13.10s, 0.763MB/s
	! aes128 (ECB decrypt): 11.51s, 0.869MB/s
	! aes128 (CBC encrypt): 15.15s, 0.660MB/s
	! aes128 (CBC decrypt): 13.10s, 0.763MB/s
	! 
	! aes192 (ECB encrypt): 15.68s, 0.638MB/s
	! aes192 (ECB decrypt): 13.59s, 0.736MB/s
	! aes192 (CBC encrypt): 17.65s, 0.567MB/s
	! aes192 (CBC decrypt): 15.31s, 0.653MB/s
	! 
	! aes256 (ECB encrypt): 17.95s, 0.557MB/s
	! aes256 (ECB decrypt): 15.90s, 0.629MB/s
	! aes256 (CBC encrypt): 20.16s, 0.496MB/s
	! aes256 (CBC decrypt): 17.47s, 0.572MB/s

	! After optimization using pre-shifted indices
	! (AES_SIDX[1-3]): 
	! aes128 (ECB encrypt): 12.46s, 0.803MB/s
	! aes128 (ECB decrypt): 10.74s, 0.931MB/s
	! aes128 (CBC encrypt): 17.74s, 0.564MB/s
	! aes128 (CBC decrypt): 12.43s, 0.805MB/s
	! 
	! aes192 (ECB encrypt): 14.59s, 0.685MB/s
	! aes192 (ECB decrypt): 12.76s, 0.784MB/s
	! aes192 (CBC encrypt): 19.97s, 0.501MB/s
	! aes192 (CBC decrypt): 14.46s, 0.692MB/s
	! 
	! aes256 (ECB encrypt): 17.00s, 0.588MB/s
	! aes256 (ECB decrypt): 14.81s, 0.675MB/s
	! aes256 (CBC encrypt): 22.65s, 0.442MB/s
	! aes256 (CBC decrypt): 16.46s, 0.608MB/s
413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445

	! After implementing double buffering
	! aes128 (ECB encrypt): 12.59s, 0.794MB/s
	! aes128 (ECB decrypt): 10.56s, 0.947MB/s
	! aes128 (CBC encrypt): 17.91s, 0.558MB/s
	! aes128 (CBC decrypt): 12.30s, 0.813MB/s
	! 
	! aes192 (ECB encrypt): 15.03s, 0.665MB/s
	! aes192 (ECB decrypt): 12.56s, 0.796MB/s
	! aes192 (CBC encrypt): 20.30s, 0.493MB/s
	! aes192 (CBC decrypt): 14.26s, 0.701MB/s
	! 
	! aes256 (ECB encrypt): 17.30s, 0.578MB/s
	! aes256 (ECB decrypt): 14.51s, 0.689MB/s
	! aes256 (CBC encrypt): 22.75s, 0.440MB/s
	! aes256 (CBC decrypt): 16.35s, 0.612MB/s
	
	! After reordering aes-encrypt.c and aes-decypt.c
	! (the order probably causes strange cache-effects):
	! aes128 (ECB encrypt): 9.21s, 1.086MB/s
	! aes128 (ECB decrypt): 11.13s, 0.898MB/s
	! aes128 (CBC encrypt): 14.12s, 0.708MB/s
	! aes128 (CBC decrypt): 13.77s, 0.726MB/s
	! 
	! aes192 (ECB encrypt): 10.86s, 0.921MB/s
	! aes192 (ECB decrypt): 13.17s, 0.759MB/s
	! aes192 (CBC encrypt): 15.74s, 0.635MB/s
	! aes192 (CBC decrypt): 15.91s, 0.629MB/s
	! 
	! aes256 (ECB encrypt): 12.71s, 0.787MB/s
	! aes256 (ECB decrypt): 15.38s, 0.650MB/s
	! aes256 (CBC encrypt): 17.49s, 0.572MB/s
	! aes256 (CBC decrypt): 17.87s, 0.560MB/s
446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463

	! After further optimizations of the initial and final loops,
	! source_loop and final_loop. 
	! aes128 (ECB encrypt): 8.07s, 1.239MB/s
	! aes128 (ECB decrypt): 9.48s, 1.055MB/s
	! aes128 (CBC encrypt): 12.76s, 0.784MB/s
	! aes128 (CBC decrypt): 12.15s, 0.823MB/s
	! 
	! aes192 (ECB encrypt): 9.43s, 1.060MB/s
	! aes192 (ECB decrypt): 11.20s, 0.893MB/s
	! aes192 (CBC encrypt): 14.19s, 0.705MB/s
	! aes192 (CBC decrypt): 13.97s, 0.716MB/s
	! 
	! aes256 (ECB encrypt): 10.81s, 0.925MB/s
	! aes256 (ECB decrypt): 12.92s, 0.774MB/s
	! aes256 (CBC encrypt): 15.59s, 0.641MB/s
	! aes256 (CBC decrypt): 15.76s, 0.635MB/s