aes.asm 12.1 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 Möller
!  
! 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
	.section	".text"
32
	.align 16
33
34
	.global _aes_crypt
	.type	_aes_crypt,#function
35
	.proc	020
36

37
! Arguments
38
define(ctx, %i0)
39
define(T, %i1)
40
define(length, %i2)
41
define(dst, %i3)
42
define(src, %i4)
43

44
! Loop invariants
45
define(wtxt, %l0)
46
define(tmp, %l1)
47
define(diff, %l2)
48
define(nrounds, %l3)
49
50

! Loop variables
51
define(round, %l4)
52
define(i, %l5)
53
define(key, %o4)
54

55
56
57
! Further loop invariants
define(T0, %l6)
define(T1, %l7)
58
define(T2, %g5)
59
define(T3, %o7)
60
61
62
63
C define(IDX1, %i5)
define(IDX1, <T + AES_SIDX1 >)
C define(IDX3, %o5)
define(IDX3, <T + AES_SIDX3 >)
64

65
! Teporaries
66
define(t0, %o0)
67
define(t1, %o1)
68
define(t2, %o2)
69
70
define(t3, %o3)

71
72
73
74
75
76
77
78
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)
79

80
_aes_crypt:
81
	save	%sp, -FRAME_SIZE, %sp
82
	cmp	length, 0
83
	be	.Lend
84
85
86
87
	! wtxt
	add	%fp, -24, wtxt
	
	add	%fp, -40, tmp
88

89
	ld	[ctx + AES_NROUNDS], nrounds
90
91
	! Compute xor, so that we can swap efficiently.
	xor	wtxt, tmp, diff
92
	! The loop variable will be multiplied by 16.
93
94
	! More loop invariants
	add	T, AES_TABLE0, T0
Niels Möller's avatar
Niels Möller committed
95
	
96
	add	T, AES_TABLE1, T1
97
98
	add	T, AES_TABLE2, T2
	add	T, AES_TABLE3, T3
99
	C add	T, AES_SIDX1, IDX1
Niels Möller's avatar
Niels Möller committed
100
	
101
	C add	T, AES_SIDX3, IDX3
102
	! Read src, and add initial subkey
103
	! Difference between ctx and src.
Niels Möller's avatar
Niels Möller committed
104
	! NOTE: These instructions are duplicated in the delay slot,
105
	! and the instruction before the branch.
106
107
108
	sub	ctx, src, %g2
	! Difference between wtxt and src
	sub	wtxt, src, %g3
109
.Lblock_loop:
110
111
	! For stop condition. Note that src is incremented in the
	! delay slot
112
	add	src, 8, %g1
113
	
114
.Lsource_loop:
115
116
117
118
	ldub	[src+3], t3
	ldub	[src+2], t2
	sll	t3, 24, t3
	ldub	[src+1], t1
119
	
120
121
122
123
	sll	t2, 16, t2
	or	t3, t2, t3
	ldub	[src], t0
	sll	t1, 8, t1
124
	
125
	! Get subkey
126
	ld	[src+%g2], t2
127
128
	or	t3, t1, t3
	or	t3, t0, t3
129
	xor	t3, t2, t3
130
	
131
	cmp	src, %g1
132
	st	t3, [src+%g3]
133
	bleu	.Lsource_loop
134
	add	src, 4, src
135
	
136
	sub	nrounds, 1, round
137
	add	ctx, 16, key
Niels Möller's avatar
Niels Möller committed
138

139
.Lround_loop:
140
	! The comments mark which j in T->table[j][ Bj(wtxt[IDXi(i)]) ]
141
142
143
144
	! the instruction is a part of. 
	!
	! The code uses the register %o[j], aka tj, as the primary 
	! register for that sub-expression. True for j==1,3.
145

146
147
	C Unrolled inner loop begins
	
148
	C i = 0
149
	ld	[IDX1+0], t1		! 1
150
151
152
	
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
153
	ld	[IDX3+0], t3		! 3
154
155
	
	sll	t1, 2, t1		! 1
156
	ld	[wtxt], t0		! 0
157
158
	! IDX2(j) = j XOR 2
	lduh	[wtxt+8], t2		! 2
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
	and	t0, 255, t0		! 0
	
	ldub	[wtxt+t3], t3		! 3
	sll	t0, 2, t0		! 0
	ld	[T0+t0], t0		! 0
	and	t2, 255, t2		! 2
	
	ld	[T1+t1], t1		! 1
	sll	t2, 2, t2		! 2
	ld	[T2+t2], t2		! 2
	sll	t3, 2, t3		! 3
	
	ld	[T3+t3], t3		! 3
	xor	t0, t1, t0		! 0, 1
	xor	t0, t2, t0		! 0, 1, 2
	! Fetch roundkey
175
	ld	[key], t1
176
177
178
	
	xor	t0, t3, t0		! 0, 1, 2, 3
	xor	t0, t1, t0
179
	st	t0, [tmp]
180
181

	C i = 1
182
	ld	[IDX1+4], t1		! 1
183
	
184
185
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
186
	ld	[IDX3+4], t3		! 3
187
	
188
	sll	t1, 2, t1		! 1
189
	ld	[wtxt+4], t0		! 0
190
191
	! IDX2(j) = j XOR 2
	lduh	[wtxt+12], t2		! 2
Niels Möller's avatar
Niels Möller committed
192
	and	t0, 255, t0		! 0
193
	
194
	ldub	[wtxt+t3], t3		! 3
195
	sll	t0, 2, t0		! 0
196
	ld	[T0+t0], t0		! 0
Niels Möller's avatar
Niels Möller committed
197
	and	t2, 255, t2		! 2
198
	
199
	ld	[T1+t1], t1		! 1
200
	sll	t2, 2, t2		! 2
201
	ld	[T2+t2], t2		! 2
Niels Möller's avatar
Niels Möller committed
202
	sll	t3, 2, t3		! 3
203
	
204
	ld	[T3+t3], t3		! 3
205
206
	xor	t0, t1, t0		! 0, 1
	xor	t0, t2, t0		! 0, 1, 2
207
	! Fetch roundkey
208
	ld	[key+4], t1
209
	
210
	xor	t0, t3, t0		! 0, 1, 2, 3
211
	xor	t0, t1, t0
212
	st	t0, [tmp+4]
213
214

	C = 2
215
	ld	[IDX1+8], t1		! 1
216
217
218
	
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
219
	ld	[IDX3+8], t3		! 3
220
221
	
	sll	t1, 2, t1		! 1
222
	ld	[wtxt+8], t0		! 0
223
224
	! IDX2(j) = j XOR 2
	lduh	[wtxt], t2		! 2
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
	and	t0, 255, t0		! 0
	
	ldub	[wtxt+t3], t3		! 3
	sll	t0, 2, t0		! 0
	ld	[T0+t0], t0		! 0
	and	t2, 255, t2		! 2
	
	ld	[T1+t1], t1		! 1
	sll	t2, 2, t2		! 2
	ld	[T2+t2], t2		! 2
	sll	t3, 2, t3		! 3
	
	ld	[T3+t3], t3		! 3
	xor	t0, t1, t0		! 0, 1
	xor	t0, t2, t0		! 0, 1, 2
	! Fetch roundkey
241
	ld	[key+8], t1
242
243
244
	
	xor	t0, t3, t0		! 0, 1, 2, 3
	xor	t0, t1, t0
245
	st	t0, [tmp+8]
246
247

	C = 3
248
	ld	[IDX1+12], t1		! 1
249
250
251
	
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
252
	ld	[IDX3+12], t3		! 3
253
254
	
	sll	t1, 2, t1		! 1
255
	ld	[wtxt+12], t0		! 0
256
257
	! IDX2(j) = j XOR 2
	lduh	[wtxt+4], t2		! 2
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
	and	t0, 255, t0		! 0
	
	ldub	[wtxt+t3], t3		! 3
	sll	t0, 2, t0		! 0
	ld	[T0+t0], t0		! 0
	and	t2, 255, t2		! 2
	
	ld	[T1+t1], t1		! 1
	sll	t2, 2, t2		! 2
	ld	[T2+t2], t2		! 2
	sll	t3, 2, t3		! 3
	
	ld	[T3+t3], t3		! 3
	xor	t0, t1, t0		! 0, 1
	xor	t0, t2, t0		! 0, 1, 2
	! Fetch roundkey
274
	ld	[key+12], t1
275
276
277
	
	xor	t0, t3, t0		! 0, 1, 2, 3
	xor	t0, t1, t0
278
	st	t0, [tmp+12]
279
			
280
	C Unrolled inner loop ends
281
	
282
	! switch roles for tmp and wtxt
283
	xor	wtxt, diff, wtxt
284
	xor	tmp, diff, tmp
285

286
	subcc	round, 1, round
287
	bne	.Lround_loop
Niels Möller's avatar
Niels Möller committed
288
	add	key, 16, key
289

Niels Möller's avatar
Niels Möller committed
290
	! Final round
291

292
	! Comments mark which j in T->sbox[Bj(wtxt[IDXj(i)])]
293
	! the instruction is part of.
294
295
296
297

	C Unrolled final loop begins

	C i = 0
298
	ld	[IDX1+0], t1 	! 1
299
300
301
	add	wtxt, t1, t1	! 1
	ldub	[t1+2], t1	! 1

302
303
304
	ld	[wtxt+0], t0	! 0
	! IDX2(j) = j XOR 2
	lduh	[wtxt+8], t2	! 2
305
	and	t0, 255, t0	! 0
306
	ld	[IDX3 + 0], t3	! 3
307
308
309
310
311
312
313
314
315
316
317
318
319
	
	and	t2, 255, t2	! 2
	ldub	[T+t1], t1	! 1
	ldub	[T+t0], t0	! 0
	sll	t1, 8, t1	! 1
	
	ldub	[wtxt+t3], t3	! 3
	or	t0, t1, t0	! 0, 1
	ldub	[T+t2], t2	! 2
	ldub	[T+t3], t3	! 3
	
	sll	t2, 16, t2	! 2
	or	t0, t2, t0	! 0, 1, 2
320
	ld	[key + 0], t2
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
	sll	t3, 24, t3	! 3
	
	or	t0, t3, t0	! 0, 1, 2, 3
	xor	t0, t2, t0
	
	srl	t0, 24, t3
	srl	t0, 16, t2
	srl	t0, 8, t1
	stb	t1, [dst+1]
	
	stb	t3, [dst+3]
	stb	t2, [dst+2]
	stb	t0, [dst]
	add	dst, 4, dst
	
	C i = 1
337
	ld	[IDX1+4], t1 	! 1
338

339
340
341
	add	wtxt, t1, t1	! 1
	ldub	[t1+2], t1	! 1

342
	ld	[wtxt+4], t0	! 0
343
344
	! IDX2(j) = j XOR 2
	lduh	[wtxt+12], t2	! 2
345
	and	t0, 255, t0	! 0
346
	ld	[IDX3 + 4], t3	! 3
347
348
349
350
351
352
353
354
355
356
357
358
359
	
	and	t2, 255, t2	! 2
	ldub	[T+t1], t1	! 1
	ldub	[T+t0], t0	! 0
	sll	t1, 8, t1	! 1
	
	ldub	[wtxt+t3], t3	! 3
	or	t0, t1, t0	! 0, 1
	ldub	[T+t2], t2	! 2
	ldub	[T+t3], t3	! 3
	
	sll	t2, 16, t2	! 2
	or	t0, t2, t0	! 0, 1, 2
360
	ld	[key + 4], t2
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
	sll	t3, 24, t3	! 3
	
	or	t0, t3, t0	! 0, 1, 2, 3
	xor	t0, t2, t0
	
	srl	t0, 24, t3
	srl	t0, 16, t2
	srl	t0, 8, t1
	stb	t1, [dst+1]
	
	stb	t3, [dst+3]
	stb	t2, [dst+2]
	stb	t0, [dst]
	add	dst, 4, dst
	
	C i = 2
377
	ld	[IDX1+8], t1 	! 1
378
	
379
380
381
	add	wtxt, t1, t1	! 1
	ldub	[t1+2], t1	! 1

382
383
384
	ld	[wtxt+8], t0	! 0
	! IDX2(j) = j XOR 2
	lduh	[wtxt+0], t2	! 2
385
	and	t0, 255, t0	! 0
386
	ld	[IDX3 + 8], t3	! 3
387
388
389
390
391
392
393
394
395
396
397
398
399
	
	and	t2, 255, t2	! 2
	ldub	[T+t1], t1	! 1
	ldub	[T+t0], t0	! 0
	sll	t1, 8, t1	! 1
	
	ldub	[wtxt+t3], t3	! 3
	or	t0, t1, t0	! 0, 1
	ldub	[T+t2], t2	! 2
	ldub	[T+t3], t3	! 3
	
	sll	t2, 16, t2	! 2
	or	t0, t2, t0	! 0, 1, 2
400
	ld	[key + 8], t2
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
	sll	t3, 24, t3	! 3
	
	or	t0, t3, t0	! 0, 1, 2, 3
	xor	t0, t2, t0
	
	srl	t0, 24, t3
	srl	t0, 16, t2
	srl	t0, 8, t1
	stb	t1, [dst+1]
	
	stb	t3, [dst+3]
	stb	t2, [dst+2]
	stb	t0, [dst]
	add	dst, 4, dst
	
	C i = 3
417
	ld	[IDX1+12], t1 	! 1
418

419
420
	add	wtxt, t1, t1	! 1
	ldub	[t1+2], t1	! 1
421

422
	ld	[wtxt+12], t0	! 0
423
424
	! IDX2(j) = j XOR 2
	lduh	[wtxt+4], t2	! 2
425
	and	t0, 255, t0	! 0
426
	ld	[IDX3 + 12], t3	! 3
427
	
428
	and	t2, 255, t2	! 2
429
	ldub	[T+t1], t1	! 1
430
	ldub	[T+t0], t0	! 0
431
	sll	t1, 8, t1	! 1
432
	
433
	ldub	[wtxt+t3], t3	! 3
434
	or	t0, t1, t0	! 0, 1
435
	ldub	[T+t2], t2	! 2
436
	ldub	[T+t3], t3	! 3
437
	
438
439
	sll	t2, 16, t2	! 2
	or	t0, t2, t0	! 0, 1, 2
440
	ld	[key + 12], t2
441
	sll	t3, 24, t3	! 3
442
	
443
	or	t0, t3, t0	! 0, 1, 2, 3
444
445
446
447
448
	xor	t0, t2, t0
	
	srl	t0, 24, t3
	srl	t0, 16, t2
	srl	t0, 8, t1
449
	stb	t1, [dst+1]
450
	
451
452
453
	stb	t3, [dst+3]
	stb	t2, [dst+2]
	stb	t0, [dst]
454
	add	dst, 4, dst
455
	
456
457
	C Unrolled final loop ends
	
458
	addcc	length, -16, length
459
	sub	ctx, src, %g2
460
	
461
	bne	.Lblock_loop
462
	sub	wtxt, src, %g3
463

464
.Lend:
465
	C add	%sp, FRAME_SIZE, %fp
466
467
	ret
	restore
468
469
470
471
.LLFE1:
.LLfe1:
	.size	_aes_crypt,.LLfe1-_aes_crypt

472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
	! 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
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557

	! 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
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575

	! 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