aes.asm 12.4 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
	!			%i0,1,2,3,4,5 (%i6=%fp, %i7 = return)
	!			%o0,1,2,3,4,5,7 (%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
define(IDX1, %i5)
61
C define(IDX1, <T + AES_SIDX1 >)
62
define(IDX3, %o5)
63

64
! Teporaries
65
define(t0, %o0)
66
define(t1, %o1)
67
define(t2, %o2)
68
69
define(t3, %o3)

70
71
72
73
74
75
76
77
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)
78

79
_aes_crypt:
80
	save	%sp, -FRAME_SIZE, %sp
81
	cmp	length, 0
82
	be	.Lend
83
84
85
86
	! wtxt
	add	%fp, -24, wtxt
	
	add	%fp, -40, tmp
87

88
	ld	[ctx + AES_NROUNDS], nrounds
89
90
	! Compute xor, so that we can swap efficiently.
	xor	wtxt, tmp, diff
91
	! The loop variable will be multiplied by 16.
92
93
	! More loop invariants
	add	T, AES_TABLE0, T0
Niels Möller's avatar
Niels Möller committed
94
	
95
	add	T, AES_TABLE1, T1
96
97
	add	T, AES_TABLE2, T2
	add	T, AES_TABLE3, T3
98
	add	T, AES_SIDX1, IDX1
Niels Möller's avatar
Niels Möller committed
99
	
100
	add	T, AES_SIDX3, IDX3
101
	! Read src, and add initial subkey
102
	! Difference between ctx and src.
Niels Möller's avatar
Niels Möller committed
103
	! NOTE: These instructions are duplicated in the delay slot,
104
	! and the instruction before the branch.
105
106
107
	sub	ctx, src, %g2
	! Difference between wtxt and src
	sub	wtxt, src, %g3
108
.Lblock_loop:
109
110
	! For stop condition. Note that src is incremented in the
	! delay slot
111
	add	src, 8, %g1
112
	
113
.Lsource_loop:
114
115
116
117
	ldub	[src+3], t3
	ldub	[src+2], t2
	sll	t3, 24, t3
	ldub	[src+1], t1
118
	
119
120
121
122
	sll	t2, 16, t2
	or	t3, t2, t3
	ldub	[src], t0
	sll	t1, 8, t1
123
	
124
	! Get subkey
125
	ld	[src+%g2], t2
126
127
	or	t3, t1, t3
	or	t3, t0, t3
128
	xor	t3, t2, t3
129
	
130
	cmp	src, %g1
131
	st	t3, [src+%g3]
132
	bleu	.Lsource_loop
133
	add	src, 4, src
134
	
135
	sub	nrounds, 1, round
136
	add	ctx, 16, key
Niels Möller's avatar
Niels Möller committed
137

138
.Lround_loop:
139
	! The comments mark which j in T->table[j][ Bj(wtxt[IDXi(i)]) ]
140
141
142
143
	! 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.
144

145
146
	C Unrolled inner loop begins
	
147
	C i = 0
148
	ld	[IDX1+0], t1		! 1
149
150
151
	
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
152
	ld	[IDX3+0], t3		! 3
153
154
	
	sll	t1, 2, t1		! 1
155
	ld	[wtxt], t0		! 0
156
157
	! IDX2(j) = j XOR 2
	lduh	[wtxt+8], t2		! 2
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
	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
174
	ld	[key], t1
175
176
177
	
	xor	t0, t3, t0		! 0, 1, 2, 3
	xor	t0, t1, t0
178
	st	t0, [tmp]
179
180

	C i = 1
181
	ld	[IDX1+4], t1		! 1
182
	
183
184
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
185
	ld	[IDX3+4], t3		! 3
186
	
187
	sll	t1, 2, t1		! 1
188
	ld	[wtxt+4], t0		! 0
189
190
	! IDX2(j) = j XOR 2
	lduh	[wtxt+12], t2		! 2
Niels Möller's avatar
Niels Möller committed
191
	and	t0, 255, t0		! 0
192
	
193
	ldub	[wtxt+t3], t3		! 3
194
	sll	t0, 2, t0		! 0
195
	ld	[T0+t0], t0		! 0
Niels Möller's avatar
Niels Möller committed
196
	and	t2, 255, t2		! 2
197
	
198
	ld	[T1+t1], t1		! 1
199
	sll	t2, 2, t2		! 2
200
	ld	[T2+t2], t2		! 2
Niels Möller's avatar
Niels Möller committed
201
	sll	t3, 2, t3		! 3
202
	
203
	ld	[T3+t3], t3		! 3
204
205
	xor	t0, t1, t0		! 0, 1
	xor	t0, t2, t0		! 0, 1, 2
206
	! Fetch roundkey
207
	ld	[key+4], t1
208
	
209
	xor	t0, t3, t0		! 0, 1, 2, 3
210
	xor	t0, t1, t0
211
	st	t0, [tmp+4]
212
213

	C = 2
214
	ld	[IDX1+8], t1		! 1
215
216
217
	
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
218
	ld	[IDX3+8], t3		! 3
219
220
	
	sll	t1, 2, t1		! 1
221
	ld	[wtxt+8], t0		! 0
222
223
	! IDX2(j) = j XOR 2
	lduh	[wtxt], t2		! 2
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
	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
240
	ld	[key+8], t1
241
242
243
	
	xor	t0, t3, t0		! 0, 1, 2, 3
	xor	t0, t1, t0
244
	st	t0, [tmp+8]
245
246

	C = 3
247
	ld	[IDX1+12], t1		! 1
248
249
250
	
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
251
	ld	[IDX3+12], t3		! 3
252
253
	
	sll	t1, 2, t1		! 1
254
	ld	[wtxt+12], t0		! 0
255
256
	! IDX2(j) = j XOR 2
	lduh	[wtxt+4], t2		! 2
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
	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
273
	ld	[key+12], t1
274
275
276
	
	xor	t0, t3, t0		! 0, 1, 2, 3
	xor	t0, t1, t0
277
	st	t0, [tmp+12]
278
			
279
	C Unrolled inner loop ends
280
	
281
	! switch roles for tmp and wtxt
282
	xor	wtxt, diff, wtxt
283
	xor	tmp, diff, tmp
284

285
	subcc	round, 1, round
286
	bne	.Lround_loop
Niels Möller's avatar
Niels Möller committed
287
	add	key, 16, key
288

289
	! final round
290
291
292
	! Use round as the loop variable, as it's already zero
undefine(<i>)
define(i, round)
293

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

	C Unrolled final loop begins
298
	! NOTE: First instruction duplicated in delay slot
299
300

	C i = 0
301
	ld	[IDX1+0], t1 	! 1
302
303
304
	add	wtxt, t1, t1	! 1
	ldub	[t1+2], t1	! 1

305
306
307
	ld	[wtxt+0], t0	! 0
	! IDX2(j) = j XOR 2
	lduh	[wtxt+8], t2	! 2
308
	and	t0, 255, t0	! 0
309
	ld	[IDX3 + 0], t3	! 3
310
311
312
313
314
315
316
317
318
319
320
321
322
	
	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
323
	ld	[key + 0], t2
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
	sll	t3, 24, t3	! 3
	
	or	t0, t3, t0	! 0, 1, 2, 3
	xor	t0, t2, t0
	add	i, 4, i
	
	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
341
	ld	[IDX1+4], t1 	! 1
342

343
344
345
346
347
	xor	i, 8, t2
	add	wtxt, t1, t1	! 1
	ldub	[t1+2], t1	! 1

	ld	[wtxt+i], t0	! 0
348
349
	! IDX2(j) = j XOR 2
	lduh	[wtxt+12], t2	! 2
350
	and	t0, 255, t0	! 0
351
	ld	[IDX3 + 4], t3	! 3
352
353
354
355
356
357
358
359
360
361
362
363
364
	
	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
365
	ld	[key + 4], t2
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
	sll	t3, 24, t3	! 3
	
	or	t0, t3, t0	! 0, 1, 2, 3
	xor	t0, t2, t0
	add	i, 4, i
	
	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
	
	ld	[IDX1+i], t1 	! 1

	C i = 2
	ld	[IDX1+i], t1 	! 1
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
413
414
415
416
417
418
419
420
421
422
423
424
425
	! IDX2(j) = j XOR 2
	xor	i, 8, t2
	add	wtxt, t1, t1	! 1
	ldub	[t1+2], t1	! 1

	ld	[wtxt+i], t0	! 0
	lduh	[wtxt+t2], t2	! 2
	and	t0, 255, t0	! 0
	ld	[IDX3 + i], t3	! 3
	
	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
	ld	[key + i], t2
	sll	t3, 24, t3	! 3
	
	or	t0, t3, t0	! 0, 1, 2, 3
	xor	t0, t2, t0
	add	i, 4, i
	
	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
	
426
	ld	[IDX1+i], t1 	! 1
427
428
429

	C i = 3
	ld	[IDX1+i], t1 	! 1
430

431
	! IDX2(j) = j XOR 2
432
	xor	i, 8, t2
433
434
	add	wtxt, t1, t1	! 1
	ldub	[t1+2], t1	! 1
435

436
	ld	[wtxt+i], t0	! 0
437
	lduh	[wtxt+t2], t2	! 2
438
	and	t0, 255, t0	! 0
439
	ld	[IDX3 + i], t3	! 3
440
	
441
	and	t2, 255, t2	! 2
442
	ldub	[T+t1], t1	! 1
443
	ldub	[T+t0], t0	! 0
444
	sll	t1, 8, t1	! 1
445
	
446
	ldub	[wtxt+t3], t3	! 3
447
	or	t0, t1, t0	! 0, 1
448
	ldub	[T+t2], t2	! 2
449
	ldub	[T+t3], t3	! 3
450
	
451
452
	sll	t2, 16, t2	! 2
	or	t0, t2, t0	! 0, 1, 2
453
	ld	[key + i], t2
454
	sll	t3, 24, t3	! 3
455
	
456
	or	t0, t3, t0	! 0, 1, 2, 3
457
458
459
460
461
462
	xor	t0, t2, t0
	add	i, 4, i
	
	srl	t0, 24, t3
	srl	t0, 16, t2
	srl	t0, 8, t1
463
	stb	t1, [dst+1]
464
	
465
466
467
	stb	t3, [dst+3]
	stb	t2, [dst+2]
	stb	t0, [dst]
468
	add	dst, 4, dst
469
	
470
	ld	[IDX1+i], t1 	! 1
471
472
473
			
	C Unrolled final loop ends
	
474
	addcc	length, -16, length
475
	sub	ctx, src, %g2
476
	
477
	bne	.Lblock_loop
478
	sub	wtxt, src, %g3
479

480
.Lend:
481
	C add	%sp, FRAME_SIZE, %fp
482
483
	ret
	restore
484
485
486
487
.LLFE1:
.LLfe1:
	.size	_aes_crypt,.LLfe1-_aes_crypt

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
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
	! 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
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573

	! 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
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591

	! 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