aes.asm 10.5 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)

Niels Möller's avatar
Niels Möller committed
66
67
68
69
70
71
72
73
74
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.
75
76
77
78
79
80
81
82
83
define(<AES_ROUND>, <
	ld	[IDX1+$1], t1		! 1
	
	ldub	[wtxt+t1], t1		! 1
	ld	[IDX3+$1], t3		! 3
	
	sll	t1, 2, t1		! 1
	ld	[wtxt+$1], t0		! 0
	! IDX2(j) = j XOR 2
Niels Möller's avatar
Niels Möller committed
84
	lduh	[wtxt+eval($1 ^ 8)], t2	! 2
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
	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
	ld	[key + $1], t1
	
	xor	t0, t3, t0		! 0, 1, 2, 3
	xor	t0, t1, t0
	st	t0, [tmp + $1]>)
	
	
108
109
110
111
112
113
114
115
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)
116

117
118
119
120
121
122
	.section	".text"
	.align 16
	.global _aes_crypt
	.type	_aes_crypt,#function
	.proc	020
	
123
_aes_crypt:
124
	save	%sp, -FRAME_SIZE, %sp
125
	cmp	length, 0
126
	be	.Lend
127
128
129
130
	! wtxt
	add	%fp, -24, wtxt
	
	add	%fp, -40, tmp
131

132
	ld	[ctx + AES_NROUNDS], nrounds
133
134
	! Compute xor, so that we can swap efficiently.
	xor	wtxt, tmp, diff
135
	! The loop variable will be multiplied by 16.
136
137
	! More loop invariants
	add	T, AES_TABLE0, T0
Niels Möller's avatar
Niels Möller committed
138
	
139
	add	T, AES_TABLE1, T1
140
141
	add	T, AES_TABLE2, T2
	add	T, AES_TABLE3, T3
Niels Möller's avatar
Niels Möller committed
142
	
143
	! Read src, and add initial subkey
144
	! Difference between ctx and src.
Niels Möller's avatar
Niels Möller committed
145
	! NOTE: These instructions are duplicated in the delay slot,
146
	! and the instruction before the branch.
147
148
149
	sub	ctx, src, %g2
	! Difference between wtxt and src
	sub	wtxt, src, %g3
150
.Lblock_loop:
151
152
	! For stop condition. Note that src is incremented in the
	! delay slot
153
	add	src, 8, %g1
154
	
155
.Lsource_loop:
156
157
158
159
	ldub	[src+3], t3
	ldub	[src+2], t2
	sll	t3, 24, t3
	ldub	[src+1], t1
160
	
161
162
163
164
	sll	t2, 16, t2
	or	t3, t2, t3
	ldub	[src], t0
	sll	t1, 8, t1
165
	
166
	! Get subkey
167
	ld	[src+%g2], t2
168
169
	or	t3, t1, t3
	or	t3, t0, t3
170
	xor	t3, t2, t3
171
	
172
	cmp	src, %g1
173
	st	t3, [src+%g3]
174
	bleu	.Lsource_loop
175
	add	src, 4, src
176
	
177
	sub	nrounds, 1, round
178
	add	ctx, 16, key
Niels Möller's avatar
Niels Möller committed
179

180
.Lround_loop:
181

Niels Möller's avatar
Niels Möller committed
182
183
184
185
	AES_ROUND(0)	! i = 0
	AES_ROUND(4)	! i = 1
	AES_ROUND(8)	! i = 2
	AES_ROUND(12)	! i = 3
186
			
187
	! switch roles for tmp and wtxt
188
	xor	wtxt, diff, wtxt
189
	xor	tmp, diff, tmp
190

191
	subcc	round, 1, round
192
	bne	.Lround_loop
Niels Möller's avatar
Niels Möller committed
193
	add	key, 16, key
194

Niels Möller's avatar
Niels Möller committed
195
	! Final round
196

197
	! Comments mark which j in T->sbox[Bj(wtxt[IDXj(i)])]
198
	! the instruction is part of.
199
200
201
202

	C Unrolled final loop begins

	C i = 0
203
	ld	[IDX1+0], t1 	! 1
204
	ldub	[wtxt+t1], t1	! 1
205

206
207
208
	ld	[wtxt+0], t0	! 0
	! IDX2(j) = j XOR 2
	lduh	[wtxt+8], t2	! 2
209
	and	t0, 255, t0	! 0
210
	ld	[IDX3 + 0], t3	! 3
211
212
213
214
215
216
217
218
219
220
221
222
223
	
	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
224
	ld	[key + 0], t2
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
	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
241
	ld	[IDX1+4], t1 	! 1
242

243
	ldub	[wtxt+t1], t1	! 1
244

245
	ld	[wtxt+4], t0	! 0
246
247
	! IDX2(j) = j XOR 2
	lduh	[wtxt+12], t2	! 2
248
	and	t0, 255, t0	! 0
249
	ld	[IDX3 + 4], t3	! 3
250
251
252
253
254
255
256
257
258
259
260
261
262
	
	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
263
	ld	[key + 4], t2
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
	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
280
	ld	[IDX1+8], t1 	! 1
281
	
282
	ldub	[wtxt+t1], t1	! 1
283

284
285
286
	ld	[wtxt+8], t0	! 0
	! IDX2(j) = j XOR 2
	lduh	[wtxt+0], t2	! 2
287
	and	t0, 255, t0	! 0
288
	ld	[IDX3 + 8], t3	! 3
289
290
291
292
293
294
295
296
297
298
299
300
301
	
	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
302
	ld	[key + 8], t2
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
	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
319
	ld	[IDX1+12], t1 	! 1
320

321
	ldub	[wtxt+t1], t1	! 1
322

323
	ld	[wtxt+12], t0	! 0
324
325
	! IDX2(j) = j XOR 2
	lduh	[wtxt+4], t2	! 2
326
	and	t0, 255, t0	! 0
327
	ld	[IDX3 + 12], t3	! 3
328
	
329
	and	t2, 255, t2	! 2
330
	ldub	[T+t1], t1	! 1
331
	ldub	[T+t0], t0	! 0
332
	sll	t1, 8, t1	! 1
333
	
334
	ldub	[wtxt+t3], t3	! 3
335
	or	t0, t1, t0	! 0, 1
336
	ldub	[T+t2], t2	! 2
337
	ldub	[T+t3], t3	! 3
338
	
339
340
	sll	t2, 16, t2	! 2
	or	t0, t2, t0	! 0, 1, 2
341
	ld	[key + 12], t2
342
	sll	t3, 24, t3	! 3
343
	
344
	or	t0, t3, t0	! 0, 1, 2, 3
345
346
347
348
349
	xor	t0, t2, t0
	
	srl	t0, 24, t3
	srl	t0, 16, t2
	srl	t0, 8, t1
350
	stb	t1, [dst+1]
351
	
352
353
354
	stb	t3, [dst+3]
	stb	t2, [dst+2]
	stb	t0, [dst]
355
	add	dst, 4, dst
356
	
357
358
	C Unrolled final loop ends
	
359
	addcc	length, -16, length
360
	sub	ctx, src, %g2
361
	
362
	bne	.Lblock_loop
363
	sub	wtxt, src, %g3
364

365
.Lend:
366
367
	ret
	restore
368
369
370
371
.LLFE1:
.LLfe1:
	.size	_aes_crypt,.LLfe1-_aes_crypt

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
413
414
415
416
417
418
419
420
421
422
423
424
	! 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
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457

	! 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
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475

	! 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