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
	!			%i0,1,2,3,4,5 (%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
61
define(IDX1, %i5)
define(IDX3, %o5)
62

63
! Teporaries
64
define(t0, %o0)
65
define(t1, %o1)
66
define(t2, %o2)
67
68
define(t3, %o3)

69
70
71
72
73
74
75
76
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)
77

78
_aes_crypt:
79
	save	%sp, -FRAME_SIZE, %sp
80
	cmp	length, 0
81
	be	.Lend
82
83
84
85
	! wtxt
	add	%fp, -24, wtxt
	
	add	%fp, -40, tmp
86

87
	ld	[ctx + AES_NROUNDS], nrounds
88
89
	! Compute xor, so that we can swap efficiently.
	xor	wtxt, tmp, diff
90
	! The loop variable will be multiplied by 16.
91
92
	! More loop invariants
	add	T, AES_TABLE0, T0
Niels Möller's avatar
Niels Möller committed
93
	
94
	add	T, AES_TABLE1, T1
95
96
	add	T, AES_TABLE2, T2
	add	T, AES_TABLE3, T3
97
	add	T, AES_SIDX1, IDX1
Niels Möller's avatar
Niels Möller committed
98
	
99
	add	T, AES_SIDX3, IDX3
100
	! Read src, and add initial subkey
101
	! Difference between ctx and src.
Niels Möller's avatar
Niels Möller committed
102
	! NOTE: These instructions are duplicated in the delay slot,
103
	! and the instruction before the branch.
104
105
106
	sub	ctx, src, %g2
	! Difference between wtxt and src
	sub	wtxt, src, %g3
107
.Lblock_loop:
108
109
	! For stop condition. Note that src is incremented in the
	! delay slot
110
	add	src, 8, %g1
111
	
112
.Lsource_loop:
113
114
115
116
	ldub	[src+3], t3
	ldub	[src+2], t2
	sll	t3, 24, t3
	ldub	[src+1], t1
117
	
118
119
120
121
	sll	t2, 16, t2
	or	t3, t2, t3
	ldub	[src], t0
	sll	t1, 8, t1
122
	
123
	! Get subkey
124
	ld	[src+%g2], t2
125
126
	or	t3, t1, t3
	or	t3, t0, t3
127
	xor	t3, t2, t3
128
	
129
	cmp	src, %g1
130
	st	t3, [src+%g3]
131
	bleu	.Lsource_loop
132
	add	src, 4, src
133
	
134
	sub	nrounds, 1, round
135
	add	ctx, 16, key
Niels Möller's avatar
Niels Möller committed
136
	nop
137
	! 4*i
138
	! NOTE: Instruction duplicated in delay slot
139
	mov	0, i
140
.Linner_loop:
141
	! The comments mark which j in T->table[j][ Bj(wtxt[IDXi(i)]) ]
142
143
144
145
	! 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.
146
147

	C i = 0
148
	ld	[IDX1], t1		! 1
149
150
	
	! IDX2(j) = j XOR 2
151
152
	mov	8, t2			! 2
	C xor	i, 8, t2		! 2
153
154
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
155
	ld	[IDX3], t3		! 3
156
157
	
	sll	t1, 2, t1		! 1
158
	ld	[wtxt], t0		! 0
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
	lduh	[wtxt+t2], t2		! 2
	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
176
	ld	[key], t1
177
178
179
	
	xor	t0, t3, t0		! 0, 1, 2, 3
	xor	t0, t1, t0
180
	st	t0, [tmp]
181
182
183
	add	i, 4, i

	C i = 1
184
	ld	[IDX1+4], t1		! 1
185
	
186
	! IDX2(j) = j XOR 2
187
188
	C xor	i, 8, t2		! 2
	mov	12, t2			! 2
189
190
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
191
	ld	[IDX3+4], t3		! 3
192
	
193
	sll	t1, 2, t1		! 1
194
	ld	[wtxt+4], t0		! 0
195
	lduh	[wtxt+t2], t2		! 2
Niels Möller's avatar
Niels Möller committed
196
	and	t0, 255, t0		! 0
197
	
198
	ldub	[wtxt+t3], t3		! 3
199
	sll	t0, 2, t0		! 0
200
	ld	[T0+t0], t0		! 0
Niels Möller's avatar
Niels Möller committed
201
	and	t2, 255, t2		! 2
202
	
203
	ld	[T1+t1], t1		! 1
204
	sll	t2, 2, t2		! 2
205
	ld	[T2+t2], t2		! 2
Niels Möller's avatar
Niels Möller committed
206
	sll	t3, 2, t3		! 3
207
	
208
	ld	[T3+t3], t3		! 3
209
210
	xor	t0, t1, t0		! 0, 1
	xor	t0, t2, t0		! 0, 1, 2
211
	! Fetch roundkey
212
	ld	[key+4], t1
213
	
214
	xor	t0, t3, t0		! 0, 1, 2, 3
215
	xor	t0, t1, t0
216
	st	t0, [tmp+4]
217
	add	i, 4, i
218
219

	C = 2
220
	ld	[IDX1+8], t1		! 1
221
222
	
	! IDX2(j) = j XOR 2
223
224
	C xor	i, 8, t2		! 2
	mov	0, t2			! 2
225
226
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
227
	ld	[IDX3+8], t3		! 3
228
229
	
	sll	t1, 2, t1		! 1
230
	ld	[wtxt+8], t0		! 0
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
	lduh	[wtxt+t2], t2		! 2
	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
248
	ld	[key+8], t1
249
250
251
	
	xor	t0, t3, t0		! 0, 1, 2, 3
	xor	t0, t1, t0
252
	st	t0, [tmp+8]
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
	add	i, 4, i

	C = 3
	ld	[IDX1+i], t1		! 1
	
	! IDX2(j) = j XOR 2
	xor	i, 8, t2
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
	ld	[IDX3+i], t3		! 3
	
	sll	t1, 2, t1		! 1
	ld	[wtxt+i], t0		! 0
	lduh	[wtxt+t2], t2		! 2
	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+i], t1
	
	xor	t0, t3, t0		! 0, 1, 2, 3
	xor	t0, t1, t0
	st	t0, [tmp+i]
	add	i, 4, i
			
	C End loop
	
292
	! switch roles for tmp and wtxt
293
	xor	wtxt, diff, wtxt
294
	xor	tmp, diff, tmp
295

296
	subcc	round, 1, round
297
	add	key, 16, key
298
299
	bne	.Linner_loop
	mov	0, i
300

301
	! final round
302
303
304
	! Use round as the loop variable, as it's already zero
undefine(<i>)
define(i, round)
305

306
307
	! Comments mark which j in T->sbox[Bj(wtxt[IDXj(i)])]
	! the instruction is part of
308
	! NOTE: First instruction duplicated in delay slot
309
	ld	[IDX1+i], t1 	! 1
310
.Lfinal_loop:
311
	! IDX2(j) = j XOR 2
312
	xor	i, 8, t2
313
314
	add	wtxt, t1, t1	! 1
	ldub	[t1+2], t1	! 1
315

316
	ld	[wtxt+i], t0	! 0
317
	lduh	[wtxt+t2], t2	! 2
318
	and	t0, 255, t0	! 0
319
	ld	[IDX3 + i], t3	! 3
320
	
321
	and	t2, 255, t2	! 2
322
	ldub	[T+t1], t1	! 1
323
	ldub	[T+t0], t0	! 0
324
	sll	t1, 8, t1	! 1
325
	
326
	ldub	[wtxt+t3], t3	! 3
327
	or	t0, t1, t0	! 0, 1
328
	ldub	[T+t2], t2	! 2
329
	ldub	[T+t3], t3	! 3
330
	
331
332
	sll	t2, 16, t2	! 2
	or	t0, t2, t0	! 0, 1, 2
333
	ld	[key + i], t2
334
	sll	t3, 24, t3	! 3
335
	
336
	or	t0, t3, t0	! 0, 1, 2, 3
337
338
339
340
341
342
343
	xor	t0, t2, t0
	add	i, 4, i
	cmp	i, 12
	
	srl	t0, 24, t3
	srl	t0, 16, t2
	srl	t0, 8, t1
344
	stb	t1, [dst+1]
345
	
346
347
348
	stb	t3, [dst+3]
	stb	t2, [dst+2]
	stb	t0, [dst]
349
	add	dst, 4, dst
350
	
351
	bleu	.Lfinal_loop
352
	ld	[IDX1+i], t1 	! 1
353
	addcc	length, -16, length
354
	sub	ctx, src, %g2
355
	
356
	bne	.Lblock_loop
357
	sub	wtxt, src, %g3
358

359
.Lend:
360
	C add	%sp, FRAME_SIZE, %fp
361
362
	ret
	restore
363
364
365
366
.LLFE1:
.LLfe1:
	.size	_aes_crypt,.LLfe1-_aes_crypt

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
413
414
415
416
417
418
419
	! 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
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
446
447
448
449
450
451
452

	! 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
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470

	! 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