aes.asm 7.62 KB
Newer Older
1
! -*- mode: asm; asm-comment-char: ?!; -*-  
2
	! Used registers:	%l0,1,2,3,4,5,6,7
3
	!			%i0,1,2,3,4 (%i6=%fp, %i7 = return)
4
	!			%o0,1,2,3 (%o6=%sp)
5
	!			%g2,3,4,5,6
6
include(`asm.m4')
Niels Möller's avatar
Niels Möller committed
7
	
8
	.file	"aes.asm"
Niels Möller's avatar
Niels Möller committed
9
	
10
	.section	".text"
11
	.align 16
12
13
	.global _aes_crypt
	.type	_aes_crypt,#function
14
	.proc	020
15

16
! Arguments
17
define(ctx, %i0)
18
define(T, %i1)
19
define(length, %i2)
20
define(dst, %i3)
21
define(src, %i4)
22

23
24
25
! Loop invariants
define(wtxt, %l0)
define(tmp, %l1)
26
define(diff, %l2)
27
define(nrounds, %l3)
28
29

! Loop variables
30
define(round, %l4)
31
define(i, %l5)
32
define(key, %g7)
33

34
35
36
37
38
! Further loop invariants
define(T0, %l6)
define(T1, %l7)
define(T2, %g5)
define(T3, %g6)
39
40
define(IDX1, %i5)
define(IDX3, %o5)
41

42
! Teporaries
43
define(t0, %o0)
44
define(t1, %o1)
45
define(t2, %o2)
46
47
define(t3, %o3)

48
_aes_crypt:
49
! Why -136?
50
	save	%sp, -136, %sp
51
	cmp	length, 0
52
	be	.Lend
53
	! wtxt
54
	add	%fp, -24, wtxt
55
	
56
	add	%fp, -40, tmp
57
	ld	[ctx + AES_NROUNDS], nrounds
58
59
	! Compute xor, so that we can swap efficiently.
	xor	wtxt, tmp, diff
60
	! The loop variable will be multiplied by 16.
61
62
	! More loop invariants
	add	T, AES_TABLE0, T0
63
	
64
65
66
	add	T, AES_TABLE1, T1
	add	T, AES_TABLE2, T2
	add	T, AES_TABLE3, T3
67
68
69
	add	T, AES_SIDX1, IDX1

	add	T, AES_SIDX3, IDX3
70
	! Read src, and add initial subkey
71
	! Difference between ctx and src.
72
73
	! NOTE: These instruction is duplicated in the delay slot,
	! and the instruction before the branch
74
75
76
	sub	ctx, src, %g2
	! Difference between wtxt and src
	sub	wtxt, src, %g3
77
.Lblock_loop:
78
79
80
	! For stop condition. Note that src is incremented in the
	! delay slot
	add	src, 8, %g4
81
	
82
.Lsource_loop:
83
84
85
86
	ldub	[src+3], t3
	ldub	[src+2], t2
	sll	t3, 24, t3
	ldub	[src+1], t1
87
	
88
89
90
91
	sll	t2, 16, t2
	or	t3, t2, t3
	ldub	[src], t0
	sll	t1, 8, t1
92
	
93
	! Get subkey
94
	ld	[src+%g2], t2
95
96
	or	t3, t1, t3
	or	t3, t0, t3
97
	xor	t3, t2, t3
98
	
99
	cmp	src, %g4
100
	st	t3, [src+%g3]
101
	bleu	.Lsource_loop
102
	add	src, 4, src
103
	
104
	sub	nrounds, 1, round
105
	add	ctx, 16, key
106
.Lround_loop:
107
	! 4*i
108
	! NOTE: Instruction duplicated in delay slot
109
	mov	0, i
110
.Linner_loop:
111
	! The comments mark which j in T->table[j][ Bj(wtxt[IDXi(i)]) ]
112
113
114
115
	! 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.
116
	
117
	ld	[IDX1+i], t1		! 1
118
	
119
120
	! IDX2(j) = j XOR 2
	xor	i, 8, t2
121
122
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
123
	ld	[IDX3+i], t3		! 3
124
	
125
	sll	t1, 2, t1		! 1
126
	ld	[wtxt+i], t0		! 0
127
	lduh	[wtxt+t2], t2		! 2
128
	and	t0, 255, t0		! 0
129
	
130
	ldub	[wtxt+t3], t3		! 3
131
	sll	t0, 2, t0		! 0
132
	ld	[T0+t0], t0		! 0
133
	and	t2, 255, t2		! 2
134
	
135
	ld	[T1+t1], t1		! 1
136
	sll	t2, 2, t2		! 2
137
	ld	[T2+t2], t2		! 2
138
	sll	t3, 2, t3		! 3
139
	
140
	ld	[T3+t3], t3		! 3
141
142
	xor	t0, t1, t0		! 0, 1
	xor	t0, t2, t0		! 0, 1, 2
143
	! Fetch roundkey
144
	ld	[key+i], t1
145
	
146
	xor	t0, t3, t0		! 0, 1, 2, 3
147
	xor	t0, t1, t0
148
	st	t0, [tmp+i]
149
	cmp	i, 8
150
	
151
	bleu	.Linner_loop
152
153
	add	i, 4, i
	! switch roles for tmp and wtxt
154
	xor	wtxt, diff, wtxt
155
156
	xor	tmp, diff, tmp

157
	subcc	round, 1, round
158
	add	key, 16, key
159
160
	bne	.Linner_loop
	mov	0, i
161

162
	! final round
163
164
165
	! Use round as the loop variable, as it's already zero
undefine(<i>)
define(i, round)
166

167
168
	! Comments mark which j in T->sbox[Bj(wtxt[IDXj(i)])]
	! the instruction is part of
169
	! NOTE: First instruction duplicated in delay slot
170
	ld	[IDX1+i], t1 	! 1
171
.Lfinal_loop:
172
	! IDX2(j) = j XOR 2
173
174
	xor	i, 8, t2
	! ld	[idx-16], t2	! 2
175
176
	add	wtxt, t1, t1	! 1
	ldub	[t1+2], t1	! 1
177

178
	ld	[wtxt+i], t0	! 0
179
	lduh	[wtxt+t2], t2	! 2
180
	and	t0, 255, t0	! 0
181
	ld	[IDX3 + i], t3	! 3
182
	
183
	and	t2, 255, t2	! 2
184
	ldub	[T+t1], t1	! 1
185
	ldub	[T+t0], t0	! 0
186
	sll	t1, 8, t1	! 1
187
	
188
	ldub	[wtxt+t3], t3	! 3
189
	or	t0, t1, t0	! 0, 1
190
	ldub	[T+t2], t2	! 2
191
	ldub	[T+t3], t3	! 3
192
	
193
194
	sll	t2, 16, t2	! 2
	or	t0, t2, t0	! 0, 1, 2
195
	ld	[key + i], t2
196
	sll	t3, 24, t3	! 3
197
	
198
	or	t0, t3, t0	! 0, 1, 2, 3
199
200
201
202
203
204
205
	xor	t0, t2, t0
	add	i, 4, i
	cmp	i, 12
	
	srl	t0, 24, t3
	srl	t0, 16, t2
	srl	t0, 8, t1
206
	stb	t1, [dst+1]
207
	
208
209
210
	stb	t3, [dst+3]
	stb	t2, [dst+2]
	stb	t0, [dst]
211
	add	dst, 4, dst
212
	
213
	bleu	.Lfinal_loop
214
	ld	[IDX1+i], t1 	! 1
215
	addcc	length, -16, length
216
	sub	ctx, src, %g2
217
	
218
	bne	.Lblock_loop
219
	sub	wtxt, src, %g3
220

221
.Lend:
222
223
	ret
	restore
224
225
226
227
.LLFE1:
.LLfe1:
	.size	_aes_crypt,.LLfe1-_aes_crypt

228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
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
	! 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
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313

	! 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
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331

	! 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