aes.asm 6.97 KB
Newer Older
1
	! Used registers:	%l0,1,2,3,4,5
2
	!			%i0,1,2,3,4,5 (%i6=%fp, %i7 = return)
3
	!			%o0,1,2,3,4,5,7 (%o6=%sp)
4
	!			%g2,3,4
5
include(`asm.m4')
Niels Möller's avatar
Niels Möller committed
6
	
7
	.file	"aes.asm"
Niels Möller's avatar
Niels Möller committed
8
	
9
10
	.section	".text"
	.align 4
11
12
	.global _aes_crypt
	.type	_aes_crypt,#function
13
	.proc	020
14

15
! Arguments
16
define(ctx, %i0)
17
define(T, %i1)
18
define(length, %i2)
19
define(dst, %i3)
20
define(src, %i4)
21

22
23
24
! Loop invariants
define(wtxt, %l0)
define(tmp, %l1)
25
define(diff, %l2)
26
define(nrounds, %l3)
27
28

! Loop variables
29
30
define(round, %l4) ! Should perhaps be 16 * round
define(i, %l5)
31
32

! Teporaries
33
define(t0, %o0)
34
define(t1, %o1)
35
define(t2, %o2)
36
define(t3, %o3)
37
define(idx, %o4)
38

39
_aes_crypt:
40
! Why -136?
41
	save	%sp, -136, %sp
42

43
	cmp	length, 0
44
	be	.Lend
45

46
	! wtxt
47
	add	%fp, -24, wtxt
48
	add	%fp, -40, tmp
49
50
51

	ld	[ctx + AES_NROUNDS], nrounds

52
53
	! Compute xor, so that we can swap efficiently.
	xor	wtxt, tmp, diff
54
55
56

	! The loop variable will be multiplied by 16.
	sll	nrounds, 4, nrounds
57

58
59
.Lblock_loop:
	! Read src, and add initial subkey
60
	mov	-4, i
61
.Lsource_loop:
62
	add	i, 4, i
63
		
64
	add	i, src, %o5
65
	ldub	[%o5+3], %g2
66

67
	ldub	[%o5+2], %g3
68
	sll	%g2, 24, %g2
69
	ldub	[%o5+1], %o0
70
71
	sll	%g3, 16, %g3
	or	%g2, %g3, %g2
72
	ldub	[src+i], %o5
73
	sll	%o0, 8, %o0
74
	ld	[ctx+i], %g3
75
	or	%g2, %o0, %g2
76
	or	%g2, %o5, %g2
77
	xor	%g2, %g3, %g2
78

79
	cmp	i, 12
80
	bleu	.Lsource_loop
81
	st	%g2, [wtxt+i]
82

83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
	! ! Read a little-endian word
	! ldub	[src+3], %g2
	! sll	%g2, 8, %g2
	! 
	! ldub	[src+2], %g3
	! or	%g3, %g2, %g2
	! sll	%g2, 8, %g2
	! 
	! ldub	[src+1], %g3
	! or	%g3, %g2, %g2
	! sll	%g2, 8, %g2
	! 
	! ldub	[src+0], %g3
	! or	%g3, %g2, %g2
	! sll	%g2, 8, %g2
	! 
99
	! ld	[ctx+%o3], %g3
100
101
102
	! xor	%g3, %g2, %g2
	! 
	! add	src, 4, src
103
	! st	%g2, [wtxt+%o4]
104
	! 
105
	! cmp	%o3, 8
106
	! bleu	.Lsource_loop
107
	! add	%o3, 4, %o3
108

109
	mov	16, round
110

111
.Lround_loop:
112
113
	! 4*i
	mov	0, i
114
	add	T, AES_SIDX3, idx
115
.Linner_loop:
116
	! The comments mark which j in T->table[j][ Bj(wtxt[IDXi(i)]) ]
117
118
119
120
	! 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.
121
	
122
	! AES_SIDX1
123
	ld	[idx-32], t1		! 1
124

125
	! AES_SIDX2
126
127
128
129
	! ld	[idx-16], t2		! 2
	! IDX2(j) = j XOR 2
	xor	i, 8, t2
	
130
	! wtxt[IDX1...]
131
132
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
133

134
	! AES_SIDX3
135
	ld	[idx], t3		! 3
136
	sll	t1, 2, t1		! 1
137
	
138
	! wtxt[i]
139
	ld	[wtxt+i], t0		! 0
140
141
	
	! wtxt[IDX2...]
142
	lduh	[wtxt+t2], t2		! 2
143
	
144
	and	t0, 255, t0		! 0
145
146

	! wtxt[IDX3...]
147
	ldub	[wtxt+t3], t3		! 3
148
	
149
150
151
	sll	t0, 2, t0		! 0
	add	t0, AES_TABLE0, t0	! 0
	ld	[T+t0], t0		! 0
152

153
	add	t1, AES_TABLE1, t1	! 1
154
	and	t2, 255, t2		! 2
155
	ld	[T+t1], t1		! 1
156
157
158
	sll	t2, 2, t2		! 2
	add	t2, AES_TABLE2, t2	! 2
	ld	[T+t2], t2		! 2
159
160
161
	sll	t3, 2, t3		! 3
	add	t3, AES_TABLE3, t3	! 3
	ld	[T+t3], t3		! 3
162
163
	xor	t0, t1, t0		! 0, 1
	xor	t0, t2, t0		! 0, 1, 2
164

165
	add	idx, 4, idx		
166

167
	! Fetch roundkey
168
169
170
	! FIXME: We could save one instruction
	! if we kept a pointer to the current subkey,
	! indexed by i.
171
172
173
	ld	[ctx+round], t1
	add	round, 4, round
	
174
	xor	t0, t3, t0		! 0, 1, 2, 3
175

176
	xor	t0, t1, t0
177
	st	t0, [tmp+i]
178
179

	cmp	i, 8
180

181
	bleu	.Linner_loop
182
	add	i, 4, i
183
	
184
	! switch roles for tmp and wtxt
185
	xor	wtxt, diff, wtxt
186

187
	cmp	round, nrounds
188
	blu	.Lround_loop
189
	xor	tmp, diff, tmp
190

191
	! final round
192
	mov	0, i
193
	add	T, 288, %g4
194
.Lfinal_loop:
195
	ld	[%g4-32], %g2
196
	sll	i, 2, %i5
197
	sll	%g2, 2, %g2
198
	add	wtxt, %g2, %g2
199
	ldub	[%g2+2], %o3
200
	add	%i5, dst, %o2
201
	ld	[%g4-16], %g3
202
203
	add	i, 1, i
	ld	[wtxt+%i5], %g2
204
	sll	%g3, 2, %g3
205
	lduh	[wtxt+%g3], %o4
206
	and	%g2, 255, %g2
207
	ld	[%g4], %o5
208
	and	%o4, 255, %o4
209
	ldub	[T+%o3], %o0
210
	sll	%o5, 2, %o5
211
	ldub	[T+%g2], %g3
212
	sll	%o0, 8, %o0
213
	ldub	[wtxt+%o5], %o3
214
	or	%g3, %o0, %g3
215
	ldub	[T+%o4], %g2
216
	cmp	i, 3
217
	ldub	[T+%o3], %o5
218
219
	sll	%g2, 16, %g2
	or	%g3, %g2, %g3
220
	ld	[ctx + round], %g2
221
222
	sll	%o5, 24, %o5
	or	%g3, %o5, %g3
223
	xor	%g3, %g2, %g3
224
	srl	%g3, 24, %o5
225
	srl	%g3, 16, %o0
226
	srl	%g3, 8, %g2
227
228
229
	stb	%g2, [%o2+1]
	stb	%o5, [%o2+3]
	stb	%o0, [%o2+2]
230
	stb	%g3, [dst+%i5]
231
	add	round, 4, round
232
	bleu	.Lfinal_loop
233
	add	%g4, 4, %g4
234
	
235
	add	src, 16, src
236
	addcc	length, -16, length
237
	bne	.Lblock_loop
238
	add	dst, 16, dst
239
.Lend:
240
241
	ret
	restore
242
243
244
245
.LLFE1:
.LLfe1:
	.size	_aes_crypt,.LLfe1-_aes_crypt

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
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
	! 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
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331

	! 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