aes.asm 7.07 KB
Newer Older
1
	! Used registers:	%l0,1,2,3,4,5,6,7
2
	!			%i0,1,2,3,4,5 (%i6=%fp, %i7 = return)
3
	!			%o0,1,2,3,4,5,7 (%o6=%sp)
4
	!			%g2,3,4,5,6
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
define(round, %l4) ! Really 16 * round
30
define(i, %l5)
31

32
33
34
35
36
37
38
! Further loop invariants
define(T0, %l6)
define(T1, %l7)
define(T2, %g5)
define(T3, %g6)
define(key, %g7)

39
! Teporaries
40
define(t0, %o0)
41
define(t1, %o1)
42
define(t2, %o2)
43
define(t3, %o3)
44
define(idx, %o4)
45

46
_aes_crypt:
47
! Why -136?
48
	save	%sp, -136, %sp
49
	cmp	length, 0
50
	be	.Lend
51
	! wtxt
52
	add	%fp, -24, wtxt
53
	
54
	add	%fp, -40, tmp
55
	ld	[ctx + AES_NROUNDS], nrounds
56
57
	! Compute xor, so that we can swap efficiently.
	xor	wtxt, tmp, diff
58
59
	! The loop variable will be multiplied by 16.
	sll	nrounds, 4, nrounds
60
61
62
63
64
65
66
	
	! More loop invariants
	add	T, AES_TABLE0, T0
	add	T, AES_TABLE1, T1
	add	T, AES_TABLE2, T2
	add	T, AES_TABLE3, T3
		
67
68
.Lblock_loop:
	! Read src, and add initial subkey
69
	mov	-4, i
70
.Lsource_loop:
71
	add	i, 4, i
72
73
74
	! add	i, src, %o5
	mov	src, %o5
	
75
76
	ldub	[%o5+3], %g2
	ldub	[%o5+2], %g3
77
	
78
	sll	%g2, 24, %g2
79
	ldub	[%o5+1], %o0
80
81
	sll	%g3, 16, %g3
	or	%g2, %g3, %g2
82
	
83
	ldub	[%o5], %o5
84
	sll	%o0, 8, %o0
85
	ld	[ctx+i], %g3
86
	or	%g2, %o0, %g2
87
	
88
	or	%g2, %o5, %g2
89
	xor	%g2, %g3, %g2
90
91
	add	src, 4, src
	cmp	i, 8
92
	bleu	.Lsource_loop
93
	
94
	st	%g2, [wtxt+i]
95

96
97
98
99
100
101
	! ! Read a little-endian word
	! ldub	[src+3], %g2
	! sll	%g2, 8, %g2
	! 
	! ldub	[src+2], %g3
	! or	%g3, %g2, %g2
102
	
103
104
105
106
107
108
109
110
111
112
	! 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
	! 
113
	! ld	[ctx+%o3], %g3
114
	
115
116
117
	! xor	%g3, %g2, %g2
	! 
	! add	src, 4, src
118
	! st	%g2, [wtxt+%o4]
119
	! 
120
	! cmp	%o3, 8
121
	
122
	! bleu	.Lsource_loop
123
	! add	%o3, 4, %o3
124

125
	mov	16, round
126
	add	ctx, 16, key
127

128
.Lround_loop:
129
130
	! 4*i
	mov	0, i
131
	add	T, AES_SIDX3, idx
132
.Linner_loop:
133
	! The comments mark which j in T->table[j][ Bj(wtxt[IDXi(i)]) ]
134
135
136
137
	! 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.
138
	
139
	! AES_SIDX1
140
	ld	[idx-32], t1		! 1
141
	! AES_SIDX2
142
143
	! IDX2(j) = j XOR 2
	xor	i, 8, t2
144
	! wtxt[IDX1...]
145
146
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
147

148
	! AES_SIDX3
149
	ld	[idx], t3		! 3
150
	sll	t1, 2, t1		! 1
151
	! wtxt[i]
152
	ld	[wtxt+i], t0		! 0
153
	! wtxt[IDX2...]
154
	lduh	[wtxt+t2], t2		! 2
155
	
156
	and	t0, 255, t0		! 0
157
	! wtxt[IDX3...]
158
	ldub	[wtxt+t3], t3		! 3
159
	sll	t0, 2, t0		! 0
160
	ld	[T0+t0], t0		! 0
161

162
	and	t2, 255, t2		! 2
163
	ld	[T1+t1], t1		! 1
164
	sll	t2, 2, t2		! 2
165
166
	ld	[T2+t2], t2		! 2

167
	sll	t3, 2, t3		! 3
168
	ld	[T3+t3], t3		! 3
169
170
	xor	t0, t1, t0		! 0, 1
	xor	t0, t2, t0		! 0, 1, 2
171

172
	add	idx, 4, idx		
173
	! Fetch roundkey
174
175
176
	! ld	[ctx+round], t1
	! add	round, 4, round
	ld	[key+i], t1
177
	xor	t0, t3, t0		! 0, 1, 2, 3
178

179
	xor	t0, t1, t0
180
	st	t0, [tmp+i]
181
	cmp	i, 8
182
	bleu	.Linner_loop
183
	
184
	add	i, 4, i
185
	
186
	! switch roles for tmp and wtxt
187
	xor	wtxt, diff, wtxt
188
189
	add	round, 16, round
	add	key, 16, key
190
	cmp	round, nrounds
191
	blu	.Lround_loop
192
	xor	tmp, diff, tmp
193

194
	! final round
195
	mov	0, i
196
	! IDX3
197
	add	T, 288, %g4
198
.Lfinal_loop:
199
	ld	[%g4-32], %g2
200
	sll	i, 2, %i5
201
	sll	%g2, 2, %g2
202
	add	wtxt, %g2, %g2
203
	ldub	[%g2+2], %o3
204
	add	%i5, dst, %o2
205
	ld	[%g4-16], %g3
206
207
	add	i, 1, i
	ld	[wtxt+%i5], %g2
208
	sll	%g3, 2, %g3
209
	lduh	[wtxt+%g3], %o4
210
	and	%g2, 255, %g2
211
	ld	[%g4], %o5
212
	and	%o4, 255, %o4
213
	ldub	[T+%o3], %o0
214
	sll	%o5, 2, %o5
215
	ldub	[T+%g2], %g3
216
	sll	%o0, 8, %o0
217
	ldub	[wtxt+%o5], %o3
218
	or	%g3, %o0, %g3
219
	ldub	[T+%o4], %g2
220
	cmp	i, 3
221
	ldub	[T+%o3], %o5
222
223
	sll	%g2, 16, %g2
	or	%g3, %g2, %g3
224
	ld	[ctx + round], %g2
225
226
	sll	%o5, 24, %o5
	or	%g3, %o5, %g3
227
	xor	%g3, %g2, %g3
228
	srl	%g3, 24, %o5
229
	srl	%g3, 16, %o0
230
	srl	%g3, 8, %g2
231
232
233
	stb	%g2, [%o2+1]
	stb	%o5, [%o2+3]
	stb	%o0, [%o2+2]
234
	stb	%g3, [dst+%i5]
235
	add	round, 4, round
236
	bleu	.Lfinal_loop
237
	add	%g4, 4, %g4
238
	
239
	! add	src, 16, src
240
	addcc	length, -16, length
241
	bne	.Lblock_loop
242
	add	dst, 16, dst
243
.Lend:
244
245
	ret
	restore
246
247
248
249
.LLFE1:
.LLfe1:
	.size	_aes_crypt,.LLfe1-_aes_crypt

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
299
300
301
302
	! 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
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
332
333
334
335

	! 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