aes.asm 6.59 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
	ldub	[src+3], t3
	ldub	[src+2], t2
75
	
76
77
78
79
	sll	t3, 24, t3
	ldub	[src+1], t1
	sll	t2, 16, t2
	or	t3, t2, t3
80
	
81
82
	ldub	[src], t0
	sll	t1, 8, t1
83
	ld	[ctx+i], %g3
84
	or	t3, t1, t3
85
	
86
87
	or	t3, t0, t3
	xor	t3, %g3, t3
88
89
	add	src, 4, src
	cmp	i, 8
90
	bleu	.Lsource_loop
91
	
92
	st	t3, [wtxt+i]
93

94
	mov	16, round
95
	add	ctx, 16, key
96

97
.Lround_loop:
98
99
	! 4*i
	mov	0, i
100
	add	T, AES_SIDX3, idx
101
.Linner_loop:
102
	! The comments mark which j in T->table[j][ Bj(wtxt[IDXi(i)]) ]
103
104
105
106
	! 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.
107
	
108
	! AES_SIDX1
109
	ld	[idx-32], t1		! 1
110
	! AES_SIDX2
111
112
	! IDX2(j) = j XOR 2
	xor	i, 8, t2
113
	! wtxt[IDX1...]
114
115
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
116

117
	! AES_SIDX3
118
	ld	[idx], t3		! 3
119
	sll	t1, 2, t1		! 1
120
	! wtxt[i]
121
	ld	[wtxt+i], t0		! 0
122
	! wtxt[IDX2...]
123
	lduh	[wtxt+t2], t2		! 2
124
	
125
	and	t0, 255, t0		! 0
126
	! wtxt[IDX3...]
127
	ldub	[wtxt+t3], t3		! 3
128
	sll	t0, 2, t0		! 0
129
	ld	[T0+t0], t0		! 0
130

131
	and	t2, 255, t2		! 2
132
	ld	[T1+t1], t1		! 1
133
	sll	t2, 2, t2		! 2
134
135
	ld	[T2+t2], t2		! 2

136
	sll	t3, 2, t3		! 3
137
	ld	[T3+t3], t3		! 3
138
139
	xor	t0, t1, t0		! 0, 1
	xor	t0, t2, t0		! 0, 1, 2
140

141
	add	idx, 4, idx		
142
	! Fetch roundkey
143
144
145
	! ld	[ctx+round], t1
	! add	round, 4, round
	ld	[key+i], t1
146
	xor	t0, t3, t0		! 0, 1, 2, 3
147

148
	xor	t0, t1, t0
149
	st	t0, [tmp+i]
150
	cmp	i, 8
151
	bleu	.Linner_loop
152
	
153
	add	i, 4, i
154
	
155
	! switch roles for tmp and wtxt
156
	xor	wtxt, diff, wtxt
157
158
	add	round, 16, round
	add	key, 16, key
159
	cmp	round, nrounds
160
	blu	.Lround_loop
161
	xor	tmp, diff, tmp
162

163
	! final round
164
	mov	0, i
165
	! IDX3
166
	add	T, 288, %g4
167
.Lfinal_loop:
168
	ld	[%g4-32], %g2
169
	sll	i, 2, %i5
170
	sll	%g2, 2, %g2
171
	add	wtxt, %g2, %g2
172
	ldub	[%g2+2], %o3
173
	add	%i5, dst, %o2
174
	ld	[%g4-16], %g3
175
176
	add	i, 1, i
	ld	[wtxt+%i5], %g2
177
	sll	%g3, 2, %g3
178
	lduh	[wtxt+%g3], %o4
179
	and	%g2, 255, %g2
180
	ld	[%g4], %o5
181
	and	%o4, 255, %o4
182
	ldub	[T+%o3], %o0
183
	sll	%o5, 2, %o5
184
	ldub	[T+%g2], %g3
185
	sll	%o0, 8, %o0
186
	ldub	[wtxt+%o5], %o3
187
	or	%g3, %o0, %g3
188
	ldub	[T+%o4], %g2
189
	cmp	i, 3
190
	ldub	[T+%o3], %o5
191
192
	sll	%g2, 16, %g2
	or	%g3, %g2, %g3
193
	ld	[ctx + round], %g2
194
195
	sll	%o5, 24, %o5
	or	%g3, %o5, %g3
196
	xor	%g3, %g2, %g3
197
	srl	%g3, 24, %o5
198
	srl	%g3, 16, %o0
199
	srl	%g3, 8, %g2
200
201
202
	stb	%g2, [%o2+1]
	stb	%o5, [%o2+3]
	stb	%o0, [%o2+2]
203
	stb	%g3, [dst+%i5]
204
	add	round, 4, round
205
	bleu	.Lfinal_loop
206
	add	%g4, 4, %g4
207
	
208
	addcc	length, -16, length
209
	bne	.Lblock_loop
210
	add	dst, 16, dst
211
.Lend:
212
213
	ret
	restore
214
215
216
217
.LLFE1:
.LLfe1:
	.size	_aes_crypt,.LLfe1-_aes_crypt

218
219
220
221
222
223
224
225
226
227
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
	! 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
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
303

	! 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