aes.asm 5.95 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
26
define(diff, %l2)
define(nround, %l3)
27
28

! Loop variables
29
30
define(round, %l4) ! Should perhaps be 16 * round
define(i, %l5)
31
32
33
34
35

! Teporaries
define(t1, %o1)
define(t3, %o3)

36
_aes_crypt:
37
! Why -136?
38
	save	%sp, -136, %sp
39

40
	cmp	length, 0
41
	be	.Lend
42

43
	! wtxt
44
	add	%fp, -24, wtxt
45
	add	%fp, -40, tmp
46
47
	! Compute xor, so that we can swap efficiently.
	xor	wtxt, tmp, diff
48
49
50
	
	ld	[ctx + AES_NROUNDS], nround

51
52
.Lblock_loop:
	! Read src, and add initial subkey
53
	mov	-4, i
54
.Lsource_loop:
55
	add	i, 4, i
56
		
57
	add	i, src, %o5
58
	ldub	[%o5+3], %g2
59

60
	ldub	[%o5+2], %g3
61
	sll	%g2, 24, %g2
62
	ldub	[%o5+1], %o0
63
64
	sll	%g3, 16, %g3
	or	%g2, %g3, %g2
65
	ldub	[src+i], %o5
66
	sll	%o0, 8, %o0
67
	ld	[ctx+i], %g3
68
	or	%g2, %o0, %g2
69
	or	%g2, %o5, %g2
70
	xor	%g2, %g3, %g2
71

72
	cmp	i, 12
73
	bleu	.Lsource_loop
74
	st	%g2, [wtxt+i]
75

76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
	! ! 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
	! 
92
	! ld	[ctx+%o3], %g3
93
94
95
	! xor	%g3, %g2, %g2
	! 
	! add	src, 4, src
96
	! st	%g2, [wtxt+%o4]
97
	! 
98
	! cmp	%o3, 8
99
	! bleu	.Lsource_loop
100
	! add	%o3, 4, %o3
101

102
	mov	1, round
103

104
105
106
	! 4*i:	i
	! This instruction copied to the delay slot of the branch here. 
	mov	0, i
107
.Lround_loop:
108
	add	T, AES_SIDX3, %o2
109
.Linner_loop:
110
	! The comments mark which j in T->table[j][ Bj(wtxt[IDXi(i)]) ]
111
112
113
114
	! 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.
115
	
116
	! AES_SIDX1
117
	ld	[%o2-32], t1		! 1
118

119
	! AES_SIDX2
120
	ld	[%o2-16], %o4		! 2
121
	! wtxt[IDX1...]
122
123
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
124

125
	! AES_SIDX3
126
127
	ld	[%o2], t3		! 3
	sll	t1, 2, t1		! 1
128
	
129
	! wtxt[i]
Niels Möller's avatar
Niels Möller committed
130
	ld	[wtxt+i], %o5		! 0
131
132
	
	! wtxt[IDX2...]
Niels Möller's avatar
Niels Möller committed
133
	lduh	[wtxt+%o4], %g3		! 2
134
	
135
	and	%o5, 255, %o5		! 0
136
137

	! wtxt[IDX3...]
138
	ldub	[wtxt+t3], t3		! 3
139
	
140
141
142
143
	sll	%o5, 2, %o5		! 0
	add	%o5, AES_TABLE0, %o5	! 0
	ld	[T+%o5], %g2		! 0

144
	add	t1, AES_TABLE1, t1	! 1
145
	and	%g3, 255, %g3		! 2
146
	ld	[T+t1], t1		! 1
147
148
149
	sll	%g3, 2, %g3		! 2
	add	%g3, AES_TABLE2, %g3	! 2
	ld	[T+%g3], %o0		! 2
150
151
152
153
	sll	t3, 2, t3		! 3
	add	t3, AES_TABLE3, t3	! 3
	ld	[T+t3], t3		! 3
	xor	%g2, t1, %g2		! 0, 1
154
155
156
	xor	%g2, %o0, %g2		! 0, 1, 2

	add	%o2, 4, %o2		
157
158
159
160
161
162

! 	! Fetch roundkey
! 	sll	round, 4, %o5
! 	add	%o5, ctx, %o5
! 	ld	[%o5], %o5
		
163
	xor	%g2, t3, %g2		! 0, 1, 2, 3
164

165
166
167
168
!	xor	%g2, %o5, %g2
	st	%g2, [tmp+i]

	cmp	i, 8
169

170
	bleu	.Linner_loop
171
	add	i, 4, i
172
	
173
	sll	round, 4, %g2
174
	add	%g2, ctx, %o0
175
176
	mov	0, i

177
.Lroundkey_loop:
178
	sll	i, 2, %g2
179
	ld	[%o0], %o5
180
181
182
	add	i, 1, i
	ld	[tmp+%g2], %g3
	cmp	i, 3
183
	xor	%g3, %o5, %g3
184
	st	%g3, [wtxt+%g2]
Niels Möller's avatar
Niels Möller committed
185
	! st	%g3, [tmp+%g2]
186
	bleu	.Lroundkey_loop
187
	add	%o0, 4, %o0
188
189

	! switch roles for tmp and wtxt
Niels Möller's avatar
Niels Möller committed
190
191
	! xor	wtxt, diff, wtxt
	! xor	tmp, diff, tmp
192

193
194
	add	round, 1, round
	cmp	round, nround
195
	blu	.Lround_loop
196
	mov	0, i
197

198
	sll	round, 4, %g2
199
200
	
	! final round
201
	add	%g2, ctx, %o7
202
	mov	0, i
203
	add	T, 288, %g4
204
.Lfinal_loop:
205
	ld	[%g4-32], %g2
206
	sll	i, 2, %i5
207
	sll	%g2, 2, %g2
208
	add	wtxt, %g2, %g2
209
	ldub	[%g2+2], %o3
210
	add	%i5, dst, %o2
211
	ld	[%g4-16], %g3
212
213
	add	i, 1, i
	ld	[wtxt+%i5], %g2
214
	sll	%g3, 2, %g3
215
	lduh	[wtxt+%g3], %o4
216
	and	%g2, 255, %g2
217
	ld	[%g4], %o5
218
	and	%o4, 255, %o4
219
	ldub	[T+%o3], %o0
220
	sll	%o5, 2, %o5
221
	ldub	[T+%g2], %g3
222
	sll	%o0, 8, %o0
223
	ldub	[wtxt+%o5], %o3
224
	or	%g3, %o0, %g3
225
	ldub	[T+%o4], %g2
226
	cmp	i, 3
227
	ldub	[T+%o3], %o5
228
229
230
	sll	%g2, 16, %g2
	or	%g3, %g2, %g3
	ld	[%o7], %g2
231
232
	sll	%o5, 24, %o5
	or	%g3, %o5, %g3
233
	xor	%g3, %g2, %g3
234
	srl	%g3, 24, %o5
235
	srl	%g3, 16, %o0
236
	srl	%g3, 8, %g2
237
238
239
	stb	%g2, [%o2+1]
	stb	%o5, [%o2+3]
	stb	%o0, [%o2+2]
240
	stb	%g3, [dst+%i5]
241
	add	%o7, 4, %o7
242
	bleu	.Lfinal_loop
243
	add	%g4, 4, %g4
244
	
245
	add	src, 16, src
246
	addcc	length, -16, length
247
	bne	.Lblock_loop
248
	add	dst, 16, dst
249
.Lend:
250
251
	ret
	restore
252
253
254
255
.LLFE1:
.LLfe1:
	.size	_aes_crypt,.LLfe1-_aes_crypt

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
303
304
305
306
307
308
	! 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