aes.asm 5.71 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
	ld	[idx-16], t2		! 2
127
	! wtxt[IDX1...]
128
129
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
130

131
	! AES_SIDX3
132
	ld	[idx], t3		! 3
133
	sll	t1, 2, t1		! 1
134
	
135
	! wtxt[i]
136
	ld	[wtxt+i], t0		! 0
137
138
	
	! wtxt[IDX2...]
139
	lduh	[wtxt+t2], t2		! 2
140
	
141
	and	t0, 255, t0		! 0
142
143

	! wtxt[IDX3...]
144
	ldub	[wtxt+t3], t3		! 3
145
	
146
147
148
	sll	t0, 2, t0		! 0
	add	t0, AES_TABLE0, t0	! 0
	ld	[T+t0], t0		! 0
149

150
	add	t1, AES_TABLE1, t1	! 1
151
	and	t2, 255, t2		! 2
152
	ld	[T+t1], t1		! 1
153
154
155
	sll	t2, 2, t2		! 2
	add	t2, AES_TABLE2, t2	! 2
	ld	[T+t2], t2		! 2
156
157
158
	sll	t3, 2, t3		! 3
	add	t3, AES_TABLE3, t3	! 3
	ld	[T+t3], t3		! 3
159
160
	xor	t0, t1, t0		! 0, 1
	xor	t0, t2, t0		! 0, 1, 2
161

162
	add	idx, 4, idx		
163

164
	! Fetch roundkey
165
	add	round, ctx, t1
166
	ld	[t1+i], t1
167
		
168
	xor	t0, t3, t0		! 0, 1, 2, 3
169

170
	xor	t0, t1, t0
171
	st	t0, [tmp+i]
172
173

	cmp	i, 8
174

175
	bleu	.Linner_loop
176
	add	i, 4, i
177
	
178
	! switch roles for tmp and wtxt
179
	xor	wtxt, diff, wtxt
180

181
182
	add	round, 16, round
	cmp	round, nrounds
183
	blu	.Lround_loop
184
	xor	tmp, diff, tmp
185

186
	! sll	round, 4, %g2
187
188
	
	! final round
189
	add	round, ctx, %o7
190
	mov	0, i
191
	add	T, 288, %g4
192
.Lfinal_loop:
193
	ld	[%g4-32], %g2
194
	sll	i, 2, %i5
195
	sll	%g2, 2, %g2
196
	add	wtxt, %g2, %g2
197
	ldub	[%g2+2], %o3
198
	add	%i5, dst, %o2
199
	ld	[%g4-16], %g3
200
201
	add	i, 1, i
	ld	[wtxt+%i5], %g2
202
	sll	%g3, 2, %g3
203
	lduh	[wtxt+%g3], %o4
204
	and	%g2, 255, %g2
205
	ld	[%g4], %o5
206
	and	%o4, 255, %o4
207
	ldub	[T+%o3], %o0
208
	sll	%o5, 2, %o5
209
	ldub	[T+%g2], %g3
210
	sll	%o0, 8, %o0
211
	ldub	[wtxt+%o5], %o3
212
	or	%g3, %o0, %g3
213
	ldub	[T+%o4], %g2
214
	cmp	i, 3
215
	ldub	[T+%o3], %o5
216
217
218
	sll	%g2, 16, %g2
	or	%g3, %g2, %g3
	ld	[%o7], %g2
219
220
	sll	%o5, 24, %o5
	or	%g3, %o5, %g3
221
	xor	%g3, %g2, %g3
222
	srl	%g3, 24, %o5
223
	srl	%g3, 16, %o0
224
	srl	%g3, 8, %g2
225
226
227
	stb	%g2, [%o2+1]
	stb	%o5, [%o2+3]
	stb	%o0, [%o2+2]
228
	stb	%g3, [dst+%i5]
229
	add	%o7, 4, %o7
230
	bleu	.Lfinal_loop
231
	add	%g4, 4, %g4
232
	
233
	add	src, 16, src
234
	addcc	length, -16, length
235
	bne	.Lblock_loop
236
	add	dst, 16, dst
237
.Lend:
238
239
	ret
	restore
240
241
242
243
.LLFE1:
.LLfe1:
	.size	_aes_crypt,.LLfe1-_aes_crypt

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
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
	! 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