aes.asm 7.01 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
72
	add	i, 4, i
	add	i, src, %o5
73
74
	ldub	[%o5+3], %g2
	ldub	[%o5+2], %g3
75
	
76
	sll	%g2, 24, %g2
77
	ldub	[%o5+1], %o0
78
79
	sll	%g3, 16, %g3
	or	%g2, %g3, %g2
80
	
81
	ldub	[src+i], %o5
82
	sll	%o0, 8, %o0
83
	ld	[ctx+i], %g3
84
	or	%g2, %o0, %g2
85
	
86
	or	%g2, %o5, %g2
87
	xor	%g2, %g3, %g2
88
	cmp	i, 12
89
	bleu	.Lsource_loop
90
	
91
	st	%g2, [wtxt+i]
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
99
	
100
101
102
103
104
105
106
107
108
109
	! 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
	! 
110
	! ld	[ctx+%o3], %g3
111
	
112
113
114
	! xor	%g3, %g2, %g2
	! 
	! add	src, 4, src
115
	! st	%g2, [wtxt+%o4]
116
	! 
117
	! cmp	%o3, 8
118
	
119
	! bleu	.Lsource_loop
120
	! add	%o3, 4, %o3
121

122
	mov	16, round
123
	! add	ctx, 16, key
124

125
.Lround_loop:
126
127
	! 4*i
	mov	0, i
128
	add	T, AES_SIDX3, idx
129
.Linner_loop:
130
	! The comments mark which j in T->table[j][ Bj(wtxt[IDXi(i)]) ]
131
132
133
134
	! 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.
135
	
136
	! AES_SIDX1
137
	ld	[idx-32], t1		! 1
138
	! AES_SIDX2
139
140
	! IDX2(j) = j XOR 2
	xor	i, 8, t2
141
	! wtxt[IDX1...]
142
143
	add	wtxt, t1, t1		! 1
	ldub	[t1+2], t1		! 1
144

145
	! AES_SIDX3
146
	ld	[idx], t3		! 3
147
	sll	t1, 2, t1		! 1
148
	! wtxt[i]
149
	ld	[wtxt+i], t0		! 0
150
	! wtxt[IDX2...]
151
	lduh	[wtxt+t2], t2		! 2
152
	
153
	and	t0, 255, t0		! 0
154
	! wtxt[IDX3...]
155
	ldub	[wtxt+t3], t3		! 3
156
	sll	t0, 2, t0		! 0
157
	ld	[T0+t0], t0		! 0
158

159
	and	t2, 255, t2		! 2
160
	ld	[T1+t1], t1		! 1
161
	sll	t2, 2, t2		! 2
162
163
	ld	[T2+t2], t2		! 2

164
	sll	t3, 2, t3		! 3
165
	ld	[T3+t3], t3		! 3
166
167
	xor	t0, t1, t0		! 0, 1
	xor	t0, t2, t0		! 0, 1, 2
168

169
	add	idx, 4, idx		
170
	! Fetch roundkey
171
172
	ld	[ctx+round], t1
	add	round, 4, round
173
	! ld	[key+i], t1
174
	xor	t0, t3, t0		! 0, 1, 2, 3
175

176
	xor	t0, t1, t0
177
	st	t0, [tmp+i]
178
	cmp	i, 8
179
	bleu	.Linner_loop
180
	
181
	add	i, 4, i
182
	
183
	! switch roles for tmp and wtxt
184
	xor	wtxt, diff, wtxt
185
	! add	key, 16, key
186
	cmp	round, nrounds
187
	blu	.Lround_loop
188
	xor	tmp, diff, tmp
189

190
	! final round
191
	mov	0, i
192
	add	T, 288, %g4
193
.Lfinal_loop:
194
	ld	[%g4-32], %g2
195
	sll	i, 2, %i5
196
	sll	%g2, 2, %g2
197
	add	wtxt, %g2, %g2
198
	ldub	[%g2+2], %o3
199
	add	%i5, dst, %o2
200
	ld	[%g4-16], %g3
201
202
	add	i, 1, i
	ld	[wtxt+%i5], %g2
203
	sll	%g3, 2, %g3
204
	lduh	[wtxt+%g3], %o4
205
	and	%g2, 255, %g2
206
	ld	[%g4], %o5
207
	and	%o4, 255, %o4
208
	ldub	[T+%o3], %o0
209
	sll	%o5, 2, %o5
210
	ldub	[T+%g2], %g3
211
	sll	%o0, 8, %o0
212
	ldub	[wtxt+%o5], %o3
213
	or	%g3, %o0, %g3
214
	ldub	[T+%o4], %g2
215
	cmp	i, 3
216
	ldub	[T+%o3], %o5
217
218
	sll	%g2, 16, %g2
	or	%g3, %g2, %g3
219
	ld	[ctx + round], %g2
220
221
	sll	%o5, 24, %o5
	or	%g3, %o5, %g3
222
	xor	%g3, %g2, %g3
223
	srl	%g3, 24, %o5
224
	srl	%g3, 16, %o0
225
	srl	%g3, 8, %g2
226
227
228
	stb	%g2, [%o2+1]
	stb	%o5, [%o2+3]
	stb	%o0, [%o2+2]
229
	stb	%g3, [dst+%i5]
230
	add	round, 4, round
231
	bleu	.Lfinal_loop
232
	add	%g4, 4, %g4
233
	
234
	add	src, 16, src
235
	addcc	length, -16, length
236
	bne	.Lblock_loop
237
	add	dst, 16, dst
238
.Lend:
239
240
	ret
	restore
241
242
243
244
.LLFE1:
.LLfe1:
	.size	_aes_crypt,.LLfe1-_aes_crypt

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
297
	! 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
298
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

	! 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