aboutsummaryrefslogtreecommitdiff
path: root/openssl/crypto/bn/asm/alpha-mont.pl
blob: 7a2cc3173b0ec49b29771bec2c24f94ada1ffd0a (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
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
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
309
310
311
312
313
314
315
316
317
#!/usr/bin/env perl
#
# ====================================================================
# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
# project. The module is, however, dual licensed under OpenSSL and
# CRYPTOGAMS licenses depending on where you obtain it. For further
# details see http://www.openssl.org/~appro/cryptogams/.
# ====================================================================
#
# On 21264 RSA sign performance improves by 70/35/20/15 percent for
# 512/1024/2048/4096 bit key lengths. This is against vendor compiler
# instructed to '-tune host' code with in-line assembler. Other
# benchmarks improve by 15-20%. To anchor it to something else, the
# code provides approximately the same performance per GHz as AMD64.
# I.e. if you compare 1GHz 21264 and 2GHz Opteron, you'll observe ~2x
# difference.

# int bn_mul_mont(
$rp="a0";	# BN_ULONG *rp,
$ap="a1";	# const BN_ULONG *ap,
$bp="a2";	# const BN_ULONG *bp,
$np="a3";	# const BN_ULONG *np,
$n0="a4";	# const BN_ULONG *n0,
$num="a5";	# int num);

$lo0="t0";
$hi0="t1";
$lo1="t2";
$hi1="t3";
$aj="t4";
$bi="t5";
$nj="t6";
$tp="t7";
$alo="t8";
$ahi="t9";
$nlo="t10";
$nhi="t11";
$tj="t12";
$i="s3";
$j="s4";
$m1="s5";

$code=<<___;
#include <asm.h>
#include <regdef.h>

.text

.set	noat
.set	noreorder

.globl	bn_mul_mont
.align	5
.ent	bn_mul_mont
bn_mul_mont:
	lda	sp,-40(sp)
	stq	ra,0(sp)
	stq	s3,8(sp)
	stq	s4,16(sp)
	stq	s5,24(sp)
	stq	fp,32(sp)
	mov	sp,fp
	.mask	0x0400f000,-40
	.frame	fp,40,ra
	.prologue 0

	.align	4
	.set	reorder
	sextl	$num,$num
	mov	0,v0
	cmplt	$num,4,AT
	bne	AT,.Lexit

	ldq	$hi0,0($ap)	# ap[0]
	s8addq	$num,16,AT
	ldq	$aj,8($ap)
	subq	sp,AT,sp
	ldq	$bi,0($bp)	# bp[0]
	mov	-4096,AT
	ldq	$n0,0($n0)
	and	sp,AT,sp

	mulq	$hi0,$bi,$lo0
	ldq	$hi1,0($np)	# np[0]
	umulh	$hi0,$bi,$hi0
	ldq	$nj,8($np)

	mulq	$lo0,$n0,$m1

	mulq	$hi1,$m1,$lo1
	umulh	$hi1,$m1,$hi1

	addq	$lo1,$lo0,$lo1
	cmpult	$lo1,$lo0,AT
	addq	$hi1,AT,$hi1

	mulq	$aj,$bi,$alo
	mov	2,$j
	umulh	$aj,$bi,$ahi
	mov	sp,$tp

	mulq	$nj,$m1,$nlo
	s8addq	$j,$ap,$aj
	umulh	$nj,$m1,$nhi
	s8addq	$j,$np,$nj
.align	4
.L1st:
	.set	noreorder
	ldq	$aj,($aj)
	addl	$j,1,$j
	ldq	$nj,($nj)
	lda	$tp,8($tp)

	addq	$alo,$hi0,$lo0
	mulq	$aj,$bi,$alo
	cmpult	$lo0,$hi0,AT
	addq	$nlo,$hi1,$lo1

	mulq	$nj,$m1,$nlo
	addq	$ahi,AT,$hi0
	cmpult	$lo1,$hi1,v0
	cmplt	$j,$num,$tj

	umulh	$aj,$bi,$ahi
	addq	$nhi,v0,$hi1
	addq	$lo1,$lo0,$lo1
	s8addq	$j,$ap,$aj

	umulh	$nj,$m1,$nhi
	cmpult	$lo1,$lo0,v0
	addq	$hi1,v0,$hi1
	s8addq	$j,$np,$nj

	stq	$lo1,-8($tp)
	nop
	unop
	bne	$tj,.L1st
	.set	reorder

	addq	$alo,$hi0,$lo0
	addq	$nlo,$hi1,$lo1
	cmpult	$lo0,$hi0,AT
	cmpult	$lo1,$hi1,v0
	addq	$ahi,AT,$hi0
	addq	$nhi,v0,$hi1

	addq	$lo1,$lo0,$lo1
	cmpult	$lo1,$lo0,v0
	addq	$hi1,v0,$hi1

	stq	$lo1,0($tp)

	addq	$hi1,$hi0,$hi1
	cmpult	$hi1,$hi0,AT
	stq	$hi1,8($tp)
	stq	AT,16($tp)

	mov	1,$i
.align	4
.Louter:
	s8addq	$i,$bp,$bi
	ldq	$hi0,($ap)
	ldq	$aj,8($ap)
	ldq	$bi,($bi)
	ldq	$hi1,($np)
	ldq	$nj,8($np)
	ldq	$tj,(sp)

	mulq	$hi0,$bi,$lo0
	umulh	$hi0,$bi,$hi0

	addq	$lo0,$tj,$lo0
	cmpult	$lo0,$tj,AT
	addq	$hi0,AT,$hi0

	mulq	$lo0,$n0,$m1

	mulq	$hi1,$m1,$lo1
	umulh	$hi1,$m1,$hi1

	addq	$lo1,$lo0,$lo1
	cmpult	$lo1,$lo0,AT
	mov	2,$j
	addq	$hi1,AT,$hi1

	mulq	$aj,$bi,$alo
	mov	sp,$tp
	umulh	$aj,$bi,$ahi

	mulq	$nj,$m1,$nlo
	s8addq	$j,$ap,$aj
	umulh	$nj,$m1,$nhi
.align	4
.Linner:
	.set	noreorder
	ldq	$tj,8($tp)	#L0
	nop			#U1
	ldq	$aj,($aj)	#L1
	s8addq	$j,$np,$nj	#U0

	ldq	$nj,($nj)	#L0
	nop			#U1
	addq	$alo,$hi0,$lo0	#L1
	lda	$tp,8($tp)

	mulq	$aj,$bi,$alo	#U1
	cmpult	$lo0,$hi0,AT	#L0
	addq	$nlo,$hi1,$lo1	#L1
	addl	$j,1,$j

	mulq	$nj,$m1,$nlo	#U1
	addq	$ahi,AT,$hi0	#L0
	addq	$lo0,$tj,$lo0	#L1
	cmpult	$lo1,$hi1,v0	#U0

	umulh	$aj,$bi,$ahi	#U1
	cmpult	$lo0,$tj,AT	#L0
	addq	$lo1,$lo0,$lo1	#L1
	addq	$nhi,v0,$hi1	#U0

	umulh	$nj,$m1,$nhi	#U1
	s8addq	$j,$ap,$aj	#L0
	cmpult	$lo1,$lo0,v0	#L1
	cmplt	$j,$num,$tj	#U0	# borrow $tj

	addq	$hi0,AT,$hi0	#L0
	addq	$hi1,v0,$hi1	#U1
	stq	$lo1,-8($tp)	#L1
	bne	$tj,.Linner	#U0
	.set	reorder

	ldq	$tj,8($tp)
	addq	$alo,$hi0,$lo0
	addq	$nlo,$hi1,$lo1
	cmpult	$lo0,$hi0,AT
	cmpult	$lo1,$hi1,v0
	addq	$ahi,AT,$hi0
	addq	$nhi,v0,$hi1

	addq	$lo0,$tj,$lo0
	cmpult	$lo0,$tj,AT
	addq	$hi0,AT,$hi0

	ldq	$tj,16($tp)
	addq	$lo1,$lo0,$j
	cmpult	$j,$lo0,v0
	addq	$hi1,v0,$hi1

	addq	$hi1,$hi0,$lo1
	stq	$j,($tp)
	cmpult	$lo1,$hi0,$hi1
	addq	$lo1,$tj,$lo1
	cmpult	$lo1,$tj,AT
	addl	$i,1,$i
	addq	$hi1,AT,$hi1
	stq	$lo1,8($tp)
	cmplt	$i,$num,$tj	# borrow $tj
	stq	$hi1,16($tp)
	bne	$tj,.Louter

	s8addq	$num,sp,$tj	# &tp[num]
	mov	$rp,$bp		# put rp aside
	mov	sp,$tp
	mov	sp,$ap
	mov	0,$hi0		# clear borrow bit

.align	4
.Lsub:	ldq	$lo0,($tp)
	ldq	$lo1,($np)
	lda	$tp,8($tp)
	lda	$np,8($np)
	subq	$lo0,$lo1,$lo1	# tp[i]-np[i]
	cmpult	$lo0,$lo1,AT
	subq	$lo1,$hi0,$lo0
	cmpult	$lo1,$lo0,$hi0
	or	$hi0,AT,$hi0
	stq	$lo0,($rp)
	cmpult	$tp,$tj,v0
	lda	$rp,8($rp)
	bne	v0,.Lsub

	subq	$hi1,$hi0,$hi0	# handle upmost overflow bit
	mov	sp,$tp
	mov	$bp,$rp		# restore rp

	and	sp,$hi0,$ap
	bic	$bp,$hi0,$bp
	bis	$bp,$ap,$ap	# ap=borrow?tp:rp

.align	4
.Lcopy:	ldq	$aj,($ap)	# copy or in-place refresh
	lda	$tp,8($tp)
	lda	$rp,8($rp)
	lda	$ap,8($ap)
	stq	zero,-8($tp)	# zap tp
	cmpult	$tp,$tj,AT
	stq	$aj,-8($rp)
	bne	AT,.Lcopy
	mov	1,v0

.Lexit:
	.set	noreorder
	mov	fp,sp
	/*ldq	ra,0(sp)*/
	ldq	s3,8(sp)
	ldq	s4,16(sp)
	ldq	s5,24(sp)
	ldq	fp,32(sp)
	lda	sp,40(sp)
	ret	(ra)
.end	bn_mul_mont
.rdata
.asciiz	"Montgomery Multiplication for Alpha, CRYPTOGAMS by <appro\@openssl.org>"
___

print $code;
close STDOUT;