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
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
|
#!/usr/bin/env perl
# ====================================================================
# Written by Andy Polyakov <appro@openssl.org> 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/.
# ====================================================================
# October 2012.
#
# SPARCv9 VIS3 Montgomery multiplicaion procedure suitable for T3 and
# onward. There are three new instructions used here: umulxhi,
# addxc[cc] and initializing store. On T3 RSA private key operations
# are 1.54/1.87/2.11/2.26 times faster for 512/1024/2048/4096-bit key
# lengths. This is without dedicated squaring procedure. On T4
# corresponding coefficients are 1.47/2.10/2.80/2.90x, which is mostly
# for reference purposes, because T4 has dedicated Montgomery
# multiplication and squaring *instructions* that deliver even more.
$bits=32;
for (@ARGV) { $bits=64 if (/\-m64/ || /\-xarch\=v9/); }
if ($bits==64) { $bias=2047; $frame=192; }
else { $bias=0; $frame=112; }
$code.=<<___ if ($bits==64);
.register %g2,#scratch
.register %g3,#scratch
___
$code.=<<___;
.section ".text",#alloc,#execinstr
___
($n0,$m0,$m1,$lo0,$hi0, $lo1,$hi1,$aj,$alo,$nj,$nlo,$tj)=
(map("%g$_",(1..5)),map("%o$_",(0..5,7)));
# int bn_mul_mont(
$rp="%o0"; # BN_ULONG *rp,
$ap="%o1"; # const BN_ULONG *ap,
$bp="%o2"; # const BN_ULONG *bp,
$np="%o3"; # const BN_ULONG *np,
$n0p="%o4"; # const BN_ULONG *n0,
$num="%o5"; # int num); # caller ensures that num is even
# and >=6
$code.=<<___;
.globl bn_mul_mont_vis3
.align 32
bn_mul_mont_vis3:
add %sp, $bias, %g4 ! real top of stack
sll $num, 2, $num ! size in bytes
add $num, 63, %g5
andn %g5, 63, %g5 ! buffer size rounded up to 64 bytes
add %g5, %g5, %g1
add %g5, %g1, %g1 ! 3*buffer size
sub %g4, %g1, %g1
andn %g1, 63, %g1 ! align at 64 byte
sub %g1, $frame, %g1 ! new top of stack
sub %g1, %g4, %g1
save %sp, %g1, %sp
___
# +-------------------------------+<----- %sp
# . .
# +-------------------------------+<----- aligned at 64 bytes
# | __int64 tmp[0] |
# +-------------------------------+
# . .
# . .
# +-------------------------------+<----- aligned at 64 bytes
# | __int64 ap[1..0] | converted ap[]
# +-------------------------------+
# | __int64 np[1..0] | converted np[]
# +-------------------------------+
# | __int64 ap[3..2] |
# . .
# . .
# +-------------------------------+
($rp,$ap,$bp,$np,$n0p,$num)=map("%i$_",(0..5));
($t0,$t1,$t2,$t3,$cnt,$tp,$bufsz,$anp)=map("%l$_",(0..7));
($ovf,$i)=($t0,$t1);
$code.=<<___;
ld [$n0p+0], $t0 ! pull n0[0..1] value
add %sp, $bias+$frame, $tp
ld [$n0p+4], $t1
add $tp, %g5, $anp
ld [$bp+0], $t2 ! m0=bp[0]
sllx $t1, 32, $n0
ld [$bp+4], $t3
or $t0, $n0, $n0
add $bp, 8, $bp
ld [$ap+0], $t0 ! ap[0]
sllx $t3, 32, $m0
ld [$ap+4], $t1
or $t2, $m0, $m0
ld [$ap+8], $t2 ! ap[1]
sllx $t1, 32, $aj
ld [$ap+12], $t3
or $t0, $aj, $aj
add $ap, 16, $ap
stxa $aj, [$anp]0xe2 ! converted ap[0]
mulx $aj, $m0, $lo0 ! ap[0]*bp[0]
umulxhi $aj, $m0, $hi0
ld [$np+0], $t0 ! np[0]
sllx $t3, 32, $aj
ld [$np+4], $t1
or $t2, $aj, $aj
ld [$np+8], $t2 ! np[1]
sllx $t1, 32, $nj
ld [$np+12], $t3
or $t0, $nj, $nj
add $np, 16, $np
stx $nj, [$anp+8] ! converted np[0]
mulx $lo0, $n0, $m1 ! "tp[0]"*n0
stx $aj, [$anp+16] ! converted ap[1]
mulx $aj, $m0, $alo ! ap[1]*bp[0]
umulxhi $aj, $m0, $aj ! ahi=aj
mulx $nj, $m1, $lo1 ! np[0]*m1
umulxhi $nj, $m1, $hi1
sllx $t3, 32, $nj
or $t2, $nj, $nj
stx $nj, [$anp+24] ! converted np[1]
add $anp, 32, $anp
addcc $lo0, $lo1, $lo1
addxc %g0, $hi1, $hi1
mulx $nj, $m1, $nlo ! np[1]*m1
umulxhi $nj, $m1, $nj ! nhi=nj
ba .L1st
sub $num, 24, $cnt ! cnt=num-3
.align 16
.L1st:
ld [$ap+0], $t0 ! ap[j]
addcc $alo, $hi0, $lo0
ld [$ap+4], $t1
addxc $aj, %g0, $hi0
sllx $t1, 32, $aj
add $ap, 8, $ap
or $t0, $aj, $aj
stxa $aj, [$anp]0xe2 ! converted ap[j]
ld [$np+0], $t2 ! np[j]
addcc $nlo, $hi1, $lo1
ld [$np+4], $t3
addxc $nj, %g0, $hi1 ! nhi=nj
sllx $t3, 32, $nj
add $np, 8, $np
mulx $aj, $m0, $alo ! ap[j]*bp[0]
or $t2, $nj, $nj
umulxhi $aj, $m0, $aj ! ahi=aj
stx $nj, [$anp+8] ! converted np[j]
add $anp, 16, $anp ! anp++
mulx $nj, $m1, $nlo ! np[j]*m1
addcc $lo0, $lo1, $lo1 ! np[j]*m1+ap[j]*bp[0]
umulxhi $nj, $m1, $nj ! nhi=nj
addxc %g0, $hi1, $hi1
stxa $lo1, [$tp]0xe2 ! tp[j-1]
add $tp, 8, $tp ! tp++
brnz,pt $cnt, .L1st
sub $cnt, 8, $cnt ! j--
!.L1st
addcc $alo, $hi0, $lo0
addxc $aj, %g0, $hi0 ! ahi=aj
addcc $nlo, $hi1, $lo1
addxc $nj, %g0, $hi1
addcc $lo0, $lo1, $lo1 ! np[j]*m1+ap[j]*bp[0]
addxc %g0, $hi1, $hi1
stxa $lo1, [$tp]0xe2 ! tp[j-1]
add $tp, 8, $tp
addcc $hi0, $hi1, $hi1
addxc %g0, %g0, $ovf ! upmost overflow bit
stxa $hi1, [$tp]0xe2
add $tp, 8, $tp
ba .Louter
sub $num, 16, $i ! i=num-2
.align 16
.Louter:
ld [$bp+0], $t2 ! m0=bp[i]
ld [$bp+4], $t3
sub $anp, $num, $anp ! rewind
sub $tp, $num, $tp
sub $anp, $num, $anp
add $bp, 8, $bp
sllx $t3, 32, $m0
ldx [$anp+0], $aj ! ap[0]
or $t2, $m0, $m0
ldx [$anp+8], $nj ! np[0]
mulx $aj, $m0, $lo0 ! ap[0]*bp[i]
ldx [$tp], $tj ! tp[0]
umulxhi $aj, $m0, $hi0
ldx [$anp+16], $aj ! ap[1]
addcc $lo0, $tj, $lo0 ! ap[0]*bp[i]+tp[0]
mulx $aj, $m0, $alo ! ap[1]*bp[i]
addxc %g0, $hi0, $hi0
mulx $lo0, $n0, $m1 ! tp[0]*n0
umulxhi $aj, $m0, $aj ! ahi=aj
mulx $nj, $m1, $lo1 ! np[0]*m1
umulxhi $nj, $m1, $hi1
ldx [$anp+24], $nj ! np[1]
add $anp, 32, $anp
addcc $lo1, $lo0, $lo1
mulx $nj, $m1, $nlo ! np[1]*m1
addxc %g0, $hi1, $hi1
umulxhi $nj, $m1, $nj ! nhi=nj
ba .Linner
sub $num, 24, $cnt ! cnt=num-3
.align 16
.Linner:
addcc $alo, $hi0, $lo0
ldx [$tp+8], $tj ! tp[j]
addxc $aj, %g0, $hi0 ! ahi=aj
ldx [$anp+0], $aj ! ap[j]
addcc $nlo, $hi1, $lo1
mulx $aj, $m0, $alo ! ap[j]*bp[i]
addxc $nj, %g0, $hi1 ! nhi=nj
ldx [$anp+8], $nj ! np[j]
add $anp, 16, $anp
umulxhi $aj, $m0, $aj ! ahi=aj
addcc $lo0, $tj, $lo0 ! ap[j]*bp[i]+tp[j]
mulx $nj, $m1, $nlo ! np[j]*m1
addxc %g0, $hi0, $hi0
umulxhi $nj, $m1, $nj ! nhi=nj
addcc $lo1, $lo0, $lo1 ! np[j]*m1+ap[j]*bp[i]+tp[j]
addxc %g0, $hi1, $hi1
stx $lo1, [$tp] ! tp[j-1]
add $tp, 8, $tp
brnz,pt $cnt, .Linner
sub $cnt, 8, $cnt
!.Linner
ldx [$tp+8], $tj ! tp[j]
addcc $alo, $hi0, $lo0
addxc $aj, %g0, $hi0 ! ahi=aj
addcc $lo0, $tj, $lo0 ! ap[j]*bp[i]+tp[j]
addxc %g0, $hi0, $hi0
addcc $nlo, $hi1, $lo1
addxc $nj, %g0, $hi1 ! nhi=nj
addcc $lo1, $lo0, $lo1 ! np[j]*m1+ap[j]*bp[i]+tp[j]
addxc %g0, $hi1, $hi1
stx $lo1, [$tp] ! tp[j-1]
subcc %g0, $ovf, %g0 ! move upmost overflow to CCR.xcc
addxccc $hi1, $hi0, $hi1
addxc %g0, %g0, $ovf
stx $hi1, [$tp+8]
add $tp, 16, $tp
brnz,pt $i, .Louter
sub $i, 8, $i
sub $anp, $num, $anp ! rewind
sub $tp, $num, $tp
sub $anp, $num, $anp
ba .Lsub
subcc $num, 8, $cnt ! cnt=num-1 and clear CCR.xcc
.align 16
.Lsub:
ldx [$tp], $tj
add $tp, 8, $tp
ldx [$anp+8], $nj
add $anp, 16, $anp
subccc $tj, $nj, $t2 ! tp[j]-np[j]
srlx $tj, 32, $tj
srlx $nj, 32, $nj
subccc $tj, $nj, $t3
add $rp, 8, $rp
st $t2, [$rp-4] ! reverse order
st $t3, [$rp-8]
brnz,pt $cnt, .Lsub
sub $cnt, 8, $cnt
sub $anp, $num, $anp ! rewind
sub $tp, $num, $tp
sub $anp, $num, $anp
sub $rp, $num, $rp
subc $ovf, %g0, $ovf ! handle upmost overflow bit
and $tp, $ovf, $ap
andn $rp, $ovf, $np
or $np, $ap, $ap ! ap=borrow?tp:rp
ba .Lcopy
sub $num, 8, $cnt
.align 16
.Lcopy: ! copy or in-place refresh
ld [$ap+0], $t2
ld [$ap+4], $t3
add $ap, 8, $ap
stx %g0, [$tp] ! zap
add $tp, 8, $tp
stx %g0, [$anp] ! zap
stx %g0, [$anp+8]
add $anp, 16, $anp
st $t3, [$rp+0] ! flip order
st $t2, [$rp+4]
add $rp, 8, $rp
brnz $cnt, .Lcopy
sub $cnt, 8, $cnt
mov 1, %o0
ret
restore
.type bn_mul_mont_vis3, #function
.size bn_mul_mont_vis3, .-bn_mul_mont_vis3
.asciz "Montgomery Multiplication for SPARCv9 VIS3, CRYPTOGAMS by <appro\@openssl.org>"
.align 4
___
# Purpose of these subroutines is to explicitly encode VIS instructions,
# so that one can compile the module without having to specify VIS
# extensions on compiler command line, e.g. -xarch=v9 vs. -xarch=v9a.
# Idea is to reserve for option to produce "universal" binary and let
# programmer detect if current CPU is VIS capable at run-time.
sub unvis3 {
my ($mnemonic,$rs1,$rs2,$rd)=@_;
my %bias = ( "g" => 0, "o" => 8, "l" => 16, "i" => 24 );
my ($ref,$opf);
my %visopf = ( "addxc" => 0x011,
"addxccc" => 0x013,
"umulxhi" => 0x016 );
$ref = "$mnemonic\t$rs1,$rs2,$rd";
if ($opf=$visopf{$mnemonic}) {
foreach ($rs1,$rs2,$rd) {
return $ref if (!/%([goli])([0-9])/);
$_=$bias{$1}+$2;
}
return sprintf ".word\t0x%08x !%s",
0x81b00000|$rd<<25|$rs1<<14|$opf<<5|$rs2,
$ref;
} else {
return $ref;
}
}
foreach (split("\n",$code)) {
s/\`([^\`]*)\`/eval $1/ge;
s/\b(umulxhi|addxc[c]{0,2})\s+(%[goli][0-7]),\s*(%[goli][0-7]),\s*(%[goli][0-7])/
&unvis3($1,$2,$3,$4)
/ge;
print $_,"\n";
}
close STDOUT;
|