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
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
|
#!/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/.
# ====================================================================
#
# May 2011
#
# The module implements bn_GF2m_mul_2x2 polynomial multiplication used
# in bn_gf2m.c. It's kind of low-hanging mechanical port from C for
# the time being... Except that it has two code paths: code suitable
# for any x86_64 CPU and PCLMULQDQ one suitable for Westmere and
# later. Improvement varies from one benchmark and �-arch to another.
# Vanilla code path is at most 20% faster than compiler-generated code
# [not very impressive], while PCLMULQDQ - whole 85%-160% better on
# 163- and 571-bit ECDH benchmarks on Intel CPUs. Keep in mind that
# these coefficients are not ones for bn_GF2m_mul_2x2 itself, as not
# all CPU time is burnt in it...
$flavour = shift;
$output = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
die "can't locate x86_64-xlate.pl";
open OUT,"| \"$^X\" $xlate $flavour $output";
*STDOUT=*OUT;
($lo,$hi)=("%rax","%rdx"); $a=$lo;
($i0,$i1)=("%rsi","%rdi");
($t0,$t1)=("%rbx","%rcx");
($b,$mask)=("%rbp","%r8");
($a1,$a2,$a4,$a8,$a12,$a48)=map("%r$_",(9..15));
($R,$Tx)=("%xmm0","%xmm1");
$code.=<<___;
.text
.type _mul_1x1,\@abi-omnipotent
.align 16
_mul_1x1:
sub \$128+8,%rsp
mov \$-1,$a1
lea ($a,$a),$i0
shr \$3,$a1
lea (,$a,4),$i1
and $a,$a1 # a1=a&0x1fffffffffffffff
lea (,$a,8),$a8
sar \$63,$a # broadcast 63rd bit
lea ($a1,$a1),$a2
sar \$63,$i0 # broadcast 62nd bit
lea (,$a1,4),$a4
and $b,$a
sar \$63,$i1 # boardcast 61st bit
mov $a,$hi # $a is $lo
shl \$63,$lo
and $b,$i0
shr \$1,$hi
mov $i0,$t1
shl \$62,$i0
and $b,$i1
shr \$2,$t1
xor $i0,$lo
mov $i1,$t0
shl \$61,$i1
xor $t1,$hi
shr \$3,$t0
xor $i1,$lo
xor $t0,$hi
mov $a1,$a12
movq \$0,0(%rsp) # tab[0]=0
xor $a2,$a12 # a1^a2
mov $a1,8(%rsp) # tab[1]=a1
mov $a4,$a48
mov $a2,16(%rsp) # tab[2]=a2
xor $a8,$a48 # a4^a8
mov $a12,24(%rsp) # tab[3]=a1^a2
xor $a4,$a1
mov $a4,32(%rsp) # tab[4]=a4
xor $a4,$a2
mov $a1,40(%rsp) # tab[5]=a1^a4
xor $a4,$a12
mov $a2,48(%rsp) # tab[6]=a2^a4
xor $a48,$a1 # a1^a4^a4^a8=a1^a8
mov $a12,56(%rsp) # tab[7]=a1^a2^a4
xor $a48,$a2 # a2^a4^a4^a8=a1^a8
mov $a8,64(%rsp) # tab[8]=a8
xor $a48,$a12 # a1^a2^a4^a4^a8=a1^a2^a8
mov $a1,72(%rsp) # tab[9]=a1^a8
xor $a4,$a1 # a1^a8^a4
mov $a2,80(%rsp) # tab[10]=a2^a8
xor $a4,$a2 # a2^a8^a4
mov $a12,88(%rsp) # tab[11]=a1^a2^a8
xor $a4,$a12 # a1^a2^a8^a4
mov $a48,96(%rsp) # tab[12]=a4^a8
mov $mask,$i0
mov $a1,104(%rsp) # tab[13]=a1^a4^a8
and $b,$i0
mov $a2,112(%rsp) # tab[14]=a2^a4^a8
shr \$4,$b
mov $a12,120(%rsp) # tab[15]=a1^a2^a4^a8
mov $mask,$i1
and $b,$i1
shr \$4,$b
movq (%rsp,$i0,8),$R # half of calculations is done in SSE2
mov $mask,$i0
and $b,$i0
shr \$4,$b
___
for ($n=1;$n<8;$n++) {
$code.=<<___;
mov (%rsp,$i1,8),$t1
mov $mask,$i1
mov $t1,$t0
shl \$`8*$n-4`,$t1
and $b,$i1
movq (%rsp,$i0,8),$Tx
shr \$`64-(8*$n-4)`,$t0
xor $t1,$lo
pslldq \$$n,$Tx
mov $mask,$i0
shr \$4,$b
xor $t0,$hi
and $b,$i0
shr \$4,$b
pxor $Tx,$R
___
}
$code.=<<___;
mov (%rsp,$i1,8),$t1
mov $t1,$t0
shl \$`8*$n-4`,$t1
movq $R,$i0
shr \$`64-(8*$n-4)`,$t0
xor $t1,$lo
psrldq \$8,$R
xor $t0,$hi
movq $R,$i1
xor $i0,$lo
xor $i1,$hi
add \$128+8,%rsp
ret
.Lend_mul_1x1:
.size _mul_1x1,.-_mul_1x1
___
($rp,$a1,$a0,$b1,$b0) = $win64? ("%rcx","%rdx","%r8", "%r9","%r10") : # Win64 order
("%rdi","%rsi","%rdx","%rcx","%r8"); # Unix order
$code.=<<___;
.extern OPENSSL_ia32cap_P
.globl bn_GF2m_mul_2x2
.type bn_GF2m_mul_2x2,\@abi-omnipotent
.align 16
bn_GF2m_mul_2x2:
mov OPENSSL_ia32cap_P(%rip),%rax
bt \$33,%rax
jnc .Lvanilla_mul_2x2
movq $a1,%xmm0
movq $b1,%xmm1
movq $a0,%xmm2
___
$code.=<<___ if ($win64);
movq 40(%rsp),%xmm3
___
$code.=<<___ if (!$win64);
movq $b0,%xmm3
___
$code.=<<___;
movdqa %xmm0,%xmm4
movdqa %xmm1,%xmm5
pclmulqdq \$0,%xmm1,%xmm0 # a1�b1
pxor %xmm2,%xmm4
pxor %xmm3,%xmm5
pclmulqdq \$0,%xmm3,%xmm2 # a0�b0
pclmulqdq \$0,%xmm5,%xmm4 # (a0+a1)�(b0+b1)
xorps %xmm0,%xmm4
xorps %xmm2,%xmm4 # (a0+a1)�(b0+b1)-a0�b0-a1�b1
movdqa %xmm4,%xmm5
pslldq \$8,%xmm4
psrldq \$8,%xmm5
pxor %xmm4,%xmm2
pxor %xmm5,%xmm0
movdqu %xmm2,0($rp)
movdqu %xmm0,16($rp)
ret
.align 16
.Lvanilla_mul_2x2:
lea -8*17(%rsp),%rsp
___
$code.=<<___ if ($win64);
mov `8*17+40`(%rsp),$b0
mov %rdi,8*15(%rsp)
mov %rsi,8*16(%rsp)
___
$code.=<<___;
mov %r14,8*10(%rsp)
mov %r13,8*11(%rsp)
mov %r12,8*12(%rsp)
mov %rbp,8*13(%rsp)
mov %rbx,8*14(%rsp)
.Lbody_mul_2x2:
mov $rp,32(%rsp) # save the arguments
mov $a1,40(%rsp)
mov $a0,48(%rsp)
mov $b1,56(%rsp)
mov $b0,64(%rsp)
mov \$0xf,$mask
mov $a1,$a
mov $b1,$b
call _mul_1x1 # a1�b1
mov $lo,16(%rsp)
mov $hi,24(%rsp)
mov 48(%rsp),$a
mov 64(%rsp),$b
call _mul_1x1 # a0�b0
mov $lo,0(%rsp)
mov $hi,8(%rsp)
mov 40(%rsp),$a
mov 56(%rsp),$b
xor 48(%rsp),$a
xor 64(%rsp),$b
call _mul_1x1 # (a0+a1)�(b0+b1)
___
@r=("%rbx","%rcx","%rdi","%rsi");
$code.=<<___;
mov 0(%rsp),@r[0]
mov 8(%rsp),@r[1]
mov 16(%rsp),@r[2]
mov 24(%rsp),@r[3]
mov 32(%rsp),%rbp
xor $hi,$lo
xor @r[1],$hi
xor @r[0],$lo
mov @r[0],0(%rbp)
xor @r[2],$hi
mov @r[3],24(%rbp)
xor @r[3],$lo
xor @r[3],$hi
xor $hi,$lo
mov $hi,16(%rbp)
mov $lo,8(%rbp)
mov 8*10(%rsp),%r14
mov 8*11(%rsp),%r13
mov 8*12(%rsp),%r12
mov 8*13(%rsp),%rbp
mov 8*14(%rsp),%rbx
___
$code.=<<___ if ($win64);
mov 8*15(%rsp),%rdi
mov 8*16(%rsp),%rsi
___
$code.=<<___;
lea 8*17(%rsp),%rsp
ret
.Lend_mul_2x2:
.size bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
.asciz "GF(2^m) Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
.align 16
___
# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
# CONTEXT *context,DISPATCHER_CONTEXT *disp)
if ($win64) {
$rec="%rcx";
$frame="%rdx";
$context="%r8";
$disp="%r9";
$code.=<<___;
.extern __imp_RtlVirtualUnwind
.type se_handler,\@abi-omnipotent
.align 16
se_handler:
push %rsi
push %rdi
push %rbx
push %rbp
push %r12
push %r13
push %r14
push %r15
pushfq
sub \$64,%rsp
mov 152($context),%rax # pull context->Rsp
mov 248($context),%rbx # pull context->Rip
lea .Lbody_mul_2x2(%rip),%r10
cmp %r10,%rbx # context->Rip<"prologue" label
jb .Lin_prologue
mov 8*10(%rax),%r14 # mimic epilogue
mov 8*11(%rax),%r13
mov 8*12(%rax),%r12
mov 8*13(%rax),%rbp
mov 8*14(%rax),%rbx
mov 8*15(%rax),%rdi
mov 8*16(%rax),%rsi
mov %rbx,144($context) # restore context->Rbx
mov %rbp,160($context) # restore context->Rbp
mov %rsi,168($context) # restore context->Rsi
mov %rdi,176($context) # restore context->Rdi
mov %r12,216($context) # restore context->R12
mov %r13,224($context) # restore context->R13
mov %r14,232($context) # restore context->R14
.Lin_prologue:
lea 8*17(%rax),%rax
mov %rax,152($context) # restore context->Rsp
mov 40($disp),%rdi # disp->ContextRecord
mov $context,%rsi # context
mov \$154,%ecx # sizeof(CONTEXT)
.long 0xa548f3fc # cld; rep movsq
mov $disp,%rsi
xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
mov 8(%rsi),%rdx # arg2, disp->ImageBase
mov 0(%rsi),%r8 # arg3, disp->ControlPc
mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
mov 40(%rsi),%r10 # disp->ContextRecord
lea 56(%rsi),%r11 # &disp->HandlerData
lea 24(%rsi),%r12 # &disp->EstablisherFrame
mov %r10,32(%rsp) # arg5
mov %r11,40(%rsp) # arg6
mov %r12,48(%rsp) # arg7
mov %rcx,56(%rsp) # arg8, (NULL)
call *__imp_RtlVirtualUnwind(%rip)
mov \$1,%eax # ExceptionContinueSearch
add \$64,%rsp
popfq
pop %r15
pop %r14
pop %r13
pop %r12
pop %rbp
pop %rbx
pop %rdi
pop %rsi
ret
.size se_handler,.-se_handler
.section .pdata
.align 4
.rva _mul_1x1
.rva .Lend_mul_1x1
.rva .LSEH_info_1x1
.rva .Lvanilla_mul_2x2
.rva .Lend_mul_2x2
.rva .LSEH_info_2x2
.section .xdata
.align 8
.LSEH_info_1x1:
.byte 0x01,0x07,0x02,0x00
.byte 0x07,0x01,0x11,0x00 # sub rsp,128+8
.LSEH_info_2x2:
.byte 9,0,0,0
.rva se_handler
___
}
$code =~ s/\`([^\`]*)\`/eval($1)/gem;
print $code;
close STDOUT;
|