summaryrefslogtreecommitdiff
path: root/src/lib/libcrypto/rc4
diff options
context:
space:
mode:
Diffstat (limited to 'src/lib/libcrypto/rc4')
-rw-r--r--src/lib/libcrypto/rc4/asm/rc4-md5-x86_64.pl515
1 files changed, 0 insertions, 515 deletions
diff --git a/src/lib/libcrypto/rc4/asm/rc4-md5-x86_64.pl b/src/lib/libcrypto/rc4/asm/rc4-md5-x86_64.pl
deleted file mode 100644
index e5e8aa08a1..0000000000
--- a/src/lib/libcrypto/rc4/asm/rc4-md5-x86_64.pl
+++ /dev/null
@@ -1,515 +0,0 @@
1#!/usr/bin/env perl
2#
3# ====================================================================
4# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
5# project. The module is, however, dual licensed under OpenSSL and
6# CRYPTOGAMS licenses depending on where you obtain it. For further
7# details see http://www.openssl.org/~appro/cryptogams/.
8# ====================================================================
9
10# June 2011
11#
12# This is RC4+MD5 "stitch" implementation. The idea, as spelled in
13# http://download.intel.com/design/intarch/papers/323686.pdf, is that
14# since both algorithms exhibit instruction-level parallelism, ILP,
15# below theoretical maximum, interleaving them would allow to utilize
16# processor resources better and achieve better performance. RC4
17# instruction sequence is virtually identical to rc4-x86_64.pl, which
18# is heavily based on submission by Maxim Perminov, Maxim Locktyukhin
19# and Jim Guilford of Intel. MD5 is fresh implementation aiming to
20# minimize register usage, which was used as "main thread" with RC4
21# weaved into it, one RC4 round per one MD5 round. In addition to the
22# stiched subroutine the script can generate standalone replacement
23# md5_block_asm_data_order and RC4. Below are performance numbers in
24# cycles per processed byte, less is better, for these the standalone
25# subroutines, sum of them, and stitched one:
26#
27# RC4 MD5 RC4+MD5 stitch gain
28# Opteron 6.5(*) 5.4 11.9 7.0 +70%(*)
29# Core2 6.5 5.8 12.3 7.7 +60%
30# Westmere 4.3 5.2 9.5 7.0 +36%
31# Sandy Bridge 4.2 5.5 9.7 6.8 +43%
32# Atom 9.3 6.5 15.8 11.1 +42%
33#
34# (*) rc4-x86_64.pl delivers 5.3 on Opteron, so real improvement
35# is +53%...
36
37my ($rc4,$md5)=(1,1); # what to generate?
38my $D="#" if (!$md5); # if set to "#", MD5 is stitched into RC4(),
39 # but its result is discarded. Idea here is
40 # to be able to use 'openssl speed rc4' for
41 # benchmarking the stitched subroutine...
42
43my $flavour = shift;
44my $output = shift;
45if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
46
47$0 =~ m/(.*[\/\\])[^\/\\]+$/; my $dir=$1; my $xlate;
48( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
49( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
50die "can't locate x86_64-xlate.pl";
51
52open OUT,"| \"$^X\" $xlate $flavour $output";
53*STDOUT=*OUT;
54
55my ($dat,$in0,$out,$ctx,$inp,$len, $func,$nargs);
56
57if ($rc4 && !$md5) {
58 ($dat,$len,$in0,$out) = ("%rdi","%rsi","%rdx","%rcx");
59 $func="RC4"; $nargs=4;
60} elsif ($md5 && !$rc4) {
61 ($ctx,$inp,$len) = ("%rdi","%rsi","%rdx");
62 $func="md5_block_asm_data_order"; $nargs=3;
63} else {
64 ($dat,$in0,$out,$ctx,$inp,$len) = ("%rdi","%rsi","%rdx","%rcx","%r8","%r9");
65 $func="rc4_md5_enc"; $nargs=6;
66 # void rc4_md5_enc(
67 # RC4_KEY *key, #
68 # const void *in0, # RC4 input
69 # void *out, # RC4 output
70 # MD5_CTX *ctx, #
71 # const void *inp, # MD5 input
72 # size_t len); # number of 64-byte blocks
73}
74
75my @K=( 0xd76aa478,0xe8c7b756,0x242070db,0xc1bdceee,
76 0xf57c0faf,0x4787c62a,0xa8304613,0xfd469501,
77 0x698098d8,0x8b44f7af,0xffff5bb1,0x895cd7be,
78 0x6b901122,0xfd987193,0xa679438e,0x49b40821,
79
80 0xf61e2562,0xc040b340,0x265e5a51,0xe9b6c7aa,
81 0xd62f105d,0x02441453,0xd8a1e681,0xe7d3fbc8,
82 0x21e1cde6,0xc33707d6,0xf4d50d87,0x455a14ed,
83 0xa9e3e905,0xfcefa3f8,0x676f02d9,0x8d2a4c8a,
84
85 0xfffa3942,0x8771f681,0x6d9d6122,0xfde5380c,
86 0xa4beea44,0x4bdecfa9,0xf6bb4b60,0xbebfbc70,
87 0x289b7ec6,0xeaa127fa,0xd4ef3085,0x04881d05,
88 0xd9d4d039,0xe6db99e5,0x1fa27cf8,0xc4ac5665,
89
90 0xf4292244,0x432aff97,0xab9423a7,0xfc93a039,
91 0x655b59c3,0x8f0ccc92,0xffeff47d,0x85845dd1,
92 0x6fa87e4f,0xfe2ce6e0,0xa3014314,0x4e0811a1,
93 0xf7537e82,0xbd3af235,0x2ad7d2bb,0xeb86d391 );
94
95my @V=("%r8d","%r9d","%r10d","%r11d"); # MD5 registers
96my $tmp="%r12d";
97
98my @XX=("%rbp","%rsi"); # RC4 registers
99my @TX=("%rax","%rbx");
100my $YY="%rcx";
101my $TY="%rdx";
102
103my $MOD=32; # 16, 32 or 64
104
105$code.=<<___;
106.text
107.align 16
108
109.globl $func
110.type $func,\@function,$nargs
111$func:
112 _CET_ENDBR
113 cmp \$0,$len
114 je .Labort
115 push %rbx
116 push %rbp
117 push %r12
118 push %r13
119 push %r14
120 push %r15
121 sub \$40,%rsp
122.Lbody:
123___
124if ($rc4) {
125$code.=<<___;
126$D#md5# mov $ctx,%r11 # reassign arguments
127 mov $len,%r12
128 mov $in0,%r13
129 mov $out,%r14
130$D#md5# mov $inp,%r15
131___
132 $ctx="%r11" if ($md5); # reassign arguments
133 $len="%r12";
134 $in0="%r13";
135 $out="%r14";
136 $inp="%r15" if ($md5);
137 $inp=$in0 if (!$md5);
138$code.=<<___;
139 xor $XX[0],$XX[0]
140 xor $YY,$YY
141
142 lea 8($dat),$dat
143 mov -8($dat),$XX[0]#b
144 mov -4($dat),$YY#b
145
146 inc $XX[0]#b
147 sub $in0,$out
148 movl ($dat,$XX[0],4),$TX[0]#d
149___
150$code.=<<___ if (!$md5);
151 xor $TX[1],$TX[1]
152 test \$-128,$len
153 jz .Loop1
154 sub $XX[0],$TX[1]
155 and \$`$MOD-1`,$TX[1]
156 jz .Loop${MOD}_is_hot
157 sub $TX[1],$len
158.Loop${MOD}_warmup:
159 add $TX[0]#b,$YY#b
160 movl ($dat,$YY,4),$TY#d
161 movl $TX[0]#d,($dat,$YY,4)
162 movl $TY#d,($dat,$XX[0],4)
163 add $TY#b,$TX[0]#b
164 inc $XX[0]#b
165 movl ($dat,$TX[0],4),$TY#d
166 movl ($dat,$XX[0],4),$TX[0]#d
167 xorb ($in0),$TY#b
168 movb $TY#b,($out,$in0)
169 lea 1($in0),$in0
170 dec $TX[1]
171 jnz .Loop${MOD}_warmup
172
173 mov $YY,$TX[1]
174 xor $YY,$YY
175 mov $TX[1]#b,$YY#b
176
177.Loop${MOD}_is_hot:
178 mov $len,32(%rsp) # save original $len
179 shr \$6,$len # number of 64-byte blocks
180___
181 if ($D && !$md5) { # stitch in dummy MD5
182 $md5=1;
183 $ctx="%r11";
184 $inp="%r15";
185 $code.=<<___;
186 mov %rsp,$ctx
187 mov $in0,$inp
188___
189 }
190}
191$code.=<<___;
192#rc4# add $TX[0]#b,$YY#b
193#rc4# lea ($dat,$XX[0],4),$XX[1]
194 shl \$6,$len
195 add $inp,$len # pointer to the end of input
196 mov $len,16(%rsp)
197
198#md5# mov $ctx,24(%rsp) # save pointer to MD5_CTX
199#md5# mov 0*4($ctx),$V[0] # load current hash value from MD5_CTX
200#md5# mov 1*4($ctx),$V[1]
201#md5# mov 2*4($ctx),$V[2]
202#md5# mov 3*4($ctx),$V[3]
203 jmp .Loop
204
205.align 16
206.Loop:
207#md5# mov $V[0],0*4(%rsp) # put aside current hash value
208#md5# mov $V[1],1*4(%rsp)
209#md5# mov $V[2],2*4(%rsp)
210#md5# mov $V[3],$tmp # forward reference
211#md5# mov $V[3],3*4(%rsp)
212___
213
214sub R0 {
215 my ($i,$a,$b,$c,$d)=@_;
216 my @rot0=(7,12,17,22);
217 my $j=$i%16;
218 my $k=$i%$MOD;
219 my $xmm="%xmm".($j&1);
220 $code.=" movdqu ($in0),%xmm2\n" if ($rc4 && $j==15);
221 $code.=" add \$$MOD,$XX[0]#b\n" if ($rc4 && $j==15 && $k==$MOD-1);
222 $code.=" pxor $xmm,$xmm\n" if ($rc4 && $j<=1);
223 $code.=<<___;
224#rc4# movl ($dat,$YY,4),$TY#d
225#md5# xor $c,$tmp
226#rc4# movl $TX[0]#d,($dat,$YY,4)
227#md5# and $b,$tmp
228#md5# add 4*`$j`($inp),$a
229#rc4# add $TY#b,$TX[0]#b
230#rc4# movl `4*(($k+1)%$MOD)`(`$k==$MOD-1?"$dat,$XX[0],4":"$XX[1]"`),$TX[1]#d
231#md5# add \$$K[$i],$a
232#md5# xor $d,$tmp
233#rc4# movz $TX[0]#b,$TX[0]#d
234#rc4# movl $TY#d,4*$k($XX[1])
235#md5# add $tmp,$a
236#rc4# add $TX[1]#b,$YY#b
237#md5# rol \$$rot0[$j%4],$a
238#md5# mov `$j==15?"$b":"$c"`,$tmp # forward reference
239#rc4# pinsrw \$`($j>>1)&7`,($dat,$TX[0],4),$xmm\n
240#md5# add $b,$a
241___
242 $code.=<<___ if ($rc4 && $j==15 && $k==$MOD-1);
243 mov $YY,$XX[1]
244 xor $YY,$YY # keyword to partial register
245 mov $XX[1]#b,$YY#b
246 lea ($dat,$XX[0],4),$XX[1]
247___
248 $code.=<<___ if ($rc4 && $j==15);
249 psllq \$8,%xmm1
250 pxor %xmm0,%xmm2
251 pxor %xmm1,%xmm2
252___
253}
254sub R1 {
255 my ($i,$a,$b,$c,$d)=@_;
256 my @rot1=(5,9,14,20);
257 my $j=$i%16;
258 my $k=$i%$MOD;
259 my $xmm="%xmm".($j&1);
260 $code.=" movdqu 16($in0),%xmm3\n" if ($rc4 && $j==15);
261 $code.=" add \$$MOD,$XX[0]#b\n" if ($rc4 && $j==15 && $k==$MOD-1);
262 $code.=" pxor $xmm,$xmm\n" if ($rc4 && $j<=1);
263 $code.=<<___;
264#rc4# movl ($dat,$YY,4),$TY#d
265#md5# xor $b,$tmp
266#rc4# movl $TX[0]#d,($dat,$YY,4)
267#md5# and $d,$tmp
268#md5# add 4*`((1+5*$j)%16)`($inp),$a
269#rc4# add $TY#b,$TX[0]#b
270#rc4# movl `4*(($k+1)%$MOD)`(`$k==$MOD-1?"$dat,$XX[0],4":"$XX[1]"`),$TX[1]#d
271#md5# add \$$K[$i],$a
272#md5# xor $c,$tmp
273#rc4# movz $TX[0]#b,$TX[0]#d
274#rc4# movl $TY#d,4*$k($XX[1])
275#md5# add $tmp,$a
276#rc4# add $TX[1]#b,$YY#b
277#md5# rol \$$rot1[$j%4],$a
278#md5# mov `$j==15?"$c":"$b"`,$tmp # forward reference
279#rc4# pinsrw \$`($j>>1)&7`,($dat,$TX[0],4),$xmm\n
280#md5# add $b,$a
281___
282 $code.=<<___ if ($rc4 && $j==15 && $k==$MOD-1);
283 mov $YY,$XX[1]
284 xor $YY,$YY # keyword to partial register
285 mov $XX[1]#b,$YY#b
286 lea ($dat,$XX[0],4),$XX[1]
287___
288 $code.=<<___ if ($rc4 && $j==15);
289 psllq \$8,%xmm1
290 pxor %xmm0,%xmm3
291 pxor %xmm1,%xmm3
292___
293}
294sub R2 {
295 my ($i,$a,$b,$c,$d)=@_;
296 my @rot2=(4,11,16,23);
297 my $j=$i%16;
298 my $k=$i%$MOD;
299 my $xmm="%xmm".($j&1);
300 $code.=" movdqu 32($in0),%xmm4\n" if ($rc4 && $j==15);
301 $code.=" add \$$MOD,$XX[0]#b\n" if ($rc4 && $j==15 && $k==$MOD-1);
302 $code.=" pxor $xmm,$xmm\n" if ($rc4 && $j<=1);
303 $code.=<<___;
304#rc4# movl ($dat,$YY,4),$TY#d
305#md5# xor $c,$tmp
306#rc4# movl $TX[0]#d,($dat,$YY,4)
307#md5# xor $b,$tmp
308#md5# add 4*`((5+3*$j)%16)`($inp),$a
309#rc4# add $TY#b,$TX[0]#b
310#rc4# movl `4*(($k+1)%$MOD)`(`$k==$MOD-1?"$dat,$XX[0],4":"$XX[1]"`),$TX[1]#d
311#md5# add \$$K[$i],$a
312#rc4# movz $TX[0]#b,$TX[0]#d
313#md5# add $tmp,$a
314#rc4# movl $TY#d,4*$k($XX[1])
315#rc4# add $TX[1]#b,$YY#b
316#md5# rol \$$rot2[$j%4],$a
317#md5# mov `$j==15?"\\\$-1":"$c"`,$tmp # forward reference
318#rc4# pinsrw \$`($j>>1)&7`,($dat,$TX[0],4),$xmm\n
319#md5# add $b,$a
320___
321 $code.=<<___ if ($rc4 && $j==15 && $k==$MOD-1);
322 mov $YY,$XX[1]
323 xor $YY,$YY # keyword to partial register
324 mov $XX[1]#b,$YY#b
325 lea ($dat,$XX[0],4),$XX[1]
326___
327 $code.=<<___ if ($rc4 && $j==15);
328 psllq \$8,%xmm1
329 pxor %xmm0,%xmm4
330 pxor %xmm1,%xmm4
331___
332}
333sub R3 {
334 my ($i,$a,$b,$c,$d)=@_;
335 my @rot3=(6,10,15,21);
336 my $j=$i%16;
337 my $k=$i%$MOD;
338 my $xmm="%xmm".($j&1);
339 $code.=" movdqu 48($in0),%xmm5\n" if ($rc4 && $j==15);
340 $code.=" add \$$MOD,$XX[0]#b\n" if ($rc4 && $j==15 && $k==$MOD-1);
341 $code.=" pxor $xmm,$xmm\n" if ($rc4 && $j<=1);
342 $code.=<<___;
343#rc4# movl ($dat,$YY,4),$TY#d
344#md5# xor $d,$tmp
345#rc4# movl $TX[0]#d,($dat,$YY,4)
346#md5# or $b,$tmp
347#md5# add 4*`((7*$j)%16)`($inp),$a
348#rc4# add $TY#b,$TX[0]#b
349#rc4# movl `4*(($k+1)%$MOD)`(`$k==$MOD-1?"$dat,$XX[0],4":"$XX[1]"`),$TX[1]#d
350#md5# add \$$K[$i],$a
351#rc4# movz $TX[0]#b,$TX[0]#d
352#md5# xor $c,$tmp
353#rc4# movl $TY#d,4*$k($XX[1])
354#md5# add $tmp,$a
355#rc4# add $TX[1]#b,$YY#b
356#md5# rol \$$rot3[$j%4],$a
357#md5# mov \$-1,$tmp # forward reference
358#rc4# pinsrw \$`($j>>1)&7`,($dat,$TX[0],4),$xmm\n
359#md5# add $b,$a
360___
361 $code.=<<___ if ($rc4 && $j==15);
362 mov $XX[0],$XX[1]
363 xor $XX[0],$XX[0] # keyword to partial register
364 mov $XX[1]#b,$XX[0]#b
365 mov $YY,$XX[1]
366 xor $YY,$YY # keyword to partial register
367 mov $XX[1]#b,$YY#b
368 lea ($dat,$XX[0],4),$XX[1]
369 psllq \$8,%xmm1
370 pxor %xmm0,%xmm5
371 pxor %xmm1,%xmm5
372___
373}
374
375my $i=0;
376for(;$i<16;$i++) { R0($i,@V); unshift(@V,pop(@V)); push(@TX,shift(@TX)); }
377for(;$i<32;$i++) { R1($i,@V); unshift(@V,pop(@V)); push(@TX,shift(@TX)); }
378for(;$i<48;$i++) { R2($i,@V); unshift(@V,pop(@V)); push(@TX,shift(@TX)); }
379for(;$i<64;$i++) { R3($i,@V); unshift(@V,pop(@V)); push(@TX,shift(@TX)); }
380
381$code.=<<___;
382#md5# add 0*4(%rsp),$V[0] # accumulate hash value
383#md5# add 1*4(%rsp),$V[1]
384#md5# add 2*4(%rsp),$V[2]
385#md5# add 3*4(%rsp),$V[3]
386
387#rc4# movdqu %xmm2,($out,$in0) # write RC4 output
388#rc4# movdqu %xmm3,16($out,$in0)
389#rc4# movdqu %xmm4,32($out,$in0)
390#rc4# movdqu %xmm5,48($out,$in0)
391#md5# lea 64($inp),$inp
392#rc4# lea 64($in0),$in0
393 cmp 16(%rsp),$inp # are we done?
394 jb .Loop
395
396#md5# mov 24(%rsp),$len # restore pointer to MD5_CTX
397#rc4# sub $TX[0]#b,$YY#b # correct $YY
398#md5# mov $V[0],0*4($len) # write MD5_CTX
399#md5# mov $V[1],1*4($len)
400#md5# mov $V[2],2*4($len)
401#md5# mov $V[3],3*4($len)
402___
403$code.=<<___ if ($rc4 && (!$md5 || $D));
404 mov 32(%rsp),$len # restore original $len
405 and \$63,$len # remaining bytes
406 jnz .Loop1
407 jmp .Ldone
408
409.align 16
410.Loop1:
411 add $TX[0]#b,$YY#b
412 movl ($dat,$YY,4),$TY#d
413 movl $TX[0]#d,($dat,$YY,4)
414 movl $TY#d,($dat,$XX[0],4)
415 add $TY#b,$TX[0]#b
416 inc $XX[0]#b
417 movl ($dat,$TX[0],4),$TY#d
418 movl ($dat,$XX[0],4),$TX[0]#d
419 xorb ($in0),$TY#b
420 movb $TY#b,($out,$in0)
421 lea 1($in0),$in0
422 dec $len
423 jnz .Loop1
424
425.Ldone:
426___
427$code.=<<___;
428#rc4# sub \$1,$XX[0]#b
429#rc4# movl $XX[0]#d,-8($dat)
430#rc4# movl $YY#d,-4($dat)
431
432 mov 40(%rsp),%r15
433 mov 48(%rsp),%r14
434 mov 56(%rsp),%r13
435 mov 64(%rsp),%r12
436 mov 72(%rsp),%rbp
437 mov 80(%rsp),%rbx
438 lea 88(%rsp),%rsp
439.Lepilogue:
440.Labort:
441 ret
442.size $func,.-$func
443___
444
445if ($rc4 && $D) { # sole purpose of this section is to provide
446 # option to use the generated module as drop-in
447 # replacement for rc4-x86_64.pl for debugging
448 # and testing purposes...
449my ($idx,$ido)=("%r8","%r9");
450my ($dat,$len,$inp)=("%rdi","%rsi","%rdx");
451
452$code.=<<___;
453.globl RC4_set_key
454.type RC4_set_key,\@function,3
455.align 16
456RC4_set_key:
457 _CET_ENDBR
458 lea 8($dat),$dat
459 lea ($inp,$len),$inp
460 neg $len
461 mov $len,%rcx
462 xor %eax,%eax
463 xor $ido,$ido
464 xor %r10,%r10
465 xor %r11,%r11
466 jmp .Lw1stloop
467
468.align 16
469.Lw1stloop:
470 mov %eax,($dat,%rax,4)
471 add \$1,%al
472 jnc .Lw1stloop
473
474 xor $ido,$ido
475 xor $idx,$idx
476.align 16
477.Lw2ndloop:
478 mov ($dat,$ido,4),%r10d
479 add ($inp,$len,1),$idx#b
480 add %r10b,$idx#b
481 add \$1,$len
482 mov ($dat,$idx,4),%r11d
483 cmovz %rcx,$len
484 mov %r10d,($dat,$idx,4)
485 mov %r11d,($dat,$ido,4)
486 add \$1,$ido#b
487 jnc .Lw2ndloop
488
489 xor %eax,%eax
490 mov %eax,-8($dat)
491 mov %eax,-4($dat)
492 ret
493.size RC4_set_key,.-RC4_set_key
494___
495}
496
497sub reg_part {
498my ($reg,$conv)=@_;
499 if ($reg =~ /%r[0-9]+/) { $reg .= $conv; }
500 elsif ($conv eq "b") { $reg =~ s/%[er]([^x]+)x?/%$1l/; }
501 elsif ($conv eq "w") { $reg =~ s/%[er](.+)/%$1/; }
502 elsif ($conv eq "d") { $reg =~ s/%[er](.+)/%e$1/; }
503 return $reg;
504}
505
506$code =~ s/(%[a-z0-9]+)#([bwd])/reg_part($1,$2)/gem;
507$code =~ s/\`([^\`]*)\`/eval $1/gem;
508$code =~ s/pinsrw\s+\$0,/movd /gm;
509
510$code =~ s/#md5#//gm if ($md5);
511$code =~ s/#rc4#//gm if ($rc4);
512
513print $code;
514
515close STDOUT;
generated by cgit v1.2.3-55-g6feb (git 2.39.1) at 2025-04-26 12:19:41 +0000