summaryrefslogtreecommitdiff
path: root/src/lib/libcrypto/rc4/asm/rc4-md5-x86_64.pl
diff options
context:
space:
mode:
Diffstat (limited to 'src/lib/libcrypto/rc4/asm/rc4-md5-x86_64.pl')
-rw-r--r--src/lib/libcrypto/rc4/asm/rc4-md5-x86_64.pl525
1 files changed, 0 insertions, 525 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 501d9e936b..0000000000
--- a/src/lib/libcrypto/rc4/asm/rc4-md5-x86_64.pl
+++ /dev/null
@@ -1,525 +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$func:
112 cmp \$0,$len
113 je .Labort
114 push %rbx
115 push %rbp
116 push %r12
117 push %r13
118 push %r14
119 push %r15
120 sub \$40,%rsp
121.Lbody:
122___
123if ($rc4) {
124$code.=<<___;
125$D#md5# mov $ctx,%r11 # reassign arguments
126 mov $len,%r12
127 mov $in0,%r13
128 mov $out,%r14
129$D#md5# mov $inp,%r15
130___
131 $ctx="%r11" if ($md5); # reassign arguments
132 $len="%r12";
133 $in0="%r13";
134 $out="%r14";
135 $inp="%r15" if ($md5);
136 $inp=$in0 if (!$md5);
137$code.=<<___;
138 xor $XX[0],$XX[0]
139 xor $YY,$YY
140
141 lea 8($dat),$dat
142 mov -8($dat),$XX[0]#b
143 mov -4($dat),$YY#b
144
145 inc $XX[0]#b
146 sub $in0,$out
147 movl ($dat,$XX[0],4),$TX[0]#d
148___
149$code.=<<___ if (!$md5);
150 xor $TX[1],$TX[1]
151 test \$-128,$len
152 jz .Loop1
153 sub $XX[0],$TX[1]
154 and \$`$MOD-1`,$TX[1]
155 jz .Loop${MOD}_is_hot
156 sub $TX[1],$len
157.Loop${MOD}_warmup:
158 add $TX[0]#b,$YY#b
159 movl ($dat,$YY,4),$TY#d
160 movl $TX[0]#d,($dat,$YY,4)
161 movl $TY#d,($dat,$XX[0],4)
162 add $TY#b,$TX[0]#b
163 inc $XX[0]#b
164 movl ($dat,$TX[0],4),$TY#d
165 movl ($dat,$XX[0],4),$TX[0]#d
166 xorb ($in0),$TY#b
167 movb $TY#b,($out,$in0)
168 lea 1($in0),$in0
169 dec $TX[1]
170 jnz .Loop${MOD}_warmup
171
172 mov $YY,$TX[1]
173 xor $YY,$YY
174 mov $TX[1]#b,$YY#b
175
176.Loop${MOD}_is_hot:
177 mov $len,32(%rsp) # save original $len
178 shr \$6,$len # number of 64-byte blocks
179___
180 if ($D && !$md5) { # stitch in dummy MD5
181 $md5=1;
182 $ctx="%r11";
183 $inp="%r15";
184 $code.=<<___;
185 mov %rsp,$ctx
186 mov $in0,$inp
187___
188 }
189}
190$code.=<<___;
191#rc4# add $TX[0]#b,$YY#b
192#rc4# lea ($dat,$XX[0],4),$XX[1]
193 shl \$6,$len
194 add $inp,$len # pointer to the end of input
195 mov $len,16(%rsp)
196
197#md5# mov $ctx,24(%rsp) # save pointer to MD5_CTX
198#md5# mov 0*4($ctx),$V[0] # load current hash value from MD5_CTX
199#md5# mov 1*4($ctx),$V[1]
200#md5# mov 2*4($ctx),$V[2]
201#md5# mov 3*4($ctx),$V[3]
202 jmp .Loop
203
204.align 16
205.Loop:
206#md5# mov $V[0],0*4(%rsp) # put aside current hash value
207#md5# mov $V[1],1*4(%rsp)
208#md5# mov $V[2],2*4(%rsp)
209#md5# mov $V[3],$tmp # forward reference
210#md5# mov $V[3],3*4(%rsp)
211___
212
213sub R0 {
214 my ($i,$a,$b,$c,$d)=@_;
215 my @rot0=(7,12,17,22);
216 my $j=$i%16;
217 my $k=$i%$MOD;
218 my $xmm="%xmm".($j&1);
219 $code.=" movdqu ($in0),%xmm2\n" if ($rc4 && $j==15);
220 $code.=" add \$$MOD,$XX[0]#b\n" if ($rc4 && $j==15 && $k==$MOD-1);
221 $code.=" pxor $xmm,$xmm\n" if ($rc4 && $j<=1);
222 $code.=<<___;
223#rc4# movl ($dat,$YY,4),$TY#d
224#md5# xor $c,$tmp
225#rc4# movl $TX[0]#d,($dat,$YY,4)
226#md5# and $b,$tmp
227#md5# add 4*`$j`($inp),$a
228#rc4# add $TY#b,$TX[0]#b
229#rc4# movl `4*(($k+1)%$MOD)`(`$k==$MOD-1?"$dat,$XX[0],4":"$XX[1]"`),$TX[1]#d
230#md5# add \$$K[$i],$a
231#md5# xor $d,$tmp
232#rc4# movz $TX[0]#b,$TX[0]#d
233#rc4# movl $TY#d,4*$k($XX[1])
234#md5# add $tmp,$a
235#rc4# add $TX[1]#b,$YY#b
236#md5# rol \$$rot0[$j%4],$a
237#md5# mov `$j==15?"$b":"$c"`,$tmp # forward reference
238#rc4# pinsrw \$`($j>>1)&7`,($dat,$TX[0],4),$xmm\n
239#md5# add $b,$a
240___
241 $code.=<<___ if ($rc4 && $j==15 && $k==$MOD-1);
242 mov $YY,$XX[1]
243 xor $YY,$YY # keyword to partial register
244 mov $XX[1]#b,$YY#b
245 lea ($dat,$XX[0],4),$XX[1]
246___
247 $code.=<<___ if ($rc4 && $j==15);
248 psllq \$8,%xmm1
249 pxor %xmm0,%xmm2
250 pxor %xmm1,%xmm2
251___
252}
253sub R1 {
254 my ($i,$a,$b,$c,$d)=@_;
255 my @rot1=(5,9,14,20);
256 my $j=$i%16;
257 my $k=$i%$MOD;
258 my $xmm="%xmm".($j&1);
259 $code.=" movdqu 16($in0),%xmm3\n" if ($rc4 && $j==15);
260 $code.=" add \$$MOD,$XX[0]#b\n" if ($rc4 && $j==15 && $k==$MOD-1);
261 $code.=" pxor $xmm,$xmm\n" if ($rc4 && $j<=1);
262 $code.=<<___;
263#rc4# movl ($dat,$YY,4),$TY#d
264#md5# xor $b,$tmp
265#rc4# movl $TX[0]#d,($dat,$YY,4)
266#md5# and $d,$tmp
267#md5# add 4*`((1+5*$j)%16)`($inp),$a
268#rc4# add $TY#b,$TX[0]#b
269#rc4# movl `4*(($k+1)%$MOD)`(`$k==$MOD-1?"$dat,$XX[0],4":"$XX[1]"`),$TX[1]#d
270#md5# add \$$K[$i],$a
271#md5# xor $c,$tmp
272#rc4# movz $TX[0]#b,$TX[0]#d
273#rc4# movl $TY#d,4*$k($XX[1])
274#md5# add $tmp,$a
275#rc4# add $TX[1]#b,$YY#b
276#md5# rol \$$rot1[$j%4],$a
277#md5# mov `$j==15?"$c":"$b"`,$tmp # forward reference
278#rc4# pinsrw \$`($j>>1)&7`,($dat,$TX[0],4),$xmm\n
279#md5# add $b,$a
280___
281 $code.=<<___ if ($rc4 && $j==15 && $k==$MOD-1);
282 mov $YY,$XX[1]
283 xor $YY,$YY # keyword to partial register
284 mov $XX[1]#b,$YY#b
285 lea ($dat,$XX[0],4),$XX[1]
286___
287 $code.=<<___ if ($rc4 && $j==15);
288 psllq \$8,%xmm1
289 pxor %xmm0,%xmm3
290 pxor %xmm1,%xmm3
291___
292}
293sub R2 {
294 my ($i,$a,$b,$c,$d)=@_;
295 my @rot2=(4,11,16,23);
296 my $j=$i%16;
297 my $k=$i%$MOD;
298 my $xmm="%xmm".($j&1);
299 $code.=" movdqu 32($in0),%xmm4\n" if ($rc4 && $j==15);
300 $code.=" add \$$MOD,$XX[0]#b\n" if ($rc4 && $j==15 && $k==$MOD-1);
301 $code.=" pxor $xmm,$xmm\n" if ($rc4 && $j<=1);
302 $code.=<<___;
303#rc4# movl ($dat,$YY,4),$TY#d
304#md5# xor $c,$tmp
305#rc4# movl $TX[0]#d,($dat,$YY,4)
306#md5# xor $b,$tmp
307#md5# add 4*`((5+3*$j)%16)`($inp),$a
308#rc4# add $TY#b,$TX[0]#b
309#rc4# movl `4*(($k+1)%$MOD)`(`$k==$MOD-1?"$dat,$XX[0],4":"$XX[1]"`),$TX[1]#d
310#md5# add \$$K[$i],$a
311#rc4# movz $TX[0]#b,$TX[0]#d
312#md5# add $tmp,$a
313#rc4# movl $TY#d,4*$k($XX[1])
314#rc4# add $TX[1]#b,$YY#b
315#md5# rol \$$rot2[$j%4],$a
316#md5# mov `$j==15?"\\\$-1":"$c"`,$tmp # forward reference
317#rc4# pinsrw \$`($j>>1)&7`,($dat,$TX[0],4),$xmm\n
318#md5# add $b,$a
319___
320 $code.=<<___ if ($rc4 && $j==15 && $k==$MOD-1);
321 mov $YY,$XX[1]
322 xor $YY,$YY # keyword to partial register
323 mov $XX[1]#b,$YY#b
324 lea ($dat,$XX[0],4),$XX[1]
325___
326 $code.=<<___ if ($rc4 && $j==15);
327 psllq \$8,%xmm1
328 pxor %xmm0,%xmm4
329 pxor %xmm1,%xmm4
330___
331}
332sub R3 {
333 my ($i,$a,$b,$c,$d)=@_;
334 my @rot3=(6,10,15,21);
335 my $j=$i%16;
336 my $k=$i%$MOD;
337 my $xmm="%xmm".($j&1);
338 $code.=" movdqu 48($in0),%xmm5\n" if ($rc4 && $j==15);
339 $code.=" add \$$MOD,$XX[0]#b\n" if ($rc4 && $j==15 && $k==$MOD-1);
340 $code.=" pxor $xmm,$xmm\n" if ($rc4 && $j<=1);
341 $code.=<<___;
342#rc4# movl ($dat,$YY,4),$TY#d
343#md5# xor $d,$tmp
344#rc4# movl $TX[0]#d,($dat,$YY,4)
345#md5# or $b,$tmp
346#md5# add 4*`((7*$j)%16)`($inp),$a
347#rc4# add $TY#b,$TX[0]#b
348#rc4# movl `4*(($k+1)%$MOD)`(`$k==$MOD-1?"$dat,$XX[0],4":"$XX[1]"`),$TX[1]#d
349#md5# add \$$K[$i],$a
350#rc4# movz $TX[0]#b,$TX[0]#d
351#md5# xor $c,$tmp
352#rc4# movl $TY#d,4*$k($XX[1])
353#md5# add $tmp,$a
354#rc4# add $TX[1]#b,$YY#b
355#md5# rol \$$rot3[$j%4],$a
356#md5# mov \$-1,$tmp # forward reference
357#rc4# pinsrw \$`($j>>1)&7`,($dat,$TX[0],4),$xmm\n
358#md5# add $b,$a
359___
360 $code.=<<___ if ($rc4 && $j==15);
361 mov $XX[0],$XX[1]
362 xor $XX[0],$XX[0] # keyword to partial register
363 mov $XX[1]#b,$XX[0]#b
364 mov $YY,$XX[1]
365 xor $YY,$YY # keyword to partial register
366 mov $XX[1]#b,$YY#b
367 lea ($dat,$XX[0],4),$XX[1]
368 psllq \$8,%xmm1
369 pxor %xmm0,%xmm5
370 pxor %xmm1,%xmm5
371___
372}
373
374my $i=0;
375for(;$i<16;$i++) { R0($i,@V); unshift(@V,pop(@V)); push(@TX,shift(@TX)); }
376for(;$i<32;$i++) { R1($i,@V); unshift(@V,pop(@V)); push(@TX,shift(@TX)); }
377for(;$i<48;$i++) { R2($i,@V); unshift(@V,pop(@V)); push(@TX,shift(@TX)); }
378for(;$i<64;$i++) { R3($i,@V); unshift(@V,pop(@V)); push(@TX,shift(@TX)); }
379
380$code.=<<___;
381#md5# add 0*4(%rsp),$V[0] # accumulate hash value
382#md5# add 1*4(%rsp),$V[1]
383#md5# add 2*4(%rsp),$V[2]
384#md5# add 3*4(%rsp),$V[3]
385
386#rc4# movdqu %xmm2,($out,$in0) # write RC4 output
387#rc4# movdqu %xmm3,16($out,$in0)
388#rc4# movdqu %xmm4,32($out,$in0)
389#rc4# movdqu %xmm5,48($out,$in0)
390#md5# lea 64($inp),$inp
391#rc4# lea 64($in0),$in0
392 cmp 16(%rsp),$inp # are we done?
393 jb .Loop
394
395#md5# mov 24(%rsp),$len # restore pointer to MD5_CTX
396#rc4# sub $TX[0]#b,$YY#b # correct $YY
397#md5# mov $V[0],0*4($len) # write MD5_CTX
398#md5# mov $V[1],1*4($len)
399#md5# mov $V[2],2*4($len)
400#md5# mov $V[3],3*4($len)
401___
402$code.=<<___ if ($rc4 && (!$md5 || $D));
403 mov 32(%rsp),$len # restore original $len
404 and \$63,$len # remaining bytes
405 jnz .Loop1
406 jmp .Ldone
407
408.align 16
409.Loop1:
410 add $TX[0]#b,$YY#b
411 movl ($dat,$YY,4),$TY#d
412 movl $TX[0]#d,($dat,$YY,4)
413 movl $TY#d,($dat,$XX[0],4)
414 add $TY#b,$TX[0]#b
415 inc $XX[0]#b
416 movl ($dat,$TX[0],4),$TY#d
417 movl ($dat,$XX[0],4),$TX[0]#d
418 xorb ($in0),$TY#b
419 movb $TY#b,($out,$in0)
420 lea 1($in0),$in0
421 dec $len
422 jnz .Loop1
423
424.Ldone:
425___
426$code.=<<___;
427#rc4# sub \$1,$XX[0]#b
428#rc4# movl $XX[0]#d,-8($dat)
429#rc4# movl $YY#d,-4($dat)
430
431 mov 40(%rsp),%r15
432 mov 48(%rsp),%r14
433 mov 56(%rsp),%r13
434 mov 64(%rsp),%r12
435 mov 72(%rsp),%rbp
436 mov 80(%rsp),%rbx
437 lea 88(%rsp),%rsp
438.Lepilogue:
439.Labort:
440 ret
441.size $func,.-$func
442___
443
444if ($rc4 && $D) { # sole purpose of this section is to provide
445 # option to use the generated module as drop-in
446 # replacement for rc4-x86_64.pl for debugging
447 # and testing purposes...
448my ($idx,$ido)=("%r8","%r9");
449my ($dat,$len,$inp)=("%rdi","%rsi","%rdx");
450
451$code.=<<___;
452.globl RC4_set_key
453.type RC4_set_key,\@function,3
454.align 16
455RC4_set_key:
456 lea 8($dat),$dat
457 lea ($inp,$len),$inp
458 neg $len
459 mov $len,%rcx
460 xor %eax,%eax
461 xor $ido,$ido
462 xor %r10,%r10
463 xor %r11,%r11
464 jmp .Lw1stloop
465
466.align 16
467.Lw1stloop:
468 mov %eax,($dat,%rax,4)
469 add \$1,%al
470 jnc .Lw1stloop
471
472 xor $ido,$ido
473 xor $idx,$idx
474.align 16
475.Lw2ndloop:
476 mov ($dat,$ido,4),%r10d
477 add ($inp,$len,1),$idx#b
478 add %r10b,$idx#b
479 add \$1,$len
480 mov ($dat,$idx,4),%r11d
481 cmovz %rcx,$len
482 mov %r10d,($dat,$idx,4)
483 mov %r11d,($dat,$ido,4)
484 add \$1,$ido#b
485 jnc .Lw2ndloop
486
487 xor %eax,%eax
488 mov %eax,-8($dat)
489 mov %eax,-4($dat)
490 ret
491.size RC4_set_key,.-RC4_set_key
492
493.globl RC4_options
494.type RC4_options,\@abi-omnipotent
495.align 16
496RC4_options:
497 lea .Lopts(%rip),%rax
498 ret
499.align 64
500.Lopts:
501.asciz "rc4(64x,int)"
502.align 64
503.size RC4_options,.-RC4_options
504___
505}
506
507sub reg_part {
508my ($reg,$conv)=@_;
509 if ($reg =~ /%r[0-9]+/) { $reg .= $conv; }
510 elsif ($conv eq "b") { $reg =~ s/%[er]([^x]+)x?/%$1l/; }
511 elsif ($conv eq "w") { $reg =~ s/%[er](.+)/%$1/; }
512 elsif ($conv eq "d") { $reg =~ s/%[er](.+)/%e$1/; }
513 return $reg;
514}
515
516$code =~ s/(%[a-z0-9]+)#([bwd])/reg_part($1,$2)/gem;
517$code =~ s/\`([^\`]*)\`/eval $1/gem;
518$code =~ s/pinsrw\s+\$0,/movd /gm;
519
520$code =~ s/#md5#//gm if ($md5);
521$code =~ s/#rc4#//gm if ($rc4);
522
523print $code;
524
525close STDOUT;
generated by cgit v1.2.3-55-g6feb (git 2.39.1) at 2025-08-03 23:26:58 +0000