2 # Copyright 2004-2016 The OpenSSL Project Authors. All Rights Reserved.
4 # Licensed under the OpenSSL license (the "License"). You may not use
5 # this file except in compliance with the License. You can obtain a copy
6 # in the file LICENSE in the source distribution or at
7 # https://www.openssl.org/source/license.html
10 # ====================================================================
11 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
12 # project. The module is, however, dual licensed under OpenSSL and
13 # CRYPTOGAMS licenses depending on where you obtain it. For further
14 # details see http://www.openssl.org/~appro/cryptogams/.
15 # ====================================================================
19 # You might fail to appreciate this module performance from the first
20 # try. If compared to "vanilla" linux-ia32-icc target, i.e. considered
21 # to be *the* best Intel C compiler without -KPIC, performance appears
22 # to be virtually identical... But try to re-configure with shared
23 # library support... Aha! Intel compiler "suddenly" lags behind by 30%
24 # [on P4, more on others]:-) And if compared to position-independent
25 # code generated by GNU C, this code performs *more* than *twice* as
26 # fast! Yes, all this buzz about PIC means that unlike other hand-
27 # coded implementations, this one was explicitly designed to be safe
28 # to use even in shared library context... This also means that this
29 # code isn't necessarily absolutely fastest "ever," because in order
30 # to achieve position independence an extra register has to be
31 # off-loaded to stack, which affects the benchmark result.
33 # Special note about instruction choice. Do you recall RC4_INT code
34 # performing poorly on P4? It might be the time to figure out why.
35 # RC4_INT code implies effective address calculations in base+offset*4
36 # form. Trouble is that it seems that offset scaling turned to be
37 # critical path... At least eliminating scaling resulted in 2.8x RC4
38 # performance improvement [as you might recall]. As AES code is hungry
39 # for scaling too, I [try to] avoid the latter by favoring off-by-2
40 # shifts and masking the result with 0xFF<<2 instead of "boring" 0xFF.
42 # As was shown by Dean Gaudet, the above note turned out to be
43 # void. Performance improvement with off-by-2 shifts was observed on
44 # intermediate implementation, which was spilling yet another register
45 # to stack... Final offset*4 code below runs just a tad faster on P4,
46 # but exhibits up to 10% improvement on other cores.
48 # Second version is "monolithic" replacement for aes_core.c, which in
49 # addition to AES_[de|en]crypt implements AES_set_[de|en]cryption_key.
50 # This made it possible to implement little-endian variant of the
51 # algorithm without modifying the base C code. Motivating factor for
52 # the undertaken effort was that it appeared that in tight IA-32
53 # register window little-endian flavor could achieve slightly higher
54 # Instruction Level Parallelism, and it indeed resulted in up to 15%
55 # better performance on most recent ยต-archs...
57 # Third version adds AES_cbc_encrypt implementation, which resulted in
58 # up to 40% performance imrovement of CBC benchmark results. 40% was
59 # observed on P4 core, where "overall" imrovement coefficient, i.e. if
60 # compared to PIC generated by GCC and in CBC mode, was observed to be
61 # as large as 4x:-) CBC performance is virtually identical to ECB now
62 # and on some platforms even better, e.g. 17.6 "small" cycles/byte on
63 # Opteron, because certain function prologues and epilogues are
64 # effectively taken out of the loop...
66 # Version 3.2 implements compressed tables and prefetch of these tables
67 # in CBC[!] mode. Former means that 3/4 of table references are now
68 # misaligned, which unfortunately has negative impact on elder IA-32
69 # implementations, Pentium suffered 30% penalty, PIII - 10%.
71 # Version 3.3 avoids L1 cache aliasing between stack frame and
72 # S-boxes, and 3.4 - L1 cache aliasing even between key schedule. The
73 # latter is achieved by copying the key schedule to controlled place in
74 # stack. This unfortunately has rather strong impact on small block CBC
75 # performance, ~2x deterioration on 16-byte block if compared to 3.3.
77 # Version 3.5 checks if there is L1 cache aliasing between user-supplied
78 # key schedule and S-boxes and abstains from copying the former if
79 # there is no. This allows end-user to consciously retain small block
80 # performance by aligning key schedule in specific manner.
82 # Version 3.6 compresses Td4 to 256 bytes and prefetches it in ECB.
84 # Current ECB performance numbers for 128-bit key in CPU cycles per
85 # processed byte [measure commonly used by AES benchmarkers] are:
87 # small footprint fully unrolled
93 # Version 3.7 reimplements outer rounds as "compact." Meaning that
94 # first and last rounds reference compact 256 bytes S-box. This means
95 # that first round consumes a lot more CPU cycles and that encrypt
96 # and decrypt performance becomes asymmetric. Encrypt performance
97 # drops by 10-12%, while decrypt - by 20-25%:-( 256 bytes S-box is
98 # aggressively pre-fetched.
100 # Version 4.0 effectively rolls back to 3.6 and instead implements
101 # additional set of functions, _[x86|sse]_AES_[en|de]crypt_compact,
102 # which use exclusively 256 byte S-box. These functions are to be
103 # called in modes not concealing plain text, such as ECB, or when
104 # we're asked to process smaller amount of data [or unconditionally
105 # on hyper-threading CPU]. Currently it's called unconditionally from
106 # AES_[en|de]crypt, which affects all modes, but CBC. CBC routine
107 # still needs to be modified to switch between slower and faster
108 # mode when appropriate... But in either case benchmark landscape
109 # changes dramatically and below numbers are CPU cycles per processed
110 # byte for 128-bit key.
112 # ECB encrypt ECB decrypt CBC large chunk
113 # P4 52[54] 83[95] 23
114 # AMD K8 46[41] 66[70] 18
115 # PIII 41[50] 60[77] 24
116 # Core 2 31[36] 45[64] 18.5
117 # Atom 76[100] 96[138] 60
120 # Version 4.1 switches to compact S-box even in key schedule setup.
122 # Version 4.2 prefetches compact S-box in every SSE round or in other
123 # words every cache-line is *guaranteed* to be accessed within ~50
124 # cycles window. Why just SSE? Because it's needed on hyper-threading
125 # CPU! Which is also why it's prefetched with 64 byte stride. Best
126 # part is that it has no negative effect on performance:-)
128 # Version 4.3 implements switch between compact and non-compact block
129 # functions in AES_cbc_encrypt depending on how much data was asked
130 # to be processed in one stroke.
132 ######################################################################
133 # Timing attacks are classified in two classes: synchronous when
134 # attacker consciously initiates cryptographic operation and collects
135 # timing data of various character afterwards, and asynchronous when
136 # malicious code is executed on same CPU simultaneously with AES,
137 # instruments itself and performs statistical analysis of this data.
139 # As far as synchronous attacks go the root to the AES timing
140 # vulnerability is twofold. Firstly, of 256 S-box elements at most 160
141 # are referred to in single 128-bit block operation. Well, in C
142 # implementation with 4 distinct tables it's actually as little as 40
143 # references per 256 elements table, but anyway... Secondly, even
144 # though S-box elements are clustered into smaller amount of cache-
145 # lines, smaller than 160 and even 40, it turned out that for certain
146 # plain-text pattern[s] or simply put chosen plain-text and given key
147 # few cache-lines remain unaccessed during block operation. Now, if
148 # attacker can figure out this access pattern, he can deduct the key
149 # [or at least part of it]. The natural way to mitigate this kind of
150 # attacks is to minimize the amount of cache-lines in S-box and/or
151 # prefetch them to ensure that every one is accessed for more uniform
152 # timing. But note that *if* plain-text was concealed in such way that
153 # input to block function is distributed *uniformly*, then attack
154 # wouldn't apply. Now note that some encryption modes, most notably
155 # CBC, do mask the plain-text in this exact way [secure cipher output
156 # is distributed uniformly]. Yes, one still might find input that
157 # would reveal the information about given key, but if amount of
158 # candidate inputs to be tried is larger than amount of possible key
159 # combinations then attack becomes infeasible. This is why revised
160 # AES_cbc_encrypt "dares" to switch to larger S-box when larger chunk
161 # of data is to be processed in one stroke. The current size limit of
162 # 512 bytes is chosen to provide same [diminishigly low] probability
163 # for cache-line to remain untouched in large chunk operation with
164 # large S-box as for single block operation with compact S-box and
165 # surely needs more careful consideration...
167 # As for asynchronous attacks. There are two flavours: attacker code
168 # being interleaved with AES on hyper-threading CPU at *instruction*
169 # level, and two processes time sharing single core. As for latter.
170 # Two vectors. 1. Given that attacker process has higher priority,
171 # yield execution to process performing AES just before timer fires
172 # off the scheduler, immediately regain control of CPU and analyze the
173 # cache state. For this attack to be efficient attacker would have to
174 # effectively slow down the operation by several *orders* of magnitute,
175 # by ratio of time slice to duration of handful of AES rounds, which
176 # unlikely to remain unnoticed. Not to mention that this also means
177 # that he would spend correspondigly more time to collect enough
178 # statistical data to mount the attack. It's probably appropriate to
179 # say that if adeversary reckons that this attack is beneficial and
180 # risks to be noticed, you probably have larger problems having him
181 # mere opportunity. In other words suggested code design expects you
182 # to preclude/mitigate this attack by overall system security design.
183 # 2. Attacker manages to make his code interrupt driven. In order for
184 # this kind of attack to be feasible, interrupt rate has to be high
185 # enough, again comparable to duration of handful of AES rounds. But
186 # is there interrupt source of such rate? Hardly, not even 1Gbps NIC
187 # generates interrupts at such raging rate...
189 # And now back to the former, hyper-threading CPU or more specifically
190 # Intel P4. Recall that asynchronous attack implies that malicious
191 # code instruments itself. And naturally instrumentation granularity
192 # has be noticeably lower than duration of codepath accessing S-box.
193 # Given that all cache-lines are accessed during that time that is.
194 # Current implementation accesses *all* cache-lines within ~50 cycles
195 # window, which is actually *less* than RDTSC latency on Intel P4!
197 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
198 push(@INC,"${dir}","${dir}../../perlasm");
205 &asm_init($ARGV[0],$x86only = $ARGV[$#ARGV] eq "386");
206 &static_label("AES_Te");
207 &static_label("AES_Td");
217 # stack frame layout in _[x86|sse]_AES_* routines, frame is allocated
219 $__ra=&DWP(0,"esp"); # return address
220 $__s0=&DWP(4,"esp"); # s0 backing store
221 $__s1=&DWP(8,"esp"); # s1 backing store
222 $__s2=&DWP(12,"esp"); # s2 backing store
223 $__s3=&DWP(16,"esp"); # s3 backing store
224 $__key=&DWP(20,"esp"); # pointer to key schedule
225 $__end=&DWP(24,"esp"); # pointer to end of key schedule
226 $__tbl=&DWP(28,"esp"); # %ebp backing store
228 # stack frame layout in AES_[en|crypt] routines, which differs from
229 # above by 4 and overlaps by %ebp backing store
230 $_tbl=&DWP(24,"esp");
231 $_esp=&DWP(28,"esp");
233 sub _data_word() { my $i; while(defined($i=shift)) { &data_word($i,$i); } }
235 $speed_limit=512; # chunks smaller than $speed_limit are
236 # processed with compact routine in CBC mode
237 $small_footprint=1; # $small_footprint=1 code is ~5% slower [on
238 # recent ยต-archs], but ~5 times smaller!
239 # I favor compact code to minimize cache
240 # contention and in hope to "collect" 5% back
241 # in real-life applications...
243 $vertical_spin=0; # shift "verticaly" defaults to 0, because of
244 # its proof-of-concept status...
245 # Note that there is no decvert(), as well as last encryption round is
246 # performed with "horizontal" shifts. This is because this "vertical"
247 # implementation [one which groups shifts on a given $s[i] to form a
248 # "column," unlike "horizontal" one, which groups shifts on different
249 # $s[i] to form a "row"] is work in progress. It was observed to run
250 # few percents faster on Intel cores, but not AMD. On AMD K8 core it's
251 # whole 12% slower:-( So we face a trade-off... Shall it be resolved
252 # some day? Till then the code is considered experimental and by
253 # default remains dormant...
257 my ($v0,$v1) = ($acc,$key);
259 &mov ($v0,$s[3]); # copy s3
260 &mov (&DWP(4,"esp"),$s[2]); # save s2
261 &mov ($v1,$s[0]); # copy s0
262 &mov (&DWP(8,"esp"),$s[1]); # save s1
264 &movz ($s[2],&HB($s[0]));
266 &mov ($s[0],&DWP(0,$te,$s[0],8)); # s0>>0
268 &mov ($s[3],&DWP(3,$te,$s[2],8)); # s0>>8
269 &movz ($s[1],&HB($v1));
271 &mov ($s[2],&DWP(2,$te,$v1,8)); # s0>>16
273 &mov ($s[1],&DWP(1,$te,$s[1],8)); # s0>>24
276 &xor ($s[3],&DWP(0,$te,$v0,8)); # s3>>0
277 &movz ($v0,&HB($v1));
279 &xor ($s[2],&DWP(3,$te,$v0,8)); # s3>>8
280 &movz ($v0,&HB($v1));
282 &xor ($s[1],&DWP(2,$te,$v1,8)); # s3>>16
283 &mov ($v1,&DWP(4,"esp")); # restore s2
284 &xor ($s[0],&DWP(1,$te,$v0,8)); # s3>>24
288 &xor ($s[2],&DWP(0,$te,$v1,8)); # s2>>0
289 &movz ($v1,&HB($v0));
291 &xor ($s[1],&DWP(3,$te,$v1,8)); # s2>>8
292 &movz ($v1,&HB($v0));
294 &xor ($s[0],&DWP(2,$te,$v0,8)); # s2>>16
295 &mov ($v0,&DWP(8,"esp")); # restore s1
296 &xor ($s[3],&DWP(1,$te,$v1,8)); # s2>>24
300 &xor ($s[1],&DWP(0,$te,$v0,8)); # s1>>0
301 &movz ($v0,&HB($v1));
303 &xor ($s[0],&DWP(3,$te,$v0,8)); # s1>>8
304 &movz ($v0,&HB($v1));
306 &xor ($s[3],&DWP(2,$te,$v1,8)); # s1>>16
307 &mov ($key,$__key); # reincarnate v1 as key
308 &xor ($s[2],&DWP(1,$te,$v0,8)); # s1>>24
311 # Another experimental routine, which features "horizontal spin," but
312 # eliminates one reference to stack. Strangely enough runs slower...
314 { my ($v0,$v1) = ($key,$acc);
316 &movz ($v0,&LB($s0)); # 3, 2, 1, 0*
317 &rotr ($s2,8); # 8,11,10, 9
318 &mov ($v1,&DWP(0,$te,$v0,8)); # 0
319 &movz ($v0,&HB($s1)); # 7, 6, 5*, 4
320 &rotr ($s3,16); # 13,12,15,14
321 &xor ($v1,&DWP(3,$te,$v0,8)); # 5
322 &movz ($v0,&HB($s2)); # 8,11,10*, 9
323 &rotr ($s0,16); # 1, 0, 3, 2
324 &xor ($v1,&DWP(2,$te,$v0,8)); # 10
325 &movz ($v0,&HB($s3)); # 13,12,15*,14
326 &xor ($v1,&DWP(1,$te,$v0,8)); # 15, t[0] collected
327 &mov ($__s0,$v1); # t[0] saved
329 &movz ($v0,&LB($s1)); # 7, 6, 5, 4*
330 &shr ($s1,16); # -, -, 7, 6
331 &mov ($v1,&DWP(0,$te,$v0,8)); # 4
332 &movz ($v0,&LB($s3)); # 13,12,15,14*
333 &xor ($v1,&DWP(2,$te,$v0,8)); # 14
334 &movz ($v0,&HB($s0)); # 1, 0, 3*, 2
335 &and ($s3,0xffff0000); # 13,12, -, -
336 &xor ($v1,&DWP(1,$te,$v0,8)); # 3
337 &movz ($v0,&LB($s2)); # 8,11,10, 9*
338 &or ($s3,$s1); # 13,12, 7, 6
339 &xor ($v1,&DWP(3,$te,$v0,8)); # 9, t[1] collected
340 &mov ($s1,$v1); # s[1]=t[1]
342 &movz ($v0,&LB($s0)); # 1, 0, 3, 2*
343 &shr ($s2,16); # -, -, 8,11
344 &mov ($v1,&DWP(2,$te,$v0,8)); # 2
345 &movz ($v0,&HB($s3)); # 13,12, 7*, 6
346 &xor ($v1,&DWP(1,$te,$v0,8)); # 7
347 &movz ($v0,&HB($s2)); # -, -, 8*,11
348 &xor ($v1,&DWP(0,$te,$v0,8)); # 8
351 &xor ($v1,&DWP(3,$te,$v0,8)); # 13, t[2] collected
353 &movz ($v0,&LB($s2)); # -, -, 8,11*
355 &mov ($s2,&DWP(1,$te,$v0,8)); # 11
356 &xor ($s2,&DWP(3,$te,$s0,8)); # 1
357 &mov ($s0,$__s0); # s[0]=t[0]
358 &movz ($v0,&LB($s3)); # 13,12, 7, 6*
359 &shr ($s3,16); # , ,13,12
360 &xor ($s2,&DWP(2,$te,$v0,8)); # 6
361 &mov ($key,$__key); # reincarnate v0 as key
362 &and ($s3,0xff); # , ,13,12*
363 &mov ($s3,&DWP(0,$te,$s3,8)); # 12
364 &xor ($s3,$s2); # s[2]=t[3] collected
365 &mov ($s2,$v1); # s[2]=t[2]
368 # More experimental code... SSE one... Even though this one eliminates
369 # *all* references to stack, it's not faster...
372 &movz ($acc,&LB("eax")); # 0
373 &mov ("ecx",&DWP(0,$tbl,$acc,8)); # 0
374 &pshufw ("mm2","mm0",0x0d); # 7, 6, 3, 2
375 &movz ("edx",&HB("eax")); # 1
376 &mov ("edx",&DWP(3,$tbl,"edx",8)); # 1
377 &shr ("eax",16); # 5, 4
379 &movz ($acc,&LB("ebx")); # 10
380 &xor ("ecx",&DWP(2,$tbl,$acc,8)); # 10
381 &pshufw ("mm6","mm4",0x08); # 13,12, 9, 8
382 &movz ($acc,&HB("ebx")); # 11
383 &xor ("edx",&DWP(1,$tbl,$acc,8)); # 11
384 &shr ("ebx",16); # 15,14
386 &movz ($acc,&HB("eax")); # 5
387 &xor ("ecx",&DWP(3,$tbl,$acc,8)); # 5
388 &movq ("mm3",QWP(16,$key));
389 &movz ($acc,&HB("ebx")); # 15
390 &xor ("ecx",&DWP(1,$tbl,$acc,8)); # 15
391 &movd ("mm0","ecx"); # t[0] collected
393 &movz ($acc,&LB("eax")); # 4
394 &mov ("ecx",&DWP(0,$tbl,$acc,8)); # 4
395 &movd ("eax","mm2"); # 7, 6, 3, 2
396 &movz ($acc,&LB("ebx")); # 14
397 &xor ("ecx",&DWP(2,$tbl,$acc,8)); # 14
398 &movd ("ebx","mm6"); # 13,12, 9, 8
400 &movz ($acc,&HB("eax")); # 3
401 &xor ("ecx",&DWP(1,$tbl,$acc,8)); # 3
402 &movz ($acc,&HB("ebx")); # 9
403 &xor ("ecx",&DWP(3,$tbl,$acc,8)); # 9
404 &movd ("mm1","ecx"); # t[1] collected
406 &movz ($acc,&LB("eax")); # 2
407 &mov ("ecx",&DWP(2,$tbl,$acc,8)); # 2
408 &shr ("eax",16); # 7, 6
409 &punpckldq ("mm0","mm1"); # t[0,1] collected
410 &movz ($acc,&LB("ebx")); # 8
411 &xor ("ecx",&DWP(0,$tbl,$acc,8)); # 8
412 &shr ("ebx",16); # 13,12
414 &movz ($acc,&HB("eax")); # 7
415 &xor ("ecx",&DWP(1,$tbl,$acc,8)); # 7
417 &movz ("eax",&LB("eax")); # 6
418 &xor ("edx",&DWP(2,$tbl,"eax",8)); # 6
419 &pshufw ("mm1","mm0",0x08); # 5, 4, 1, 0
420 &movz ($acc,&HB("ebx")); # 13
421 &xor ("ecx",&DWP(3,$tbl,$acc,8)); # 13
422 &xor ("ecx",&DWP(24,$key)); # t[2]
423 &movd ("mm4","ecx"); # t[2] collected
424 &movz ("ebx",&LB("ebx")); # 12
425 &xor ("edx",&DWP(0,$tbl,"ebx",8)); # 12
427 &movd ("eax","mm1"); # 5, 4, 1, 0
428 &mov ("ebx",&DWP(28,$key)); # t[3]
430 &movd ("mm5","ebx"); # t[3] collected
431 &and ("ebx",0xffff0000);
434 &punpckldq ("mm4","mm5"); # t[2,3] collected
437 ######################################################################
438 # "Compact" block function
439 ######################################################################
443 while ($#_>5) { pop(@_); $Fn=sub{}; }
446 my $out = $i==3?$s[0]:$acc;
448 # $Fn is used in first compact round and its purpose is to
449 # void restoration of some values from stack, so that after
450 # 4xenccompact with extra argument $key value is left there...
451 if ($i==3) { &$Fn ($key,$__key); }##%edx
452 else { &mov ($out,$s[0]); }
454 if ($i==1) { &shr ($s[0],16); }#%ebx[1]
455 if ($i==2) { &shr ($s[0],24); }#%ecx[2]
456 &movz ($out,&BP(-128,$te,$out,1));
458 if ($i==3) { $tmp=$s[1]; }##%eax
459 &movz ($tmp,&HB($s[1]));
460 &movz ($tmp,&BP(-128,$te,$tmp,1));
464 if ($i==3) { $tmp=$s[2]; &mov ($s[1],$__s0); }##%ebx
465 else { &mov ($tmp,$s[2]);
467 if ($i==2) { &and ($s[1],0xFF); }#%edx[2]
469 &movz ($tmp,&BP(-128,$te,$tmp,1));
473 if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }##%ecx
474 elsif($i==2){ &movz ($tmp,&HB($s[3])); }#%ebx[2]
475 else { &mov ($tmp,$s[3]);
477 &movz ($tmp,&BP(-128,$te,$tmp,1));
480 if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); }
481 if ($i==3) { &mov ($s[3],$acc); }
486 { my @s = ($s0,$s1,$s2,$s3);
492 &lea ($r2,&DWP(0,$s[$i],$s[$i]));
495 &and ($r2,0xfefefefe);
498 &and ($acc,0x1b1b1b1b);
500 &xor ($acc,$r2); # r2
503 &xor ($s[$i],$acc); # r0 ^ r2
508 &mov ($tmp,0x80808080) if ($i!=1);
509 &xor ($s[$i],$acc); # ROTATE(r2^r0,24) ^ r2
512 &function_begin_B("_x86_AES_encrypt_compact");
513 # note that caller is expected to allocate stack frame for me!
514 &mov ($__key,$key); # save key
516 &xor ($s0,&DWP(0,$key)); # xor with key
517 &xor ($s1,&DWP(4,$key));
518 &xor ($s2,&DWP(8,$key));
519 &xor ($s3,&DWP(12,$key));
521 &mov ($acc,&DWP(240,$key)); # load key->rounds
522 &lea ($acc,&DWP(-2,$acc,$acc));
523 &lea ($acc,&DWP(0,$key,$acc,8));
524 &mov ($__end,$acc); # end of key schedule
527 &mov ($key,&DWP(0-128,$tbl));
528 &mov ($acc,&DWP(32-128,$tbl));
529 &mov ($key,&DWP(64-128,$tbl));
530 &mov ($acc,&DWP(96-128,$tbl));
531 &mov ($key,&DWP(128-128,$tbl));
532 &mov ($acc,&DWP(160-128,$tbl));
533 &mov ($key,&DWP(192-128,$tbl));
534 &mov ($acc,&DWP(224-128,$tbl));
536 &set_label("loop",16);
538 &enccompact(0,$tbl,$s0,$s1,$s2,$s3,1);
539 &enccompact(1,$tbl,$s1,$s2,$s3,$s0,1);
540 &enccompact(2,$tbl,$s2,$s3,$s0,$s1,1);
541 &enccompact(3,$tbl,$s3,$s0,$s1,$s2,1);
542 &mov ($tbl,0x80808080);
549 &add ($key,16); # advance rd_key
550 &xor ($s0,&DWP(0,$key));
551 &xor ($s1,&DWP(4,$key));
552 &xor ($s2,&DWP(8,$key));
553 &xor ($s3,&DWP(12,$key));
557 &jb (&label("loop"));
559 &enccompact(0,$tbl,$s0,$s1,$s2,$s3);
560 &enccompact(1,$tbl,$s1,$s2,$s3,$s0);
561 &enccompact(2,$tbl,$s2,$s3,$s0,$s1);
562 &enccompact(3,$tbl,$s3,$s0,$s1,$s2);
564 &xor ($s0,&DWP(16,$key));
565 &xor ($s1,&DWP(20,$key));
566 &xor ($s2,&DWP(24,$key));
567 &xor ($s3,&DWP(28,$key));
570 &function_end_B("_x86_AES_encrypt_compact");
572 ######################################################################
573 # "Compact" SSE block function.
574 ######################################################################
576 # Performance is not actually extraordinary in comparison to pure
577 # x86 code. In particular encrypt performance is virtually the same.
578 # Decrypt performance on the other hand is 15-20% better on newer
579 # ยต-archs [but we're thankful for *any* improvement here], and ~50%
580 # better on PIII:-) And additionally on the pros side this code
581 # eliminates redundant references to stack and thus relieves/
582 # minimizes the pressure on the memory bus.
584 # MMX register layout lsb
585 # +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
587 # +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
588 # | s3 | s2 | s1 | s0 |
589 # +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
590 # |15|14|13|12|11|10| 9| 8| 7| 6| 5| 4| 3| 2| 1| 0|
591 # +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
593 # Indexes translate as s[N/4]>>(8*(N%4)), e.g. 5 means s1>>8.
594 # In this terms encryption and decryption "compact" permutation
595 # matrices can be depicted as following:
597 # encryption lsb # decryption lsb
598 # +----++----+----+----+----+ # +----++----+----+----+----+
599 # | t0 || 15 | 10 | 5 | 0 | # | t0 || 7 | 10 | 13 | 0 |
600 # +----++----+----+----+----+ # +----++----+----+----+----+
601 # | t1 || 3 | 14 | 9 | 4 | # | t1 || 11 | 14 | 1 | 4 |
602 # +----++----+----+----+----+ # +----++----+----+----+----+
603 # | t2 || 7 | 2 | 13 | 8 | # | t2 || 15 | 2 | 5 | 8 |
604 # +----++----+----+----+----+ # +----++----+----+----+----+
605 # | t3 || 11 | 6 | 1 | 12 | # | t3 || 3 | 6 | 9 | 12 |
606 # +----++----+----+----+----+ # +----++----+----+----+----+
608 ######################################################################
609 # Why not xmm registers? Short answer. It was actually tested and
610 # was not any faster, but *contrary*, most notably on Intel CPUs.
611 # Longer answer. Main advantage of using mm registers is that movd
612 # latency is lower, especially on Intel P4. While arithmetic
613 # instructions are twice as many, they can be scheduled every cycle
614 # and not every second one when they are operating on xmm register,
615 # so that "arithmetic throughput" remains virtually the same. And
616 # finally the code can be executed even on elder SSE-only CPUs:-)
620 &pshufw ("mm1","mm0",0x08); # 5, 4, 1, 0
621 &pshufw ("mm5","mm4",0x0d); # 15,14,11,10
622 &movd ("eax","mm1"); # 5, 4, 1, 0
623 &movd ("ebx","mm5"); # 15,14,11,10
626 &movz ($acc,&LB("eax")); # 0
627 &movz ("edx",&HB("eax")); # 1
628 &pshufw ("mm2","mm0",0x0d); # 7, 6, 3, 2
629 &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 0
630 &movz ($key,&LB("ebx")); # 10
631 &movz ("edx",&BP(-128,$tbl,"edx",1)); # 1
632 &shr ("eax",16); # 5, 4
635 &movz ($acc,&BP(-128,$tbl,$key,1)); # 10
636 &movz ($key,&HB("ebx")); # 11
638 &pshufw ("mm6","mm4",0x08); # 13,12, 9, 8
639 &or ("ecx",$acc); # 10
640 &movz ($acc,&BP(-128,$tbl,$key,1)); # 11
641 &movz ($key,&HB("eax")); # 5
643 &shr ("ebx",16); # 15,14
644 &or ("edx",$acc); # 11
646 &movz ($acc,&BP(-128,$tbl,$key,1)); # 5
647 &movz ($key,&HB("ebx")); # 15
649 &or ("ecx",$acc); # 5
650 &movz ($acc,&BP(-128,$tbl,$key,1)); # 15
651 &movz ($key,&LB("eax")); # 4
653 &or ("ecx",$acc); # 15
655 &movz ($acc,&BP(-128,$tbl,$key,1)); # 4
656 &movz ($key,&LB("ebx")); # 14
657 &movd ("eax","mm2"); # 7, 6, 3, 2
658 &movd ("mm0","ecx"); # t[0] collected
659 &movz ("ecx",&BP(-128,$tbl,$key,1)); # 14
660 &movz ($key,&HB("eax")); # 3
661 &shl ("ecx",16); # 14
662 &movd ("ebx","mm6"); # 13,12, 9, 8
663 &or ("ecx",$acc); # 14
665 &movz ($acc,&BP(-128,$tbl,$key,1)); # 3
666 &movz ($key,&HB("ebx")); # 9
668 &or ("ecx",$acc); # 3
669 &movz ($acc,&BP(-128,$tbl,$key,1)); # 9
670 &movz ($key,&LB("ebx")); # 8
672 &shr ("ebx",16); # 13,12
673 &or ("ecx",$acc); # 9
675 &movz ($acc,&BP(-128,$tbl,$key,1)); # 8
676 &movz ($key,&LB("eax")); # 2
677 &shr ("eax",16); # 7, 6
678 &movd ("mm1","ecx"); # t[1] collected
679 &movz ("ecx",&BP(-128,$tbl,$key,1)); # 2
680 &movz ($key,&HB("eax")); # 7
682 &and ("eax",0xff); # 6
683 &or ("ecx",$acc); # 2
685 &punpckldq ("mm0","mm1"); # t[0,1] collected
687 &movz ($acc,&BP(-128,$tbl,$key,1)); # 7
688 &movz ($key,&HB("ebx")); # 13
690 &and ("ebx",0xff); # 12
691 &movz ("eax",&BP(-128,$tbl,"eax",1)); # 6
692 &or ("ecx",$acc); # 7
694 &movz ($acc,&BP(-128,$tbl,$key,1)); # 13
695 &or ("edx","eax"); # 6
697 &movz ("ebx",&BP(-128,$tbl,"ebx",1)); # 12
698 &or ("ecx",$acc); # 13
699 &or ("edx","ebx"); # 12
701 &movd ("mm4","ecx"); # t[2] collected
702 &movd ("mm5","edx"); # t[3] collected
704 &punpckldq ("mm4","mm5"); # t[2,3] collected
708 &function_begin_B("_sse_AES_encrypt_compact");
709 &pxor ("mm0",&QWP(0,$key)); # 7, 6, 5, 4, 3, 2, 1, 0
710 &pxor ("mm4",&QWP(8,$key)); # 15,14,13,12,11,10, 9, 8
712 # note that caller is expected to allocate stack frame for me!
713 &mov ($acc,&DWP(240,$key)); # load key->rounds
714 &lea ($acc,&DWP(-2,$acc,$acc));
715 &lea ($acc,&DWP(0,$key,$acc,8));
716 &mov ($__end,$acc); # end of key schedule
718 &mov ($s0,0x1b1b1b1b); # magic constant
719 &mov (&DWP(8,"esp"),$s0);
720 &mov (&DWP(12,"esp"),$s0);
723 &mov ($s0,&DWP(0-128,$tbl));
724 &mov ($s1,&DWP(32-128,$tbl));
725 &mov ($s2,&DWP(64-128,$tbl));
726 &mov ($s3,&DWP(96-128,$tbl));
727 &mov ($s0,&DWP(128-128,$tbl));
728 &mov ($s1,&DWP(160-128,$tbl));
729 &mov ($s2,&DWP(192-128,$tbl));
730 &mov ($s3,&DWP(224-128,$tbl));
732 &set_label("loop",16);
738 &movq ("mm2",&QWP(8,"esp"));
739 &pxor ("mm3","mm3"); &pxor ("mm7","mm7");
740 &movq ("mm1","mm0"); &movq ("mm5","mm4"); # r0
741 &pcmpgtb("mm3","mm0"); &pcmpgtb("mm7","mm4");
742 &pand ("mm3","mm2"); &pand ("mm7","mm2");
743 &pshufw ("mm2","mm0",0xb1); &pshufw ("mm6","mm4",0xb1);# ROTATE(r0,16)
744 &paddb ("mm0","mm0"); &paddb ("mm4","mm4");
745 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # = r2
746 &pshufw ("mm3","mm2",0xb1); &pshufw ("mm7","mm6",0xb1);# r0
747 &pxor ("mm1","mm0"); &pxor ("mm5","mm4"); # r0^r2
748 &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= ROTATE(r0,16)
750 &movq ("mm2","mm3"); &movq ("mm6","mm7");
751 &pslld ("mm3",8); &pslld ("mm7",8);
752 &psrld ("mm2",24); &psrld ("mm6",24);
753 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= r0<<8
754 &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= r0>>24
756 &movq ("mm3","mm1"); &movq ("mm7","mm5");
757 &movq ("mm2",&QWP(0,$key)); &movq ("mm6",&QWP(8,$key));
758 &psrld ("mm1",8); &psrld ("mm5",8);
759 &mov ($s0,&DWP(0-128,$tbl));
760 &pslld ("mm3",24); &pslld ("mm7",24);
761 &mov ($s1,&DWP(64-128,$tbl));
762 &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= (r2^r0)<<8
763 &mov ($s2,&DWP(128-128,$tbl));
764 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= (r2^r0)>>24
765 &mov ($s3,&DWP(192-128,$tbl));
767 &pxor ("mm0","mm2"); &pxor ("mm4","mm6");
768 &jmp (&label("loop"));
770 &set_label("out",16);
771 &pxor ("mm0",&QWP(0,$key));
772 &pxor ("mm4",&QWP(8,$key));
775 &function_end_B("_sse_AES_encrypt_compact");
778 ######################################################################
779 # Vanilla block function.
780 ######################################################################
783 { my ($i,$te,@s) = @_;
785 my $out = $i==3?$s[0]:$acc;
787 # lines marked with #%e?x[i] denote "reordered" instructions...
788 if ($i==3) { &mov ($key,$__key); }##%edx
789 else { &mov ($out,$s[0]);
791 if ($i==1) { &shr ($s[0],16); }#%ebx[1]
792 if ($i==2) { &shr ($s[0],24); }#%ecx[2]
793 &mov ($out,&DWP(0,$te,$out,8));
795 if ($i==3) { $tmp=$s[1]; }##%eax
796 &movz ($tmp,&HB($s[1]));
797 &xor ($out,&DWP(3,$te,$tmp,8));
799 if ($i==3) { $tmp=$s[2]; &mov ($s[1],$__s0); }##%ebx
800 else { &mov ($tmp,$s[2]);
802 if ($i==2) { &and ($s[1],0xFF); }#%edx[2]
804 &xor ($out,&DWP(2,$te,$tmp,8));
806 if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }##%ecx
807 elsif($i==2){ &movz ($tmp,&HB($s[3])); }#%ebx[2]
808 else { &mov ($tmp,$s[3]);
810 &xor ($out,&DWP(1,$te,$tmp,8));
811 if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); }
812 if ($i==3) { &mov ($s[3],$acc); }
819 my $out = $i==3?$s[0]:$acc;
821 if ($i==3) { &mov ($key,$__key); }##%edx
822 else { &mov ($out,$s[0]); }
824 if ($i==1) { &shr ($s[0],16); }#%ebx[1]
825 if ($i==2) { &shr ($s[0],24); }#%ecx[2]
826 &mov ($out,&DWP(2,$te,$out,8));
827 &and ($out,0x000000ff);
829 if ($i==3) { $tmp=$s[1]; }##%eax
830 &movz ($tmp,&HB($s[1]));
831 &mov ($tmp,&DWP(0,$te,$tmp,8));
832 &and ($tmp,0x0000ff00);
835 if ($i==3) { $tmp=$s[2]; &mov ($s[1],$__s0); }##%ebx
836 else { &mov ($tmp,$s[2]);
838 if ($i==2) { &and ($s[1],0xFF); }#%edx[2]
840 &mov ($tmp,&DWP(0,$te,$tmp,8));
841 &and ($tmp,0x00ff0000);
844 if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }##%ecx
845 elsif($i==2){ &movz ($tmp,&HB($s[3])); }#%ebx[2]
846 else { &mov ($tmp,$s[3]);
848 &mov ($tmp,&DWP(2,$te,$tmp,8));
849 &and ($tmp,0xff000000);
851 if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); }
852 if ($i==3) { &mov ($s[3],$acc); }
855 &function_begin_B("_x86_AES_encrypt");
856 if ($vertical_spin) {
857 # I need high parts of volatile registers to be accessible...
858 &exch ($s1="edi",$key="ebx");
859 &mov ($s2="esi",$acc="ecx");
862 # note that caller is expected to allocate stack frame for me!
863 &mov ($__key,$key); # save key
865 &xor ($s0,&DWP(0,$key)); # xor with key
866 &xor ($s1,&DWP(4,$key));
867 &xor ($s2,&DWP(8,$key));
868 &xor ($s3,&DWP(12,$key));
870 &mov ($acc,&DWP(240,$key)); # load key->rounds
872 if ($small_footprint) {
873 &lea ($acc,&DWP(-2,$acc,$acc));
874 &lea ($acc,&DWP(0,$key,$acc,8));
875 &mov ($__end,$acc); # end of key schedule
877 &set_label("loop",16);
878 if ($vertical_spin) {
879 &encvert($tbl,$s0,$s1,$s2,$s3);
881 &encstep(0,$tbl,$s0,$s1,$s2,$s3);
882 &encstep(1,$tbl,$s1,$s2,$s3,$s0);
883 &encstep(2,$tbl,$s2,$s3,$s0,$s1);
884 &encstep(3,$tbl,$s3,$s0,$s1,$s2);
886 &add ($key,16); # advance rd_key
887 &xor ($s0,&DWP(0,$key));
888 &xor ($s1,&DWP(4,$key));
889 &xor ($s2,&DWP(8,$key));
890 &xor ($s3,&DWP(12,$key));
893 &jb (&label("loop"));
897 &jle (&label("10rounds"));
899 &jle (&label("12rounds"));
901 &set_label("14rounds",4);
902 for ($i=1;$i<3;$i++) {
903 if ($vertical_spin) {
904 &encvert($tbl,$s0,$s1,$s2,$s3);
906 &encstep(0,$tbl,$s0,$s1,$s2,$s3);
907 &encstep(1,$tbl,$s1,$s2,$s3,$s0);
908 &encstep(2,$tbl,$s2,$s3,$s0,$s1);
909 &encstep(3,$tbl,$s3,$s0,$s1,$s2);
911 &xor ($s0,&DWP(16*$i+0,$key));
912 &xor ($s1,&DWP(16*$i+4,$key));
913 &xor ($s2,&DWP(16*$i+8,$key));
914 &xor ($s3,&DWP(16*$i+12,$key));
917 &mov ($__key,$key); # advance rd_key
918 &set_label("12rounds",4);
919 for ($i=1;$i<3;$i++) {
920 if ($vertical_spin) {
921 &encvert($tbl,$s0,$s1,$s2,$s3);
923 &encstep(0,$tbl,$s0,$s1,$s2,$s3);
924 &encstep(1,$tbl,$s1,$s2,$s3,$s0);
925 &encstep(2,$tbl,$s2,$s3,$s0,$s1);
926 &encstep(3,$tbl,$s3,$s0,$s1,$s2);
928 &xor ($s0,&DWP(16*$i+0,$key));
929 &xor ($s1,&DWP(16*$i+4,$key));
930 &xor ($s2,&DWP(16*$i+8,$key));
931 &xor ($s3,&DWP(16*$i+12,$key));
934 &mov ($__key,$key); # advance rd_key
935 &set_label("10rounds",4);
936 for ($i=1;$i<10;$i++) {
937 if ($vertical_spin) {
938 &encvert($tbl,$s0,$s1,$s2,$s3);
940 &encstep(0,$tbl,$s0,$s1,$s2,$s3);
941 &encstep(1,$tbl,$s1,$s2,$s3,$s0);
942 &encstep(2,$tbl,$s2,$s3,$s0,$s1);
943 &encstep(3,$tbl,$s3,$s0,$s1,$s2);
945 &xor ($s0,&DWP(16*$i+0,$key));
946 &xor ($s1,&DWP(16*$i+4,$key));
947 &xor ($s2,&DWP(16*$i+8,$key));
948 &xor ($s3,&DWP(16*$i+12,$key));
952 if ($vertical_spin) {
953 # "reincarnate" some registers for "horizontal" spin...
954 &mov ($s1="ebx",$key="edi");
955 &mov ($s2="ecx",$acc="esi");
957 &enclast(0,$tbl,$s0,$s1,$s2,$s3);
958 &enclast(1,$tbl,$s1,$s2,$s3,$s0);
959 &enclast(2,$tbl,$s2,$s3,$s0,$s1);
960 &enclast(3,$tbl,$s3,$s0,$s1,$s2);
962 &add ($key,$small_footprint?16:160);
963 &xor ($s0,&DWP(0,$key));
964 &xor ($s1,&DWP(4,$key));
965 &xor ($s2,&DWP(8,$key));
966 &xor ($s3,&DWP(12,$key));
970 &set_label("AES_Te",64); # Yes! I keep it in the code segment!
971 &_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6);
972 &_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591);
973 &_data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56);
974 &_data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec);
975 &_data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa);
976 &_data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb);
977 &_data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45);
978 &_data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b);
979 &_data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c);
980 &_data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83);
981 &_data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9);
982 &_data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a);
983 &_data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d);
984 &_data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f);
985 &_data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df);
986 &_data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea);
987 &_data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34);
988 &_data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b);
989 &_data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d);
990 &_data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413);
991 &_data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1);
992 &_data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6);
993 &_data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972);
994 &_data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85);
995 &_data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed);
996 &_data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511);
997 &_data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe);
998 &_data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b);
999 &_data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05);
1000 &_data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1);
1001 &_data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142);
1002 &_data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf);
1003 &_data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3);
1004 &_data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e);
1005 &_data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a);
1006 &_data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6);
1007 &_data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3);
1008 &_data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b);
1009 &_data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428);
1010 &_data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad);
1011 &_data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14);
1012 &_data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8);
1013 &_data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4);
1014 &_data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2);
1015 &_data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda);
1016 &_data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949);
1017 &_data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf);
1018 &_data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810);
1019 &_data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c);
1020 &_data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697);
1021 &_data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e);
1022 &_data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f);
1023 &_data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc);
1024 &_data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c);
1025 &_data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969);
1026 &_data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27);
1027 &_data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122);
1028 &_data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433);
1029 &_data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9);
1030 &_data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5);
1031 &_data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a);
1032 &_data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0);
1033 &_data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e);
1034 &_data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c);
1036 #Te4 # four copies of Te4 to choose from to avoid L1 aliasing
1037 &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
1038 &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
1039 &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
1040 &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
1041 &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
1042 &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
1043 &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
1044 &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
1045 &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
1046 &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
1047 &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
1048 &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
1049 &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
1050 &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
1051 &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
1052 &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
1053 &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
1054 &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
1055 &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
1056 &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
1057 &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
1058 &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
1059 &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
1060 &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
1061 &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
1062 &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
1063 &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
1064 &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
1065 &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
1066 &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
1067 &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
1068 &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
1070 &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
1071 &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
1072 &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
1073 &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
1074 &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
1075 &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
1076 &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
1077 &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
1078 &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
1079 &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
1080 &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
1081 &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
1082 &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
1083 &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
1084 &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
1085 &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
1086 &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
1087 &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
1088 &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
1089 &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
1090 &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
1091 &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
1092 &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
1093 &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
1094 &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
1095 &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
1096 &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
1097 &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
1098 &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
1099 &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
1100 &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
1101 &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
1103 &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
1104 &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
1105 &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
1106 &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
1107 &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
1108 &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
1109 &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
1110 &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
1111 &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
1112 &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
1113 &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
1114 &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
1115 &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
1116 &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
1117 &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
1118 &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
1119 &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
1120 &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
1121 &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
1122 &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
1123 &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
1124 &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
1125 &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
1126 &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
1127 &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
1128 &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
1129 &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
1130 &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
1131 &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
1132 &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
1133 &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
1134 &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
1136 &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
1137 &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
1138 &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
1139 &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
1140 &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
1141 &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
1142 &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
1143 &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
1144 &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
1145 &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
1146 &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
1147 &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
1148 &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
1149 &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
1150 &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
1151 &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
1152 &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
1153 &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
1154 &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
1155 &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
1156 &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
1157 &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
1158 &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
1159 &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
1160 &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
1161 &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
1162 &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
1163 &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
1164 &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
1165 &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
1166 &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
1167 &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
1169 &data_word(0x00000001, 0x00000002, 0x00000004, 0x00000008);
1170 &data_word(0x00000010, 0x00000020, 0x00000040, 0x00000080);
1171 &data_word(0x0000001b, 0x00000036, 0x00000000, 0x00000000);
1172 &data_word(0x00000000, 0x00000000, 0x00000000, 0x00000000);
1173 &function_end_B("_x86_AES_encrypt");
1175 # void AES_encrypt (const void *inp,void *out,const AES_KEY *key);
1176 &function_begin("AES_encrypt");
1177 &mov ($acc,&wparam(0)); # load inp
1178 &mov ($key,&wparam(2)); # load key
1182 &and ("esp",-64); # align to cache-line
1184 # place stack frame just "above" the key schedule
1185 &lea ($s1,&DWP(-64-63,$key));
1188 &and ($s1,0x3C0); # modulo 1024, but aligned to cache-line
1190 &add ("esp",4); # 4 is reserved for caller's return address
1191 &mov ($_esp,$s0); # save stack pointer
1193 &call (&label("pic_point")); # make it PIC!
1194 &set_label("pic_point");
1196 &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if (!$x86only);
1197 &lea ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
1199 # pick Te4 copy which can't "overlap" with stack frame or key schedule
1200 &lea ($s1,&DWP(768-4,"esp"));
1203 &lea ($tbl,&DWP(2048+128,$tbl,$s1));
1206 &bt (&DWP(0,$s0),25); # check for SSE bit
1207 &jnc (&label("x86"));
1209 &movq ("mm0",&QWP(0,$acc));
1210 &movq ("mm4",&QWP(8,$acc));
1211 &call ("_sse_AES_encrypt_compact");
1212 &mov ("esp",$_esp); # restore stack pointer
1213 &mov ($acc,&wparam(1)); # load out
1214 &movq (&QWP(0,$acc),"mm0"); # write output data
1215 &movq (&QWP(8,$acc),"mm4");
1219 &set_label("x86",16);
1221 &mov ($s0,&DWP(0,$acc)); # load input data
1222 &mov ($s1,&DWP(4,$acc));
1223 &mov ($s2,&DWP(8,$acc));
1224 &mov ($s3,&DWP(12,$acc));
1225 &call ("_x86_AES_encrypt_compact");
1226 &mov ("esp",$_esp); # restore stack pointer
1227 &mov ($acc,&wparam(1)); # load out
1228 &mov (&DWP(0,$acc),$s0); # write output data
1229 &mov (&DWP(4,$acc),$s1);
1230 &mov (&DWP(8,$acc),$s2);
1231 &mov (&DWP(12,$acc),$s3);
1232 &function_end("AES_encrypt");
1234 #--------------------------------------------------------------------#
1236 ######################################################################
1237 # "Compact" block function
1238 ######################################################################
1242 while ($#_>5) { pop(@_); $Fn=sub{}; }
1245 my $out = $i==3?$s[0]:$acc;
1247 # $Fn is used in first compact round and its purpose is to
1248 # void restoration of some values from stack, so that after
1249 # 4xdeccompact with extra argument $key, $s0 and $s1 values
1251 if($i==3) { &$Fn ($key,$__key); }
1252 else { &mov ($out,$s[0]); }
1254 &movz ($out,&BP(-128,$td,$out,1));
1256 if ($i==3) { $tmp=$s[1]; }
1257 &movz ($tmp,&HB($s[1]));
1258 &movz ($tmp,&BP(-128,$td,$tmp,1));
1262 if ($i==3) { $tmp=$s[2]; &mov ($s[1],$acc); }
1263 else { mov ($tmp,$s[2]); }
1266 &movz ($tmp,&BP(-128,$td,$tmp,1));
1270 if ($i==3) { $tmp=$s[3]; &$Fn ($s[2],$__s1); }
1271 else { &mov ($tmp,$s[3]); }
1273 &movz ($tmp,&BP(-128,$td,$tmp,1));
1276 if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); }
1277 if ($i==3) { &$Fn ($s[3],$__s0); }
1280 # must be called with 2,3,0,1 as argument sequence!!!
1282 { my @s = ($s0,$s1,$s2,$s3);
1285 my $tp2 = @s[($i+2)%4]; $tp2 = @s[2] if ($i==1);
1286 my $tp4 = @s[($i+3)%4]; $tp4 = @s[3] if ($i==1);
1289 &mov ($tmp,0x80808080);
1293 &lea ($tp2,&DWP(0,$s[$i],$s[$i]));
1295 &and ($tp2,0xfefefefe);
1296 &and ($acc,0x1b1b1b1b);
1298 &mov ($tmp,0x80808080);
1303 &lea ($tp4,&DWP(0,$tp2,$tp2));
1305 &and ($tp4,0xfefefefe);
1306 &and ($acc,0x1b1b1b1b);
1307 &xor ($tp2,$s[$i]); # tp2^tp1
1309 &mov ($tmp,0x80808080);
1314 &lea ($tp8,&DWP(0,$tp4,$tp4));
1316 &and ($tp8,0xfefefefe);
1317 &and ($acc,0x1b1b1b1b);
1318 &xor ($tp4,$s[$i]); # tp4^tp1
1319 &rotl ($s[$i],8); # = ROTATE(tp1,8)
1327 &xor ($s[$i],$tp8); # ^= tp8^(tp4^tp1)^(tp2^tp1)
1329 &xor ($s[$i],$tp2); # ^= ROTATE(tp8^tp2^tp1,24)
1331 &xor ($s[$i],$tp4); # ^= ROTATE(tp8^tp4^tp1,16)
1332 &mov ($s[0],$__s0) if($i==2); #prefetch $s0
1333 &mov ($s[1],$__s1) if($i==3); #prefetch $s1
1334 &mov ($s[2],$__s2) if($i==1);
1335 &xor ($s[$i],$tp8); # ^= ROTATE(tp8,8)
1337 &mov ($s[3],$__s3) if($i==1);
1338 &mov (&DWP(4+4*$i,"esp"),$s[$i]) if($i>=2);
1341 &function_begin_B("_x86_AES_decrypt_compact");
1342 # note that caller is expected to allocate stack frame for me!
1343 &mov ($__key,$key); # save key
1345 &xor ($s0,&DWP(0,$key)); # xor with key
1346 &xor ($s1,&DWP(4,$key));
1347 &xor ($s2,&DWP(8,$key));
1348 &xor ($s3,&DWP(12,$key));
1350 &mov ($acc,&DWP(240,$key)); # load key->rounds
1352 &lea ($acc,&DWP(-2,$acc,$acc));
1353 &lea ($acc,&DWP(0,$key,$acc,8));
1354 &mov ($__end,$acc); # end of key schedule
1357 &mov ($key,&DWP(0-128,$tbl));
1358 &mov ($acc,&DWP(32-128,$tbl));
1359 &mov ($key,&DWP(64-128,$tbl));
1360 &mov ($acc,&DWP(96-128,$tbl));
1361 &mov ($key,&DWP(128-128,$tbl));
1362 &mov ($acc,&DWP(160-128,$tbl));
1363 &mov ($key,&DWP(192-128,$tbl));
1364 &mov ($acc,&DWP(224-128,$tbl));
1366 &set_label("loop",16);
1368 &deccompact(0,$tbl,$s0,$s3,$s2,$s1,1);
1369 &deccompact(1,$tbl,$s1,$s0,$s3,$s2,1);
1370 &deccompact(2,$tbl,$s2,$s1,$s0,$s3,1);
1371 &deccompact(3,$tbl,$s3,$s2,$s1,$s0,1);
1378 &add ($key,16); # advance rd_key
1379 &xor ($s0,&DWP(0,$key));
1380 &xor ($s1,&DWP(4,$key));
1381 &xor ($s2,&DWP(8,$key));
1382 &xor ($s3,&DWP(12,$key));
1386 &jb (&label("loop"));
1388 &deccompact(0,$tbl,$s0,$s3,$s2,$s1);
1389 &deccompact(1,$tbl,$s1,$s0,$s3,$s2);
1390 &deccompact(2,$tbl,$s2,$s1,$s0,$s3);
1391 &deccompact(3,$tbl,$s3,$s2,$s1,$s0);
1393 &xor ($s0,&DWP(16,$key));
1394 &xor ($s1,&DWP(20,$key));
1395 &xor ($s2,&DWP(24,$key));
1396 &xor ($s3,&DWP(28,$key));
1399 &function_end_B("_x86_AES_decrypt_compact");
1401 ######################################################################
1402 # "Compact" SSE block function.
1403 ######################################################################
1405 sub sse_deccompact()
1407 &pshufw ("mm1","mm0",0x0c); # 7, 6, 1, 0
1408 &pshufw ("mm5","mm4",0x09); # 13,12,11,10
1409 &movd ("eax","mm1"); # 7, 6, 1, 0
1410 &movd ("ebx","mm5"); # 13,12,11,10
1413 &movz ($acc,&LB("eax")); # 0
1414 &movz ("edx",&HB("eax")); # 1
1415 &pshufw ("mm2","mm0",0x06); # 3, 2, 5, 4
1416 &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 0
1417 &movz ($key,&LB("ebx")); # 10
1418 &movz ("edx",&BP(-128,$tbl,"edx",1)); # 1
1419 &shr ("eax",16); # 7, 6
1422 &movz ($acc,&BP(-128,$tbl,$key,1)); # 10
1423 &movz ($key,&HB("ebx")); # 11
1424 &shl ($acc,16); # 10
1425 &pshufw ("mm6","mm4",0x03); # 9, 8,15,14
1426 &or ("ecx",$acc); # 10
1427 &movz ($acc,&BP(-128,$tbl,$key,1)); # 11
1428 &movz ($key,&HB("eax")); # 7
1429 &shl ($acc,24); # 11
1430 &shr ("ebx",16); # 13,12
1431 &or ("edx",$acc); # 11
1433 &movz ($acc,&BP(-128,$tbl,$key,1)); # 7
1434 &movz ($key,&HB("ebx")); # 13
1436 &or ("ecx",$acc); # 7
1437 &movz ($acc,&BP(-128,$tbl,$key,1)); # 13
1438 &movz ($key,&LB("eax")); # 6
1440 &movd ("eax","mm2"); # 3, 2, 5, 4
1441 &or ("ecx",$acc); # 13
1443 &movz ($acc,&BP(-128,$tbl,$key,1)); # 6
1444 &movz ($key,&LB("ebx")); # 12
1446 &movd ("ebx","mm6"); # 9, 8,15,14
1447 &movd ("mm0","ecx"); # t[0] collected
1448 &movz ("ecx",&BP(-128,$tbl,$key,1)); # 12
1449 &movz ($key,&LB("eax")); # 4
1450 &or ("ecx",$acc); # 12
1452 &movz ($acc,&BP(-128,$tbl,$key,1)); # 4
1453 &movz ($key,&LB("ebx")); # 14
1454 &or ("edx",$acc); # 4
1455 &movz ($acc,&BP(-128,$tbl,$key,1)); # 14
1456 &movz ($key,&HB("eax")); # 5
1457 &shl ($acc,16); # 14
1458 &shr ("eax",16); # 3, 2
1459 &or ("edx",$acc); # 14
1461 &movz ($acc,&BP(-128,$tbl,$key,1)); # 5
1462 &movz ($key,&HB("ebx")); # 15
1463 &shr ("ebx",16); # 9, 8
1465 &movd ("mm1","edx"); # t[1] collected
1466 &movz ("edx",&BP(-128,$tbl,$key,1)); # 15
1467 &movz ($key,&HB("ebx")); # 9
1468 &shl ("edx",24); # 15
1469 &and ("ebx",0xff); # 8
1470 &or ("edx",$acc); # 15
1472 &punpckldq ("mm0","mm1"); # t[0,1] collected
1474 &movz ($acc,&BP(-128,$tbl,$key,1)); # 9
1475 &movz ($key,&LB("eax")); # 2
1477 &movz ("eax",&HB("eax")); # 3
1478 &movz ("ebx",&BP(-128,$tbl,"ebx",1)); # 8
1479 &or ("ecx",$acc); # 9
1480 &movz ($acc,&BP(-128,$tbl,$key,1)); # 2
1481 &or ("edx","ebx"); # 8
1483 &movz ("eax",&BP(-128,$tbl,"eax",1)); # 3
1484 &or ("edx",$acc); # 2
1485 &shl ("eax",24); # 3
1486 &or ("ecx","eax"); # 3
1488 &movd ("mm4","edx"); # t[2] collected
1489 &movd ("mm5","ecx"); # t[3] collected
1491 &punpckldq ("mm4","mm5"); # t[2,3] collected
1495 &function_begin_B("_sse_AES_decrypt_compact");
1496 &pxor ("mm0",&QWP(0,$key)); # 7, 6, 5, 4, 3, 2, 1, 0
1497 &pxor ("mm4",&QWP(8,$key)); # 15,14,13,12,11,10, 9, 8
1499 # note that caller is expected to allocate stack frame for me!
1500 &mov ($acc,&DWP(240,$key)); # load key->rounds
1501 &lea ($acc,&DWP(-2,$acc,$acc));
1502 &lea ($acc,&DWP(0,$key,$acc,8));
1503 &mov ($__end,$acc); # end of key schedule
1505 &mov ($s0,0x1b1b1b1b); # magic constant
1506 &mov (&DWP(8,"esp"),$s0);
1507 &mov (&DWP(12,"esp"),$s0);
1510 &mov ($s0,&DWP(0-128,$tbl));
1511 &mov ($s1,&DWP(32-128,$tbl));
1512 &mov ($s2,&DWP(64-128,$tbl));
1513 &mov ($s3,&DWP(96-128,$tbl));
1514 &mov ($s0,&DWP(128-128,$tbl));
1515 &mov ($s1,&DWP(160-128,$tbl));
1516 &mov ($s2,&DWP(192-128,$tbl));
1517 &mov ($s3,&DWP(224-128,$tbl));
1519 &set_label("loop",16);
1523 &ja (&label("out"));
1525 # ROTATE(x^y,N) == ROTATE(x,N)^ROTATE(y,N)
1526 &movq ("mm3","mm0"); &movq ("mm7","mm4");
1527 &movq ("mm2","mm0",1); &movq ("mm6","mm4",1);
1528 &movq ("mm1","mm0"); &movq ("mm5","mm4");
1529 &pshufw ("mm0","mm0",0xb1); &pshufw ("mm4","mm4",0xb1);# = ROTATE(tp0,16)
1530 &pslld ("mm2",8); &pslld ("mm6",8);
1531 &psrld ("mm3",8); &psrld ("mm7",8);
1532 &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= tp0<<8
1533 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp0>>8
1534 &pslld ("mm2",16); &pslld ("mm6",16);
1535 &psrld ("mm3",16); &psrld ("mm7",16);
1536 &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= tp0<<24
1537 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp0>>24
1539 &movq ("mm3",&QWP(8,"esp"));
1540 &pxor ("mm2","mm2"); &pxor ("mm6","mm6");
1541 &pcmpgtb("mm2","mm1"); &pcmpgtb("mm6","mm5");
1542 &pand ("mm2","mm3"); &pand ("mm6","mm3");
1543 &paddb ("mm1","mm1"); &paddb ("mm5","mm5");
1544 &pxor ("mm1","mm2"); &pxor ("mm5","mm6"); # tp2
1545 &movq ("mm3","mm1"); &movq ("mm7","mm5");
1546 &movq ("mm2","mm1"); &movq ("mm6","mm5");
1547 &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp2
1548 &pslld ("mm3",24); &pslld ("mm7",24);
1549 &psrld ("mm2",8); &psrld ("mm6",8);
1550 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp2<<24
1551 &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= tp2>>8
1553 &movq ("mm2",&QWP(8,"esp"));
1554 &pxor ("mm3","mm3"); &pxor ("mm7","mm7");
1555 &pcmpgtb("mm3","mm1"); &pcmpgtb("mm7","mm5");
1556 &pand ("mm3","mm2"); &pand ("mm7","mm2");
1557 &paddb ("mm1","mm1"); &paddb ("mm5","mm5");
1558 &pxor ("mm1","mm3"); &pxor ("mm5","mm7"); # tp4
1559 &pshufw ("mm3","mm1",0xb1); &pshufw ("mm7","mm5",0xb1);
1560 &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp4
1561 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= ROTATE(tp4,16)
1563 &pxor ("mm3","mm3"); &pxor ("mm7","mm7");
1564 &pcmpgtb("mm3","mm1"); &pcmpgtb("mm7","mm5");
1565 &pand ("mm3","mm2"); &pand ("mm7","mm2");
1566 &paddb ("mm1","mm1"); &paddb ("mm5","mm5");
1567 &pxor ("mm1","mm3"); &pxor ("mm5","mm7"); # tp8
1568 &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp8
1569 &movq ("mm3","mm1"); &movq ("mm7","mm5");
1570 &pshufw ("mm2","mm1",0xb1); &pshufw ("mm6","mm5",0xb1);
1571 &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= ROTATE(tp8,16)
1572 &pslld ("mm1",8); &pslld ("mm5",8);
1573 &psrld ("mm3",8); &psrld ("mm7",8);
1574 &movq ("mm2",&QWP(0,$key)); &movq ("mm6",&QWP(8,$key));
1575 &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp8<<8
1576 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp8>>8
1577 &mov ($s0,&DWP(0-128,$tbl));
1578 &pslld ("mm1",16); &pslld ("mm5",16);
1579 &mov ($s1,&DWP(64-128,$tbl));
1580 &psrld ("mm3",16); &psrld ("mm7",16);
1581 &mov ($s2,&DWP(128-128,$tbl));
1582 &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp8<<24
1583 &mov ($s3,&DWP(192-128,$tbl));
1584 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp8>>24
1586 &pxor ("mm0","mm2"); &pxor ("mm4","mm6");
1587 &jmp (&label("loop"));
1589 &set_label("out",16);
1590 &pxor ("mm0",&QWP(0,$key));
1591 &pxor ("mm4",&QWP(8,$key));
1594 &function_end_B("_sse_AES_decrypt_compact");
1597 ######################################################################
1598 # Vanilla block function.
1599 ######################################################################
1602 { my ($i,$td,@s) = @_;
1604 my $out = $i==3?$s[0]:$acc;
1606 # no instructions are reordered, as performance appears
1607 # optimal... or rather that all attempts to reorder didn't
1608 # result in better performance [which by the way is not a
1609 # bit lower than ecryption].
1610 if($i==3) { &mov ($key,$__key); }
1611 else { &mov ($out,$s[0]); }
1613 &mov ($out,&DWP(0,$td,$out,8));
1615 if ($i==3) { $tmp=$s[1]; }
1616 &movz ($tmp,&HB($s[1]));
1617 &xor ($out,&DWP(3,$td,$tmp,8));
1619 if ($i==3) { $tmp=$s[2]; &mov ($s[1],$acc); }
1620 else { &mov ($tmp,$s[2]); }
1623 &xor ($out,&DWP(2,$td,$tmp,8));
1625 if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }
1626 else { &mov ($tmp,$s[3]); }
1628 &xor ($out,&DWP(1,$td,$tmp,8));
1629 if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); }
1630 if ($i==3) { &mov ($s[3],$__s0); }
1635 { my ($i,$td,@s)=@_;
1637 my $out = $i==3?$s[0]:$acc;
1639 if($i==0) { &lea ($td,&DWP(2048+128,$td));
1640 &mov ($tmp,&DWP(0-128,$td));
1641 &mov ($acc,&DWP(32-128,$td));
1642 &mov ($tmp,&DWP(64-128,$td));
1643 &mov ($acc,&DWP(96-128,$td));
1644 &mov ($tmp,&DWP(128-128,$td));
1645 &mov ($acc,&DWP(160-128,$td));
1646 &mov ($tmp,&DWP(192-128,$td));
1647 &mov ($acc,&DWP(224-128,$td));
1648 &lea ($td,&DWP(-128,$td)); }
1649 if($i==3) { &mov ($key,$__key); }
1650 else { &mov ($out,$s[0]); }
1652 &movz ($out,&BP(0,$td,$out,1));
1654 if ($i==3) { $tmp=$s[1]; }
1655 &movz ($tmp,&HB($s[1]));
1656 &movz ($tmp,&BP(0,$td,$tmp,1));
1660 if ($i==3) { $tmp=$s[2]; &mov ($s[1],$acc); }
1661 else { mov ($tmp,$s[2]); }
1664 &movz ($tmp,&BP(0,$td,$tmp,1));
1668 if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }
1669 else { &mov ($tmp,$s[3]); }
1671 &movz ($tmp,&BP(0,$td,$tmp,1));
1674 if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); }
1675 if ($i==3) { &mov ($s[3],$__s0);
1676 &lea ($td,&DWP(-2048,$td)); }
1679 &function_begin_B("_x86_AES_decrypt");
1680 # note that caller is expected to allocate stack frame for me!
1681 &mov ($__key,$key); # save key
1683 &xor ($s0,&DWP(0,$key)); # xor with key
1684 &xor ($s1,&DWP(4,$key));
1685 &xor ($s2,&DWP(8,$key));
1686 &xor ($s3,&DWP(12,$key));
1688 &mov ($acc,&DWP(240,$key)); # load key->rounds
1690 if ($small_footprint) {
1691 &lea ($acc,&DWP(-2,$acc,$acc));
1692 &lea ($acc,&DWP(0,$key,$acc,8));
1693 &mov ($__end,$acc); # end of key schedule
1694 &set_label("loop",16);
1695 &decstep(0,$tbl,$s0,$s3,$s2,$s1);
1696 &decstep(1,$tbl,$s1,$s0,$s3,$s2);
1697 &decstep(2,$tbl,$s2,$s1,$s0,$s3);
1698 &decstep(3,$tbl,$s3,$s2,$s1,$s0);
1699 &add ($key,16); # advance rd_key
1700 &xor ($s0,&DWP(0,$key));
1701 &xor ($s1,&DWP(4,$key));
1702 &xor ($s2,&DWP(8,$key));
1703 &xor ($s3,&DWP(12,$key));
1706 &jb (&label("loop"));
1710 &jle (&label("10rounds"));
1712 &jle (&label("12rounds"));
1714 &set_label("14rounds",4);
1715 for ($i=1;$i<3;$i++) {
1716 &decstep(0,$tbl,$s0,$s3,$s2,$s1);
1717 &decstep(1,$tbl,$s1,$s0,$s3,$s2);
1718 &decstep(2,$tbl,$s2,$s1,$s0,$s3);
1719 &decstep(3,$tbl,$s3,$s2,$s1,$s0);
1720 &xor ($s0,&DWP(16*$i+0,$key));
1721 &xor ($s1,&DWP(16*$i+4,$key));
1722 &xor ($s2,&DWP(16*$i+8,$key));
1723 &xor ($s3,&DWP(16*$i+12,$key));
1726 &mov ($__key,$key); # advance rd_key
1727 &set_label("12rounds",4);
1728 for ($i=1;$i<3;$i++) {
1729 &decstep(0,$tbl,$s0,$s3,$s2,$s1);
1730 &decstep(1,$tbl,$s1,$s0,$s3,$s2);
1731 &decstep(2,$tbl,$s2,$s1,$s0,$s3);
1732 &decstep(3,$tbl,$s3,$s2,$s1,$s0);
1733 &xor ($s0,&DWP(16*$i+0,$key));
1734 &xor ($s1,&DWP(16*$i+4,$key));
1735 &xor ($s2,&DWP(16*$i+8,$key));
1736 &xor ($s3,&DWP(16*$i+12,$key));
1739 &mov ($__key,$key); # advance rd_key
1740 &set_label("10rounds",4);
1741 for ($i=1;$i<10;$i++) {
1742 &decstep(0,$tbl,$s0,$s3,$s2,$s1);
1743 &decstep(1,$tbl,$s1,$s0,$s3,$s2);
1744 &decstep(2,$tbl,$s2,$s1,$s0,$s3);
1745 &decstep(3,$tbl,$s3,$s2,$s1,$s0);
1746 &xor ($s0,&DWP(16*$i+0,$key));
1747 &xor ($s1,&DWP(16*$i+4,$key));
1748 &xor ($s2,&DWP(16*$i+8,$key));
1749 &xor ($s3,&DWP(16*$i+12,$key));
1753 &declast(0,$tbl,$s0,$s3,$s2,$s1);
1754 &declast(1,$tbl,$s1,$s0,$s3,$s2);
1755 &declast(2,$tbl,$s2,$s1,$s0,$s3);
1756 &declast(3,$tbl,$s3,$s2,$s1,$s0);
1758 &add ($key,$small_footprint?16:160);
1759 &xor ($s0,&DWP(0,$key));
1760 &xor ($s1,&DWP(4,$key));
1761 &xor ($s2,&DWP(8,$key));
1762 &xor ($s3,&DWP(12,$key));
1766 &set_label("AES_Td",64); # Yes! I keep it in the code segment!
1767 &_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a);
1768 &_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b);
1769 &_data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5);
1770 &_data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5);
1771 &_data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d);
1772 &_data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b);
1773 &_data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295);
1774 &_data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e);
1775 &_data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927);
1776 &_data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d);
1777 &_data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362);
1778 &_data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9);
1779 &_data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52);
1780 &_data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566);
1781 &_data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3);
1782 &_data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed);
1783 &_data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e);
1784 &_data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4);
1785 &_data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4);
1786 &_data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd);
1787 &_data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d);
1788 &_data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060);
1789 &_data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967);
1790 &_data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879);
1791 &_data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000);
1792 &_data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c);
1793 &_data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36);
1794 &_data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624);
1795 &_data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b);
1796 &_data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c);
1797 &_data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12);
1798 &_data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14);
1799 &_data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3);
1800 &_data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b);
1801 &_data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8);
1802 &_data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684);
1803 &_data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7);
1804 &_data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177);
1805 &_data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947);
1806 &_data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322);
1807 &_data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498);
1808 &_data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f);
1809 &_data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54);
1810 &_data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382);
1811 &_data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf);
1812 &_data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb);
1813 &_data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83);
1814 &_data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef);
1815 &_data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029);
1816 &_data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235);
1817 &_data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733);
1818 &_data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117);
1819 &_data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4);
1820 &_data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546);
1821 &_data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb);
1822 &_data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d);
1823 &_data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb);
1824 &_data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a);
1825 &_data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773);
1826 &_data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478);
1827 &_data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2);
1828 &_data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff);
1829 &_data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664);
1830 &_data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0);
1832 #Td4: # four copies of Td4 to choose from to avoid L1 aliasing
1833 &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
1834 &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
1835 &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
1836 &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
1837 &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
1838 &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
1839 &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
1840 &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
1841 &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
1842 &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
1843 &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
1844 &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
1845 &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
1846 &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
1847 &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
1848 &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
1849 &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
1850 &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
1851 &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
1852 &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
1853 &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
1854 &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
1855 &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
1856 &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
1857 &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
1858 &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
1859 &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
1860 &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
1861 &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
1862 &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
1863 &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
1864 &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
1866 &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
1867 &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
1868 &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
1869 &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
1870 &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
1871 &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
1872 &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
1873 &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
1874 &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
1875 &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
1876 &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
1877 &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
1878 &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
1879 &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
1880 &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
1881 &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
1882 &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
1883 &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
1884 &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
1885 &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
1886 &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
1887 &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
1888 &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
1889 &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
1890 &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
1891 &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
1892 &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
1893 &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
1894 &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
1895 &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
1896 &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
1897 &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
1899 &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
1900 &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
1901 &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
1902 &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
1903 &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
1904 &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
1905 &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
1906 &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
1907 &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
1908 &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
1909 &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
1910 &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
1911 &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
1912 &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
1913 &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
1914 &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
1915 &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
1916 &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
1917 &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
1918 &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
1919 &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
1920 &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
1921 &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
1922 &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
1923 &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
1924 &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
1925 &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
1926 &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
1927 &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
1928 &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
1929 &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
1930 &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
1932 &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
1933 &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
1934 &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
1935 &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
1936 &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
1937 &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
1938 &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
1939 &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
1940 &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
1941 &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
1942 &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
1943 &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
1944 &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
1945 &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
1946 &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
1947 &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
1948 &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
1949 &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
1950 &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
1951 &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
1952 &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
1953 &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
1954 &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
1955 &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
1956 &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
1957 &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
1958 &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
1959 &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
1960 &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
1961 &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
1962 &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
1963 &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
1964 &function_end_B("_x86_AES_decrypt");
1966 # void AES_decrypt (const void *inp,void *out,const AES_KEY *key);
1967 &function_begin("AES_decrypt");
1968 &mov ($acc,&wparam(0)); # load inp
1969 &mov ($key,&wparam(2)); # load key
1973 &and ("esp",-64); # align to cache-line
1975 # place stack frame just "above" the key schedule
1976 &lea ($s1,&DWP(-64-63,$key));
1979 &and ($s1,0x3C0); # modulo 1024, but aligned to cache-line
1981 &add ("esp",4); # 4 is reserved for caller's return address
1982 &mov ($_esp,$s0); # save stack pointer
1984 &call (&label("pic_point")); # make it PIC!
1985 &set_label("pic_point");
1987 &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only);
1988 &lea ($tbl,&DWP(&label("AES_Td")."-".&label("pic_point"),$tbl));
1990 # pick Td4 copy which can't "overlap" with stack frame or key schedule
1991 &lea ($s1,&DWP(768-4,"esp"));
1994 &lea ($tbl,&DWP(2048+128,$tbl,$s1));
1997 &bt (&DWP(0,$s0),25); # check for SSE bit
1998 &jnc (&label("x86"));
2000 &movq ("mm0",&QWP(0,$acc));
2001 &movq ("mm4",&QWP(8,$acc));
2002 &call ("_sse_AES_decrypt_compact");
2003 &mov ("esp",$_esp); # restore stack pointer
2004 &mov ($acc,&wparam(1)); # load out
2005 &movq (&QWP(0,$acc),"mm0"); # write output data
2006 &movq (&QWP(8,$acc),"mm4");
2010 &set_label("x86",16);
2012 &mov ($s0,&DWP(0,$acc)); # load input data
2013 &mov ($s1,&DWP(4,$acc));
2014 &mov ($s2,&DWP(8,$acc));
2015 &mov ($s3,&DWP(12,$acc));
2016 &call ("_x86_AES_decrypt_compact");
2017 &mov ("esp",$_esp); # restore stack pointer
2018 &mov ($acc,&wparam(1)); # load out
2019 &mov (&DWP(0,$acc),$s0); # write output data
2020 &mov (&DWP(4,$acc),$s1);
2021 &mov (&DWP(8,$acc),$s2);
2022 &mov (&DWP(12,$acc),$s3);
2023 &function_end("AES_decrypt");
2025 # void AES_cbc_encrypt (const void char *inp, unsigned char *out,
2026 # size_t length, const AES_KEY *key,
2027 # unsigned char *ivp,const int enc);
2029 # stack frame layout
2030 # -4(%esp) # return address 0(%esp)
2031 # 0(%esp) # s0 backing store 4(%esp)
2032 # 4(%esp) # s1 backing store 8(%esp)
2033 # 8(%esp) # s2 backing store 12(%esp)
2034 # 12(%esp) # s3 backing store 16(%esp)
2035 # 16(%esp) # key backup 20(%esp)
2036 # 20(%esp) # end of key schedule 24(%esp)
2037 # 24(%esp) # %ebp backup 28(%esp)
2038 # 28(%esp) # %esp backup
2039 my $_inp=&DWP(32,"esp"); # copy of wparam(0)
2040 my $_out=&DWP(36,"esp"); # copy of wparam(1)
2041 my $_len=&DWP(40,"esp"); # copy of wparam(2)
2042 my $_key=&DWP(44,"esp"); # copy of wparam(3)
2043 my $_ivp=&DWP(48,"esp"); # copy of wparam(4)
2044 my $_tmp=&DWP(52,"esp"); # volatile variable
2046 my $ivec=&DWP(60,"esp"); # ivec[16]
2047 my $aes_key=&DWP(76,"esp"); # copy of aes_key
2048 my $mark=&DWP(76+240,"esp"); # copy of aes_key->rounds
2050 &function_begin("AES_cbc_encrypt");
2051 &mov ($s2 eq "ecx"? $s2 : "",&wparam(2)); # load len
2053 &je (&label("drop_out"));
2055 &call (&label("pic_point")); # make it PIC!
2056 &set_label("pic_point");
2058 &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only);
2060 &cmp (&wparam(5),0);
2061 &lea ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
2062 &jne (&label("picked_te"));
2063 &lea ($tbl,&DWP(&label("AES_Td")."-".&label("AES_Te"),$tbl));
2064 &set_label("picked_te");
2066 # one can argue if this is required
2070 &cmp ($s2,$speed_limit);
2071 &jb (&label("slow_way"));
2073 &jnz (&label("slow_way"));
2075 &bt (&DWP(0,$s0),28); # check for hyper-threading bit
2076 &jc (&label("slow_way"));
2078 # pre-allocate aligned stack frame...
2079 &lea ($acc,&DWP(-80-244,"esp"));
2082 # ... and make sure it doesn't alias with $tbl modulo 4096
2084 &lea ($s1,&DWP(2048+256,$tbl));
2086 &and ($s0,0xfff); # s = %ebp&0xfff
2087 &and ($s1,0xfff); # e = (%ebp+2048+256)&0xfff
2088 &and ($s3,0xfff); # p = %esp&0xfff
2090 &cmp ($s3,$s1); # if (p>=e) %esp =- (p-e);
2091 &jb (&label("tbl_break_out"));
2094 &jmp (&label("tbl_ok"));
2095 &set_label("tbl_break_out",4); # else %esp -= (p-s)&0xfff + framesz;
2100 &set_label("tbl_ok",4);
2102 &lea ($s3,&wparam(0)); # obtain pointer to parameter block
2103 &exch ("esp",$acc); # allocate stack frame
2104 &add ("esp",4); # reserve for return address!
2105 &mov ($_tbl,$tbl); # save %ebp
2106 &mov ($_esp,$acc); # save %esp
2108 &mov ($s0,&DWP(0,$s3)); # load inp
2109 &mov ($s1,&DWP(4,$s3)); # load out
2110 #&mov ($s2,&DWP(8,$s3)); # load len
2111 &mov ($key,&DWP(12,$s3)); # load key
2112 &mov ($acc,&DWP(16,$s3)); # load ivp
2113 &mov ($s3,&DWP(20,$s3)); # load enc flag
2115 &mov ($_inp,$s0); # save copy of inp
2116 &mov ($_out,$s1); # save copy of out
2117 &mov ($_len,$s2); # save copy of len
2118 &mov ($_key,$key); # save copy of key
2119 &mov ($_ivp,$acc); # save copy of ivp
2121 &mov ($mark,0); # copy of aes_key->rounds = 0;
2122 # do we copy key schedule to stack?
2123 &mov ($s1 eq "ebx" ? $s1 : "",$key);
2124 &mov ($s2 eq "ecx" ? $s2 : "",244/4);
2128 &lea ("edi",$aes_key);
2129 &cmp ($s1,2048+256);
2130 &jb (&label("do_copy"));
2131 &cmp ($s1,4096-244);
2132 &jb (&label("skip_copy"));
2133 &set_label("do_copy",4);
2135 &data_word(0xA5F3F689); # rep movsd
2136 &set_label("skip_copy");
2139 &set_label("prefetch_tbl",4);
2140 &mov ($s0,&DWP(0,$tbl));
2141 &mov ($s1,&DWP(32,$tbl));
2142 &mov ($s2,&DWP(64,$tbl));
2143 &mov ($acc,&DWP(96,$tbl));
2144 &lea ($tbl,&DWP(128,$tbl));
2146 &jnz (&label("prefetch_tbl"));
2153 &je (&label("fast_decrypt"));
2155 #----------------------------- ENCRYPT -----------------------------#
2156 &mov ($s0,&DWP(0,$key)); # load iv
2157 &mov ($s1,&DWP(4,$key));
2159 &set_label("fast_enc_loop",16);
2160 &mov ($s2,&DWP(8,$key));
2161 &mov ($s3,&DWP(12,$key));
2163 &xor ($s0,&DWP(0,$acc)); # xor input data
2164 &xor ($s1,&DWP(4,$acc));
2165 &xor ($s2,&DWP(8,$acc));
2166 &xor ($s3,&DWP(12,$acc));
2168 &mov ($key,$_key); # load key
2169 &call ("_x86_AES_encrypt");
2171 &mov ($acc,$_inp); # load inp
2172 &mov ($key,$_out); # load out
2174 &mov (&DWP(0,$key),$s0); # save output data
2175 &mov (&DWP(4,$key),$s1);
2176 &mov (&DWP(8,$key),$s2);
2177 &mov (&DWP(12,$key),$s3);
2179 &lea ($acc,&DWP(16,$acc)); # advance inp
2180 &mov ($s2,$_len); # load len
2181 &mov ($_inp,$acc); # save inp
2182 &lea ($s3,&DWP(16,$key)); # advance out
2183 &mov ($_out,$s3); # save out
2184 &sub ($s2,16); # decrease len
2185 &mov ($_len,$s2); # save len
2186 &jnz (&label("fast_enc_loop"));
2187 &mov ($acc,$_ivp); # load ivp
2188 &mov ($s2,&DWP(8,$key)); # restore last 2 dwords
2189 &mov ($s3,&DWP(12,$key));
2190 &mov (&DWP(0,$acc),$s0); # save ivec
2191 &mov (&DWP(4,$acc),$s1);
2192 &mov (&DWP(8,$acc),$s2);
2193 &mov (&DWP(12,$acc),$s3);
2195 &cmp ($mark,0); # was the key schedule copied?
2197 &je (&label("skip_ezero"));
2198 # zero copy of key schedule
2202 &data_word(0xABF3F689); # rep stosd
2203 &set_label("skip_ezero");
2206 &set_label("drop_out");
2208 &pushf (); # kludge, never executed
2210 #----------------------------- DECRYPT -----------------------------#
2211 &set_label("fast_decrypt",16);
2214 &je (&label("fast_dec_in_place")); # in-place processing...
2219 &set_label("fast_dec_loop",16);
2220 &mov ($s0,&DWP(0,$acc)); # read input
2221 &mov ($s1,&DWP(4,$acc));
2222 &mov ($s2,&DWP(8,$acc));
2223 &mov ($s3,&DWP(12,$acc));
2225 &mov ($key,$_key); # load key
2226 &call ("_x86_AES_decrypt");
2228 &mov ($key,$_tmp); # load ivp
2229 &mov ($acc,$_len); # load len
2230 &xor ($s0,&DWP(0,$key)); # xor iv
2231 &xor ($s1,&DWP(4,$key));
2232 &xor ($s2,&DWP(8,$key));
2233 &xor ($s3,&DWP(12,$key));
2235 &mov ($key,$_out); # load out
2236 &mov ($acc,$_inp); # load inp
2238 &mov (&DWP(0,$key),$s0); # write output
2239 &mov (&DWP(4,$key),$s1);
2240 &mov (&DWP(8,$key),$s2);
2241 &mov (&DWP(12,$key),$s3);
2243 &mov ($s2,$_len); # load len
2244 &mov ($_tmp,$acc); # save ivp
2245 &lea ($acc,&DWP(16,$acc)); # advance inp
2246 &mov ($_inp,$acc); # save inp
2247 &lea ($key,&DWP(16,$key)); # advance out
2248 &mov ($_out,$key); # save out
2249 &sub ($s2,16); # decrease len
2250 &mov ($_len,$s2); # save len
2251 &jnz (&label("fast_dec_loop"));
2252 &mov ($key,$_tmp); # load temp ivp
2253 &mov ($acc,$_ivp); # load user ivp
2254 &mov ($s0,&DWP(0,$key)); # load iv
2255 &mov ($s1,&DWP(4,$key));
2256 &mov ($s2,&DWP(8,$key));
2257 &mov ($s3,&DWP(12,$key));
2258 &mov (&DWP(0,$acc),$s0); # copy back to user
2259 &mov (&DWP(4,$acc),$s1);
2260 &mov (&DWP(8,$acc),$s2);
2261 &mov (&DWP(12,$acc),$s3);
2262 &jmp (&label("fast_dec_out"));
2264 &set_label("fast_dec_in_place",16);
2265 &set_label("fast_dec_in_place_loop");
2266 &mov ($s0,&DWP(0,$acc)); # read input
2267 &mov ($s1,&DWP(4,$acc));
2268 &mov ($s2,&DWP(8,$acc));
2269 &mov ($s3,&DWP(12,$acc));
2272 &mov (&DWP(0,$key),$s0); # copy to temp
2273 &mov (&DWP(4,$key),$s1);
2274 &mov (&DWP(8,$key),$s2);
2275 &mov (&DWP(12,$key),$s3);
2277 &mov ($key,$_key); # load key
2278 &call ("_x86_AES_decrypt");
2280 &mov ($key,$_ivp); # load ivp
2281 &mov ($acc,$_out); # load out
2282 &xor ($s0,&DWP(0,$key)); # xor iv
2283 &xor ($s1,&DWP(4,$key));
2284 &xor ($s2,&DWP(8,$key));
2285 &xor ($s3,&DWP(12,$key));
2287 &mov (&DWP(0,$acc),$s0); # write output
2288 &mov (&DWP(4,$acc),$s1);
2289 &mov (&DWP(8,$acc),$s2);
2290 &mov (&DWP(12,$acc),$s3);
2292 &lea ($acc,&DWP(16,$acc)); # advance out
2293 &mov ($_out,$acc); # save out
2296 &mov ($s0,&DWP(0,$acc)); # read temp
2297 &mov ($s1,&DWP(4,$acc));
2298 &mov ($s2,&DWP(8,$acc));
2299 &mov ($s3,&DWP(12,$acc));
2301 &mov (&DWP(0,$key),$s0); # copy iv
2302 &mov (&DWP(4,$key),$s1);
2303 &mov (&DWP(8,$key),$s2);
2304 &mov (&DWP(12,$key),$s3);
2306 &mov ($acc,$_inp); # load inp
2307 &mov ($s2,$_len); # load len
2308 &lea ($acc,&DWP(16,$acc)); # advance inp
2309 &mov ($_inp,$acc); # save inp
2310 &sub ($s2,16); # decrease len
2311 &mov ($_len,$s2); # save len
2312 &jnz (&label("fast_dec_in_place_loop"));
2314 &set_label("fast_dec_out",4);
2315 &cmp ($mark,0); # was the key schedule copied?
2317 &je (&label("skip_dzero"));
2318 # zero copy of key schedule
2322 &data_word(0xABF3F689); # rep stosd
2323 &set_label("skip_dzero");
2327 &pushf (); # kludge, never executed
2329 #--------------------------- SLOW ROUTINE ---------------------------#
2330 &set_label("slow_way",16);
2332 &mov ($s0,&DWP(0,$s0)) if (!$x86only);# load OPENSSL_ia32cap
2333 &mov ($key,&wparam(3)); # load key
2335 # pre-allocate aligned stack frame...
2336 &lea ($acc,&DWP(-80,"esp"));
2339 # ... and make sure it doesn't alias with $key modulo 1024
2340 &lea ($s1,&DWP(-80-63,$key));
2343 &and ($s1,0x3C0); # modulo 1024, but aligned to cache-line
2346 # pick S-box copy which can't overlap with stack frame or $key
2347 &lea ($s1,&DWP(768,$acc));
2350 &lea ($tbl,&DWP(2048+128,$tbl,$s1));
2352 &lea ($s3,&wparam(0)); # pointer to parameter block
2355 &add ("esp",4); # reserve for return address!
2356 &mov ($_tbl,$tbl); # save %ebp
2357 &mov ($_esp,$acc); # save %esp
2358 &mov ($_tmp,$s0); # save OPENSSL_ia32cap
2360 &mov ($s0,&DWP(0,$s3)); # load inp
2361 &mov ($s1,&DWP(4,$s3)); # load out
2362 #&mov ($s2,&DWP(8,$s3)); # load len
2363 #&mov ($key,&DWP(12,$s3)); # load key
2364 &mov ($acc,&DWP(16,$s3)); # load ivp
2365 &mov ($s3,&DWP(20,$s3)); # load enc flag
2367 &mov ($_inp,$s0); # save copy of inp
2368 &mov ($_out,$s1); # save copy of out
2369 &mov ($_len,$s2); # save copy of len
2370 &mov ($_key,$key); # save copy of key
2371 &mov ($_ivp,$acc); # save copy of ivp
2377 &je (&label("slow_decrypt"));
2379 #--------------------------- SLOW ENCRYPT ---------------------------#
2382 &jb (&label("slow_enc_tail"));
2385 &bt ($_tmp,25); # check for SSE bit
2386 &jnc (&label("slow_enc_x86"));
2388 &movq ("mm0",&QWP(0,$key)); # load iv
2389 &movq ("mm4",&QWP(8,$key));
2391 &set_label("slow_enc_loop_sse",16);
2392 &pxor ("mm0",&QWP(0,$acc)); # xor input data
2393 &pxor ("mm4",&QWP(8,$acc));
2396 &call ("_sse_AES_encrypt_compact");
2398 &mov ($acc,$_inp); # load inp
2399 &mov ($key,$_out); # load out
2400 &mov ($s2,$_len); # load len
2402 &movq (&QWP(0,$key),"mm0"); # save output data
2403 &movq (&QWP(8,$key),"mm4");
2405 &lea ($acc,&DWP(16,$acc)); # advance inp
2406 &mov ($_inp,$acc); # save inp
2407 &lea ($s3,&DWP(16,$key)); # advance out
2408 &mov ($_out,$s3); # save out
2409 &sub ($s2,16); # decrease len
2411 &mov ($_len,$s2); # save len
2412 &jae (&label("slow_enc_loop_sse"));
2414 &jnz (&label("slow_enc_tail"));
2415 &mov ($acc,$_ivp); # load ivp
2416 &movq (&QWP(0,$acc),"mm0"); # save ivec
2417 &movq (&QWP(8,$acc),"mm4");
2422 &pushf (); # kludge, never executed
2424 &set_label("slow_enc_x86",16);
2425 &mov ($s0,&DWP(0,$key)); # load iv
2426 &mov ($s1,&DWP(4,$key));
2428 &set_label("slow_enc_loop_x86",4);
2429 &mov ($s2,&DWP(8,$key));
2430 &mov ($s3,&DWP(12,$key));
2432 &xor ($s0,&DWP(0,$acc)); # xor input data
2433 &xor ($s1,&DWP(4,$acc));
2434 &xor ($s2,&DWP(8,$acc));
2435 &xor ($s3,&DWP(12,$acc));
2437 &mov ($key,$_key); # load key
2438 &call ("_x86_AES_encrypt_compact");
2440 &mov ($acc,$_inp); # load inp
2441 &mov ($key,$_out); # load out
2443 &mov (&DWP(0,$key),$s0); # save output data
2444 &mov (&DWP(4,$key),$s1);
2445 &mov (&DWP(8,$key),$s2);
2446 &mov (&DWP(12,$key),$s3);
2448 &mov ($s2,$_len); # load len
2449 &lea ($acc,&DWP(16,$acc)); # advance inp
2450 &mov ($_inp,$acc); # save inp
2451 &lea ($s3,&DWP(16,$key)); # advance out
2452 &mov ($_out,$s3); # save out
2453 &sub ($s2,16); # decrease len
2455 &mov ($_len,$s2); # save len
2456 &jae (&label("slow_enc_loop_x86"));
2458 &jnz (&label("slow_enc_tail"));
2459 &mov ($acc,$_ivp); # load ivp
2460 &mov ($s2,&DWP(8,$key)); # restore last dwords
2461 &mov ($s3,&DWP(12,$key));
2462 &mov (&DWP(0,$acc),$s0); # save ivec
2463 &mov (&DWP(4,$acc),$s1);
2464 &mov (&DWP(8,$acc),$s2);
2465 &mov (&DWP(12,$acc),$s3);
2470 &pushf (); # kludge, never executed
2472 &set_label("slow_enc_tail",16);
2473 &emms () if (!$x86only);
2474 &mov ($key eq "edi"? $key:"",$s3); # load out to edi
2477 &cmp ($key,$acc eq "esi"? $acc:""); # compare with inp
2478 &je (&label("enc_in_place"));
2480 &data_word(0xA4F3F689); # rep movsb # copy input
2481 &jmp (&label("enc_skip_in_place"));
2482 &set_label("enc_in_place");
2483 &lea ($key,&DWP(0,$key,$s2));
2484 &set_label("enc_skip_in_place");
2488 &data_word(0xAAF3F689); # rep stosb # zero tail
2490 &mov ($key,$_ivp); # restore ivp
2491 &mov ($acc,$s3); # output as input
2492 &mov ($s0,&DWP(0,$key));
2493 &mov ($s1,&DWP(4,$key));
2494 &mov ($_len,16); # len=16
2495 &jmp (&label("slow_enc_loop_x86")); # one more spin...
2497 #--------------------------- SLOW DECRYPT ---------------------------#
2498 &set_label("slow_decrypt",16);
2500 &bt ($_tmp,25); # check for SSE bit
2501 &jnc (&label("slow_dec_loop_x86"));
2503 &set_label("slow_dec_loop_sse",4);
2504 &movq ("mm0",&QWP(0,$acc)); # read input
2505 &movq ("mm4",&QWP(8,$acc));
2508 &call ("_sse_AES_decrypt_compact");
2510 &mov ($acc,$_inp); # load inp
2512 &mov ($s1,$_out); # load out
2513 &mov ($s2,$_len); # load len
2514 &mov ($key,$_ivp); # load ivp
2516 &movq ("mm1",&QWP(0,$acc)); # re-read input
2517 &movq ("mm5",&QWP(8,$acc));
2519 &pxor ("mm0",&QWP(0,$key)); # xor iv
2520 &pxor ("mm4",&QWP(8,$key));
2522 &movq (&QWP(0,$key),"mm1"); # copy input to iv
2523 &movq (&QWP(8,$key),"mm5");
2525 &sub ($s2,16); # decrease len
2526 &jc (&label("slow_dec_partial_sse"));
2528 &movq (&QWP(0,$s1),"mm0"); # write output
2529 &movq (&QWP(8,$s1),"mm4");
2531 &lea ($s1,&DWP(16,$s1)); # advance out
2532 &mov ($_out,$s1); # save out
2533 &lea ($acc,&DWP(16,$acc)); # advance inp
2534 &mov ($_inp,$acc); # save inp
2535 &mov ($_len,$s2); # save len
2536 &jnz (&label("slow_dec_loop_sse"));
2541 &pushf (); # kludge, never executed
2543 &set_label("slow_dec_partial_sse",16);
2544 &movq (&QWP(0,$s0),"mm0"); # save output to temp
2545 &movq (&QWP(8,$s0),"mm4");
2548 &add ($s2 eq "ecx" ? "ecx":"",16);
2549 &mov ("edi",$s1); # out
2550 &mov ("esi",$s0); # temp
2552 &data_word(0xA4F3F689); # rep movsb # copy partial output
2557 &pushf (); # kludge, never executed
2559 &set_label("slow_dec_loop_x86",16);
2560 &mov ($s0,&DWP(0,$acc)); # read input
2561 &mov ($s1,&DWP(4,$acc));
2562 &mov ($s2,&DWP(8,$acc));
2563 &mov ($s3,&DWP(12,$acc));
2566 &mov (&DWP(0,$key),$s0); # copy to temp
2567 &mov (&DWP(4,$key),$s1);
2568 &mov (&DWP(8,$key),$s2);
2569 &mov (&DWP(12,$key),$s3);
2571 &mov ($key,$_key); # load key
2572 &call ("_x86_AES_decrypt_compact");
2574 &mov ($key,$_ivp); # load ivp
2575 &mov ($acc,$_len); # load len
2576 &xor ($s0,&DWP(0,$key)); # xor iv
2577 &xor ($s1,&DWP(4,$key));
2578 &xor ($s2,&DWP(8,$key));
2579 &xor ($s3,&DWP(12,$key));
2582 &jc (&label("slow_dec_partial_x86"));
2584 &mov ($_len,$acc); # save len
2585 &mov ($acc,$_out); # load out
2587 &mov (&DWP(0,$acc),$s0); # write output
2588 &mov (&DWP(4,$acc),$s1);
2589 &mov (&DWP(8,$acc),$s2);
2590 &mov (&DWP(12,$acc),$s3);
2592 &lea ($acc,&DWP(16,$acc)); # advance out
2593 &mov ($_out,$acc); # save out
2596 &mov ($s0,&DWP(0,$acc)); # read temp
2597 &mov ($s1,&DWP(4,$acc));
2598 &mov ($s2,&DWP(8,$acc));
2599 &mov ($s3,&DWP(12,$acc));
2601 &mov (&DWP(0,$key),$s0); # copy it to iv
2602 &mov (&DWP(4,$key),$s1);
2603 &mov (&DWP(8,$key),$s2);
2604 &mov (&DWP(12,$key),$s3);
2606 &mov ($acc,$_inp); # load inp
2607 &lea ($acc,&DWP(16,$acc)); # advance inp
2608 &mov ($_inp,$acc); # save inp
2609 &jnz (&label("slow_dec_loop_x86"));
2613 &pushf (); # kludge, never executed
2615 &set_label("slow_dec_partial_x86",16);
2617 &mov (&DWP(0,$acc),$s0); # save output to temp
2618 &mov (&DWP(4,$acc),$s1);
2619 &mov (&DWP(8,$acc),$s2);
2620 &mov (&DWP(12,$acc),$s3);
2623 &mov ($s0,&DWP(0,$acc)); # re-read input
2624 &mov ($s1,&DWP(4,$acc));
2625 &mov ($s2,&DWP(8,$acc));
2626 &mov ($s3,&DWP(12,$acc));
2628 &mov (&DWP(0,$key),$s0); # copy it to iv
2629 &mov (&DWP(4,$key),$s1);
2630 &mov (&DWP(8,$key),$s2);
2631 &mov (&DWP(12,$key),$s3);
2637 &data_word(0xA4F3F689); # rep movsb # copy partial output
2641 &function_end("AES_cbc_encrypt");
2644 #------------------------------------------------------------------#
2648 &movz ("esi",&LB("edx")); # rk[i]>>0
2649 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2650 &movz ("esi",&HB("edx")); # rk[i]>>8
2654 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2656 &movz ("esi",&LB("edx")); # rk[i]>>16
2659 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2660 &movz ("esi",&HB("edx")); # rk[i]>>24
2664 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2668 &xor ("eax",&DWP(1024-128,$tbl,"ecx",4)); # rcon
2671 &function_begin("_x86_AES_set_encrypt_key");
2672 &mov ("esi",&wparam(1)); # user supplied key
2673 &mov ("edi",&wparam(3)); # private key schedule
2676 &jz (&label("badpointer"));
2678 &jz (&label("badpointer"));
2680 &call (&label("pic_point"));
2681 &set_label("pic_point");
2683 &lea ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
2684 &lea ($tbl,&DWP(2048+128,$tbl));
2687 &mov ("eax",&DWP(0-128,$tbl));
2688 &mov ("ebx",&DWP(32-128,$tbl));
2689 &mov ("ecx",&DWP(64-128,$tbl));
2690 &mov ("edx",&DWP(96-128,$tbl));
2691 &mov ("eax",&DWP(128-128,$tbl));
2692 &mov ("ebx",&DWP(160-128,$tbl));
2693 &mov ("ecx",&DWP(192-128,$tbl));
2694 &mov ("edx",&DWP(224-128,$tbl));
2696 &mov ("ecx",&wparam(2)); # number of bits in key
2698 &je (&label("10rounds"));
2700 &je (&label("12rounds"));
2702 &je (&label("14rounds"));
2703 &mov ("eax",-2); # invalid number of bits
2704 &jmp (&label("exit"));
2706 &set_label("10rounds");
2707 &mov ("eax",&DWP(0,"esi")); # copy first 4 dwords
2708 &mov ("ebx",&DWP(4,"esi"));
2709 &mov ("ecx",&DWP(8,"esi"));
2710 &mov ("edx",&DWP(12,"esi"));
2711 &mov (&DWP(0,"edi"),"eax");
2712 &mov (&DWP(4,"edi"),"ebx");
2713 &mov (&DWP(8,"edi"),"ecx");
2714 &mov (&DWP(12,"edi"),"edx");
2717 &jmp (&label("10shortcut"));
2720 &set_label("10loop");
2721 &mov ("eax",&DWP(0,"edi")); # rk[0]
2722 &mov ("edx",&DWP(12,"edi")); # rk[3]
2723 &set_label("10shortcut");
2726 &mov (&DWP(16,"edi"),"eax"); # rk[4]
2727 &xor ("eax",&DWP(4,"edi"));
2728 &mov (&DWP(20,"edi"),"eax"); # rk[5]
2729 &xor ("eax",&DWP(8,"edi"));
2730 &mov (&DWP(24,"edi"),"eax"); # rk[6]
2731 &xor ("eax",&DWP(12,"edi"));
2732 &mov (&DWP(28,"edi"),"eax"); # rk[7]
2736 &jl (&label("10loop"));
2738 &mov (&DWP(80,"edi"),10); # setup number of rounds
2740 &jmp (&label("exit"));
2742 &set_label("12rounds");
2743 &mov ("eax",&DWP(0,"esi")); # copy first 6 dwords
2744 &mov ("ebx",&DWP(4,"esi"));
2745 &mov ("ecx",&DWP(8,"esi"));
2746 &mov ("edx",&DWP(12,"esi"));
2747 &mov (&DWP(0,"edi"),"eax");
2748 &mov (&DWP(4,"edi"),"ebx");
2749 &mov (&DWP(8,"edi"),"ecx");
2750 &mov (&DWP(12,"edi"),"edx");
2751 &mov ("ecx",&DWP(16,"esi"));
2752 &mov ("edx",&DWP(20,"esi"));
2753 &mov (&DWP(16,"edi"),"ecx");
2754 &mov (&DWP(20,"edi"),"edx");
2757 &jmp (&label("12shortcut"));
2760 &set_label("12loop");
2761 &mov ("eax",&DWP(0,"edi")); # rk[0]
2762 &mov ("edx",&DWP(20,"edi")); # rk[5]
2763 &set_label("12shortcut");
2766 &mov (&DWP(24,"edi"),"eax"); # rk[6]
2767 &xor ("eax",&DWP(4,"edi"));
2768 &mov (&DWP(28,"edi"),"eax"); # rk[7]
2769 &xor ("eax",&DWP(8,"edi"));
2770 &mov (&DWP(32,"edi"),"eax"); # rk[8]
2771 &xor ("eax",&DWP(12,"edi"));
2772 &mov (&DWP(36,"edi"),"eax"); # rk[9]
2775 &je (&label("12break"));
2778 &xor ("eax",&DWP(16,"edi"));
2779 &mov (&DWP(40,"edi"),"eax"); # rk[10]
2780 &xor ("eax",&DWP(20,"edi"));
2781 &mov (&DWP(44,"edi"),"eax"); # rk[11]
2784 &jmp (&label("12loop"));
2786 &set_label("12break");
2787 &mov (&DWP(72,"edi"),12); # setup number of rounds
2789 &jmp (&label("exit"));
2791 &set_label("14rounds");
2792 &mov ("eax",&DWP(0,"esi")); # copy first 8 dwords
2793 &mov ("ebx",&DWP(4,"esi"));
2794 &mov ("ecx",&DWP(8,"esi"));
2795 &mov ("edx",&DWP(12,"esi"));
2796 &mov (&DWP(0,"edi"),"eax");
2797 &mov (&DWP(4,"edi"),"ebx");
2798 &mov (&DWP(8,"edi"),"ecx");
2799 &mov (&DWP(12,"edi"),"edx");
2800 &mov ("eax",&DWP(16,"esi"));
2801 &mov ("ebx",&DWP(20,"esi"));
2802 &mov ("ecx",&DWP(24,"esi"));
2803 &mov ("edx",&DWP(28,"esi"));
2804 &mov (&DWP(16,"edi"),"eax");
2805 &mov (&DWP(20,"edi"),"ebx");
2806 &mov (&DWP(24,"edi"),"ecx");
2807 &mov (&DWP(28,"edi"),"edx");
2810 &jmp (&label("14shortcut"));
2813 &set_label("14loop");
2814 &mov ("edx",&DWP(28,"edi")); # rk[7]
2815 &set_label("14shortcut");
2816 &mov ("eax",&DWP(0,"edi")); # rk[0]
2820 &mov (&DWP(32,"edi"),"eax"); # rk[8]
2821 &xor ("eax",&DWP(4,"edi"));
2822 &mov (&DWP(36,"edi"),"eax"); # rk[9]
2823 &xor ("eax",&DWP(8,"edi"));
2824 &mov (&DWP(40,"edi"),"eax"); # rk[10]
2825 &xor ("eax",&DWP(12,"edi"));
2826 &mov (&DWP(44,"edi"),"eax"); # rk[11]
2829 &je (&label("14break"));
2833 &mov ("eax",&DWP(16,"edi")); # rk[4]
2834 &movz ("esi",&LB("edx")); # rk[11]>>0
2835 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2836 &movz ("esi",&HB("edx")); # rk[11]>>8
2839 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2842 &movz ("esi",&LB("edx")); # rk[11]>>16
2845 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2846 &movz ("esi",&HB("edx")); # rk[11]>>24
2850 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2854 &mov (&DWP(48,"edi"),"eax"); # rk[12]
2855 &xor ("eax",&DWP(20,"edi"));
2856 &mov (&DWP(52,"edi"),"eax"); # rk[13]
2857 &xor ("eax",&DWP(24,"edi"));
2858 &mov (&DWP(56,"edi"),"eax"); # rk[14]
2859 &xor ("eax",&DWP(28,"edi"));
2860 &mov (&DWP(60,"edi"),"eax"); # rk[15]
2863 &jmp (&label("14loop"));
2865 &set_label("14break");
2866 &mov (&DWP(48,"edi"),14); # setup number of rounds
2868 &jmp (&label("exit"));
2870 &set_label("badpointer");
2873 &function_end("_x86_AES_set_encrypt_key");
2875 # int AES_set_encrypt_key(const unsigned char *userKey, const int bits,
2877 &function_begin_B("AES_set_encrypt_key");
2878 &call ("_x86_AES_set_encrypt_key");
2880 &function_end_B("AES_set_encrypt_key");
2883 { my ($i,$key,$tp1,$tp2,$tp4,$tp8) = @_;
2886 &mov ($tmp,0x80808080);
2888 &lea ($tp2,&DWP(0,$tp1,$tp1));
2892 &and ($tp2,0xfefefefe);
2893 &and ($acc,0x1b1b1b1b);
2895 &mov ($tmp,0x80808080);
2898 &lea ($tp4,&DWP(0,$tp2,$tp2));
2902 &and ($tp4,0xfefefefe);
2903 &and ($acc,0x1b1b1b1b);
2904 &xor ($tp2,$tp1); # tp2^tp1
2906 &mov ($tmp,0x80808080);
2909 &lea ($tp8,&DWP(0,$tp4,$tp4));
2912 &xor ($tp4,$tp1); # tp4^tp1
2914 &and ($tp8,0xfefefefe);
2915 &and ($acc,0x1b1b1b1b);
2916 &rotl ($tp1,8); # = ROTATE(tp1,8)
2919 &mov ($tmp,&DWP(4*($i+1),$key)); # modulo-scheduled load
2926 &xor ($tp1,$tp8); # ^= tp8^(tp4^tp1)^(tp2^tp1)
2928 &xor ($tp1,$tp2); # ^= ROTATE(tp8^tp2^tp1,24)
2930 &xor ($tp1,$tp4); # ^= ROTATE(tp8^tp4^tp1,16)
2932 &xor ($tp1,$tp8); # ^= ROTATE(tp8,8)
2934 &mov (&DWP(4*$i,$key),$tp1);
2937 # int AES_set_decrypt_key(const unsigned char *userKey, const int bits,
2939 &function_begin_B("AES_set_decrypt_key");
2940 &call ("_x86_AES_set_encrypt_key");
2942 &je (&label("proceed"));
2945 &set_label("proceed");
2951 &mov ("esi",&wparam(2));
2952 &mov ("ecx",&DWP(240,"esi")); # pull number of rounds
2953 &lea ("ecx",&DWP(0,"","ecx",4));
2954 &lea ("edi",&DWP(0,"esi","ecx",4)); # pointer to last chunk
2956 &set_label("invert",4); # invert order of chunks
2957 &mov ("eax",&DWP(0,"esi"));
2958 &mov ("ebx",&DWP(4,"esi"));
2959 &mov ("ecx",&DWP(0,"edi"));
2960 &mov ("edx",&DWP(4,"edi"));
2961 &mov (&DWP(0,"edi"),"eax");
2962 &mov (&DWP(4,"edi"),"ebx");
2963 &mov (&DWP(0,"esi"),"ecx");
2964 &mov (&DWP(4,"esi"),"edx");
2965 &mov ("eax",&DWP(8,"esi"));
2966 &mov ("ebx",&DWP(12,"esi"));
2967 &mov ("ecx",&DWP(8,"edi"));
2968 &mov ("edx",&DWP(12,"edi"));
2969 &mov (&DWP(8,"edi"),"eax");
2970 &mov (&DWP(12,"edi"),"ebx");
2971 &mov (&DWP(8,"esi"),"ecx");
2972 &mov (&DWP(12,"esi"),"edx");
2976 &jne (&label("invert"));
2978 &mov ($key,&wparam(2));
2979 &mov ($acc,&DWP(240,$key)); # pull number of rounds
2980 &lea ($acc,&DWP(-2,$acc,$acc));
2981 &lea ($acc,&DWP(0,$key,$acc,8));
2982 &mov (&wparam(2),$acc);
2984 &mov ($s0,&DWP(16,$key)); # modulo-scheduled load
2985 &set_label("permute",4); # permute the key schedule
2987 &deckey (0,$key,$s0,$s1,$s2,$s3);
2988 &deckey (1,$key,$s1,$s2,$s3,$s0);
2989 &deckey (2,$key,$s2,$s3,$s0,$s1);
2990 &deckey (3,$key,$s3,$s0,$s1,$s2);
2991 &cmp ($key,&wparam(2));
2992 &jb (&label("permute"));
2994 &xor ("eax","eax"); # return success
2995 &function_end("AES_set_decrypt_key");
2996 &asciz("AES for x86, CRYPTOGAMS by <appro\@openssl.org>");
projects / openssl.git / blob