3 # ====================================================================
4 # Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
5 # project. Rights for redistribution and usage in source and binary
6 # forms are granted according to the OpenSSL license.
7 # ====================================================================
11 # You might fail to appreciate this module performance from the first
12 # try. If compared to "vanilla" linux-ia32-icc target, i.e. considered
13 # to be *the* best Intel C compiler without -KPIC, performance appears
14 # to be virtually identical... But try to re-configure with shared
15 # library support... Aha! Intel compiler "suddenly" lags behind by 30%
16 # [on P4, more on others]:-) And if compared to position-independent
17 # code generated by GNU C, this code performs *more* than *twice* as
18 # fast! Yes, all this buzz about PIC means that unlike other hand-
19 # coded implementations, this one was explicitly designed to be safe
20 # to use even in shared library context... This also means that this
21 # code isn't necessarily absolutely fastest "ever," because in order
22 # to achieve position independence an extra register has to be
23 # off-loaded to stack, which affects the benchmark result.
25 # Special note about instruction choice. Do you recall RC4_INT code
26 # performing poorly on P4? It might be the time to figure out why.
27 # RC4_INT code implies effective address calculations in base+offset*4
28 # form. Trouble is that it seems that offset scaling turned to be
29 # critical path... At least eliminating scaling resulted in 2.8x RC4
30 # performance improvement [as you might recall]. As AES code is hungry
31 # for scaling too, I [try to] avoid the latter by favoring off-by-2
32 # shifts and masking the result with 0xFF<<2 instead of "boring" 0xFF.
34 # As was shown by Dean Gaudet <dean@arctic.org>, the above note turned
35 # void. Performance improvement with off-by-2 shifts was observed on
36 # intermediate implementation, which was spilling yet another register
37 # to stack... Final offset*4 code below runs just a tad faster on P4,
38 # but exhibits up to 10% improvement on other cores.
40 # Second version is "monolithic" replacement for aes_core.c, which in
41 # addition to AES_[de|en]crypt implements AES_set_[de|en]cryption_key.
42 # This made it possible to implement little-endian variant of the
43 # algorithm without modifying the base C code. Motivating factor for
44 # the undertaken effort was that it appeared that in tight IA-32
45 # register window little-endian flavor could achieve slightly higher
46 # Instruction Level Parallelism, and it indeed resulted in up to 15%
47 # better performance on most recent µ-archs...
49 # Third version adds AES_cbc_encrypt implementation, which resulted in
50 # up to 40% performance imrovement of CBC benchmark results. 40% was
51 # observed on P4 core, where "overall" imrovement coefficient, i.e. if
52 # compared to PIC generated by GCC and in CBC mode, was observed to be
53 # as large as 4x:-) CBC performance is virtually identical to ECB now
54 # and on some platforms even better, e.g. 17.6 "small" cycles/byte on
55 # Opteron, because certain function prologues and epilogues are
56 # effectively taken out of the loop...
58 # Version 3.2 implements compressed tables and prefetch of these tables
59 # in CBC[!] mode. Former means that 3/4 of table references are now
60 # misaligned, which unfortunately has negative impact on elder IA-32
61 # implementations, Pentium suffered 30% penalty, PIII - 10%.
63 # Version 3.3 avoids L1 cache aliasing between stack frame and
64 # S-boxes, and 3.4 - L1 cache aliasing even between key schedule. The
65 # latter is achieved by copying the key schedule to controlled place in
66 # stack. This unfortunately has rather strong impact on small block CBC
67 # performance, ~2x deterioration on 16-byte block if compared to 3.3.
69 # Version 3.5 checks if there is L1 cache aliasing between user-supplied
70 # key schedule and S-boxes and abstains from copying the former if
71 # there is no. This allows end-user to consciously retain small block
72 # performance by aligning key schedule in specific manner.
74 # Version 3.6 compresses Td4 to 256 bytes and prefetches it in ECB.
76 # Current ECB performance numbers for 128-bit key in CPU cycles per
77 # processed byte [measure commonly used by AES benchmarkers] are:
79 # small footprint fully unrolled
85 # Version 3.7 reimplements outer rounds as "compact." Meaning that
86 # first and last rounds reference compact 256 bytes S-box. This means
87 # that first round consumes a lot more CPU cycles and that encrypt
88 # and decrypt performance becomes asymmetric. Encrypt performance
89 # drops by 10-12%, while decrypt - by 20-25%:-( 256 bytes S-box is
90 # aggressively pre-fetched.
92 # Version 4.0 effectively rolls back to 3.6 and instead implements
93 # additional set of functions, _[x86|sse]_AES_[en|de]crypt_compact,
94 # which use exclusively 256 byte S-box. These functions are to be
95 # called in modes not concealing plain text, such as ECB, or when
96 # we're asked to process smaller amount of data [or unconditionally
97 # on hyper-threading CPU]. Currently it's called unconditionally from
98 # AES_[en|de]crypt, which affects all modes, but CBC. CBC routine
99 # still needs to be modified to switch between slower and faster
100 # mode when appropriate... But in either case benchmark landscape
101 # changes dramatically and below numbers are CPU cycles per processed
102 # byte for 128-bit key.
104 # ECB encrypt ECB decrypt CBC large chunk
105 # P4 56[60] 84[100] 23
106 # AMD K8 48[44] 70[79] 18
107 # PIII 41[50] 61[91] 24
110 # Version 4.1 switches to compact S-box even in key schedule setup.
112 # Version 4.2 prefetches compact S-box in every SSE round or in other
113 # words every cache-line is *guaranteed* to be accessed within ~50
114 # cycles window. Why just SSE? Because it's needed on hyper-threading
115 # CPU! Which is also why it's prefetched with 64 byte stride. Best
116 # part is that it has no negative effect on performance:-)
118 # Version 4.3 implements switch between compact and non-compact block
119 # functions in AES_cbc_encrypt depending on how much data was asked
120 # to process in one stroke.
122 # Timing attacks are classified in two classes: synchronous when
123 # attacker consciously initiates cryptographic operation and collect
124 # timing data of various character afterwards, and asynchronous when
125 # malicious code is executed on same CPU simultaneously with AES,
126 # instruments itself and performs statistical analysis of this data.
128 # As far as synchronous attacks go the root to the AES timing
129 # vulnerability is twofold. Firstly, of 256 S-box elements at most 160
130 # are referred to in single 128-bit block operation. Well, in C
131 # implementation with 4 distinct tables it's actually as little as 40
132 # references per 256 elements table, but anyway... Secondly, even
133 # though S-box elements are clustered into smaller amount of cache-
134 # lines, smaller than 160 and even 40, it turned out that for certain
135 # plain-text pattern[s] or simply put chosen plain-text and given key
136 # few cache-lines remain unaccessed during block operation. Now, if
137 # attacker can figure out this access pattern, he can deduct the key
138 # [or at least part of it]. The natural way to mitigate this kind of
139 # attacks is to minimize the amount of cache-lines in S-box and/or
140 # prefetch them to ensure that every one is accessed for more uniform
141 # timing. But note that *if* plain-text was concealed in such way that
142 # input to block function is distributed *uniformly*, then attack
143 # wouldn't apply. Now note that some encryption modes, most notably
144 # CBC, do masks the plain-text in this exact way [secure cipher output
145 # is distributed uniformly]. Yes, one still might find input that
146 # would reveal the information about given key, but if amount of
147 # candidate inputs to be tried is larger than amount possible key
148 # combinations then attack becomes infeasible. This is why revised
149 # AES_cbc_encrypt "dares" to switch to larger S-box when larger chunk
150 # of data is to be processed in one stroke. The current size limit of
151 # 512 bytes is chosen to provide same [diminishigly low] probability
152 # for cache-line to remain untouched in large chunk operation with
153 # large S-box as for single block operation with compact S-box and
154 # surely needs more careful consideration...
156 # As for asynchronous attacks. There are two flavours: attacker code
157 # being interleaved with AES on hyper-threading CPU at *instruction*
158 # level, and two processes time sharing single core. As for latter.
159 # Two vectors. 1. Given that attacker process has higher priority,
160 # yield execution to process performing AES just before timer fires
161 # off the scheduler, immediately regain control of CPU and analyze the
162 # cache state. For this attack to be efficient attacker would have to
163 # effectively slow down the operation by several *orders* of magnitute,
164 # by ratio of time slice to duration of handful of AES rounds, which
165 # unlikely to remain unnoticed. Not to mention that this also means
166 # that he would spend correspondigly more time to collect enough
167 # statistical data to mount the attack. It's probably appropriate to
168 # say that if adeversary reckons that this attack is beneficial and
169 # risks to be noticed, you probably have larger problems having him
170 # mere opportunity. In other words suggested code design expects you
171 # to preclude/mitigate this attack by overall system security design.
172 # 2. Attacker manages to make his code interrupt driven. In order for
173 # this kind of attack to be feasible, interrupt rate has to be high
174 # enough, again comparable to duration of handful of AES rounds. But
175 # is there interrupt source of such rate? Hardly, not even 1Gbps NIC
176 # generates interrupts at such raging rate...
178 # And now back to the former, hyper-threading CPU or more specifically
179 # Intel P4. Recall that asynchronous attack implies that malicious
180 # code instruments itself. And naturally instrumentation granularity
181 # has be noticeably lower than duration of codepath accessing S-box.
182 # Given that all cache-lines are accessed during that time that is.
183 # Current implementation accesses *all* cache-lines within ~50 cycles
184 # window, which is actually *less* than RDTSC latency on Intel P4!
186 push(@INC,"perlasm","../../perlasm");
189 &asm_init($ARGV[0],"aes-586.pl",$x86only = $ARGV[$#ARGV] eq "386");
199 # stack frame layout in _[x86|sse]_AES_* routines, frame is allocated
201 $__ra=&DWP(0,"esp"); # return address
202 $__s0=&DWP(4,"esp"); # s0 backing store
203 $__s1=&DWP(8,"esp"); # s1 backing store
204 $__s2=&DWP(12,"esp"); # s2 backing store
205 $__s3=&DWP(16,"esp"); # s3 backing store
206 $__key=&DWP(20,"esp"); # pointer to key schedule
207 $__end=&DWP(24,"esp"); # pointer to end of key schedule
208 $__tbl=&DWP(28,"esp"); # %ebp backing store
210 # stack frame layout in AES_[en|crypt] routines, which differs from
211 # above by 4 and overlaps by %ebp backing store
212 $_tbl=&DWP(24,"esp");
213 $_esp=&DWP(28,"esp");
215 sub _data_word() { my $i; while(defined($i=shift)) { &data_word($i,$i); } }
217 $speed_limit=512; # chunks smaller than $speed_limit are
218 # processed with compact routine in CBC mode
219 $small_footprint=1; # $small_footprint=1 code is ~5% slower [on
220 # recent µ-archs], but ~5 times smaller!
221 # I favor compact code to minimize cache
222 # contention and in hope to "collect" 5% back
223 # in real-life applications...
225 $vertical_spin=0; # shift "verticaly" defaults to 0, because of
226 # its proof-of-concept status...
227 # Note that there is no decvert(), as well as last encryption round is
228 # performed with "horizontal" shifts. This is because this "vertical"
229 # implementation [one which groups shifts on a given $s[i] to form a
230 # "column," unlike "horizontal" one, which groups shifts on different
231 # $s[i] to form a "row"] is work in progress. It was observed to run
232 # few percents faster on Intel cores, but not AMD. On AMD K8 core it's
233 # whole 12% slower:-( So we face a trade-off... Shall it be resolved
234 # some day? Till then the code is considered experimental and by
235 # default remains dormant...
239 my $v0 = $acc, $v1 = $key;
241 &mov ($v0,$s[3]); # copy s3
242 &mov (&DWP(4,"esp"),$s[2]); # save s2
243 &mov ($v1,$s[0]); # copy s0
244 &mov (&DWP(8,"esp"),$s[1]); # save s1
246 &movz ($s[2],&HB($s[0]));
248 &mov ($s[0],&DWP(0,$te,$s[0],8)); # s0>>0
250 &mov ($s[3],&DWP(3,$te,$s[2],8)); # s0>>8
251 &movz ($s[1],&HB($v1));
253 &mov ($s[2],&DWP(2,$te,$v1,8)); # s0>>16
255 &mov ($s[1],&DWP(1,$te,$s[1],8)); # s0>>24
258 &xor ($s[3],&DWP(0,$te,$v0,8)); # s3>>0
259 &movz ($v0,&HB($v1));
261 &xor ($s[2],&DWP(3,$te,$v0,8)); # s3>>8
262 &movz ($v0,&HB($v1));
264 &xor ($s[1],&DWP(2,$te,$v1,8)); # s3>>16
265 &mov ($v1,&DWP(4,"esp")); # restore s2
266 &xor ($s[0],&DWP(1,$te,$v0,8)); # s3>>24
270 &xor ($s[2],&DWP(0,$te,$v1,8)); # s2>>0
271 &movz ($v1,&HB($v0));
273 &xor ($s[1],&DWP(3,$te,$v1,8)); # s2>>8
274 &movz ($v1,&HB($v0));
276 &xor ($s[0],&DWP(2,$te,$v0,8)); # s2>>16
277 &mov ($v0,&DWP(8,"esp")); # restore s1
278 &xor ($s[3],&DWP(1,$te,$v1,8)); # s2>>24
282 &xor ($s[1],&DWP(0,$te,$v0,8)); # s1>>0
283 &movz ($v0,&HB($v1));
285 &xor ($s[0],&DWP(3,$te,$v0,8)); # s1>>8
286 &movz ($v0,&HB($v1));
288 &xor ($s[3],&DWP(2,$te,$v1,8)); # s1>>16
289 &mov ($key,$__key); # reincarnate v1 as key
290 &xor ($s[2],&DWP(1,$te,$v0,8)); # s1>>24
293 # Another experimental routine, which features "horizontal spin," but
294 # eliminates one reference to stack. Strangely enough runs slower...
296 { my $v0 = $key, $v1 = $acc;
298 &movz ($v0,&LB($s0)); # 3, 2, 1, 0*
299 &rotr ($s2,8); # 8,11,10, 9
300 &mov ($v1,&DWP(0,$te,$v0,8)); # 0
301 &movz ($v0,&HB($s1)); # 7, 6, 5*, 4
302 &rotr ($s3,16); # 13,12,15,14
303 &xor ($v1,&DWP(3,$te,$v0,8)); # 5
304 &movz ($v0,&HB($s2)); # 8,11,10*, 9
305 &rotr ($s0,16); # 1, 0, 3, 2
306 &xor ($v1,&DWP(2,$te,$v0,8)); # 10
307 &movz ($v0,&HB($s3)); # 13,12,15*,14
308 &xor ($v1,&DWP(1,$te,$v0,8)); # 15, t[0] collected
309 &mov ($__s0,$v1); # t[0] saved
311 &movz ($v0,&LB($s1)); # 7, 6, 5, 4*
312 &shr ($s1,16); # -, -, 7, 6
313 &mov ($v1,&DWP(0,$te,$v0,8)); # 4
314 &movz ($v0,&LB($s3)); # 13,12,15,14*
315 &xor ($v1,&DWP(2,$te,$v0,8)); # 14
316 &movz ($v0,&HB($s0)); # 1, 0, 3*, 2
317 &and ($s3,0xffff0000); # 13,12, -, -
318 &xor ($v1,&DWP(1,$te,$v0,8)); # 3
319 &movz ($v0,&LB($s2)); # 8,11,10, 9*
320 &or ($s3,$s1); # 13,12, 7, 6
321 &xor ($v1,&DWP(3,$te,$v0,8)); # 9, t[1] collected
322 &mov ($s1,$v1); # s[1]=t[1]
324 &movz ($v0,&LB($s0)); # 1, 0, 3, 2*
325 &shr ($s2,16); # -, -, 8,11
326 &mov ($v1,&DWP(2,$te,$v0,8)); # 2
327 &movz ($v0,&HB($s3)); # 13,12, 7*, 6
328 &xor ($v1,&DWP(1,$te,$v0,8)); # 7
329 &movz ($v0,&HB($s2)); # -, -, 8*,11
330 &xor ($v1,&DWP(0,$te,$v0,8)); # 8
333 &xor ($v1,&DWP(3,$te,$v0,8)); # 13, t[2] collected
335 &movz ($v0,&LB($s2)); # -, -, 8,11*
337 &mov ($s2,&DWP(1,$te,$v0,8)); # 11
338 &xor ($s2,&DWP(3,$te,$s0,8)); # 1
339 &mov ($s0,$__s0); # s[0]=t[0]
340 &movz ($v0,&LB($s3)); # 13,12, 7, 6*
341 &shr ($s3,16); # , ,13,12
342 &xor ($s2,&DWP(2,$te,$v0,8)); # 6
343 &mov ($key,$__key); # reincarnate v0 as key
344 &and ($s3,0xff); # , ,13,12*
345 &mov ($s3,&DWP(0,$te,$s3,8)); # 12
346 &xor ($s3,$s2); # s[2]=t[3] collected
347 &mov ($s2,$v1); # s[2]=t[2]
350 # More experimental code... SSE one... Even though this one eliminates
351 # *all* references to stack, it's not faster...
354 &movz ($acc,&LB("eax")); # 0
355 &mov ("ecx",&DWP(0,$tbl,$acc,8)); # 0
356 &pshufw ("mm2","mm0",0x0d); # 7, 6, 3, 2
357 &movz ("edx",&HB("eax")); # 1
358 &mov ("edx",&DWP(3,$tbl,"edx",8)); # 1
359 &shr ("eax",16); # 5, 4
361 &movz ($acc,&LB("ebx")); # 10
362 &xor ("ecx",&DWP(2,$tbl,$acc,8)); # 10
363 &pshufw ("mm6","mm4",0x08); # 13,12, 9, 8
364 &movz ($acc,&HB("ebx")); # 11
365 &xor ("edx",&DWP(1,$tbl,$acc,8)); # 11
366 &shr ("ebx",16); # 15,14
368 &movz ($acc,&HB("eax")); # 5
369 &xor ("ecx",&DWP(3,$tbl,$acc,8)); # 5
370 &movq ("mm3",QWP(16,$key));
371 &movz ($acc,&HB("ebx")); # 15
372 &xor ("ecx",&DWP(1,$tbl,$acc,8)); # 15
373 &movd ("mm0","ecx"); # t[0] collected
375 &movz ($acc,&LB("eax")); # 4
376 &mov ("ecx",&DWP(0,$tbl,$acc,8)); # 4
377 &movd ("eax","mm2"); # 7, 6, 3, 2
378 &movz ($acc,&LB("ebx")); # 14
379 &xor ("ecx",&DWP(2,$tbl,$acc,8)); # 14
380 &movd ("ebx","mm6"); # 13,12, 9, 8
382 &movz ($acc,&HB("eax")); # 3
383 &xor ("ecx",&DWP(1,$tbl,$acc,8)); # 3
384 &movz ($acc,&HB("ebx")); # 9
385 &xor ("ecx",&DWP(3,$tbl,$acc,8)); # 9
386 &movd ("mm1","ecx"); # t[1] collected
388 &movz ($acc,&LB("eax")); # 2
389 &mov ("ecx",&DWP(2,$tbl,$acc,8)); # 2
390 &shr ("eax",16); # 7, 6
391 &punpckldq ("mm0","mm1"); # t[0,1] collected
392 &movz ($acc,&LB("ebx")); # 8
393 &xor ("ecx",&DWP(0,$tbl,$acc,8)); # 8
394 &shr ("ebx",16); # 13,12
396 &movz ($acc,&HB("eax")); # 7
397 &xor ("ecx",&DWP(1,$tbl,$acc,8)); # 7
399 &movz ("eax",&LB("eax")); # 6
400 &xor ("edx",&DWP(2,$tbl,"eax",8)); # 6
401 &pshufw ("mm1","mm0",0x08); # 5, 4, 1, 0
402 &movz ($acc,&HB("ebx")); # 13
403 &xor ("ecx",&DWP(3,$tbl,$acc,8)); # 13
404 &xor ("ecx",&DWP(24,$key)); # t[2]
405 &movd ("mm4","ecx"); # t[2] collected
406 &movz ("ebx",&LB("ebx")); # 12
407 &xor ("edx",&DWP(0,$tbl,"ebx",8)); # 12
409 &movd ("eax","mm1"); # 5, 4, 1, 0
410 &mov ("ebx",&DWP(28,$key)); # t[3]
412 &movd ("mm5","ebx"); # t[3] collected
413 &and ("ebx",0xffff0000);
416 &punpckldq ("mm4","mm5"); # t[2,3] collected
419 ######################################################################
420 # "Compact" block function
421 ######################################################################
425 while ($#_>5) { pop(@_); $Fn=sub{}; }
428 my $out = $i==3?$s[0]:$acc;
430 # $Fn is used in first compact round and its purpose is to
431 # void restoration of some values from stack, so that after
432 # 4xenccompact with extra argument $key value is left there...
433 if ($i==3) { &$Fn ($key,$__key); }##%edx
434 else { &mov ($out,$s[0]); }
436 if ($i==1) { &shr ($s[0],16); }#%ebx[1]
437 if ($i==2) { &shr ($s[0],24); }#%ecx[2]
438 &movz ($out,&BP(-128,$te,$out,1));
440 if ($i==3) { $tmp=$s[1]; }##%eax
441 &movz ($tmp,&HB($s[1]));
442 &movz ($tmp,&BP(-128,$te,$tmp,1));
446 if ($i==3) { $tmp=$s[2]; &mov ($s[1],$__s0); }##%ebx
447 else { &mov ($tmp,$s[2]);
449 if ($i==2) { &and ($s[1],0xFF); }#%edx[2]
451 &movz ($tmp,&BP(-128,$te,$tmp,1));
455 if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }##%ecx
456 elsif($i==2){ &movz ($tmp,&HB($s[3])); }#%ebx[2]
457 else { &mov ($tmp,$s[3]);
459 &movz ($tmp,&BP(-128,$te,$tmp,1));
462 if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); }
463 if ($i==3) { &mov ($s[3],$acc); }
468 { my @s = ($s0,$s1,$s2,$s3);
474 &and ($acc,0x80808080);
478 &and ($r2,0x7f7f7f7f);
480 &lea ($r2,&DWP(0,$r2,$r2));
481 &and ($acc,0x1b1b1b1b);
483 &xor ($acc,$r2); # r2
485 &xor ($s[$i],$acc); # r0 ^ r2
487 &xor ($s[$i],$acc) # ROTATE(r2^r0,24) ^ r2
494 &public_label("AES_Te");
495 &function_begin_B("_x86_AES_encrypt_compact");
496 # note that caller is expected to allocate stack frame for me!
497 &mov ($__key,$key); # save key
499 &xor ($s0,&DWP(0,$key)); # xor with key
500 &xor ($s1,&DWP(4,$key));
501 &xor ($s2,&DWP(8,$key));
502 &xor ($s3,&DWP(12,$key));
504 &mov ($acc,&DWP(240,$key)); # load key->rounds
505 &lea ($acc,&DWP(-2,$acc,$acc));
506 &lea ($acc,&DWP(0,$key,$acc,8));
507 &mov ($__end,$acc); # end of key schedule
510 &mov ($key,&DWP(0-128,$tbl));
511 &mov ($acc,&DWP(32-128,$tbl));
512 &mov ($key,&DWP(64-128,$tbl));
513 &mov ($acc,&DWP(96-128,$tbl));
514 &mov ($key,&DWP(128-128,$tbl));
515 &mov ($acc,&DWP(160-128,$tbl));
516 &mov ($key,&DWP(192-128,$tbl));
517 &mov ($acc,&DWP(224-128,$tbl));
519 &set_label("loop",16);
521 &enccompact(0,$tbl,$s0,$s1,$s2,$s3,1);
522 &enccompact(1,$tbl,$s1,$s2,$s3,$s0,1);
523 &enccompact(2,$tbl,$s2,$s3,$s0,$s1,1);
524 &enccompact(3,$tbl,$s3,$s0,$s1,$s2,1);
531 &add ($key,16); # advance rd_key
532 &xor ($s0,&DWP(0,$key));
533 &xor ($s1,&DWP(4,$key));
534 &xor ($s2,&DWP(8,$key));
535 &xor ($s3,&DWP(12,$key));
539 &jb (&label("loop"));
541 &enccompact(0,$tbl,$s0,$s1,$s2,$s3);
542 &enccompact(1,$tbl,$s1,$s2,$s3,$s0);
543 &enccompact(2,$tbl,$s2,$s3,$s0,$s1);
544 &enccompact(3,$tbl,$s3,$s0,$s1,$s2);
546 &xor ($s0,&DWP(16,$key));
547 &xor ($s1,&DWP(20,$key));
548 &xor ($s2,&DWP(24,$key));
549 &xor ($s3,&DWP(28,$key));
552 &function_end_B("_x86_AES_encrypt_compact");
554 ######################################################################
555 # "Compact" SSE block function.
556 ######################################################################
558 # Performance is not actually extraordinary in comparison to pure
559 # x86 code. In particular encrypt performance is virtually the same.
560 # Decrypt performance on the other hand is 15-20% better on newer
561 # µ-archs [but we're thankful for *any* improvement here], and ~50%
562 # better on PIII:-) And additionally on the pros side this code
563 # eliminates redundant references to stack and thus relieves/
564 # minimizes the pressure on the memory bus.
566 # MMX register layout lsb
567 # +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
569 # +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
570 # | s3 | s2 | s1 | s0 |
571 # +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
572 # |15|14|13|12|11|10| 9| 8| 7| 6| 5| 4| 3| 2| 1| 0|
573 # +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
575 # Indexes translate as s[N/4]>>(8*(N%4)), e.g. 5 means s1>>8.
576 # In this terms encryption and decryption "compact" permutation
577 # matrices can be depicted as following:
579 # encryption lsb # decryption lsb
580 # +----++----+----+----+----+ # +----++----+----+----+----+
581 # | t0 || 15 | 10 | 5 | 0 | # | t0 || 7 | 10 | 13 | 0 |
582 # +----++----+----+----+----+ # +----++----+----+----+----+
583 # | t1 || 3 | 14 | 9 | 4 | # | t1 || 11 | 14 | 1 | 4 |
584 # +----++----+----+----+----+ # +----++----+----+----+----+
585 # | t2 || 7 | 2 | 13 | 8 | # | t2 || 15 | 2 | 5 | 8 |
586 # +----++----+----+----+----+ # +----++----+----+----+----+
587 # | t3 || 11 | 6 | 1 | 12 | # | t3 || 3 | 6 | 9 | 12 |
588 # +----++----+----+----+----+ # +----++----+----+----+----+
590 ######################################################################
591 # Why not xmm registers? Short answer. It was actually tested and
592 # was not any faster, but *contrary*, most notably on Intel CPUs.
593 # Longer answer. Main advantage of using mm registers is that movd
594 # latency is lower, especially on Intel P4. While arithmetic
595 # instructions are twice as many, they can be scheduled every cycle
596 # and not every second one when they are operating on xmm register,
597 # so that "arithmetic throughput" remains virtually the same. And
598 # finally the code can be executed even on elder SSE-only CPUs:-)
602 &pshufw ("mm1","mm0",0x08); # 5, 4, 1, 0
603 &pshufw ("mm5","mm4",0x0d); # 15,14,11,10
604 &movd ("eax","mm1"); # 5, 4, 1, 0
605 &movd ("ebx","mm5"); # 15,14,11,10
607 &movz ($acc,&LB("eax")); # 0
608 &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 0
609 &pshufw ("mm2","mm0",0x0d); # 7, 6, 3, 2
610 &movz ("edx",&HB("eax")); # 1
611 &movz ("edx",&BP(-128,$tbl,"edx",1)); # 1
613 &shr ("eax",16); # 5, 4
615 &movz ($acc,&LB("ebx")); # 10
616 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 10
618 &or ("ecx",$acc); # 10
619 &pshufw ("mm6","mm4",0x08); # 13,12, 9, 8
620 &movz ($acc,&HB("ebx")); # 11
621 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 11
623 &or ("edx",$acc); # 11
624 &shr ("ebx",16); # 15,14
626 &movz ($acc,&HB("eax")); # 5
627 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 5
629 &or ("ecx",$acc); # 5
630 &movz ($acc,&HB("ebx")); # 15
631 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 15
633 &or ("ecx",$acc); # 15
634 &movd ("mm0","ecx"); # t[0] collected
636 &movz ($acc,&LB("eax")); # 4
637 &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 4
638 &movd ("eax","mm2"); # 7, 6, 3, 2
639 &movz ($acc,&LB("ebx")); # 14
640 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 14
642 &or ("ecx",$acc); # 14
644 &movd ("ebx","mm6"); # 13,12, 9, 8
645 &movz ($acc,&HB("eax")); # 3
646 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 3
648 &or ("ecx",$acc); # 3
649 &movz ($acc,&HB("ebx")); # 9
650 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 9
652 &or ("ecx",$acc); # 9
653 &movd ("mm1","ecx"); # t[1] collected
655 &movz ($acc,&LB("ebx")); # 8
656 &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 8
657 &shr ("ebx",16); # 13,12
658 &movz ($acc,&LB("eax")); # 2
659 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 2
661 &or ("ecx",$acc); # 2
662 &shr ("eax",16); # 7, 6
664 &punpckldq ("mm0","mm1"); # t[0,1] collected
666 &movz ($acc,&HB("eax")); # 7
667 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 7
669 &or ("ecx",$acc); # 7
670 &and ("eax",0xff); # 6
671 &movz ("eax",&BP(-128,$tbl,"eax",1)); # 6
673 &or ("edx","eax"); # 6
674 &movz ($acc,&HB("ebx")); # 13
675 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 13
677 &or ("ecx",$acc); # 13
678 &movd ("mm4","ecx"); # t[2] collected
679 &and ("ebx",0xff); # 12
680 &movz ("ebx",&BP(-128,$tbl,"ebx",1)); # 12
681 &or ("edx","ebx"); # 12
682 &movd ("mm5","edx"); # t[3] collected
684 &punpckldq ("mm4","mm5"); # t[2,3] collected
688 &public_label("AES_Te");
689 &function_begin_B("_sse_AES_encrypt_compact");
690 &pxor ("mm0",&QWP(0,$key)); # 7, 6, 5, 4, 3, 2, 1, 0
691 &pxor ("mm4",&QWP(8,$key)); # 15,14,13,12,11,10, 9, 8
693 # note that caller is expected to allocate stack frame for me!
694 &mov ($acc,&DWP(240,$key)); # load key->rounds
695 &lea ($acc,&DWP(-2,$acc,$acc));
696 &lea ($acc,&DWP(0,$key,$acc,8));
697 &mov ($__end,$acc); # end of key schedule
699 &mov ($s0,0x1b1b1b1b); # magic constant
700 &mov (&DWP(8,"esp"),$s0);
701 &mov (&DWP(12,"esp"),$s0);
704 &mov ($s0,&DWP(0-128,$tbl));
705 &mov ($s1,&DWP(32-128,$tbl));
706 &mov ($s2,&DWP(64-128,$tbl));
707 &mov ($s3,&DWP(96-128,$tbl));
708 &mov ($s0,&DWP(128-128,$tbl));
709 &mov ($s1,&DWP(160-128,$tbl));
710 &mov ($s2,&DWP(192-128,$tbl));
711 &mov ($s3,&DWP(224-128,$tbl));
713 &set_label("loop",16);
719 &movq ("mm2",&QWP(8,"esp"));
720 &pxor ("mm3","mm3"); &pxor ("mm7","mm7");
721 &movq ("mm1","mm0"); &movq ("mm5","mm4"); # r0
722 &pcmpgtb("mm3","mm0"); &pcmpgtb("mm7","mm4");
723 &pand ("mm3","mm2"); &pand ("mm7","mm2");
724 &pshufw ("mm2","mm0",0xb1); &pshufw ("mm6","mm4",0xb1);# ROTATE(r0,16)
725 &paddb ("mm0","mm0"); &paddb ("mm4","mm4");
726 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # = r2
727 &pshufw ("mm3","mm2",0xb1); &pshufw ("mm7","mm6",0xb1);# r0
728 &pxor ("mm1","mm0"); &pxor ("mm5","mm4"); # r0^r2
729 &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= ROTATE(r0,16)
731 &movq ("mm2","mm3"); &movq ("mm6","mm7");
732 &pslld ("mm3",8); &pslld ("mm7",8);
733 &psrld ("mm2",24); &psrld ("mm6",24);
734 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= r0<<8
735 &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= r0>>24
737 &movq ("mm3","mm1"); &movq ("mm7","mm5");
738 &movq ("mm2",&QWP(0,$key)); &movq ("mm6",&QWP(8,$key));
739 &psrld ("mm1",8); &psrld ("mm5",8);
740 &mov ($s0,&DWP(0-128,$tbl));
741 &pslld ("mm3",24); &pslld ("mm7",24);
742 &mov ($s1,&DWP(64-128,$tbl));
743 &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= (r2^r0)<<8
744 &mov ($s2,&DWP(128-128,$tbl));
745 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= (r2^r0)>>24
746 &mov ($s3,&DWP(192-128,$tbl));
748 &pxor ("mm0","mm2"); &pxor ("mm4","mm6");
749 &jmp (&label("loop"));
751 &set_label("out",16);
752 &pxor ("mm0",&QWP(0,$key));
753 &pxor ("mm4",&QWP(8,$key));
756 &function_end_B("_sse_AES_encrypt_compact");
759 ######################################################################
760 # Vanilla block function.
761 ######################################################################
764 { my ($i,$te,@s) = @_;
766 my $out = $i==3?$s[0]:$acc;
768 # lines marked with #%e?x[i] denote "reordered" instructions...
769 if ($i==3) { &mov ($key,$__key); }##%edx
770 else { &mov ($out,$s[0]);
772 if ($i==1) { &shr ($s[0],16); }#%ebx[1]
773 if ($i==2) { &shr ($s[0],24); }#%ecx[2]
774 &mov ($out,&DWP(0,$te,$out,8));
776 if ($i==3) { $tmp=$s[1]; }##%eax
777 &movz ($tmp,&HB($s[1]));
778 &xor ($out,&DWP(3,$te,$tmp,8));
780 if ($i==3) { $tmp=$s[2]; &mov ($s[1],$__s0); }##%ebx
781 else { &mov ($tmp,$s[2]);
783 if ($i==2) { &and ($s[1],0xFF); }#%edx[2]
785 &xor ($out,&DWP(2,$te,$tmp,8));
787 if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }##%ecx
788 elsif($i==2){ &movz ($tmp,&HB($s[3])); }#%ebx[2]
789 else { &mov ($tmp,$s[3]);
791 &xor ($out,&DWP(1,$te,$tmp,8));
792 if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); }
793 if ($i==3) { &mov ($s[3],$acc); }
800 my $out = $i==3?$s[0]:$acc;
802 if ($i==3) { &mov ($key,$__key); }##%edx
803 else { &mov ($out,$s[0]); }
805 if ($i==1) { &shr ($s[0],16); }#%ebx[1]
806 if ($i==2) { &shr ($s[0],24); }#%ecx[2]
807 &mov ($out,&DWP(2,$te,$out,8));
808 &and ($out,0x000000ff);
810 if ($i==3) { $tmp=$s[1]; }##%eax
811 &movz ($tmp,&HB($s[1]));
812 &mov ($tmp,&DWP(0,$te,$tmp,8));
813 &and ($tmp,0x0000ff00);
816 if ($i==3) { $tmp=$s[2]; &mov ($s[1],$__s0); }##%ebx
817 else { &mov ($tmp,$s[2]);
819 if ($i==2) { &and ($s[1],0xFF); }#%edx[2]
821 &mov ($tmp,&DWP(0,$te,$tmp,8));
822 &and ($tmp,0x00ff0000);
825 if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }##%ecx
826 elsif($i==2){ &movz ($tmp,&HB($s[3])); }#%ebx[2]
827 else { &mov ($tmp,$s[3]);
829 &mov ($tmp,&DWP(2,$te,$tmp,8));
830 &and ($tmp,0xff000000);
832 if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); }
833 if ($i==3) { &mov ($s[3],$acc); }
836 &public_label("AES_Te");
837 &function_begin_B("_x86_AES_encrypt");
838 if ($vertical_spin) {
839 # I need high parts of volatile registers to be accessible...
840 &exch ($s1="edi",$key="ebx");
841 &mov ($s2="esi",$acc="ecx");
844 # note that caller is expected to allocate stack frame for me!
845 &mov ($__key,$key); # save key
847 &xor ($s0,&DWP(0,$key)); # xor with key
848 &xor ($s1,&DWP(4,$key));
849 &xor ($s2,&DWP(8,$key));
850 &xor ($s3,&DWP(12,$key));
852 &mov ($acc,&DWP(240,$key)); # load key->rounds
854 if ($small_footprint) {
855 &lea ($acc,&DWP(-2,$acc,$acc));
856 &lea ($acc,&DWP(0,$key,$acc,8));
857 &mov ($__end,$acc); # end of key schedule
859 &set_label("loop",16);
860 if ($vertical_spin) {
861 &encvert($tbl,$s0,$s1,$s2,$s3);
863 &encstep(0,$tbl,$s0,$s1,$s2,$s3);
864 &encstep(1,$tbl,$s1,$s2,$s3,$s0);
865 &encstep(2,$tbl,$s2,$s3,$s0,$s1);
866 &encstep(3,$tbl,$s3,$s0,$s1,$s2);
868 &add ($key,16); # advance rd_key
869 &xor ($s0,&DWP(0,$key));
870 &xor ($s1,&DWP(4,$key));
871 &xor ($s2,&DWP(8,$key));
872 &xor ($s3,&DWP(12,$key));
875 &jb (&label("loop"));
879 &jle (&label("10rounds"));
881 &jle (&label("12rounds"));
883 &set_label("14rounds",4);
884 for ($i=1;$i<3;$i++) {
885 if ($vertical_spin) {
886 &encvert($tbl,$s0,$s1,$s2,$s3);
888 &encstep(0,$tbl,$s0,$s1,$s2,$s3);
889 &encstep(1,$tbl,$s1,$s2,$s3,$s0);
890 &encstep(2,$tbl,$s2,$s3,$s0,$s1);
891 &encstep(3,$tbl,$s3,$s0,$s1,$s2);
893 &xor ($s0,&DWP(16*$i+0,$key));
894 &xor ($s1,&DWP(16*$i+4,$key));
895 &xor ($s2,&DWP(16*$i+8,$key));
896 &xor ($s3,&DWP(16*$i+12,$key));
899 &mov ($__key,$key); # advance rd_key
900 &set_label("12rounds",4);
901 for ($i=1;$i<3;$i++) {
902 if ($vertical_spin) {
903 &encvert($tbl,$s0,$s1,$s2,$s3);
905 &encstep(0,$tbl,$s0,$s1,$s2,$s3);
906 &encstep(1,$tbl,$s1,$s2,$s3,$s0);
907 &encstep(2,$tbl,$s2,$s3,$s0,$s1);
908 &encstep(3,$tbl,$s3,$s0,$s1,$s2);
910 &xor ($s0,&DWP(16*$i+0,$key));
911 &xor ($s1,&DWP(16*$i+4,$key));
912 &xor ($s2,&DWP(16*$i+8,$key));
913 &xor ($s3,&DWP(16*$i+12,$key));
916 &mov ($__key,$key); # advance rd_key
917 &set_label("10rounds",4);
918 for ($i=1;$i<10;$i++) {
919 if ($vertical_spin) {
920 &encvert($tbl,$s0,$s1,$s2,$s3);
922 &encstep(0,$tbl,$s0,$s1,$s2,$s3);
923 &encstep(1,$tbl,$s1,$s2,$s3,$s0);
924 &encstep(2,$tbl,$s2,$s3,$s0,$s1);
925 &encstep(3,$tbl,$s3,$s0,$s1,$s2);
927 &xor ($s0,&DWP(16*$i+0,$key));
928 &xor ($s1,&DWP(16*$i+4,$key));
929 &xor ($s2,&DWP(16*$i+8,$key));
930 &xor ($s3,&DWP(16*$i+12,$key));
934 if ($vertical_spin) {
935 # "reincarnate" some registers for "horizontal" spin...
936 &mov ($s1="ebx",$key="edi");
937 &mov ($s2="ecx",$acc="esi");
939 &enclast(0,$tbl,$s0,$s1,$s2,$s3);
940 &enclast(1,$tbl,$s1,$s2,$s3,$s0);
941 &enclast(2,$tbl,$s2,$s3,$s0,$s1);
942 &enclast(3,$tbl,$s3,$s0,$s1,$s2);
944 &add ($key,$small_footprint?16:160);
945 &xor ($s0,&DWP(0,$key));
946 &xor ($s1,&DWP(4,$key));
947 &xor ($s2,&DWP(8,$key));
948 &xor ($s3,&DWP(12,$key));
952 &set_label("AES_Te",64); # Yes! I keep it in the code segment!
953 &_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6);
954 &_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591);
955 &_data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56);
956 &_data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec);
957 &_data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa);
958 &_data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb);
959 &_data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45);
960 &_data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b);
961 &_data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c);
962 &_data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83);
963 &_data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9);
964 &_data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a);
965 &_data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d);
966 &_data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f);
967 &_data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df);
968 &_data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea);
969 &_data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34);
970 &_data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b);
971 &_data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d);
972 &_data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413);
973 &_data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1);
974 &_data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6);
975 &_data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972);
976 &_data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85);
977 &_data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed);
978 &_data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511);
979 &_data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe);
980 &_data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b);
981 &_data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05);
982 &_data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1);
983 &_data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142);
984 &_data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf);
985 &_data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3);
986 &_data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e);
987 &_data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a);
988 &_data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6);
989 &_data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3);
990 &_data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b);
991 &_data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428);
992 &_data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad);
993 &_data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14);
994 &_data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8);
995 &_data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4);
996 &_data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2);
997 &_data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda);
998 &_data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949);
999 &_data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf);
1000 &_data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810);
1001 &_data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c);
1002 &_data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697);
1003 &_data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e);
1004 &_data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f);
1005 &_data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc);
1006 &_data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c);
1007 &_data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969);
1008 &_data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27);
1009 &_data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122);
1010 &_data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433);
1011 &_data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9);
1012 &_data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5);
1013 &_data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a);
1014 &_data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0);
1015 &_data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e);
1016 &_data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c);
1018 #Te4 # four copies of Te4 to choose from to avoid L1 aliasing
1019 &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
1020 &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
1021 &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
1022 &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
1023 &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
1024 &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
1025 &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
1026 &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
1027 &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
1028 &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
1029 &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
1030 &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
1031 &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
1032 &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
1033 &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
1034 &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
1035 &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
1036 &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
1037 &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
1038 &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
1039 &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
1040 &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
1041 &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
1042 &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
1043 &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
1044 &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
1045 &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
1046 &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
1047 &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
1048 &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
1049 &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
1050 &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
1052 &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
1053 &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
1054 &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
1055 &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
1056 &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
1057 &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
1058 &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
1059 &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
1060 &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
1061 &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
1062 &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
1063 &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
1064 &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
1065 &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
1066 &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
1067 &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
1068 &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
1069 &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
1070 &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
1071 &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
1072 &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
1073 &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
1074 &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
1075 &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
1076 &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
1077 &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
1078 &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
1079 &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
1080 &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
1081 &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
1082 &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
1083 &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
1085 &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
1086 &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
1087 &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
1088 &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
1089 &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
1090 &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
1091 &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
1092 &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
1093 &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
1094 &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
1095 &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
1096 &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
1097 &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
1098 &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
1099 &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
1100 &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
1101 &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
1102 &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
1103 &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
1104 &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
1105 &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
1106 &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
1107 &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
1108 &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
1109 &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
1110 &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
1111 &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
1112 &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
1113 &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
1114 &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
1115 &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
1116 &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
1118 &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
1119 &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
1120 &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
1121 &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
1122 &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
1123 &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
1124 &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
1125 &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
1126 &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
1127 &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
1128 &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
1129 &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
1130 &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
1131 &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
1132 &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
1133 &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
1134 &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
1135 &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
1136 &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
1137 &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
1138 &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
1139 &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
1140 &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
1141 &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
1142 &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
1143 &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
1144 &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
1145 &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
1146 &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
1147 &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
1148 &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
1149 &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
1151 &data_word(0x00000001, 0x00000002, 0x00000004, 0x00000008);
1152 &data_word(0x00000010, 0x00000020, 0x00000040, 0x00000080);
1153 &data_word(0x0000001b, 0x00000036, 0x00000000, 0x00000000);
1154 &data_word(0x00000000, 0x00000000, 0x00000000, 0x00000000);
1155 &function_end_B("_x86_AES_encrypt");
1157 # void AES_encrypt (const void *inp,void *out,const AES_KEY *key);
1158 &public_label("AES_Te");
1159 &function_begin("AES_encrypt");
1160 &mov ($acc,&wparam(0)); # load inp
1161 &mov ($key,&wparam(2)); # load key
1165 &and ("esp",-64); # align to cache-line
1167 # place stack frame just "above" the key schedule
1168 &lea ($s1,&DWP(-64-63,$key));
1171 &and ($s1,0x3C0); # modulo 1024, but aligned to cache-line
1173 &add ("esp",4); # 4 is reserved for caller's return address
1174 &mov ($_esp,$s0); # save stack pointer
1176 &call (&label("pic_point")); # make it PIC!
1177 &set_label("pic_point");
1179 &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if (!$x86only);
1180 &lea ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
1182 # pick Te4 copy which can't "overlap" with stack frame or key schedule
1183 &lea ($s1,&DWP(768-4,"esp"));
1186 &lea ($tbl,&DWP(2048+128,$tbl,$s1));
1189 &bt (&DWP(0,$s0),25); # check for SSE bit
1190 &jnc (&label("x86"));
1192 &movq ("mm0",&QWP(0,$acc));
1193 &movq ("mm4",&QWP(8,$acc));
1194 &call ("_sse_AES_encrypt_compact");
1195 &mov ("esp",$_esp); # restore stack pointer
1196 &mov ($acc,&wparam(1)); # load out
1197 &movq (&QWP(0,$acc),"mm0"); # write output data
1198 &movq (&QWP(8,$acc),"mm4");
1202 &set_label("x86",16);
1204 &mov ($s0,&DWP(0,$acc)); # load input data
1205 &mov ($s1,&DWP(4,$acc));
1206 &mov ($s2,&DWP(8,$acc));
1207 &mov ($s3,&DWP(12,$acc));
1208 &call ("_x86_AES_encrypt_compact");
1209 &mov ("esp",$_esp); # restore stack pointer
1210 &mov ($acc,&wparam(1)); # load out
1211 &mov (&DWP(0,$acc),$s0); # write output data
1212 &mov (&DWP(4,$acc),$s1);
1213 &mov (&DWP(8,$acc),$s2);
1214 &mov (&DWP(12,$acc),$s3);
1215 &function_end("AES_encrypt");
1217 #--------------------------------------------------------------------#
1219 ######################################################################
1220 # "Compact" block function
1221 ######################################################################
1225 while ($#_>5) { pop(@_); $Fn=sub{}; }
1228 my $out = $i==3?$s[0]:$acc;
1230 # $Fn is used in first compact round and its purpose is to
1231 # void restoration of some values from stack, so that after
1232 # 4xdeccompact with extra argument $key, $s0 and $s1 values
1234 if($i==3) { &$Fn ($key,$__key); }
1235 else { &mov ($out,$s[0]); }
1237 &movz ($out,&BP(-128,$td,$out,1));
1239 if ($i==3) { $tmp=$s[1]; }
1240 &movz ($tmp,&HB($s[1]));
1241 &movz ($tmp,&BP(-128,$td,$tmp,1));
1245 if ($i==3) { $tmp=$s[2]; &mov ($s[1],$acc); }
1246 else { mov ($tmp,$s[2]); }
1249 &movz ($tmp,&BP(-128,$td,$tmp,1));
1253 if ($i==3) { $tmp=$s[3]; &$Fn ($s[2],$__s1); }
1254 else { &mov ($tmp,$s[3]); }
1256 &movz ($tmp,&BP(-128,$td,$tmp,1));
1259 if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); }
1260 if ($i==3) { &$Fn ($s[3],$__s0); }
1263 # must be called with 2,3,0,1 as argument sequence!!!
1265 { my @s = ($s0,$s1,$s2,$s3);
1268 my $tp2 = @s[($i+2)%4]; $tp2 = @s[2] if ($i==1);
1269 my $tp4 = @s[($i+3)%4]; $tp4 = @s[3] if ($i==1);
1273 &and ($acc,0x80808080);
1277 &and ($tp2,0x7f7f7f7f);
1280 &and ($acc,0x1b1b1b1b);
1284 &and ($acc,0x80808080);
1287 &xor ($tp2,$s[$i]); # tp2^tp1
1289 &and ($tp4,0x7f7f7f7f);
1292 &and ($acc,0x1b1b1b1b);
1296 &and ($acc,0x80808080);
1299 &xor ($tp4,$s[$i]); # tp4^tp1
1301 &and ($tp8,0x7f7f7f7f);
1304 &and ($acc,0x1b1b1b1b);
1305 &rotl ($s[$i],8); # = ROTATE(tp1,8)
1313 &xor ($s[$i],$tp8); # ^= tp8^(tp4^tp1)^(tp2^tp1)
1315 &xor ($s[$i],$tp2); # ^= ROTATE(tp8^tp2^tp1,24)
1317 &xor ($s[$i],$tp4); # ^= ROTATE(tp8^tp4^tp1,16)
1318 &xor ($s[$i],$tp8); # ^= ROTATE(tp8,8)
1320 &mov ($s[0],$__s0) if($i==2); #prefetch $s0
1321 &mov ($s[1],$__s1) if($i==3); #prefetch $s1
1322 &mov ($s[2],$__s2) if($i==1);
1323 &mov ($s[3],$__s3) if($i==1);
1324 &mov (&DWP(4+4*$i,"esp"),$s[$i]) if($i>=2);
1327 &public_label("AES_Td");
1328 &function_begin_B("_x86_AES_decrypt_compact");
1329 # note that caller is expected to allocate stack frame for me!
1330 &mov ($__key,$key); # save key
1332 &xor ($s0,&DWP(0,$key)); # xor with key
1333 &xor ($s1,&DWP(4,$key));
1334 &xor ($s2,&DWP(8,$key));
1335 &xor ($s3,&DWP(12,$key));
1337 &mov ($acc,&DWP(240,$key)); # load key->rounds
1339 &lea ($acc,&DWP(-2,$acc,$acc));
1340 &lea ($acc,&DWP(0,$key,$acc,8));
1341 &mov ($__end,$acc); # end of key schedule
1344 &mov ($key,&DWP(0-128,$tbl));
1345 &mov ($acc,&DWP(32-128,$tbl));
1346 &mov ($key,&DWP(64-128,$tbl));
1347 &mov ($acc,&DWP(96-128,$tbl));
1348 &mov ($key,&DWP(128-128,$tbl));
1349 &mov ($acc,&DWP(160-128,$tbl));
1350 &mov ($key,&DWP(192-128,$tbl));
1351 &mov ($acc,&DWP(224-128,$tbl));
1353 &set_label("loop",16);
1355 &deccompact(0,$tbl,$s0,$s3,$s2,$s1,1);
1356 &deccompact(1,$tbl,$s1,$s0,$s3,$s2,1);
1357 &deccompact(2,$tbl,$s2,$s1,$s0,$s3,1);
1358 &deccompact(3,$tbl,$s3,$s2,$s1,$s0,1);
1365 &add ($key,16); # advance rd_key
1366 &xor ($s0,&DWP(0,$key));
1367 &xor ($s1,&DWP(4,$key));
1368 &xor ($s2,&DWP(8,$key));
1369 &xor ($s3,&DWP(12,$key));
1373 &jb (&label("loop"));
1375 &deccompact(0,$tbl,$s0,$s3,$s2,$s1);
1376 &deccompact(1,$tbl,$s1,$s0,$s3,$s2);
1377 &deccompact(2,$tbl,$s2,$s1,$s0,$s3);
1378 &deccompact(3,$tbl,$s3,$s2,$s1,$s0);
1380 &xor ($s0,&DWP(16,$key));
1381 &xor ($s1,&DWP(20,$key));
1382 &xor ($s2,&DWP(24,$key));
1383 &xor ($s3,&DWP(28,$key));
1386 &function_end_B("_x86_AES_decrypt_compact");
1388 ######################################################################
1389 # "Compact" SSE block function.
1390 ######################################################################
1392 sub sse_deccompact()
1394 &pshufw ("mm1","mm0",0x0c); # 7, 6, 1, 0
1395 &movd ("eax","mm1"); # 7, 6, 1, 0
1397 &pshufw ("mm5","mm4",0x09); # 13,12,11,10
1398 &movz ($acc,&LB("eax")); # 0
1399 &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 0
1400 &movd ("ebx","mm5"); # 13,12,11,10
1401 &movz ("edx",&HB("eax")); # 1
1402 &movz ("edx",&BP(-128,$tbl,"edx",1)); # 1
1405 &pshufw ("mm2","mm0",0x06); # 3, 2, 5, 4
1406 &movz ($acc,&LB("ebx")); # 10
1407 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 10
1408 &shl ($acc,16); # 10
1409 &or ("ecx",$acc); # 10
1410 &shr ("eax",16); # 7, 6
1411 &movz ($acc,&HB("ebx")); # 11
1412 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 11
1413 &shl ($acc,24); # 11
1414 &or ("edx",$acc); # 11
1415 &shr ("ebx",16); # 13,12
1417 &pshufw ("mm6","mm4",0x03); # 9, 8,15,14
1418 &movz ($acc,&HB("eax")); # 7
1419 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 7
1421 &or ("ecx",$acc); # 7
1422 &movz ($acc,&HB("ebx")); # 13
1423 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 13
1425 &or ("ecx",$acc); # 13
1426 &movd ("mm0","ecx"); # t[0] collected
1428 &movz ($acc,&LB("eax")); # 6
1429 &movd ("eax","mm2"); # 3, 2, 5, 4
1430 &movz ("ecx",&BP(-128,$tbl,$acc,1)); # 6
1431 &shl ("ecx",16); # 6
1432 &movz ($acc,&LB("ebx")); # 12
1433 &movd ("ebx","mm6"); # 9, 8,15,14
1434 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 12
1435 &or ("ecx",$acc); # 12
1437 &movz ($acc,&LB("eax")); # 4
1438 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 4
1439 &or ("edx",$acc); # 4
1440 &movz ($acc,&LB("ebx")); # 14
1441 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 14
1442 &shl ($acc,16); # 14
1443 &or ("edx",$acc); # 14
1444 &movd ("mm1","edx"); # t[1] collected
1446 &movz ($acc,&HB("eax")); # 5
1447 &movz ("edx",&BP(-128,$tbl,$acc,1)); # 5
1449 &movz ($acc,&HB("ebx")); # 15
1450 &shr ("eax",16); # 3, 2
1451 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 15
1452 &shl ($acc,24); # 15
1453 &or ("edx",$acc); # 15
1454 &shr ("ebx",16); # 9, 8
1456 &punpckldq ("mm0","mm1"); # t[0,1] collected
1458 &movz ($acc,&HB("ebx")); # 9
1459 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 9
1461 &or ("ecx",$acc); # 9
1462 &and ("ebx",0xff); # 8
1463 &movz ("ebx",&BP(-128,$tbl,"ebx",1)); # 8
1464 &or ("edx","ebx"); # 8
1465 &movz ($acc,&LB("eax")); # 2
1466 &movz ($acc,&BP(-128,$tbl,$acc,1)); # 2
1468 &or ("edx",$acc); # 2
1469 &movd ("mm4","edx"); # t[2] collected
1470 &movz ("eax",&HB("eax")); # 3
1471 &movz ("eax",&BP(-128,$tbl,"eax",1)); # 3
1472 &shl ("eax",24); # 3
1473 &or ("ecx","eax"); # 3
1474 &movd ("mm5","ecx"); # t[3] collected
1476 &punpckldq ("mm4","mm5"); # t[2,3] collected
1480 &public_label("AES_Td");
1481 &function_begin_B("_sse_AES_decrypt_compact");
1482 &pxor ("mm0",&QWP(0,$key)); # 7, 6, 5, 4, 3, 2, 1, 0
1483 &pxor ("mm4",&QWP(8,$key)); # 15,14,13,12,11,10, 9, 8
1485 # note that caller is expected to allocate stack frame for me!
1486 &mov ($acc,&DWP(240,$key)); # load key->rounds
1487 &lea ($acc,&DWP(-2,$acc,$acc));
1488 &lea ($acc,&DWP(0,$key,$acc,8));
1489 &mov ($__end,$acc); # end of key schedule
1491 &mov ($s0,0x1b1b1b1b); # magic constant
1492 &mov (&DWP(8,"esp"),$s0);
1493 &mov (&DWP(12,"esp"),$s0);
1496 &mov ($s0,&DWP(0-128,$tbl));
1497 &mov ($s1,&DWP(32-128,$tbl));
1498 &mov ($s2,&DWP(64-128,$tbl));
1499 &mov ($s3,&DWP(96-128,$tbl));
1500 &mov ($s0,&DWP(128-128,$tbl));
1501 &mov ($s1,&DWP(160-128,$tbl));
1502 &mov ($s2,&DWP(192-128,$tbl));
1503 &mov ($s3,&DWP(224-128,$tbl));
1505 &set_label("loop",16);
1509 &ja (&label("out"));
1511 # ROTATE(x^y,N) == ROTATE(x,N)^ROTATE(y,N)
1512 &movq ("mm3","mm0"); &movq ("mm7","mm4");
1513 &movq ("mm2","mm0",1); &movq ("mm6","mm4",1);
1514 &movq ("mm1","mm0"); &movq ("mm5","mm4");
1515 &pshufw ("mm0","mm0",0xb1); &pshufw ("mm4","mm4",0xb1);# = ROTATE(tp0,16)
1516 &pslld ("mm2",8); &pslld ("mm6",8);
1517 &psrld ("mm3",8); &psrld ("mm7",8);
1518 &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= tp0<<8
1519 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp0>>8
1520 &pslld ("mm2",16); &pslld ("mm6",16);
1521 &psrld ("mm3",16); &psrld ("mm7",16);
1522 &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= tp0<<24
1523 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp0>>24
1525 &movq ("mm3",&QWP(8,"esp"));
1526 &pxor ("mm2","mm2"); &pxor ("mm6","mm6");
1527 &pcmpgtb("mm2","mm1"); &pcmpgtb("mm6","mm5");
1528 &pand ("mm2","mm3"); &pand ("mm6","mm3");
1529 &paddb ("mm1","mm1"); &paddb ("mm5","mm5");
1530 &pxor ("mm1","mm2"); &pxor ("mm5","mm6"); # tp2
1531 &movq ("mm3","mm1"); &movq ("mm7","mm5");
1532 &movq ("mm2","mm1"); &movq ("mm6","mm5");
1533 &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp2
1534 &pslld ("mm3",24); &pslld ("mm7",24);
1535 &psrld ("mm2",8); &psrld ("mm6",8);
1536 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp2<<24
1537 &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= tp2>>8
1539 &movq ("mm2",&QWP(8,"esp"));
1540 &pxor ("mm3","mm3"); &pxor ("mm7","mm7");
1541 &pcmpgtb("mm3","mm1"); &pcmpgtb("mm7","mm5");
1542 &pand ("mm3","mm2"); &pand ("mm7","mm2");
1543 &paddb ("mm1","mm1"); &paddb ("mm5","mm5");
1544 &pxor ("mm1","mm3"); &pxor ("mm5","mm7"); # tp4
1545 &pshufw ("mm3","mm1",0xb1); &pshufw ("mm7","mm5",0xb1);
1546 &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp4
1547 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= ROTATE(tp4,16)
1549 &pxor ("mm3","mm3"); &pxor ("mm7","mm7");
1550 &pcmpgtb("mm3","mm1"); &pcmpgtb("mm7","mm5");
1551 &pand ("mm3","mm2"); &pand ("mm7","mm2");
1552 &paddb ("mm1","mm1"); &paddb ("mm5","mm5");
1553 &pxor ("mm1","mm3"); &pxor ("mm5","mm7"); # tp8
1554 &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp8
1555 &movq ("mm3","mm1"); &movq ("mm7","mm5");
1556 &pshufw ("mm2","mm1",0xb1); &pshufw ("mm6","mm5",0xb1);
1557 &pxor ("mm0","mm2"); &pxor ("mm4","mm6"); # ^= ROTATE(tp8,16)
1558 &pslld ("mm1",8); &pslld ("mm5",8);
1559 &psrld ("mm3",8); &psrld ("mm7",8);
1560 &movq ("mm2",&QWP(0,$key)); &movq ("mm6",&QWP(8,$key));
1561 &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp8<<8
1562 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp8>>8
1563 &mov ($s0,&DWP(0-128,$tbl));
1564 &pslld ("mm1",16); &pslld ("mm5",16);
1565 &mov ($s1,&DWP(64-128,$tbl));
1566 &psrld ("mm3",16); &psrld ("mm7",16);
1567 &mov ($s2,&DWP(128-128,$tbl));
1568 &pxor ("mm0","mm1"); &pxor ("mm4","mm5"); # ^= tp8<<24
1569 &mov ($s3,&DWP(192-128,$tbl));
1570 &pxor ("mm0","mm3"); &pxor ("mm4","mm7"); # ^= tp8>>24
1572 &pxor ("mm0","mm2"); &pxor ("mm4","mm6");
1573 &jmp (&label("loop"));
1575 &set_label("out",16);
1576 &pxor ("mm0",&QWP(0,$key));
1577 &pxor ("mm4",&QWP(8,$key));
1580 &function_end_B("_sse_AES_decrypt_compact");
1583 ######################################################################
1584 # Vanilla block function.
1585 ######################################################################
1588 { my ($i,$td,@s) = @_;
1590 my $out = $i==3?$s[0]:$acc;
1592 # no instructions are reordered, as performance appears
1593 # optimal... or rather that all attempts to reorder didn't
1594 # result in better performance [which by the way is not a
1595 # bit lower than ecryption].
1596 if($i==3) { &mov ($key,$__key); }
1597 else { &mov ($out,$s[0]); }
1599 &mov ($out,&DWP(0,$td,$out,8));
1601 if ($i==3) { $tmp=$s[1]; }
1602 &movz ($tmp,&HB($s[1]));
1603 &xor ($out,&DWP(3,$td,$tmp,8));
1605 if ($i==3) { $tmp=$s[2]; &mov ($s[1],$acc); }
1606 else { &mov ($tmp,$s[2]); }
1609 &xor ($out,&DWP(2,$td,$tmp,8));
1611 if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }
1612 else { &mov ($tmp,$s[3]); }
1614 &xor ($out,&DWP(1,$td,$tmp,8));
1615 if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); }
1616 if ($i==3) { &mov ($s[3],$__s0); }
1621 { my ($i,$td,@s)=@_;
1623 my $out = $i==3?$s[0]:$acc;
1625 if($i==0) { &lea ($td,&DWP(2048+128,$td));
1626 &mov ($tmp,&DWP(0-128,$td));
1627 &mov ($acc,&DWP(32-128,$td));
1628 &mov ($tmp,&DWP(64-128,$td));
1629 &mov ($acc,&DWP(96-128,$td));
1630 &mov ($tmp,&DWP(128-128,$td));
1631 &mov ($acc,&DWP(160-128,$td));
1632 &mov ($tmp,&DWP(192-128,$td));
1633 &mov ($acc,&DWP(224-128,$td));
1634 &lea ($td,&DWP(-128,$td)); }
1635 if($i==3) { &mov ($key,$__key); }
1636 else { &mov ($out,$s[0]); }
1638 &movz ($out,&BP(0,$td,$out,1));
1640 if ($i==3) { $tmp=$s[1]; }
1641 &movz ($tmp,&HB($s[1]));
1642 &movz ($tmp,&BP(0,$td,$tmp,1));
1646 if ($i==3) { $tmp=$s[2]; &mov ($s[1],$acc); }
1647 else { mov ($tmp,$s[2]); }
1650 &movz ($tmp,&BP(0,$td,$tmp,1));
1654 if ($i==3) { $tmp=$s[3]; &mov ($s[2],$__s1); }
1655 else { &mov ($tmp,$s[3]); }
1657 &movz ($tmp,&BP(0,$td,$tmp,1));
1660 if ($i<2) { &mov (&DWP(4+4*$i,"esp"),$out); }
1661 if ($i==3) { &mov ($s[3],$__s0);
1662 &lea ($td,&DWP(-2048,$td)); }
1665 &public_label("AES_Td");
1666 &function_begin_B("_x86_AES_decrypt");
1667 # note that caller is expected to allocate stack frame for me!
1668 &mov ($__key,$key); # save key
1670 &xor ($s0,&DWP(0,$key)); # xor with key
1671 &xor ($s1,&DWP(4,$key));
1672 &xor ($s2,&DWP(8,$key));
1673 &xor ($s3,&DWP(12,$key));
1675 &mov ($acc,&DWP(240,$key)); # load key->rounds
1677 if ($small_footprint) {
1678 &lea ($acc,&DWP(-2,$acc,$acc));
1679 &lea ($acc,&DWP(0,$key,$acc,8));
1680 &mov ($__end,$acc); # end of key schedule
1681 &set_label("loop",16);
1682 &decstep(0,$tbl,$s0,$s3,$s2,$s1);
1683 &decstep(1,$tbl,$s1,$s0,$s3,$s2);
1684 &decstep(2,$tbl,$s2,$s1,$s0,$s3);
1685 &decstep(3,$tbl,$s3,$s2,$s1,$s0);
1686 &add ($key,16); # advance rd_key
1687 &xor ($s0,&DWP(0,$key));
1688 &xor ($s1,&DWP(4,$key));
1689 &xor ($s2,&DWP(8,$key));
1690 &xor ($s3,&DWP(12,$key));
1693 &jb (&label("loop"));
1697 &jle (&label("10rounds"));
1699 &jle (&label("12rounds"));
1701 &set_label("14rounds",4);
1702 for ($i=1;$i<3;$i++) {
1703 &decstep(0,$tbl,$s0,$s3,$s2,$s1);
1704 &decstep(1,$tbl,$s1,$s0,$s3,$s2);
1705 &decstep(2,$tbl,$s2,$s1,$s0,$s3);
1706 &decstep(3,$tbl,$s3,$s2,$s1,$s0);
1707 &xor ($s0,&DWP(16*$i+0,$key));
1708 &xor ($s1,&DWP(16*$i+4,$key));
1709 &xor ($s2,&DWP(16*$i+8,$key));
1710 &xor ($s3,&DWP(16*$i+12,$key));
1713 &mov ($__key,$key); # advance rd_key
1714 &set_label("12rounds",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,$key); # advance rd_key
1727 &set_label("10rounds",4);
1728 for ($i=1;$i<10;$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));
1740 &declast(0,$tbl,$s0,$s3,$s2,$s1);
1741 &declast(1,$tbl,$s1,$s0,$s3,$s2);
1742 &declast(2,$tbl,$s2,$s1,$s0,$s3);
1743 &declast(3,$tbl,$s3,$s2,$s1,$s0);
1745 &add ($key,$small_footprint?16:160);
1746 &xor ($s0,&DWP(0,$key));
1747 &xor ($s1,&DWP(4,$key));
1748 &xor ($s2,&DWP(8,$key));
1749 &xor ($s3,&DWP(12,$key));
1753 &set_label("AES_Td",64); # Yes! I keep it in the code segment!
1754 &_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a);
1755 &_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b);
1756 &_data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5);
1757 &_data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5);
1758 &_data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d);
1759 &_data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b);
1760 &_data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295);
1761 &_data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e);
1762 &_data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927);
1763 &_data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d);
1764 &_data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362);
1765 &_data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9);
1766 &_data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52);
1767 &_data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566);
1768 &_data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3);
1769 &_data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed);
1770 &_data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e);
1771 &_data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4);
1772 &_data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4);
1773 &_data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd);
1774 &_data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d);
1775 &_data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060);
1776 &_data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967);
1777 &_data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879);
1778 &_data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000);
1779 &_data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c);
1780 &_data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36);
1781 &_data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624);
1782 &_data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b);
1783 &_data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c);
1784 &_data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12);
1785 &_data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14);
1786 &_data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3);
1787 &_data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b);
1788 &_data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8);
1789 &_data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684);
1790 &_data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7);
1791 &_data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177);
1792 &_data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947);
1793 &_data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322);
1794 &_data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498);
1795 &_data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f);
1796 &_data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54);
1797 &_data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382);
1798 &_data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf);
1799 &_data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb);
1800 &_data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83);
1801 &_data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef);
1802 &_data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029);
1803 &_data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235);
1804 &_data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733);
1805 &_data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117);
1806 &_data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4);
1807 &_data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546);
1808 &_data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb);
1809 &_data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d);
1810 &_data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb);
1811 &_data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a);
1812 &_data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773);
1813 &_data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478);
1814 &_data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2);
1815 &_data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff);
1816 &_data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664);
1817 &_data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0);
1819 #Td4: # four copies of Td4 to choose from to avoid L1 aliasing
1820 &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
1821 &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
1822 &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
1823 &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
1824 &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
1825 &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
1826 &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
1827 &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
1828 &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
1829 &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
1830 &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
1831 &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
1832 &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
1833 &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
1834 &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
1835 &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
1836 &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
1837 &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
1838 &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
1839 &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
1840 &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
1841 &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
1842 &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
1843 &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
1844 &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
1845 &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
1846 &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
1847 &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
1848 &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
1849 &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
1850 &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
1851 &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
1853 &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
1854 &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
1855 &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
1856 &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
1857 &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
1858 &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
1859 &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
1860 &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
1861 &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
1862 &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
1863 &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
1864 &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
1865 &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
1866 &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
1867 &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
1868 &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
1869 &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
1870 &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
1871 &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
1872 &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
1873 &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
1874 &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
1875 &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
1876 &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
1877 &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
1878 &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
1879 &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
1880 &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
1881 &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
1882 &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
1883 &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
1884 &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
1886 &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
1887 &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
1888 &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
1889 &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
1890 &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
1891 &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
1892 &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
1893 &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
1894 &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
1895 &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
1896 &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
1897 &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
1898 &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
1899 &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
1900 &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
1901 &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
1902 &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
1903 &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
1904 &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
1905 &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
1906 &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
1907 &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
1908 &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
1909 &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
1910 &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
1911 &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
1912 &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
1913 &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
1914 &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
1915 &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
1916 &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
1917 &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
1919 &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
1920 &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
1921 &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
1922 &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
1923 &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
1924 &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
1925 &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
1926 &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
1927 &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
1928 &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
1929 &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
1930 &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
1931 &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
1932 &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
1933 &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
1934 &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
1935 &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
1936 &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
1937 &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
1938 &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
1939 &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
1940 &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
1941 &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
1942 &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
1943 &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
1944 &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
1945 &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
1946 &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
1947 &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
1948 &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
1949 &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
1950 &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
1951 &function_end_B("_x86_AES_decrypt");
1953 # void AES_decrypt (const void *inp,void *out,const AES_KEY *key);
1954 &public_label("AES_Td");
1955 &function_begin("AES_decrypt");
1956 &mov ($acc,&wparam(0)); # load inp
1957 &mov ($key,&wparam(2)); # load key
1961 &and ("esp",-64); # align to cache-line
1963 # place stack frame just "above" the key schedule
1964 &lea ($s1,&DWP(-64-63,$key));
1967 &and ($s1,0x3C0); # modulo 1024, but aligned to cache-line
1969 &add ("esp",4); # 4 is reserved for caller's return address
1970 &mov ($_esp,$s0); # save stack pointer
1972 &call (&label("pic_point")); # make it PIC!
1973 &set_label("pic_point");
1975 &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only);
1976 &lea ($tbl,&DWP(&label("AES_Td")."-".&label("pic_point"),$tbl));
1978 # pick Td4 copy which can't "overlap" with stack frame or key schedule
1979 &lea ($s1,&DWP(768-4,"esp"));
1982 &lea ($tbl,&DWP(2048+128,$tbl,$s1));
1985 &bt (&DWP(0,$s0),25); # check for SSE bit
1986 &jnc (&label("x86"));
1988 &movq ("mm0",&QWP(0,$acc));
1989 &movq ("mm4",&QWP(8,$acc));
1990 &call ("_sse_AES_decrypt_compact");
1991 &mov ("esp",$_esp); # restore stack pointer
1992 &mov ($acc,&wparam(1)); # load out
1993 &movq (&QWP(0,$acc),"mm0"); # write output data
1994 &movq (&QWP(8,$acc),"mm4");
1998 &set_label("x86",16);
2000 &mov ($s0,&DWP(0,$acc)); # load input data
2001 &mov ($s1,&DWP(4,$acc));
2002 &mov ($s2,&DWP(8,$acc));
2003 &mov ($s3,&DWP(12,$acc));
2004 &call ("_x86_AES_decrypt_compact");
2005 &mov ("esp",$_esp); # restore stack pointer
2006 &mov ($acc,&wparam(1)); # load out
2007 &mov (&DWP(0,$acc),$s0); # write output data
2008 &mov (&DWP(4,$acc),$s1);
2009 &mov (&DWP(8,$acc),$s2);
2010 &mov (&DWP(12,$acc),$s3);
2011 &function_end("AES_decrypt");
2013 # void AES_cbc_encrypt (const void char *inp, unsigned char *out,
2014 # size_t length, const AES_KEY *key,
2015 # unsigned char *ivp,const int enc);
2017 # stack frame layout
2018 # -4(%esp) # return address 0(%esp)
2019 # 0(%esp) # s0 backing store 4(%esp)
2020 # 4(%esp) # s1 backing store 8(%esp)
2021 # 8(%esp) # s2 backing store 12(%esp)
2022 # 12(%esp) # s3 backing store 16(%esp)
2023 # 16(%esp) # key backup 20(%esp)
2024 # 20(%esp) # end of key schedule 24(%esp)
2025 # 24(%esp) # %ebp backup 28(%esp)
2026 # 28(%esp) # %esp backup
2027 my $_inp=&DWP(32,"esp"); # copy of wparam(0)
2028 my $_out=&DWP(36,"esp"); # copy of wparam(1)
2029 my $_len=&DWP(40,"esp"); # copy of wparam(2)
2030 my $_key=&DWP(44,"esp"); # copy of wparam(3)
2031 my $_ivp=&DWP(48,"esp"); # copy of wparam(4)
2032 my $_tmp=&DWP(52,"esp"); # volatile variable
2034 my $ivec=&DWP(60,"esp"); # ivec[16]
2035 my $aes_key=&DWP(76,"esp"); # copy of aes_key
2036 my $mark=&DWP(76+240,"esp"); # copy of aes_key->rounds
2038 &public_label("AES_Te");
2039 &public_label("AES_Td");
2040 &function_begin("AES_cbc_encrypt");
2041 &mov ($s2 eq "ecx"? $s2 : "",&wparam(2)); # load len
2043 &je (&label("drop_out"));
2045 &call (&label("pic_point")); # make it PIC!
2046 &set_label("pic_point");
2048 &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only);
2050 &cmp (&wparam(5),0);
2051 &lea ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
2052 &jne (&label("picked_te"));
2053 &lea ($tbl,&DWP(&label("AES_Td")."-".&label("AES_Te"),$tbl));
2054 &set_label("picked_te");
2056 # one can argue if this is required
2060 &cmp ($s2,$speed_limit);
2061 &jb (&label("slow_way"));
2063 &jnz (&label("slow_way"));
2065 &bt (&DWP(0,$s0),28); # check for hyper-threading bit
2066 &jc (&label("slow_way"));
2068 # pre-allocate aligned stack frame...
2069 &lea ($acc,&DWP(-80-244,"esp"));
2072 # ... and make sure it doesn't alias with $tbl modulo 4096
2074 &lea ($s1,&DWP(2048+256,$tbl));
2076 &and ($s0,0xfff); # s = %ebp&0xfff
2077 &and ($s1,0xfff); # e = (%ebp+2048+256)&0xfff
2078 &and ($s3,0xfff); # p = %esp&0xfff
2080 &cmp ($s3,$s1); # if (p>=e) %esp =- (p-e);
2081 &jb (&label("tbl_break_out"));
2084 &jmp (&label("tbl_ok"));
2085 &set_label("tbl_break_out",4); # else %esp -= (p-s)&0xfff + framesz;
2090 &set_label("tbl_ok",4);
2092 &lea ($s3,&wparam(0)); # obtain pointer to parameter block
2093 &exch ("esp",$acc); # allocate stack frame
2094 &add ("esp",4); # reserve for return address!
2095 &mov ($_tbl,$tbl); # save %ebp
2096 &mov ($_esp,$acc); # save %esp
2098 &mov ($s0,&DWP(0,$s3)); # load inp
2099 &mov ($s1,&DWP(4,$s3)); # load out
2100 #&mov ($s2,&DWP(8,$s3)); # load len
2101 &mov ($key,&DWP(12,$s3)); # load key
2102 &mov ($acc,&DWP(16,$s3)); # load ivp
2103 &mov ($s3,&DWP(20,$s3)); # load enc flag
2105 &mov ($_inp,$s0); # save copy of inp
2106 &mov ($_out,$s1); # save copy of out
2107 &mov ($_len,$s2); # save copy of len
2108 &mov ($_key,$key); # save copy of key
2109 &mov ($_ivp,$acc); # save copy of ivp
2111 &mov ($mark,0); # copy of aes_key->rounds = 0;
2112 # do we copy key schedule to stack?
2113 &mov ($s1 eq "ebx" ? $s1 : "",$key);
2114 &mov ($s2 eq "ecx" ? $s2 : "",244/4);
2118 &lea ("edi",$aes_key);
2119 &cmp ($s1,2048+256);
2120 &jb (&label("do_copy"));
2121 &cmp ($s1,4096-244);
2122 &jb (&label("skip_copy"));
2123 &set_label("do_copy",4);
2125 &data_word(0xA5F3F689); # rep movsd
2126 &set_label("skip_copy");
2129 &set_label("prefetch_tbl",4);
2130 &mov ($s0,&DWP(0,$tbl));
2131 &mov ($s1,&DWP(32,$tbl));
2132 &mov ($s2,&DWP(64,$tbl));
2133 &mov ($acc,&DWP(96,$tbl));
2134 &lea ($tbl,&DWP(128,$tbl));
2136 &jnz (&label("prefetch_tbl"));
2143 &je (&label("fast_decrypt"));
2145 #----------------------------- ENCRYPT -----------------------------#
2146 &mov ($s0,&DWP(0,$key)); # load iv
2147 &mov ($s1,&DWP(4,$key));
2149 &set_label("fast_enc_loop",16);
2150 &mov ($s2,&DWP(8,$key));
2151 &mov ($s3,&DWP(12,$key));
2153 &xor ($s0,&DWP(0,$acc)); # xor input data
2154 &xor ($s1,&DWP(4,$acc));
2155 &xor ($s2,&DWP(8,$acc));
2156 &xor ($s3,&DWP(12,$acc));
2158 &mov ($key,$_key); # load key
2159 &call ("_x86_AES_encrypt");
2161 &mov ($acc,$_inp); # load inp
2162 &mov ($key,$_out); # load out
2164 &mov (&DWP(0,$key),$s0); # save output data
2165 &mov (&DWP(4,$key),$s1);
2166 &mov (&DWP(8,$key),$s2);
2167 &mov (&DWP(12,$key),$s3);
2169 &lea ($acc,&DWP(16,$acc)); # advance inp
2170 &mov ($s2,$_len); # load len
2171 &mov ($_inp,$acc); # save inp
2172 &lea ($s3,&DWP(16,$key)); # advance out
2173 &mov ($_out,$s3); # save out
2174 &sub ($s2,16); # decrease len
2175 &mov ($_len,$s2); # save len
2176 &jnz (&label("fast_enc_loop"));
2177 &mov ($acc,$_ivp); # load ivp
2178 &mov ($s2,&DWP(8,$key)); # restore last 2 dwords
2179 &mov ($s3,&DWP(12,$key));
2180 &mov (&DWP(0,$acc),$s0); # save ivec
2181 &mov (&DWP(4,$acc),$s1);
2182 &mov (&DWP(8,$acc),$s2);
2183 &mov (&DWP(12,$acc),$s3);
2185 &cmp ($mark,0); # was the key schedule copied?
2187 &je (&label("skip_ezero"));
2188 # zero copy of key schedule
2192 &data_word(0xABF3F689); # rep stosd
2193 &set_label("skip_ezero")
2196 &set_label("drop_out");
2198 &pushf (); # kludge, never executed
2200 #----------------------------- DECRYPT -----------------------------#
2201 &set_label("fast_decrypt",16);
2204 &je (&label("fast_dec_in_place")); # in-place processing...
2209 &set_label("fast_dec_loop",16);
2210 &mov ($s0,&DWP(0,$acc)); # read input
2211 &mov ($s1,&DWP(4,$acc));
2212 &mov ($s2,&DWP(8,$acc));
2213 &mov ($s3,&DWP(12,$acc));
2215 &mov ($key,$_key); # load key
2216 &call ("_x86_AES_decrypt");
2218 &mov ($key,$_tmp); # load ivp
2219 &mov ($acc,$_len); # load len
2220 &xor ($s0,&DWP(0,$key)); # xor iv
2221 &xor ($s1,&DWP(4,$key));
2222 &xor ($s2,&DWP(8,$key));
2223 &xor ($s3,&DWP(12,$key));
2225 &mov ($key,$_out); # load out
2226 &mov ($acc,$_inp); # load inp
2228 &mov (&DWP(0,$key),$s0); # write output
2229 &mov (&DWP(4,$key),$s1);
2230 &mov (&DWP(8,$key),$s2);
2231 &mov (&DWP(12,$key),$s3);
2233 &mov ($s2,$_len); # load len
2234 &mov ($_tmp,$acc); # save ivp
2235 &lea ($acc,&DWP(16,$acc)); # advance inp
2236 &mov ($_inp,$acc); # save inp
2237 &lea ($key,&DWP(16,$key)); # advance out
2238 &mov ($_out,$key); # save out
2239 &sub ($s2,16); # decrease len
2240 &mov ($_len,$s2); # save len
2241 &jnz (&label("fast_dec_loop"));
2242 &mov ($key,$_tmp); # load temp ivp
2243 &mov ($acc,$_ivp); # load user ivp
2244 &mov ($s0,&DWP(0,$key)); # load iv
2245 &mov ($s1,&DWP(4,$key));
2246 &mov ($s2,&DWP(8,$key));
2247 &mov ($s3,&DWP(12,$key));
2248 &mov (&DWP(0,$acc),$s0); # copy back to user
2249 &mov (&DWP(4,$acc),$s1);
2250 &mov (&DWP(8,$acc),$s2);
2251 &mov (&DWP(12,$acc),$s3);
2252 &jmp (&label("fast_dec_out"));
2254 &set_label("fast_dec_in_place",16);
2255 &set_label("fast_dec_in_place_loop");
2256 &mov ($s0,&DWP(0,$acc)); # read input
2257 &mov ($s1,&DWP(4,$acc));
2258 &mov ($s2,&DWP(8,$acc));
2259 &mov ($s3,&DWP(12,$acc));
2262 &mov (&DWP(0,$key),$s0); # copy to temp
2263 &mov (&DWP(4,$key),$s1);
2264 &mov (&DWP(8,$key),$s2);
2265 &mov (&DWP(12,$key),$s3);
2267 &mov ($key,$_key); # load key
2268 &call ("_x86_AES_decrypt");
2270 &mov ($key,$_ivp); # load ivp
2271 &mov ($acc,$_out); # load out
2272 &xor ($s0,&DWP(0,$key)); # xor iv
2273 &xor ($s1,&DWP(4,$key));
2274 &xor ($s2,&DWP(8,$key));
2275 &xor ($s3,&DWP(12,$key));
2277 &mov (&DWP(0,$acc),$s0); # write output
2278 &mov (&DWP(4,$acc),$s1);
2279 &mov (&DWP(8,$acc),$s2);
2280 &mov (&DWP(12,$acc),$s3);
2282 &lea ($acc,&DWP(16,$acc)); # advance out
2283 &mov ($_out,$acc); # save out
2286 &mov ($s0,&DWP(0,$acc)); # read temp
2287 &mov ($s1,&DWP(4,$acc));
2288 &mov ($s2,&DWP(8,$acc));
2289 &mov ($s3,&DWP(12,$acc));
2291 &mov (&DWP(0,$key),$s0); # copy iv
2292 &mov (&DWP(4,$key),$s1);
2293 &mov (&DWP(8,$key),$s2);
2294 &mov (&DWP(12,$key),$s3);
2296 &mov ($acc,$_inp); # load inp
2297 &mov ($s2,$_len); # load len
2298 &lea ($acc,&DWP(16,$acc)); # advance inp
2299 &mov ($_inp,$acc); # save inp
2300 &sub ($s2,16); # decrease len
2301 &mov ($_len,$s2); # save len
2302 &jnz (&label("fast_dec_in_place_loop"));
2304 &set_label("fast_dec_out",4);
2305 &cmp ($mark,0); # was the key schedule copied?
2307 &je (&label("skip_dzero"));
2308 # zero copy of key schedule
2312 &data_word(0xABF3F689); # rep stosd
2313 &set_label("skip_dzero")
2317 &pushf (); # kludge, never executed
2319 #--------------------------- SLOW ROUTINE ---------------------------#
2320 &set_label("slow_way",16);
2322 &mov ($s0,&DWP(0,$s0)) if (!$x86only);# load OPENSSL_ia32cap
2323 &mov ($key,&wparam(3)); # load key
2325 # pre-allocate aligned stack frame...
2326 &lea ($acc,&DWP(-80,"esp"));
2329 # ... and make sure it doesn't alias with $key modulo 1024
2330 &lea ($s1,&DWP(-80-63,$key));
2333 &and ($s1,0x3C0); # modulo 1024, but aligned to cache-line
2336 # pick S-box copy which can't overlap with stack frame or $key
2337 &lea ($s1,&DWP(768,$acc));
2340 &lea ($tbl,&DWP(2048+128,$tbl,$s1));
2342 &lea ($s3,&wparam(0)); # pointer to parameter block
2345 &add ("esp",4); # reserve for return address!
2346 &mov ($_tbl,$tbl); # save %ebp
2347 &mov ($_esp,$acc); # save %esp
2348 &mov ($_tmp,$s0); # save OPENSSL_ia32cap
2350 &mov ($s0,&DWP(0,$s3)); # load inp
2351 &mov ($s1,&DWP(4,$s3)); # load out
2352 #&mov ($s2,&DWP(8,$s3)); # load len
2353 #&mov ($key,&DWP(12,$s3)); # load key
2354 &mov ($acc,&DWP(16,$s3)); # load ivp
2355 &mov ($s3,&DWP(20,$s3)); # load enc flag
2357 &mov ($_inp,$s0); # save copy of inp
2358 &mov ($_out,$s1); # save copy of out
2359 &mov ($_len,$s2); # save copy of len
2360 &mov ($_key,$key); # save copy of key
2361 &mov ($_ivp,$acc); # save copy of ivp
2367 &je (&label("slow_decrypt"));
2369 #--------------------------- SLOW ENCRYPT ---------------------------#
2371 &jb (&label("slow_enc_tail"));
2374 &bt ($_tmp,25); # check for SSE bit
2375 &jnc (&label("slow_enc_x86"));
2377 &movq ("mm0",&QWP(0,$key)); # load iv
2378 &movq ("mm4",&QWP(8,$key));
2380 &set_label("slow_enc_loop_sse",16);
2381 &pxor ("mm0",&QWP(0,$acc)); # xor input data
2382 &pxor ("mm4",&QWP(8,$acc));
2385 &call ("_sse_AES_encrypt_compact");
2387 &mov ($acc,$_inp); # load inp
2388 &mov ($key,$_out); # load out
2389 &mov ($s2,$_len); # load len
2391 &movq (&QWP(0,$key),"mm0"); # save output data
2392 &movq (&QWP(8,$key),"mm4");
2394 &lea ($acc,&DWP(16,$acc)); # advance inp
2395 &mov ($_inp,$acc); # save inp
2396 &lea ($s3,&DWP(16,$key)); # advance out
2397 &mov ($_out,$s3); # save out
2398 &sub ($s2,16); # decrease len
2400 &mov ($_len,$s2); # save len
2401 &jae (&label("slow_enc_loop_sse"));
2403 &jnz (&label("slow_enc_tail"));
2404 &mov ($acc,$_ivp); # load ivp
2405 &movq (&QWP(0,$acc),"mm0"); # save ivec
2406 &movq (&QWP(8,$acc),"mm4");
2411 &pushf (); # kludge, never executed
2413 &set_label("slow_enc_x86",16);
2414 &mov ($s0,&DWP(0,$key)); # load iv
2415 &mov ($s1,&DWP(4,$key));
2417 &set_label("slow_enc_loop_x86",4);
2418 &mov ($s2,&DWP(8,$key));
2419 &mov ($s3,&DWP(12,$key));
2421 &xor ($s0,&DWP(0,$acc)); # xor input data
2422 &xor ($s1,&DWP(4,$acc));
2423 &xor ($s2,&DWP(8,$acc));
2424 &xor ($s3,&DWP(12,$acc));
2426 &mov ($key,$_key); # load key
2427 &call ("_x86_AES_encrypt_compact");
2429 &mov ($acc,$_inp); # load inp
2430 &mov ($key,$_out); # load out
2432 &mov (&DWP(0,$key),$s0); # save output data
2433 &mov (&DWP(4,$key),$s1);
2434 &mov (&DWP(8,$key),$s2);
2435 &mov (&DWP(12,$key),$s3);
2437 &mov ($s2,$_len); # load len
2438 &lea ($acc,&DWP(16,$acc)); # advance inp
2439 &mov ($_inp,$acc); # save inp
2440 &lea ($s3,&DWP(16,$key)); # advance out
2441 &mov ($_out,$s3); # save out
2442 &sub ($s2,16); # decrease len
2444 &mov ($_len,$s2); # save len
2445 &jae (&label("slow_enc_loop_x86"));
2447 &jnz (&label("slow_enc_tail"));
2448 &mov ($acc,$_ivp); # load ivp
2449 &mov ($s2,&DWP(8,$key)); # restore last dwords
2450 &mov ($s3,&DWP(12,$key));
2451 &mov (&DWP(0,$acc),$s0); # save ivec
2452 &mov (&DWP(4,$acc),$s1);
2453 &mov (&DWP(8,$acc),$s2);
2454 &mov (&DWP(12,$acc),$s3);
2459 &pushf (); # kludge, never executed
2461 &set_label("slow_enc_tail",16);
2463 &mov ($key eq "edi"? $key:"",$s3); # load out to edi
2466 &cmp ($key,$acc eq "esi"? $acc:""); # compare with inp
2467 &je (&label("enc_in_place"));
2469 &data_word(0xA4F3F689); # rep movsb # copy input
2470 &jmp (&label("enc_skip_in_place"));
2471 &set_label("enc_in_place");
2472 &lea ($key,&DWP(0,$key,$s2));
2473 &set_label("enc_skip_in_place");
2477 &data_word(0xAAF3F689); # rep stosb # zero tail
2479 &lea ($key,&DWP(-16,$s3)); # restore ivp
2480 &mov ($acc,$s3); # output as input
2481 &mov ($s0,&DWP(0,$key));
2482 &mov ($s1,&DWP(4,$key));
2483 &mov ($_len,16); # len=16
2484 &jmp (&label("slow_enc_loop_x86")); # one more spin...
2486 #--------------------------- SLOW DECRYPT ---------------------------#
2487 &set_label("slow_decrypt",16);
2489 &bt ($_tmp,25); # check for SSE bit
2490 &jnc (&label("slow_dec_loop_x86"));
2492 &set_label("slow_dec_loop_sse",4);
2493 &movq ("mm0",&QWP(0,$acc)); # read input
2494 &movq ("mm4",&QWP(8,$acc));
2497 &call ("_sse_AES_decrypt_compact");
2499 &mov ($acc,$_inp); # load inp
2501 &mov ($s1,$_out); # load out
2502 &mov ($s2,$_len); # load len
2503 &mov ($key,$_ivp); # load ivp
2505 &movq ("mm1",&QWP(0,$acc)); # re-read input
2506 &movq ("mm5",&QWP(8,$acc));
2508 &pxor ("mm0",&QWP(0,$key)); # xor iv
2509 &pxor ("mm4",&QWP(8,$key));
2511 &movq (&QWP(0,$key),"mm1"); # copy input to iv
2512 &movq (&QWP(8,$key),"mm5");
2514 &sub ($s2,16); # decrease len
2515 &jc (&label("slow_dec_partial_sse"));
2517 &movq (&QWP(0,$s1),"mm0"); # write output
2518 &movq (&QWP(8,$s1),"mm4");
2520 &lea ($s1,&DWP(16,$s1)); # advance out
2521 &mov ($_out,$s1); # save out
2522 &lea ($acc,&DWP(16,$acc)); # advance inp
2523 &mov ($_inp,$acc); # save inp
2524 &mov ($_len,$s2); # save len
2525 &jnz (&label("slow_dec_loop_sse"));
2530 &pushf (); # kludge, never executed
2532 &set_label("slow_dec_partial_sse",16);
2533 &movq (&QWP(0,$s0),"mm0"); # save output to temp
2534 &movq (&QWP(8,$s0),"mm4");
2537 &add ($s2 eq "ecx" ? "ecx":"",16);
2538 &mov ("edi",$s1); # out
2539 &mov ("esi",$s0); # temp
2541 &data_word(0xA4F3F689); # rep movsb # copy partial output
2546 &pushf (); # kludge, never executed
2548 &set_label("slow_dec_loop_x86",16);
2549 &mov ($s0,&DWP(0,$acc)); # read input
2550 &mov ($s1,&DWP(4,$acc));
2551 &mov ($s2,&DWP(8,$acc));
2552 &mov ($s3,&DWP(12,$acc));
2555 &mov (&DWP(0,$key),$s0); # copy to temp
2556 &mov (&DWP(4,$key),$s1);
2557 &mov (&DWP(8,$key),$s2);
2558 &mov (&DWP(12,$key),$s3);
2560 &mov ($key,$_key); # load key
2561 &call ("_x86_AES_decrypt_compact");
2563 &mov ($key,$_ivp); # load ivp
2564 &mov ($acc,$_len); # load len
2565 &xor ($s0,&DWP(0,$key)); # xor iv
2566 &xor ($s1,&DWP(4,$key));
2567 &xor ($s2,&DWP(8,$key));
2568 &xor ($s3,&DWP(12,$key));
2571 &jc (&label("slow_dec_partial_x86"));
2573 &mov ($_len,$acc); # save len
2574 &mov ($acc,$_out); # load out
2576 &mov (&DWP(0,$acc),$s0); # write output
2577 &mov (&DWP(4,$acc),$s1);
2578 &mov (&DWP(8,$acc),$s2);
2579 &mov (&DWP(12,$acc),$s3);
2581 &lea ($acc,&DWP(16,$acc)); # advance out
2582 &mov ($_out,$acc); # save out
2585 &mov ($s0,&DWP(0,$acc)); # read temp
2586 &mov ($s1,&DWP(4,$acc));
2587 &mov ($s2,&DWP(8,$acc));
2588 &mov ($s3,&DWP(12,$acc));
2590 &mov (&DWP(0,$key),$s0); # copy it to iv
2591 &mov (&DWP(4,$key),$s1);
2592 &mov (&DWP(8,$key),$s2);
2593 &mov (&DWP(12,$key),$s3);
2595 &mov ($acc,$_inp); # load inp
2596 &lea ($acc,&DWP(16,$acc)); # advance inp
2597 &mov ($_inp,$acc); # save inp
2598 &mov ($_len,$s2); # save len
2599 &jnz (&label("slow_dec_loop_x86"));
2603 &pushf (); # kludge, never executed
2605 &set_label("slow_dec_partial_x86",16);
2607 &mov (&DWP(0,$acc),$s0); # save output to temp
2608 &mov (&DWP(4,$acc),$s1);
2609 &mov (&DWP(8,$acc),$s2);
2610 &mov (&DWP(12,$acc),$s3);
2613 &mov ($s0,&DWP(0,$acc)); # re-read input
2614 &mov ($s1,&DWP(4,$acc));
2615 &mov ($s2,&DWP(8,$acc));
2616 &mov ($s3,&DWP(12,$acc));
2618 &mov (&DWP(0,$key),$s0); # copy it to iv
2619 &mov (&DWP(4,$key),$s1);
2620 &mov (&DWP(8,$key),$s2);
2621 &mov (&DWP(12,$key),$s3);
2627 &data_word(0xA4F3F689); # rep movsb # copy partial output
2631 &function_end("AES_cbc_encrypt");
2634 #------------------------------------------------------------------#
2638 &movz ("esi",&LB("edx")); # rk[i]>>0
2639 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2640 &movz ("esi",&HB("edx")); # rk[i]>>8
2644 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2646 &movz ("esi",&LB("edx")); # rk[i]>>16
2649 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2650 &movz ("esi",&HB("edx")); # rk[i]>>24
2654 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2658 &xor ("eax",&DWP(1024-128,$tbl,"ecx",4)); # rcon
2661 # int AES_set_encrypt_key(const unsigned char *userKey, const int bits,
2663 &public_label("AES_Te");
2664 &function_begin("AES_set_encrypt_key");
2665 &mov ("esi",&wparam(0)); # user supplied key
2666 &mov ("edi",&wparam(2)); # private key schedule
2669 &jz (&label("badpointer"));
2671 &jz (&label("badpointer"));
2673 &call (&label("pic_point"));
2674 &set_label("pic_point");
2676 &lea ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
2677 &lea ($tbl,&DWP(2048+128,$tbl));
2680 &mov ("eax",&DWP(0-128,$tbl));
2681 &mov ("ebx",&DWP(32-128,$tbl));
2682 &mov ("ecx",&DWP(64-128,$tbl));
2683 &mov ("edx",&DWP(96-128,$tbl));
2684 &mov ("eax",&DWP(128-128,$tbl));
2685 &mov ("ebx",&DWP(160-128,$tbl));
2686 &mov ("ecx",&DWP(192-128,$tbl));
2687 &mov ("edx",&DWP(224-128,$tbl));
2689 &mov ("ecx",&wparam(1)); # number of bits in key
2691 &je (&label("10rounds"));
2693 &je (&label("12rounds"));
2695 &je (&label("14rounds"));
2696 &mov ("eax",-2); # invalid number of bits
2697 &jmp (&label("exit"));
2699 &set_label("10rounds");
2700 &mov ("eax",&DWP(0,"esi")); # copy first 4 dwords
2701 &mov ("ebx",&DWP(4,"esi"));
2702 &mov ("ecx",&DWP(8,"esi"));
2703 &mov ("edx",&DWP(12,"esi"));
2704 &mov (&DWP(0,"edi"),"eax");
2705 &mov (&DWP(4,"edi"),"ebx");
2706 &mov (&DWP(8,"edi"),"ecx");
2707 &mov (&DWP(12,"edi"),"edx");
2710 &jmp (&label("10shortcut"));
2713 &set_label("10loop");
2714 &mov ("eax",&DWP(0,"edi")); # rk[0]
2715 &mov ("edx",&DWP(12,"edi")); # rk[3]
2716 &set_label("10shortcut");
2719 &mov (&DWP(16,"edi"),"eax"); # rk[4]
2720 &xor ("eax",&DWP(4,"edi"));
2721 &mov (&DWP(20,"edi"),"eax"); # rk[5]
2722 &xor ("eax",&DWP(8,"edi"));
2723 &mov (&DWP(24,"edi"),"eax"); # rk[6]
2724 &xor ("eax",&DWP(12,"edi"));
2725 &mov (&DWP(28,"edi"),"eax"); # rk[7]
2729 &jl (&label("10loop"));
2731 &mov (&DWP(80,"edi"),10); # setup number of rounds
2733 &jmp (&label("exit"));
2735 &set_label("12rounds");
2736 &mov ("eax",&DWP(0,"esi")); # copy first 6 dwords
2737 &mov ("ebx",&DWP(4,"esi"));
2738 &mov ("ecx",&DWP(8,"esi"));
2739 &mov ("edx",&DWP(12,"esi"));
2740 &mov (&DWP(0,"edi"),"eax");
2741 &mov (&DWP(4,"edi"),"ebx");
2742 &mov (&DWP(8,"edi"),"ecx");
2743 &mov (&DWP(12,"edi"),"edx");
2744 &mov ("ecx",&DWP(16,"esi"));
2745 &mov ("edx",&DWP(20,"esi"));
2746 &mov (&DWP(16,"edi"),"ecx");
2747 &mov (&DWP(20,"edi"),"edx");
2750 &jmp (&label("12shortcut"));
2753 &set_label("12loop");
2754 &mov ("eax",&DWP(0,"edi")); # rk[0]
2755 &mov ("edx",&DWP(20,"edi")); # rk[5]
2756 &set_label("12shortcut");
2759 &mov (&DWP(24,"edi"),"eax"); # rk[6]
2760 &xor ("eax",&DWP(4,"edi"));
2761 &mov (&DWP(28,"edi"),"eax"); # rk[7]
2762 &xor ("eax",&DWP(8,"edi"));
2763 &mov (&DWP(32,"edi"),"eax"); # rk[8]
2764 &xor ("eax",&DWP(12,"edi"));
2765 &mov (&DWP(36,"edi"),"eax"); # rk[9]
2768 &je (&label("12break"));
2771 &xor ("eax",&DWP(16,"edi"));
2772 &mov (&DWP(40,"edi"),"eax"); # rk[10]
2773 &xor ("eax",&DWP(20,"edi"));
2774 &mov (&DWP(44,"edi"),"eax"); # rk[11]
2777 &jmp (&label("12loop"));
2779 &set_label("12break");
2780 &mov (&DWP(72,"edi"),12); # setup number of rounds
2782 &jmp (&label("exit"));
2784 &set_label("14rounds");
2785 &mov ("eax",&DWP(0,"esi")); # copy first 8 dwords
2786 &mov ("ebx",&DWP(4,"esi"));
2787 &mov ("ecx",&DWP(8,"esi"));
2788 &mov ("edx",&DWP(12,"esi"));
2789 &mov (&DWP(0,"edi"),"eax");
2790 &mov (&DWP(4,"edi"),"ebx");
2791 &mov (&DWP(8,"edi"),"ecx");
2792 &mov (&DWP(12,"edi"),"edx");
2793 &mov ("eax",&DWP(16,"esi"));
2794 &mov ("ebx",&DWP(20,"esi"));
2795 &mov ("ecx",&DWP(24,"esi"));
2796 &mov ("edx",&DWP(28,"esi"));
2797 &mov (&DWP(16,"edi"),"eax");
2798 &mov (&DWP(20,"edi"),"ebx");
2799 &mov (&DWP(24,"edi"),"ecx");
2800 &mov (&DWP(28,"edi"),"edx");
2803 &jmp (&label("14shortcut"));
2806 &set_label("14loop");
2807 &mov ("edx",&DWP(28,"edi")); # rk[7]
2808 &set_label("14shortcut");
2809 &mov ("eax",&DWP(0,"edi")); # rk[0]
2813 &mov (&DWP(32,"edi"),"eax"); # rk[8]
2814 &xor ("eax",&DWP(4,"edi"));
2815 &mov (&DWP(36,"edi"),"eax"); # rk[9]
2816 &xor ("eax",&DWP(8,"edi"));
2817 &mov (&DWP(40,"edi"),"eax"); # rk[10]
2818 &xor ("eax",&DWP(12,"edi"));
2819 &mov (&DWP(44,"edi"),"eax"); # rk[11]
2822 &je (&label("14break"));
2826 &mov ("eax",&DWP(16,"edi")); # rk[4]
2827 &movz ("esi",&LB("edx")); # rk[11]>>0
2828 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2829 &movz ("esi",&HB("edx")); # rk[11]>>8
2832 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2835 &movz ("esi",&LB("edx")); # rk[11]>>16
2838 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2839 &movz ("esi",&HB("edx")); # rk[11]>>24
2843 &movz ("ebx",&BP(-128,$tbl,"esi",1));
2847 &mov (&DWP(48,"edi"),"eax"); # rk[12]
2848 &xor ("eax",&DWP(20,"edi"));
2849 &mov (&DWP(52,"edi"),"eax"); # rk[13]
2850 &xor ("eax",&DWP(24,"edi"));
2851 &mov (&DWP(56,"edi"),"eax"); # rk[14]
2852 &xor ("eax",&DWP(28,"edi"));
2853 &mov (&DWP(60,"edi"),"eax"); # rk[15]
2856 &jmp (&label("14loop"));
2858 &set_label("14break");
2859 &mov (&DWP(48,"edi"),14); # setup number of rounds
2861 &jmp (&label("exit"));
2863 &set_label("badpointer");
2866 &function_end("AES_set_encrypt_key");
2869 { my ($i,$key,$tp1,$tp2,$tp4,$tp8) = @_;
2873 &and ($acc,0x80808080);
2877 &and ($tp2,0x7f7f7f7f);
2880 &and ($acc,0x1b1b1b1b);
2884 &and ($acc,0x80808080);
2887 &xor ($tp2,$tp1); # tp2^tp1
2889 &and ($tp4,0x7f7f7f7f);
2892 &and ($acc,0x1b1b1b1b);
2896 &and ($acc,0x80808080);
2899 &xor ($tp4,$tp1); # tp4^tp1
2901 &and ($tp8,0x7f7f7f7f);
2904 &and ($acc,0x1b1b1b1b);
2905 &rotl ($tp1,8); # = ROTATE(tp1,8)
2908 &mov ($tmp,&DWP(4*($i+1),$key)); # modulo-scheduled load
2915 &xor ($tp1,$tp8); # ^= tp8^(tp4^tp1)^(tp2^tp1)
2917 &xor ($tp1,$tp2); # ^= ROTATE(tp8^tp2^tp1,24)
2919 &xor ($tp1,$tp4); # ^= ROTATE(tp8^tp4^tp1,16)
2921 &xor ($tp1,$tp8); # ^= ROTATE(tp8,8)
2923 &mov (&DWP(4*$i,$key),$tp1);
2926 # int AES_set_decrypt_key(const unsigned char *userKey, const int bits,
2928 &public_label("AES_Td");
2929 &public_label("AES_Te");
2930 &function_begin_B("AES_set_decrypt_key");
2931 &mov ("eax",&wparam(0));
2932 &mov ("ecx",&wparam(1));
2933 &mov ("edx",&wparam(2));
2935 &mov (&DWP(0,"esp"),"eax");
2936 &mov (&DWP(4,"esp"),"ecx");
2937 &mov (&DWP(8,"esp"),"edx");
2938 &call ("AES_set_encrypt_key");
2941 &je (&label("proceed"));
2944 &set_label("proceed");
2950 &mov ("esi",&wparam(2));
2951 &mov ("ecx",&DWP(240,"esi")); # pull number of rounds
2952 &lea ("ecx",&DWP(0,"","ecx",4));
2953 &lea ("edi",&DWP(0,"esi","ecx",4)); # pointer to last chunk
2955 &set_label("invert",4); # invert order of chunks
2956 &mov ("eax",&DWP(0,"esi"));
2957 &mov ("ebx",&DWP(4,"esi"));
2958 &mov ("ecx",&DWP(0,"edi"));
2959 &mov ("edx",&DWP(4,"edi"));
2960 &mov (&DWP(0,"edi"),"eax");
2961 &mov (&DWP(4,"edi"),"ebx");
2962 &mov (&DWP(0,"esi"),"ecx");
2963 &mov (&DWP(4,"esi"),"edx");
2964 &mov ("eax",&DWP(8,"esi"));
2965 &mov ("ebx",&DWP(12,"esi"));
2966 &mov ("ecx",&DWP(8,"edi"));
2967 &mov ("edx",&DWP(12,"edi"));
2968 &mov (&DWP(8,"edi"),"eax");
2969 &mov (&DWP(12,"edi"),"ebx");
2970 &mov (&DWP(8,"esi"),"ecx");
2971 &mov (&DWP(12,"esi"),"edx");
2975 &jne (&label("invert"));
2977 &mov ($key,&wparam(2));
2978 &mov ($acc,&DWP(240,$key)); # pull number of rounds
2979 &lea ($acc,&DWP(-2,$acc,$acc));
2980 &lea ($acc,&DWP(0,$key,$acc,8));
2981 &mov (&wparam(2),$acc);
2983 &mov ($s0,&DWP(16,$key)); # modulo-scheduled load
2984 &set_label("permute",4); # permute the key schedule
2986 &deckey (0,$key,$s0,$s1,$s2,$s3);
2987 &deckey (1,$key,$s1,$s2,$s3,$s0);
2988 &deckey (2,$key,$s2,$s3,$s0,$s1);
2989 &deckey (3,$key,$s3,$s0,$s1,$s2);
2990 &cmp ($key,&wparam(2));
2991 &jb (&label("permute"));
2993 &xor ("eax","eax"); # return success
2994 &function_end("AES_set_decrypt_key");
projects / openssl.git / blob