29059edf8b7a6bfacf9dfa02769e630a9db27f2a
[openssl.git] / crypto / aes / asm / aes-586.pl
1 #! /usr/bin/env perl
2 # Copyright 2004-2016 The OpenSSL Project Authors. All Rights Reserved.
3 #
4 # Licensed under the OpenSSL license (the "License").  You may not use
5 # this file except in compliance with the License.  You can obtain a copy
6 # in the file LICENSE in the source distribution or at
7 # https://www.openssl.org/source/license.html
8
9 #
10 # ====================================================================
11 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
12 # project. The module is, however, dual licensed under OpenSSL and
13 # CRYPTOGAMS licenses depending on where you obtain it. For further
14 # details see http://www.openssl.org/~appro/cryptogams/.
15 # ====================================================================
16 #
17 # Version 4.3.
18 #
19 # You might fail to appreciate this module performance from the first
20 # try. If compared to "vanilla" linux-ia32-icc target, i.e. considered
21 # to be *the* best Intel C compiler without -KPIC, performance appears
22 # to be virtually identical... But try to re-configure with shared
23 # library support... Aha! Intel compiler "suddenly" lags behind by 30%
24 # [on P4, more on others]:-) And if compared to position-independent
25 # code generated by GNU C, this code performs *more* than *twice* as
26 # fast! Yes, all this buzz about PIC means that unlike other hand-
27 # coded implementations, this one was explicitly designed to be safe
28 # to use even in shared library context... This also means that this
29 # code isn't necessarily absolutely fastest "ever," because in order
30 # to achieve position independence an extra register has to be
31 # off-loaded to stack, which affects the benchmark result.
32 #
33 # Special note about instruction choice. Do you recall RC4_INT code
34 # performing poorly on P4? It might be the time to figure out why.
35 # RC4_INT code implies effective address calculations in base+offset*4
36 # form. Trouble is that it seems that offset scaling turned to be
37 # critical path... At least eliminating scaling resulted in 2.8x RC4
38 # performance improvement [as you might recall]. As AES code is hungry
39 # for scaling too, I [try to] avoid the latter by favoring off-by-2
40 # shifts and masking the result with 0xFF<<2 instead of "boring" 0xFF.
41 #
42 # As was shown by Dean Gaudet, the above note turned out to be
43 # void. Performance improvement with off-by-2 shifts was observed on
44 # intermediate implementation, which was spilling yet another register
45 # to stack... Final offset*4 code below runs just a tad faster on P4,
46 # but exhibits up to 10% improvement on other cores.
47 #
48 # Second version is "monolithic" replacement for aes_core.c, which in
49 # addition to AES_[de|en]crypt implements AES_set_[de|en]cryption_key.
50 # This made it possible to implement little-endian variant of the
51 # algorithm without modifying the base C code. Motivating factor for
52 # the undertaken effort was that it appeared that in tight IA-32
53 # register window little-endian flavor could achieve slightly higher
54 # Instruction Level Parallelism, and it indeed resulted in up to 15%
55 # better performance on most recent µ-archs...
56 #
57 # Third version adds AES_cbc_encrypt implementation, which resulted in
58 # up to 40% performance improvement of CBC benchmark results. 40% was
59 # observed on P4 core, where "overall" improvement coefficient, i.e. if
60 # compared to PIC generated by GCC and in CBC mode, was observed to be
61 # as large as 4x:-) CBC performance is virtually identical to ECB now
62 # and on some platforms even better, e.g. 17.6 "small" cycles/byte on
63 # Opteron, because certain function prologues and epilogues are
64 # effectively taken out of the loop...
65 #
66 # Version 3.2 implements compressed tables and prefetch of these tables
67 # in CBC[!] mode. Former means that 3/4 of table references are now
68 # misaligned, which unfortunately has negative impact on elder IA-32
69 # implementations, Pentium suffered 30% penalty, PIII - 10%.
70 #
71 # Version 3.3 avoids L1 cache aliasing between stack frame and
72 # S-boxes, and 3.4 - L1 cache aliasing even between key schedule. The
73 # latter is achieved by copying the key schedule to controlled place in
74 # stack. This unfortunately has rather strong impact on small block CBC
75 # performance, ~2x deterioration on 16-byte block if compared to 3.3.
76 #
77 # Version 3.5 checks if there is L1 cache aliasing between user-supplied
78 # key schedule and S-boxes and abstains from copying the former if
79 # there is no. This allows end-user to consciously retain small block
80 # performance by aligning key schedule in specific manner.
81 #
82 # Version 3.6 compresses Td4 to 256 bytes and prefetches it in ECB.
83 #
84 # Current ECB performance numbers for 128-bit key in CPU cycles per
85 # processed byte [measure commonly used by AES benchmarkers] are:
86 #
87 #               small footprint         fully unrolled
88 # P4            24                      22
89 # AMD K8        20                      19
90 # PIII          25                      23
91 # Pentium       81                      78
92 #
93 # Version 3.7 reimplements outer rounds as "compact." Meaning that
94 # first and last rounds reference compact 256 bytes S-box. This means
95 # that first round consumes a lot more CPU cycles and that encrypt
96 # and decrypt performance becomes asymmetric. Encrypt performance
97 # drops by 10-12%, while decrypt - by 20-25%:-( 256 bytes S-box is
98 # aggressively pre-fetched.
99 #
100 # Version 4.0 effectively rolls back to 3.6 and instead implements
101 # additional set of functions, _[x86|sse]_AES_[en|de]crypt_compact,
102 # which use exclusively 256 byte S-box. These functions are to be
103 # called in modes not concealing plain text, such as ECB, or when
104 # we're asked to process smaller amount of data [or unconditionally
105 # on hyper-threading CPU]. Currently it's called unconditionally from
106 # AES_[en|de]crypt, which affects all modes, but CBC. CBC routine
107 # still needs to be modified to switch between slower and faster
108 # mode when appropriate... But in either case benchmark landscape
109 # changes dramatically and below numbers are CPU cycles per processed
110 # byte for 128-bit key.
111 #
112 #               ECB encrypt     ECB decrypt     CBC large chunk
113 # P4            52[54]          83[95]          23
114 # AMD K8        46[41]          66[70]          18
115 # PIII          41[50]          60[77]          24
116 # Core 2        31[36]          45[64]          18.5
117 # Atom          76[100]         96[138]         60
118 # Pentium       115             150             77
119 #
120 # Version 4.1 switches to compact S-box even in key schedule setup.
121 #
122 # Version 4.2 prefetches compact S-box in every SSE round or in other
123 # words every cache-line is *guaranteed* to be accessed within ~50
124 # cycles window. Why just SSE? Because it's needed on hyper-threading
125 # CPU! Which is also why it's prefetched with 64 byte stride. Best
126 # part is that it has no negative effect on performance:-)
127 #
128 # Version 4.3 implements switch between compact and non-compact block
129 # functions in AES_cbc_encrypt depending on how much data was asked
130 # to be processed in one stroke.
131 #
132 ######################################################################
133 # Timing attacks are classified in two classes: synchronous when
134 # attacker consciously initiates cryptographic operation and collects
135 # timing data of various character afterwards, and asynchronous when
136 # malicious code is executed on same CPU simultaneously with AES,
137 # instruments itself and performs statistical analysis of this data.
138 #
139 # As far as synchronous attacks go the root to the AES timing
140 # vulnerability is twofold. Firstly, of 256 S-box elements at most 160
141 # are referred to in single 128-bit block operation. Well, in C
142 # implementation with 4 distinct tables it's actually as little as 40
143 # references per 256 elements table, but anyway... Secondly, even
144 # though S-box elements are clustered into smaller amount of cache-
145 # lines, smaller than 160 and even 40, it turned out that for certain
146 # plain-text pattern[s] or simply put chosen plain-text and given key
147 # few cache-lines remain unaccessed during block operation. Now, if
148 # attacker can figure out this access pattern, he can deduct the key
149 # [or at least part of it]. The natural way to mitigate this kind of
150 # attacks is to minimize the amount of cache-lines in S-box and/or
151 # prefetch them to ensure that every one is accessed for more uniform
152 # timing. But note that *if* plain-text was concealed in such way that
153 # input to block function is distributed *uniformly*, then attack
154 # wouldn't apply. Now note that some encryption modes, most notably
155 # CBC, do mask the plain-text in this exact way [secure cipher output
156 # is distributed uniformly]. Yes, one still might find input that
157 # would reveal the information about given key, but if amount of
158 # candidate inputs to be tried is larger than amount of possible key
159 # combinations then attack becomes infeasible. This is why revised
160 # AES_cbc_encrypt "dares" to switch to larger S-box when larger chunk
161 # of data is to be processed in one stroke. The current size limit of
162 # 512 bytes is chosen to provide same [diminishingly low] probability
163 # for cache-line to remain untouched in large chunk operation with
164 # large S-box as for single block operation with compact S-box and
165 # surely needs more careful consideration...
166 #
167 # As for asynchronous attacks. There are two flavours: attacker code
168 # being interleaved with AES on hyper-threading CPU at *instruction*
169 # level, and two processes time sharing single core. As for latter.
170 # Two vectors. 1. Given that attacker process has higher priority,
171 # yield execution to process performing AES just before timer fires
172 # off the scheduler, immediately regain control of CPU and analyze the
173 # cache state. For this attack to be efficient attacker would have to
174 # effectively slow down the operation by several *orders* of magnitude,
175 # by ratio of time slice to duration of handful of AES rounds, which
176 # unlikely to remain unnoticed. Not to mention that this also means
177 # that he would spend correspondingly more time to collect enough
178 # statistical data to mount the attack. It's probably appropriate to
179 # say that if adversary reckons that this attack is beneficial and
180 # risks to be noticed, you probably have larger problems having him
181 # mere opportunity. In other words suggested code design expects you
182 # to preclude/mitigate this attack by overall system security design.
183 # 2. Attacker manages to make his code interrupt driven. In order for
184 # this kind of attack to be feasible, interrupt rate has to be high
185 # enough, again comparable to duration of handful of AES rounds. But
186 # is there interrupt source of such rate? Hardly, not even 1Gbps NIC
187 # generates interrupts at such raging rate...
188 #
189 # And now back to the former, hyper-threading CPU or more specifically
190 # Intel P4. Recall that asynchronous attack implies that malicious
191 # code instruments itself. And naturally instrumentation granularity
192 # has be noticeably lower than duration of codepath accessing S-box.
193 # Given that all cache-lines are accessed during that time that is.
194 # Current implementation accesses *all* cache-lines within ~50 cycles
195 # window, which is actually *less* than RDTSC latency on Intel P4!
196
197 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
198 push(@INC,"${dir}","${dir}../../perlasm");
199 require "x86asm.pl";
200
201 $output = pop;
202 open OUT,">$output";
203 *STDOUT=*OUT;
204
205 &asm_init($ARGV[0],$x86only = $ARGV[$#ARGV] eq "386");
206 &static_label("AES_Te");
207 &static_label("AES_Td");
208
209 $s0="eax";
210 $s1="ebx";
211 $s2="ecx";
212 $s3="edx";
213 $key="edi";
214 $acc="esi";
215 $tbl="ebp";
216
217 # stack frame layout in _[x86|sse]_AES_* routines, frame is allocated
218 # by caller
219 $__ra=&DWP(0,"esp");    # return address
220 $__s0=&DWP(4,"esp");    # s0 backing store
221 $__s1=&DWP(8,"esp");    # s1 backing store
222 $__s2=&DWP(12,"esp");   # s2 backing store
223 $__s3=&DWP(16,"esp");   # s3 backing store
224 $__key=&DWP(20,"esp");  # pointer to key schedule
225 $__end=&DWP(24,"esp");  # pointer to end of key schedule
226 $__tbl=&DWP(28,"esp");  # %ebp backing store
227
228 # stack frame layout in AES_[en|crypt] routines, which differs from
229 # above by 4 and overlaps by %ebp backing store
230 $_tbl=&DWP(24,"esp");
231 $_esp=&DWP(28,"esp");
232
233 sub _data_word() { my $i; while(defined($i=shift)) { &data_word($i,$i); } }
234
235 $speed_limit=512;       # chunks smaller than $speed_limit are
236                         # processed with compact routine in CBC mode
237 $small_footprint=1;     # $small_footprint=1 code is ~5% slower [on
238                         # recent µ-archs], but ~5 times smaller!
239                         # I favor compact code to minimize cache
240                         # contention and in hope to "collect" 5% back
241                         # in real-life applications...
242
243 $vertical_spin=0;       # shift "vertically" defaults to 0, because of
244                         # its proof-of-concept status...
245 # Note that there is no decvert(), as well as last encryption round is
246 # performed with "horizontal" shifts. This is because this "vertical"
247 # implementation [one which groups shifts on a given $s[i] to form a
248 # "column," unlike "horizontal" one, which groups shifts on different
249 # $s[i] to form a "row"] is work in progress. It was observed to run
250 # few percents faster on Intel cores, but not AMD. On AMD K8 core it's
251 # whole 12% slower:-( So we face a trade-off... Shall it be resolved
252 # some day? Till then the code is considered experimental and by
253 # default remains dormant...
254
255 sub encvert()
256 { my ($te,@s) = @_;
257   my ($v0,$v1) = ($acc,$key);
258
259         &mov    ($v0,$s[3]);                            # copy s3
260         &mov    (&DWP(4,"esp"),$s[2]);                  # save s2
261         &mov    ($v1,$s[0]);                            # copy s0
262         &mov    (&DWP(8,"esp"),$s[1]);                  # save s1
263
264         &movz   ($s[2],&HB($s[0]));
265         &and    ($s[0],0xFF);
266         &mov    ($s[0],&DWP(0,$te,$s[0],8));            # s0>>0
267         &shr    ($v1,16);
268         &mov    ($s[3],&DWP(3,$te,$s[2],8));            # s0>>8
269         &movz   ($s[1],&HB($v1));
270         &and    ($v1,0xFF);
271         &mov    ($s[2],&DWP(2,$te,$v1,8));              # s0>>16
272          &mov   ($v1,$v0);
273         &mov    ($s[1],&DWP(1,$te,$s[1],8));            # s0>>24
274
275         &and    ($v0,0xFF);
276         &xor    ($s[3],&DWP(0,$te,$v0,8));              # s3>>0
277         &movz   ($v0,&HB($v1));
278         &shr    ($v1,16);
279         &xor    ($s[2],&DWP(3,$te,$v0,8));              # s3>>8
280         &movz   ($v0,&HB($v1));
281         &and    ($v1,0xFF);
282         &xor    ($s[1],&DWP(2,$te,$v1,8));              # s3>>16
283          &mov   ($v1,&DWP(4,"esp"));                    # restore s2
284         &xor    ($s[0],&DWP(1,$te,$v0,8));              # s3>>24
285
286         &mov    ($v0,$v1);
287         &and    ($v1,0xFF);
288         &xor    ($s[2],&DWP(0,$te,$v1,8));              # s2>>0
289         &movz   ($v1,&HB($v0));
290         &shr    ($v0,16);
291         &xor    ($s[1],&DWP(3,$te,$v1,8));              # s2>>8
292         &movz   ($v1,&HB($v0));
293         &and    ($v0,0xFF);
294         &xor    ($s[0],&DWP(2,$te,$v0,8));              # s2>>16
295          &mov   ($v0,&DWP(8,"esp"));                    # restore s1
296         &xor    ($s[3],&DWP(1,$te,$v1,8));              # s2>>24
297
298         &mov    ($v1,$v0);
299         &and    ($v0,0xFF);
300         &xor    ($s[1],&DWP(0,$te,$v0,8));              # s1>>0
301         &movz   ($v0,&HB($v1));
302         &shr    ($v1,16);
303         &xor    ($s[0],&DWP(3,$te,$v0,8));              # s1>>8
304         &movz   ($v0,&HB($v1));
305         &and    ($v1,0xFF);
306         &xor    ($s[3],&DWP(2,$te,$v1,8));              # s1>>16
307          &mov   ($key,$__key);                          # reincarnate v1 as key
308         &xor    ($s[2],&DWP(1,$te,$v0,8));              # s1>>24
309 }
310
311 # Another experimental routine, which features "horizontal spin," but
312 # eliminates one reference to stack. Strangely enough runs slower...
313 sub enchoriz()
314 { my ($v0,$v1) = ($key,$acc);
315
316         &movz   ($v0,&LB($s0));                 #  3, 2, 1, 0*
317         &rotr   ($s2,8);                        #  8,11,10, 9
318         &mov    ($v1,&DWP(0,$te,$v0,8));        #  0
319         &movz   ($v0,&HB($s1));                 #  7, 6, 5*, 4
320         &rotr   ($s3,16);                       # 13,12,15,14
321         &xor    ($v1,&DWP(3,$te,$v0,8));        #  5
322         &movz   ($v0,&HB($s2));                 #  8,11,10*, 9
323         &rotr   ($s0,16);                       #  1, 0, 3, 2
324         &xor    ($v1,&DWP(2,$te,$v0,8));        # 10
325         &movz   ($v0,&HB($s3));                 # 13,12,15*,14
326         &xor    ($v1,&DWP(1,$te,$v0,8));        # 15, t[0] collected
327         &mov    ($__s0,$v1);                    # t[0] saved
328
329         &movz   ($v0,&LB($s1));                 #  7, 6, 5, 4*
330         &shr    ($s1,16);                       #  -, -, 7, 6
331         &mov    ($v1,&DWP(0,$te,$v0,8));        #  4
332         &movz   ($v0,&LB($s3));                 # 13,12,15,14*
333         &xor    ($v1,&DWP(2,$te,$v0,8));        # 14
334         &movz   ($v0,&HB($s0));                 #  1, 0, 3*, 2
335         &and    ($s3,0xffff0000);               # 13,12, -, -
336         &xor    ($v1,&DWP(1,$te,$v0,8));        #  3
337         &movz   ($v0,&LB($s2));                 #  8,11,10, 9*
338         &or     ($s3,$s1);                      # 13,12, 7, 6
339         &xor    ($v1,&DWP(3,$te,$v0,8));        #  9, t[1] collected
340         &mov    ($s1,$v1);                      #  s[1]=t[1]
341
342         &movz   ($v0,&LB($s0));                 #  1, 0, 3, 2*
343         &shr    ($s2,16);                       #  -, -, 8,11
344         &mov    ($v1,&DWP(2,$te,$v0,8));        #  2
345         &movz   ($v0,&HB($s3));                 # 13,12, 7*, 6
346         &xor    ($v1,&DWP(1,$te,$v0,8));        #  7
347         &movz   ($v0,&HB($s2));                 #  -, -, 8*,11
348         &xor    ($v1,&DWP(0,$te,$v0,8));        #  8
349         &mov    ($v0,$s3);
350         &shr    ($v0,24);                       # 13
351         &xor    ($v1,&DWP(3,$te,$v0,8));        # 13, t[2] collected
352
353         &movz   ($v0,&LB($s2));                 #  -, -, 8,11*
354         &shr    ($s0,24);                       #  1*
355         &mov    ($s2,&DWP(1,$te,$v0,8));        # 11
356         &xor    ($s2,&DWP(3,$te,$s0,8));        #  1
357         &mov    ($s0,$__s0);                    # s[0]=t[0]
358         &movz   ($v0,&LB($s3));                 # 13,12, 7, 6*
359         &shr    ($s3,16);                       #   ,  ,13,12
360         &xor    ($s2,&DWP(2,$te,$v0,8));        #  6
361         &mov    ($key,$__key);                  # reincarnate v0 as key
362         &and    ($s3,0xff);                     #   ,  ,13,12*
363         &mov    ($s3,&DWP(0,$te,$s3,8));        # 12
364         &xor    ($s3,$s2);                      # s[2]=t[3] collected
365         &mov    ($s2,$v1);                      # s[2]=t[2]
366 }
367
368 # More experimental code... SSE one... Even though this one eliminates
369 # *all* references to stack, it's not faster...
370 sub sse_encbody()
371 {
372         &movz   ($acc,&LB("eax"));              #  0
373         &mov    ("ecx",&DWP(0,$tbl,$acc,8));    #  0
374         &pshufw ("mm2","mm0",0x0d);             #  7, 6, 3, 2
375         &movz   ("edx",&HB("eax"));             #  1
376         &mov    ("edx",&DWP(3,$tbl,"edx",8));   #  1
377         &shr    ("eax",16);                     #  5, 4
378
379         &movz   ($acc,&LB("ebx"));              # 10
380         &xor    ("ecx",&DWP(2,$tbl,$acc,8));    # 10
381         &pshufw ("mm6","mm4",0x08);             # 13,12, 9, 8
382         &movz   ($acc,&HB("ebx"));              # 11
383         &xor    ("edx",&DWP(1,$tbl,$acc,8));    # 11
384         &shr    ("ebx",16);                     # 15,14
385
386         &movz   ($acc,&HB("eax"));              #  5
387         &xor    ("ecx",&DWP(3,$tbl,$acc,8));    #  5
388         &movq   ("mm3",QWP(16,$key));
389         &movz   ($acc,&HB("ebx"));              # 15
390         &xor    ("ecx",&DWP(1,$tbl,$acc,8));    # 15
391         &movd   ("mm0","ecx");                  # t[0] collected
392
393         &movz   ($acc,&LB("eax"));              #  4
394         &mov    ("ecx",&DWP(0,$tbl,$acc,8));    #  4
395         &movd   ("eax","mm2");                  #  7, 6, 3, 2
396         &movz   ($acc,&LB("ebx"));              # 14
397         &xor    ("ecx",&DWP(2,$tbl,$acc,8));    # 14
398         &movd   ("ebx","mm6");                  # 13,12, 9, 8
399
400         &movz   ($acc,&HB("eax"));              #  3
401         &xor    ("ecx",&DWP(1,$tbl,$acc,8));    #  3
402         &movz   ($acc,&HB("ebx"));              #  9
403         &xor    ("ecx",&DWP(3,$tbl,$acc,8));    #  9
404         &movd   ("mm1","ecx");                  # t[1] collected
405
406         &movz   ($acc,&LB("eax"));              #  2
407         &mov    ("ecx",&DWP(2,$tbl,$acc,8));    #  2
408         &shr    ("eax",16);                     #  7, 6
409         &punpckldq      ("mm0","mm1");          # t[0,1] collected
410         &movz   ($acc,&LB("ebx"));              #  8
411         &xor    ("ecx",&DWP(0,$tbl,$acc,8));    #  8
412         &shr    ("ebx",16);                     # 13,12
413
414         &movz   ($acc,&HB("eax"));              #  7
415         &xor    ("ecx",&DWP(1,$tbl,$acc,8));    #  7
416         &pxor   ("mm0","mm3");
417         &movz   ("eax",&LB("eax"));             #  6
418         &xor    ("edx",&DWP(2,$tbl,"eax",8));   #  6
419         &pshufw ("mm1","mm0",0x08);             #  5, 4, 1, 0
420         &movz   ($acc,&HB("ebx"));              # 13
421         &xor    ("ecx",&DWP(3,$tbl,$acc,8));    # 13
422         &xor    ("ecx",&DWP(24,$key));          # t[2]
423         &movd   ("mm4","ecx");                  # t[2] collected
424         &movz   ("ebx",&LB("ebx"));             # 12
425         &xor    ("edx",&DWP(0,$tbl,"ebx",8));   # 12
426         &shr    ("ecx",16);
427         &movd   ("eax","mm1");                  #  5, 4, 1, 0
428         &mov    ("ebx",&DWP(28,$key));          # t[3]
429         &xor    ("ebx","edx");
430         &movd   ("mm5","ebx");                  # t[3] collected
431         &and    ("ebx",0xffff0000);
432         &or     ("ebx","ecx");
433
434         &punpckldq      ("mm4","mm5");          # t[2,3] collected
435 }
436
437 ######################################################################
438 # "Compact" block function
439 ######################################################################
440
441 sub enccompact()
442 { my $Fn = \&mov;
443   while ($#_>5) { pop(@_); $Fn=sub{}; }
444   my ($i,$te,@s)=@_;
445   my $tmp = $key;
446   my $out = $i==3?$s[0]:$acc;
447
448         # $Fn is used in first compact round and its purpose is to
449         # void restoration of some values from stack, so that after
450         # 4xenccompact with extra argument $key value is left there...
451         if ($i==3)  {   &$Fn    ($key,$__key);                  }##%edx
452         else        {   &mov    ($out,$s[0]);                   }
453                         &and    ($out,0xFF);
454         if ($i==1)  {   &shr    ($s[0],16);                     }#%ebx[1]
455         if ($i==2)  {   &shr    ($s[0],24);                     }#%ecx[2]
456                         &movz   ($out,&BP(-128,$te,$out,1));
457
458         if ($i==3)  {   $tmp=$s[1];                             }##%eax
459                         &movz   ($tmp,&HB($s[1]));
460                         &movz   ($tmp,&BP(-128,$te,$tmp,1));
461                         &shl    ($tmp,8);
462                         &xor    ($out,$tmp);
463
464         if ($i==3)  {   $tmp=$s[2]; &mov ($s[1],$__s0);         }##%ebx
465         else        {   &mov    ($tmp,$s[2]);
466                         &shr    ($tmp,16);                      }
467         if ($i==2)  {   &and    ($s[1],0xFF);                   }#%edx[2]
468                         &and    ($tmp,0xFF);
469                         &movz   ($tmp,&BP(-128,$te,$tmp,1));
470                         &shl    ($tmp,16);
471                         &xor    ($out,$tmp);
472
473         if ($i==3)  {   $tmp=$s[3]; &mov ($s[2],$__s1);         }##%ecx
474         elsif($i==2){   &movz   ($tmp,&HB($s[3]));              }#%ebx[2]
475         else        {   &mov    ($tmp,$s[3]);
476                         &shr    ($tmp,24);                      }
477                         &movz   ($tmp,&BP(-128,$te,$tmp,1));
478                         &shl    ($tmp,24);
479                         &xor    ($out,$tmp);
480         if ($i<2)   {   &mov    (&DWP(4+4*$i,"esp"),$out);      }
481         if ($i==3)  {   &mov    ($s[3],$acc);                   }
482         &comment();
483 }
484
485 sub enctransform()
486 { my @s = ($s0,$s1,$s2,$s3);
487   my $i = shift;
488   my $tmp = $tbl;
489   my $r2  = $key ;
490
491         &and    ($tmp,$s[$i]);
492         &lea    ($r2,&DWP(0,$s[$i],$s[$i]));
493         &mov    ($acc,$tmp);
494         &shr    ($tmp,7);
495         &and    ($r2,0xfefefefe);
496         &sub    ($acc,$tmp);
497         &mov    ($tmp,$s[$i]);
498         &and    ($acc,0x1b1b1b1b);
499         &rotr   ($tmp,16);
500         &xor    ($acc,$r2);     # r2
501         &mov    ($r2,$s[$i]);
502
503         &xor    ($s[$i],$acc);  # r0 ^ r2
504         &rotr   ($r2,16+8);
505         &xor    ($acc,$tmp);
506         &rotl   ($s[$i],24);
507         &xor    ($acc,$r2);
508         &mov    ($tmp,0x80808080)       if ($i!=1);
509         &xor    ($s[$i],$acc);  # ROTATE(r2^r0,24) ^ r2
510 }
511
512 &function_begin_B("_x86_AES_encrypt_compact");
513         # note that caller is expected to allocate stack frame for me!
514         &mov    ($__key,$key);                  # save key
515
516         &xor    ($s0,&DWP(0,$key));             # xor with key
517         &xor    ($s1,&DWP(4,$key));
518         &xor    ($s2,&DWP(8,$key));
519         &xor    ($s3,&DWP(12,$key));
520
521         &mov    ($acc,&DWP(240,$key));          # load key->rounds
522         &lea    ($acc,&DWP(-2,$acc,$acc));
523         &lea    ($acc,&DWP(0,$key,$acc,8));
524         &mov    ($__end,$acc);                  # end of key schedule
525
526         # prefetch Te4
527         &mov    ($key,&DWP(0-128,$tbl));
528         &mov    ($acc,&DWP(32-128,$tbl));
529         &mov    ($key,&DWP(64-128,$tbl));
530         &mov    ($acc,&DWP(96-128,$tbl));
531         &mov    ($key,&DWP(128-128,$tbl));
532         &mov    ($acc,&DWP(160-128,$tbl));
533         &mov    ($key,&DWP(192-128,$tbl));
534         &mov    ($acc,&DWP(224-128,$tbl));
535
536         &set_label("loop",16);
537
538                 &enccompact(0,$tbl,$s0,$s1,$s2,$s3,1);
539                 &enccompact(1,$tbl,$s1,$s2,$s3,$s0,1);
540                 &enccompact(2,$tbl,$s2,$s3,$s0,$s1,1);
541                 &enccompact(3,$tbl,$s3,$s0,$s1,$s2,1);
542                 &mov    ($tbl,0x80808080);
543                 &enctransform(2);
544                 &enctransform(3);
545                 &enctransform(0);
546                 &enctransform(1);
547                 &mov    ($key,$__key);
548                 &mov    ($tbl,$__tbl);
549                 &add    ($key,16);              # advance rd_key
550                 &xor    ($s0,&DWP(0,$key));
551                 &xor    ($s1,&DWP(4,$key));
552                 &xor    ($s2,&DWP(8,$key));
553                 &xor    ($s3,&DWP(12,$key));
554
555         &cmp    ($key,$__end);
556         &mov    ($__key,$key);
557         &jb     (&label("loop"));
558
559         &enccompact(0,$tbl,$s0,$s1,$s2,$s3);
560         &enccompact(1,$tbl,$s1,$s2,$s3,$s0);
561         &enccompact(2,$tbl,$s2,$s3,$s0,$s1);
562         &enccompact(3,$tbl,$s3,$s0,$s1,$s2);
563
564         &xor    ($s0,&DWP(16,$key));
565         &xor    ($s1,&DWP(20,$key));
566         &xor    ($s2,&DWP(24,$key));
567         &xor    ($s3,&DWP(28,$key));
568
569         &ret    ();
570 &function_end_B("_x86_AES_encrypt_compact");
571
572 ######################################################################
573 # "Compact" SSE block function.
574 ######################################################################
575 #
576 # Performance is not actually extraordinary in comparison to pure
577 # x86 code. In particular encrypt performance is virtually the same.
578 # Decrypt performance on the other hand is 15-20% better on newer
579 # µ-archs [but we're thankful for *any* improvement here], and ~50%
580 # better on PIII:-) And additionally on the pros side this code
581 # eliminates redundant references to stack and thus relieves/
582 # minimizes the pressure on the memory bus.
583 #
584 # MMX register layout                           lsb
585 # +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
586 # |          mm4          |          mm0          |
587 # +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
588 # |     s3    |     s2    |     s1    |     s0    |
589 # +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
590 # |15|14|13|12|11|10| 9| 8| 7| 6| 5| 4| 3| 2| 1| 0|
591 # +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
592 #
593 # Indexes translate as s[N/4]>>(8*(N%4)), e.g. 5 means s1>>8.
594 # In this terms encryption and decryption "compact" permutation
595 # matrices can be depicted as following:
596 #
597 # encryption              lsb   # decryption              lsb
598 # +----++----+----+----+----+   # +----++----+----+----+----+
599 # | t0 || 15 | 10 |  5 |  0 |   # | t0 ||  7 | 10 | 13 |  0 |
600 # +----++----+----+----+----+   # +----++----+----+----+----+
601 # | t1 ||  3 | 14 |  9 |  4 |   # | t1 || 11 | 14 |  1 |  4 |
602 # +----++----+----+----+----+   # +----++----+----+----+----+
603 # | t2 ||  7 |  2 | 13 |  8 |   # | t2 || 15 |  2 |  5 |  8 |
604 # +----++----+----+----+----+   # +----++----+----+----+----+
605 # | t3 || 11 |  6 |  1 | 12 |   # | t3 ||  3 |  6 |  9 | 12 |
606 # +----++----+----+----+----+   # +----++----+----+----+----+
607 #
608 ######################################################################
609 # Why not xmm registers? Short answer. It was actually tested and
610 # was not any faster, but *contrary*, most notably on Intel CPUs.
611 # Longer answer. Main advantage of using mm registers is that movd
612 # latency is lower, especially on Intel P4. While arithmetic
613 # instructions are twice as many, they can be scheduled every cycle
614 # and not every second one when they are operating on xmm register,
615 # so that "arithmetic throughput" remains virtually the same. And
616 # finally the code can be executed even on elder SSE-only CPUs:-)
617
618 sub sse_enccompact()
619 {
620         &pshufw ("mm1","mm0",0x08);             #  5, 4, 1, 0
621         &pshufw ("mm5","mm4",0x0d);             # 15,14,11,10
622         &movd   ("eax","mm1");                  #  5, 4, 1, 0
623         &movd   ("ebx","mm5");                  # 15,14,11,10
624         &mov    ($__key,$key);
625
626         &movz   ($acc,&LB("eax"));              #  0
627         &movz   ("edx",&HB("eax"));             #  1
628         &pshufw ("mm2","mm0",0x0d);             #  7, 6, 3, 2
629         &movz   ("ecx",&BP(-128,$tbl,$acc,1));  #  0
630         &movz   ($key,&LB("ebx"));              # 10
631         &movz   ("edx",&BP(-128,$tbl,"edx",1)); #  1
632         &shr    ("eax",16);                     #  5, 4
633         &shl    ("edx",8);                      #  1
634
635         &movz   ($acc,&BP(-128,$tbl,$key,1));   # 10
636         &movz   ($key,&HB("ebx"));              # 11
637         &shl    ($acc,16);                      # 10
638         &pshufw ("mm6","mm4",0x08);             # 13,12, 9, 8
639         &or     ("ecx",$acc);                   # 10
640         &movz   ($acc,&BP(-128,$tbl,$key,1));   # 11
641         &movz   ($key,&HB("eax"));              #  5
642         &shl    ($acc,24);                      # 11
643         &shr    ("ebx",16);                     # 15,14
644         &or     ("edx",$acc);                   # 11
645
646         &movz   ($acc,&BP(-128,$tbl,$key,1));   #  5
647         &movz   ($key,&HB("ebx"));              # 15
648         &shl    ($acc,8);                       #  5
649         &or     ("ecx",$acc);                   #  5
650         &movz   ($acc,&BP(-128,$tbl,$key,1));   # 15
651         &movz   ($key,&LB("eax"));              #  4
652         &shl    ($acc,24);                      # 15
653         &or     ("ecx",$acc);                   # 15
654
655         &movz   ($acc,&BP(-128,$tbl,$key,1));   #  4
656         &movz   ($key,&LB("ebx"));              # 14
657         &movd   ("eax","mm2");                  #  7, 6, 3, 2
658         &movd   ("mm0","ecx");                  # t[0] collected
659         &movz   ("ecx",&BP(-128,$tbl,$key,1));  # 14
660         &movz   ($key,&HB("eax"));              #  3
661         &shl    ("ecx",16);                     # 14
662         &movd   ("ebx","mm6");                  # 13,12, 9, 8
663         &or     ("ecx",$acc);                   # 14
664
665         &movz   ($acc,&BP(-128,$tbl,$key,1));   #  3
666         &movz   ($key,&HB("ebx"));              #  9
667         &shl    ($acc,24);                      #  3
668         &or     ("ecx",$acc);                   #  3
669         &movz   ($acc,&BP(-128,$tbl,$key,1));   #  9
670         &movz   ($key,&LB("ebx"));              #  8
671         &shl    ($acc,8);                       #  9
672         &shr    ("ebx",16);                     # 13,12
673         &or     ("ecx",$acc);                   #  9
674
675         &movz   ($acc,&BP(-128,$tbl,$key,1));   #  8
676         &movz   ($key,&LB("eax"));              #  2
677         &shr    ("eax",16);                     #  7, 6
678         &movd   ("mm1","ecx");                  # t[1] collected
679         &movz   ("ecx",&BP(-128,$tbl,$key,1));  #  2
680         &movz   ($key,&HB("eax"));              #  7
681         &shl    ("ecx",16);                     #  2
682         &and    ("eax",0xff);                   #  6
683         &or     ("ecx",$acc);                   #  2
684
685         &punpckldq      ("mm0","mm1");          # t[0,1] collected
686
687         &movz   ($acc,&BP(-128,$tbl,$key,1));   #  7
688         &movz   ($key,&HB("ebx"));              # 13
689         &shl    ($acc,24);                      #  7
690         &and    ("ebx",0xff);                   # 12
691         &movz   ("eax",&BP(-128,$tbl,"eax",1)); #  6
692         &or     ("ecx",$acc);                   #  7
693         &shl    ("eax",16);                     #  6
694         &movz   ($acc,&BP(-128,$tbl,$key,1));   # 13
695         &or     ("edx","eax");                  #  6
696         &shl    ($acc,8);                       # 13
697         &movz   ("ebx",&BP(-128,$tbl,"ebx",1)); # 12
698         &or     ("ecx",$acc);                   # 13
699         &or     ("edx","ebx");                  # 12
700         &mov    ($key,$__key);
701         &movd   ("mm4","ecx");                  # t[2] collected
702         &movd   ("mm5","edx");                  # t[3] collected
703
704         &punpckldq      ("mm4","mm5");          # t[2,3] collected
705 }
706
707                                         if (!$x86only) {
708 &function_begin_B("_sse_AES_encrypt_compact");
709         &pxor   ("mm0",&QWP(0,$key));   #  7, 6, 5, 4, 3, 2, 1, 0
710         &pxor   ("mm4",&QWP(8,$key));   # 15,14,13,12,11,10, 9, 8
711
712         # note that caller is expected to allocate stack frame for me!
713         &mov    ($acc,&DWP(240,$key));          # load key->rounds
714         &lea    ($acc,&DWP(-2,$acc,$acc));
715         &lea    ($acc,&DWP(0,$key,$acc,8));
716         &mov    ($__end,$acc);                  # end of key schedule
717
718         &mov    ($s0,0x1b1b1b1b);               # magic constant
719         &mov    (&DWP(8,"esp"),$s0);
720         &mov    (&DWP(12,"esp"),$s0);
721
722         # prefetch Te4
723         &mov    ($s0,&DWP(0-128,$tbl));
724         &mov    ($s1,&DWP(32-128,$tbl));
725         &mov    ($s2,&DWP(64-128,$tbl));
726         &mov    ($s3,&DWP(96-128,$tbl));
727         &mov    ($s0,&DWP(128-128,$tbl));
728         &mov    ($s1,&DWP(160-128,$tbl));
729         &mov    ($s2,&DWP(192-128,$tbl));
730         &mov    ($s3,&DWP(224-128,$tbl));
731
732         &set_label("loop",16);
733                 &sse_enccompact();
734                 &add    ($key,16);
735                 &cmp    ($key,$__end);
736                 &ja     (&label("out"));
737
738                 &movq   ("mm2",&QWP(8,"esp"));
739                 &pxor   ("mm3","mm3");          &pxor   ("mm7","mm7");
740                 &movq   ("mm1","mm0");          &movq   ("mm5","mm4");  # r0
741                 &pcmpgtb("mm3","mm0");          &pcmpgtb("mm7","mm4");
742                 &pand   ("mm3","mm2");          &pand   ("mm7","mm2");
743                 &pshufw ("mm2","mm0",0xb1);     &pshufw ("mm6","mm4",0xb1);# ROTATE(r0,16)
744                 &paddb  ("mm0","mm0");          &paddb  ("mm4","mm4");
745                 &pxor   ("mm0","mm3");          &pxor   ("mm4","mm7");  # = r2
746                 &pshufw ("mm3","mm2",0xb1);     &pshufw ("mm7","mm6",0xb1);# r0
747                 &pxor   ("mm1","mm0");          &pxor   ("mm5","mm4");  # r0^r2
748                 &pxor   ("mm0","mm2");          &pxor   ("mm4","mm6");  # ^= ROTATE(r0,16)
749
750                 &movq   ("mm2","mm3");          &movq   ("mm6","mm7");
751                 &pslld  ("mm3",8);              &pslld  ("mm7",8);
752                 &psrld  ("mm2",24);             &psrld  ("mm6",24);
753                 &pxor   ("mm0","mm3");          &pxor   ("mm4","mm7");  # ^= r0<<8
754                 &pxor   ("mm0","mm2");          &pxor   ("mm4","mm6");  # ^= r0>>24
755
756                 &movq   ("mm3","mm1");          &movq   ("mm7","mm5");
757                 &movq   ("mm2",&QWP(0,$key));   &movq   ("mm6",&QWP(8,$key));
758                 &psrld  ("mm1",8);              &psrld  ("mm5",8);
759                 &mov    ($s0,&DWP(0-128,$tbl));
760                 &pslld  ("mm3",24);             &pslld  ("mm7",24);
761                 &mov    ($s1,&DWP(64-128,$tbl));
762                 &pxor   ("mm0","mm1");          &pxor   ("mm4","mm5");  # ^= (r2^r0)<<8
763                 &mov    ($s2,&DWP(128-128,$tbl));
764                 &pxor   ("mm0","mm3");          &pxor   ("mm4","mm7");  # ^= (r2^r0)>>24
765                 &mov    ($s3,&DWP(192-128,$tbl));
766
767                 &pxor   ("mm0","mm2");          &pxor   ("mm4","mm6");
768         &jmp    (&label("loop"));
769
770         &set_label("out",16);
771         &pxor   ("mm0",&QWP(0,$key));
772         &pxor   ("mm4",&QWP(8,$key));
773
774         &ret    ();
775 &function_end_B("_sse_AES_encrypt_compact");
776                                         }
777
778 ######################################################################
779 # Vanilla block function.
780 ######################################################################
781
782 sub encstep()
783 { my ($i,$te,@s) = @_;
784   my $tmp = $key;
785   my $out = $i==3?$s[0]:$acc;
786
787         # lines marked with #%e?x[i] denote "reordered" instructions...
788         if ($i==3)  {   &mov    ($key,$__key);                  }##%edx
789         else        {   &mov    ($out,$s[0]);
790                         &and    ($out,0xFF);                    }
791         if ($i==1)  {   &shr    ($s[0],16);                     }#%ebx[1]
792         if ($i==2)  {   &shr    ($s[0],24);                     }#%ecx[2]
793                         &mov    ($out,&DWP(0,$te,$out,8));
794
795         if ($i==3)  {   $tmp=$s[1];                             }##%eax
796                         &movz   ($tmp,&HB($s[1]));
797                         &xor    ($out,&DWP(3,$te,$tmp,8));
798
799         if ($i==3)  {   $tmp=$s[2]; &mov ($s[1],$__s0);         }##%ebx
800         else        {   &mov    ($tmp,$s[2]);
801                         &shr    ($tmp,16);                      }
802         if ($i==2)  {   &and    ($s[1],0xFF);                   }#%edx[2]
803                         &and    ($tmp,0xFF);
804                         &xor    ($out,&DWP(2,$te,$tmp,8));
805
806         if ($i==3)  {   $tmp=$s[3]; &mov ($s[2],$__s1);         }##%ecx
807         elsif($i==2){   &movz   ($tmp,&HB($s[3]));              }#%ebx[2]
808         else        {   &mov    ($tmp,$s[3]);
809                         &shr    ($tmp,24)                       }
810                         &xor    ($out,&DWP(1,$te,$tmp,8));
811         if ($i<2)   {   &mov    (&DWP(4+4*$i,"esp"),$out);      }
812         if ($i==3)  {   &mov    ($s[3],$acc);                   }
813                         &comment();
814 }
815
816 sub enclast()
817 { my ($i,$te,@s)=@_;
818   my $tmp = $key;
819   my $out = $i==3?$s[0]:$acc;
820
821         if ($i==3)  {   &mov    ($key,$__key);                  }##%edx
822         else        {   &mov    ($out,$s[0]);                   }
823                         &and    ($out,0xFF);
824         if ($i==1)  {   &shr    ($s[0],16);                     }#%ebx[1]
825         if ($i==2)  {   &shr    ($s[0],24);                     }#%ecx[2]
826                         &mov    ($out,&DWP(2,$te,$out,8));
827                         &and    ($out,0x000000ff);
828
829         if ($i==3)  {   $tmp=$s[1];                             }##%eax
830                         &movz   ($tmp,&HB($s[1]));
831                         &mov    ($tmp,&DWP(0,$te,$tmp,8));
832                         &and    ($tmp,0x0000ff00);
833                         &xor    ($out,$tmp);
834
835         if ($i==3)  {   $tmp=$s[2]; &mov ($s[1],$__s0);         }##%ebx
836         else        {   &mov    ($tmp,$s[2]);
837                         &shr    ($tmp,16);                      }
838         if ($i==2)  {   &and    ($s[1],0xFF);                   }#%edx[2]
839                         &and    ($tmp,0xFF);
840                         &mov    ($tmp,&DWP(0,$te,$tmp,8));
841                         &and    ($tmp,0x00ff0000);
842                         &xor    ($out,$tmp);
843
844         if ($i==3)  {   $tmp=$s[3]; &mov ($s[2],$__s1);         }##%ecx
845         elsif($i==2){   &movz   ($tmp,&HB($s[3]));              }#%ebx[2]
846         else        {   &mov    ($tmp,$s[3]);
847                         &shr    ($tmp,24);                      }
848                         &mov    ($tmp,&DWP(2,$te,$tmp,8));
849                         &and    ($tmp,0xff000000);
850                         &xor    ($out,$tmp);
851         if ($i<2)   {   &mov    (&DWP(4+4*$i,"esp"),$out);      }
852         if ($i==3)  {   &mov    ($s[3],$acc);                   }
853 }
854
855 &function_begin_B("_x86_AES_encrypt");
856         if ($vertical_spin) {
857                 # I need high parts of volatile registers to be accessible...
858                 &exch   ($s1="edi",$key="ebx");
859                 &mov    ($s2="esi",$acc="ecx");
860         }
861
862         # note that caller is expected to allocate stack frame for me!
863         &mov    ($__key,$key);                  # save key
864
865         &xor    ($s0,&DWP(0,$key));             # xor with key
866         &xor    ($s1,&DWP(4,$key));
867         &xor    ($s2,&DWP(8,$key));
868         &xor    ($s3,&DWP(12,$key));
869
870         &mov    ($acc,&DWP(240,$key));          # load key->rounds
871
872         if ($small_footprint) {
873             &lea        ($acc,&DWP(-2,$acc,$acc));
874             &lea        ($acc,&DWP(0,$key,$acc,8));
875             &mov        ($__end,$acc);          # end of key schedule
876
877             &set_label("loop",16);
878                 if ($vertical_spin) {
879                     &encvert($tbl,$s0,$s1,$s2,$s3);
880                 } else {
881                     &encstep(0,$tbl,$s0,$s1,$s2,$s3);
882                     &encstep(1,$tbl,$s1,$s2,$s3,$s0);
883                     &encstep(2,$tbl,$s2,$s3,$s0,$s1);
884                     &encstep(3,$tbl,$s3,$s0,$s1,$s2);
885                 }
886                 &add    ($key,16);              # advance rd_key
887                 &xor    ($s0,&DWP(0,$key));
888                 &xor    ($s1,&DWP(4,$key));
889                 &xor    ($s2,&DWP(8,$key));
890                 &xor    ($s3,&DWP(12,$key));
891             &cmp        ($key,$__end);
892             &mov        ($__key,$key);
893             &jb         (&label("loop"));
894         }
895         else {
896             &cmp        ($acc,10);
897             &jle        (&label("10rounds"));
898             &cmp        ($acc,12);
899             &jle        (&label("12rounds"));
900
901         &set_label("14rounds",4);
902             for ($i=1;$i<3;$i++) {
903                 if ($vertical_spin) {
904                     &encvert($tbl,$s0,$s1,$s2,$s3);
905                 } else {
906                     &encstep(0,$tbl,$s0,$s1,$s2,$s3);
907                     &encstep(1,$tbl,$s1,$s2,$s3,$s0);
908                     &encstep(2,$tbl,$s2,$s3,$s0,$s1);
909                     &encstep(3,$tbl,$s3,$s0,$s1,$s2);
910                 }
911                 &xor    ($s0,&DWP(16*$i+0,$key));
912                 &xor    ($s1,&DWP(16*$i+4,$key));
913                 &xor    ($s2,&DWP(16*$i+8,$key));
914                 &xor    ($s3,&DWP(16*$i+12,$key));
915             }
916             &add        ($key,32);
917             &mov        ($__key,$key);          # advance rd_key
918         &set_label("12rounds",4);
919             for ($i=1;$i<3;$i++) {
920                 if ($vertical_spin) {
921                     &encvert($tbl,$s0,$s1,$s2,$s3);
922                 } else {
923                     &encstep(0,$tbl,$s0,$s1,$s2,$s3);
924                     &encstep(1,$tbl,$s1,$s2,$s3,$s0);
925                     &encstep(2,$tbl,$s2,$s3,$s0,$s1);
926                     &encstep(3,$tbl,$s3,$s0,$s1,$s2);
927                 }
928                 &xor    ($s0,&DWP(16*$i+0,$key));
929                 &xor    ($s1,&DWP(16*$i+4,$key));
930                 &xor    ($s2,&DWP(16*$i+8,$key));
931                 &xor    ($s3,&DWP(16*$i+12,$key));
932             }
933             &add        ($key,32);
934             &mov        ($__key,$key);          # advance rd_key
935         &set_label("10rounds",4);
936             for ($i=1;$i<10;$i++) {
937                 if ($vertical_spin) {
938                     &encvert($tbl,$s0,$s1,$s2,$s3);
939                 } else {
940                     &encstep(0,$tbl,$s0,$s1,$s2,$s3);
941                     &encstep(1,$tbl,$s1,$s2,$s3,$s0);
942                     &encstep(2,$tbl,$s2,$s3,$s0,$s1);
943                     &encstep(3,$tbl,$s3,$s0,$s1,$s2);
944                 }
945                 &xor    ($s0,&DWP(16*$i+0,$key));
946                 &xor    ($s1,&DWP(16*$i+4,$key));
947                 &xor    ($s2,&DWP(16*$i+8,$key));
948                 &xor    ($s3,&DWP(16*$i+12,$key));
949             }
950         }
951
952         if ($vertical_spin) {
953             # "reincarnate" some registers for "horizontal" spin...
954             &mov        ($s1="ebx",$key="edi");
955             &mov        ($s2="ecx",$acc="esi");
956         }
957         &enclast(0,$tbl,$s0,$s1,$s2,$s3);
958         &enclast(1,$tbl,$s1,$s2,$s3,$s0);
959         &enclast(2,$tbl,$s2,$s3,$s0,$s1);
960         &enclast(3,$tbl,$s3,$s0,$s1,$s2);
961
962         &add    ($key,$small_footprint?16:160);
963         &xor    ($s0,&DWP(0,$key));
964         &xor    ($s1,&DWP(4,$key));
965         &xor    ($s2,&DWP(8,$key));
966         &xor    ($s3,&DWP(12,$key));
967
968         &ret    ();
969
970 &set_label("AES_Te",64);        # Yes! I keep it in the code segment!
971         &_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6);
972         &_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591);
973         &_data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56);
974         &_data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec);
975         &_data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa);
976         &_data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb);
977         &_data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45);
978         &_data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b);
979         &_data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c);
980         &_data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83);
981         &_data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9);
982         &_data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a);
983         &_data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d);
984         &_data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f);
985         &_data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df);
986         &_data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea);
987         &_data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34);
988         &_data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b);
989         &_data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d);
990         &_data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413);
991         &_data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1);
992         &_data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6);
993         &_data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972);
994         &_data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85);
995         &_data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed);
996         &_data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511);
997         &_data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe);
998         &_data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b);
999         &_data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05);
1000         &_data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1);
1001         &_data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142);
1002         &_data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf);
1003         &_data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3);
1004         &_data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e);
1005         &_data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a);
1006         &_data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6);
1007         &_data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3);
1008         &_data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b);
1009         &_data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428);
1010         &_data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad);
1011         &_data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14);
1012         &_data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8);
1013         &_data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4);
1014         &_data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2);
1015         &_data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda);
1016         &_data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949);
1017         &_data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf);
1018         &_data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810);
1019         &_data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c);
1020         &_data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697);
1021         &_data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e);
1022         &_data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f);
1023         &_data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc);
1024         &_data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c);
1025         &_data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969);
1026         &_data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27);
1027         &_data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122);
1028         &_data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433);
1029         &_data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9);
1030         &_data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5);
1031         &_data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a);
1032         &_data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0);
1033         &_data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e);
1034         &_data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c);
1035
1036 #Te4    # four copies of Te4 to choose from to avoid L1 aliasing
1037         &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
1038         &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
1039         &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
1040         &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
1041         &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
1042         &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
1043         &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
1044         &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
1045         &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
1046         &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
1047         &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
1048         &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
1049         &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
1050         &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
1051         &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
1052         &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
1053         &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
1054         &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
1055         &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
1056         &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
1057         &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
1058         &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
1059         &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
1060         &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
1061         &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
1062         &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
1063         &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
1064         &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
1065         &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
1066         &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
1067         &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
1068         &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
1069
1070         &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
1071         &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
1072         &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
1073         &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
1074         &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
1075         &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
1076         &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
1077         &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
1078         &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
1079         &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
1080         &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
1081         &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
1082         &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
1083         &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
1084         &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
1085         &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
1086         &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
1087         &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
1088         &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
1089         &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
1090         &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
1091         &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
1092         &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
1093         &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
1094         &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
1095         &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
1096         &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
1097         &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
1098         &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
1099         &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
1100         &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
1101         &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
1102
1103         &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
1104         &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
1105         &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
1106         &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
1107         &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
1108         &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
1109         &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
1110         &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
1111         &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
1112         &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
1113         &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
1114         &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
1115         &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
1116         &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
1117         &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
1118         &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
1119         &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
1120         &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
1121         &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
1122         &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
1123         &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
1124         &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
1125         &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
1126         &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
1127         &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
1128         &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
1129         &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
1130         &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
1131         &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
1132         &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
1133         &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
1134         &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
1135
1136         &data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
1137         &data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
1138         &data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
1139         &data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
1140         &data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
1141         &data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
1142         &data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
1143         &data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
1144         &data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
1145         &data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
1146         &data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
1147         &data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
1148         &data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
1149         &data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
1150         &data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
1151         &data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
1152         &data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
1153         &data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
1154         &data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
1155         &data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
1156         &data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
1157         &data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
1158         &data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
1159         &data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
1160         &data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
1161         &data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
1162         &data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
1163         &data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
1164         &data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
1165         &data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
1166         &data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
1167         &data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
1168 #rcon:
1169         &data_word(0x00000001, 0x00000002, 0x00000004, 0x00000008);
1170         &data_word(0x00000010, 0x00000020, 0x00000040, 0x00000080);
1171         &data_word(0x0000001b, 0x00000036, 0x00000000, 0x00000000);
1172         &data_word(0x00000000, 0x00000000, 0x00000000, 0x00000000);
1173 &function_end_B("_x86_AES_encrypt");
1174
1175 # void AES_encrypt (const void *inp,void *out,const AES_KEY *key);
1176 &function_begin("AES_encrypt");
1177         &mov    ($acc,&wparam(0));              # load inp
1178         &mov    ($key,&wparam(2));              # load key
1179
1180         &mov    ($s0,"esp");
1181         &sub    ("esp",36);
1182         &and    ("esp",-64);                    # align to cache-line
1183
1184         # place stack frame just "above" the key schedule
1185         &lea    ($s1,&DWP(-64-63,$key));
1186         &sub    ($s1,"esp");
1187         &neg    ($s1);
1188         &and    ($s1,0x3C0);    # modulo 1024, but aligned to cache-line
1189         &sub    ("esp",$s1);
1190         &add    ("esp",4);      # 4 is reserved for caller's return address
1191         &mov    ($_esp,$s0);                    # save stack pointer
1192
1193         &call   (&label("pic_point"));          # make it PIC!
1194         &set_label("pic_point");
1195         &blindpop($tbl);
1196         &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if (!$x86only);
1197         &lea    ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
1198
1199         # pick Te4 copy which can't "overlap" with stack frame or key schedule
1200         &lea    ($s1,&DWP(768-4,"esp"));
1201         &sub    ($s1,$tbl);
1202         &and    ($s1,0x300);
1203         &lea    ($tbl,&DWP(2048+128,$tbl,$s1));
1204
1205                                         if (!$x86only) {
1206         &bt     (&DWP(0,$s0),25);       # check for SSE bit
1207         &jnc    (&label("x86"));
1208
1209         &movq   ("mm0",&QWP(0,$acc));
1210         &movq   ("mm4",&QWP(8,$acc));
1211         &call   ("_sse_AES_encrypt_compact");
1212         &mov    ("esp",$_esp);                  # restore stack pointer
1213         &mov    ($acc,&wparam(1));              # load out
1214         &movq   (&QWP(0,$acc),"mm0");           # write output data
1215         &movq   (&QWP(8,$acc),"mm4");
1216         &emms   ();
1217         &function_end_A();
1218                                         }
1219         &set_label("x86",16);
1220         &mov    ($_tbl,$tbl);
1221         &mov    ($s0,&DWP(0,$acc));             # load input data
1222         &mov    ($s1,&DWP(4,$acc));
1223         &mov    ($s2,&DWP(8,$acc));
1224         &mov    ($s3,&DWP(12,$acc));
1225         &call   ("_x86_AES_encrypt_compact");
1226         &mov    ("esp",$_esp);                  # restore stack pointer
1227         &mov    ($acc,&wparam(1));              # load out
1228         &mov    (&DWP(0,$acc),$s0);             # write output data
1229         &mov    (&DWP(4,$acc),$s1);
1230         &mov    (&DWP(8,$acc),$s2);
1231         &mov    (&DWP(12,$acc),$s3);
1232 &function_end("AES_encrypt");
1233
1234 #--------------------------------------------------------------------#
1235
1236 ######################################################################
1237 # "Compact" block function
1238 ######################################################################
1239
1240 sub deccompact()
1241 { my $Fn = \&mov;
1242   while ($#_>5) { pop(@_); $Fn=sub{}; }
1243   my ($i,$td,@s)=@_;
1244   my $tmp = $key;
1245   my $out = $i==3?$s[0]:$acc;
1246
1247         # $Fn is used in first compact round and its purpose is to
1248         # void restoration of some values from stack, so that after
1249         # 4xdeccompact with extra argument $key, $s0 and $s1 values
1250         # are left there...
1251         if($i==3)   {   &$Fn    ($key,$__key);                  }
1252         else        {   &mov    ($out,$s[0]);                   }
1253                         &and    ($out,0xFF);
1254                         &movz   ($out,&BP(-128,$td,$out,1));
1255
1256         if ($i==3)  {   $tmp=$s[1];                             }
1257                         &movz   ($tmp,&HB($s[1]));
1258                         &movz   ($tmp,&BP(-128,$td,$tmp,1));
1259                         &shl    ($tmp,8);
1260                         &xor    ($out,$tmp);
1261
1262         if ($i==3)  {   $tmp=$s[2]; &mov ($s[1],$acc);          }
1263         else        {   mov     ($tmp,$s[2]);                   }
1264                         &shr    ($tmp,16);
1265                         &and    ($tmp,0xFF);
1266                         &movz   ($tmp,&BP(-128,$td,$tmp,1));
1267                         &shl    ($tmp,16);
1268                         &xor    ($out,$tmp);
1269
1270         if ($i==3)  {   $tmp=$s[3]; &$Fn ($s[2],$__s1);         }
1271         else        {   &mov    ($tmp,$s[3]);                   }
1272                         &shr    ($tmp,24);
1273                         &movz   ($tmp,&BP(-128,$td,$tmp,1));
1274                         &shl    ($tmp,24);
1275                         &xor    ($out,$tmp);
1276         if ($i<2)   {   &mov    (&DWP(4+4*$i,"esp"),$out);      }
1277         if ($i==3)  {   &$Fn    ($s[3],$__s0);                  }
1278 }
1279
1280 # must be called with 2,3,0,1 as argument sequence!!!
1281 sub dectransform()
1282 { my @s = ($s0,$s1,$s2,$s3);
1283   my $i = shift;
1284   my $tmp = $key;
1285   my $tp2 = @s[($i+2)%4]; $tp2 = @s[2] if ($i==1);
1286   my $tp4 = @s[($i+3)%4]; $tp4 = @s[3] if ($i==1);
1287   my $tp8 = $tbl;
1288
1289         &mov    ($tmp,0x80808080);
1290         &and    ($tmp,$s[$i]);
1291         &mov    ($acc,$tmp);
1292         &shr    ($tmp,7);
1293         &lea    ($tp2,&DWP(0,$s[$i],$s[$i]));
1294         &sub    ($acc,$tmp);
1295         &and    ($tp2,0xfefefefe);
1296         &and    ($acc,0x1b1b1b1b);
1297         &xor    ($tp2,$acc);
1298         &mov    ($tmp,0x80808080);
1299
1300         &and    ($tmp,$tp2);
1301         &mov    ($acc,$tmp);
1302         &shr    ($tmp,7);
1303         &lea    ($tp4,&DWP(0,$tp2,$tp2));
1304         &sub    ($acc,$tmp);
1305         &and    ($tp4,0xfefefefe);
1306         &and    ($acc,0x1b1b1b1b);
1307          &xor   ($tp2,$s[$i]);  # tp2^tp1
1308         &xor    ($tp4,$acc);
1309         &mov    ($tmp,0x80808080);
1310
1311         &and    ($tmp,$tp4);
1312         &mov    ($acc,$tmp);
1313         &shr    ($tmp,7);
1314         &lea    ($tp8,&DWP(0,$tp4,$tp4));
1315         &sub    ($acc,$tmp);
1316         &and    ($tp8,0xfefefefe);
1317         &and    ($acc,0x1b1b1b1b);
1318          &xor   ($tp4,$s[$i]);  # tp4^tp1
1319          &rotl  ($s[$i],8);     # = ROTATE(tp1,8)
1320         &xor    ($tp8,$acc);
1321
1322         &xor    ($s[$i],$tp2);
1323         &xor    ($tp2,$tp8);
1324         &xor    ($s[$i],$tp4);
1325         &xor    ($tp4,$tp8);
1326         &rotl   ($tp2,24);
1327         &xor    ($s[$i],$tp8);  # ^= tp8^(tp4^tp1)^(tp2^tp1)
1328         &rotl   ($tp4,16);
1329         &xor    ($s[$i],$tp2);  # ^= ROTATE(tp8^tp2^tp1,24)
1330         &rotl   ($tp8,8);
1331         &xor    ($s[$i],$tp4);  # ^= ROTATE(tp8^tp4^tp1,16)
1332          &mov   ($s[0],$__s0)                   if($i==2); #prefetch $s0
1333          &mov   ($s[1],$__s1)                   if($i==3); #prefetch $s1
1334          &mov   ($s[2],$__s2)                   if($i==1);
1335         &xor    ($s[$i],$tp8);  # ^= ROTATE(tp8,8)
1336
1337         &mov    ($s[3],$__s3)                   if($i==1);
1338         &mov    (&DWP(4+4*$i,"esp"),$s[$i])     if($i>=2);
1339 }
1340
1341 &function_begin_B("_x86_AES_decrypt_compact");
1342         # note that caller is expected to allocate stack frame for me!
1343         &mov    ($__key,$key);                  # save key
1344
1345         &xor    ($s0,&DWP(0,$key));             # xor with key
1346         &xor    ($s1,&DWP(4,$key));
1347         &xor    ($s2,&DWP(8,$key));
1348         &xor    ($s3,&DWP(12,$key));
1349
1350         &mov    ($acc,&DWP(240,$key));          # load key->rounds
1351
1352         &lea    ($acc,&DWP(-2,$acc,$acc));
1353         &lea    ($acc,&DWP(0,$key,$acc,8));
1354         &mov    ($__end,$acc);                  # end of key schedule
1355
1356         # prefetch Td4
1357         &mov    ($key,&DWP(0-128,$tbl));
1358         &mov    ($acc,&DWP(32-128,$tbl));
1359         &mov    ($key,&DWP(64-128,$tbl));
1360         &mov    ($acc,&DWP(96-128,$tbl));
1361         &mov    ($key,&DWP(128-128,$tbl));
1362         &mov    ($acc,&DWP(160-128,$tbl));
1363         &mov    ($key,&DWP(192-128,$tbl));
1364         &mov    ($acc,&DWP(224-128,$tbl));
1365
1366         &set_label("loop",16);
1367
1368                 &deccompact(0,$tbl,$s0,$s3,$s2,$s1,1);
1369                 &deccompact(1,$tbl,$s1,$s0,$s3,$s2,1);
1370                 &deccompact(2,$tbl,$s2,$s1,$s0,$s3,1);
1371                 &deccompact(3,$tbl,$s3,$s2,$s1,$s0,1);
1372                 &dectransform(2);
1373                 &dectransform(3);
1374                 &dectransform(0);
1375                 &dectransform(1);
1376                 &mov    ($key,$__key);
1377                 &mov    ($tbl,$__tbl);
1378                 &add    ($key,16);              # advance rd_key
1379                 &xor    ($s0,&DWP(0,$key));
1380                 &xor    ($s1,&DWP(4,$key));
1381                 &xor    ($s2,&DWP(8,$key));
1382                 &xor    ($s3,&DWP(12,$key));
1383
1384         &cmp    ($key,$__end);
1385         &mov    ($__key,$key);
1386         &jb     (&label("loop"));
1387
1388         &deccompact(0,$tbl,$s0,$s3,$s2,$s1);
1389         &deccompact(1,$tbl,$s1,$s0,$s3,$s2);
1390         &deccompact(2,$tbl,$s2,$s1,$s0,$s3);
1391         &deccompact(3,$tbl,$s3,$s2,$s1,$s0);
1392
1393         &xor    ($s0,&DWP(16,$key));
1394         &xor    ($s1,&DWP(20,$key));
1395         &xor    ($s2,&DWP(24,$key));
1396         &xor    ($s3,&DWP(28,$key));
1397
1398         &ret    ();
1399 &function_end_B("_x86_AES_decrypt_compact");
1400
1401 ######################################################################
1402 # "Compact" SSE block function.
1403 ######################################################################
1404
1405 sub sse_deccompact()
1406 {
1407         &pshufw ("mm1","mm0",0x0c);             #  7, 6, 1, 0
1408         &pshufw ("mm5","mm4",0x09);             # 13,12,11,10
1409         &movd   ("eax","mm1");                  #  7, 6, 1, 0
1410         &movd   ("ebx","mm5");                  # 13,12,11,10
1411         &mov    ($__key,$key);
1412
1413         &movz   ($acc,&LB("eax"));              #  0
1414         &movz   ("edx",&HB("eax"));             #  1
1415         &pshufw ("mm2","mm0",0x06);             #  3, 2, 5, 4
1416         &movz   ("ecx",&BP(-128,$tbl,$acc,1));  #  0
1417         &movz   ($key,&LB("ebx"));              # 10
1418         &movz   ("edx",&BP(-128,$tbl,"edx",1)); #  1
1419         &shr    ("eax",16);                     #  7, 6
1420         &shl    ("edx",8);                      #  1
1421
1422         &movz   ($acc,&BP(-128,$tbl,$key,1));   # 10
1423         &movz   ($key,&HB("ebx"));              # 11
1424         &shl    ($acc,16);                      # 10
1425         &pshufw ("mm6","mm4",0x03);             # 9, 8,15,14
1426         &or     ("ecx",$acc);                   # 10
1427         &movz   ($acc,&BP(-128,$tbl,$key,1));   # 11
1428         &movz   ($key,&HB("eax"));              #  7
1429         &shl    ($acc,24);                      # 11
1430         &shr    ("ebx",16);                     # 13,12
1431         &or     ("edx",$acc);                   # 11
1432
1433         &movz   ($acc,&BP(-128,$tbl,$key,1));   #  7
1434         &movz   ($key,&HB("ebx"));              # 13
1435         &shl    ($acc,24);                      #  7
1436         &or     ("ecx",$acc);                   #  7
1437         &movz   ($acc,&BP(-128,$tbl,$key,1));   # 13
1438         &movz   ($key,&LB("eax"));              #  6
1439         &shl    ($acc,8);                       # 13
1440         &movd   ("eax","mm2");                  #  3, 2, 5, 4
1441         &or     ("ecx",$acc);                   # 13
1442
1443         &movz   ($acc,&BP(-128,$tbl,$key,1));   #  6
1444         &movz   ($key,&LB("ebx"));              # 12
1445         &shl    ($acc,16);                      #  6
1446         &movd   ("ebx","mm6");                  #  9, 8,15,14
1447         &movd   ("mm0","ecx");                  # t[0] collected
1448         &movz   ("ecx",&BP(-128,$tbl,$key,1));  # 12
1449         &movz   ($key,&LB("eax"));              #  4
1450         &or     ("ecx",$acc);                   # 12
1451
1452         &movz   ($acc,&BP(-128,$tbl,$key,1));   #  4
1453         &movz   ($key,&LB("ebx"));              # 14
1454         &or     ("edx",$acc);                   #  4
1455         &movz   ($acc,&BP(-128,$tbl,$key,1));   # 14
1456         &movz   ($key,&HB("eax"));              #  5
1457         &shl    ($acc,16);                      # 14
1458         &shr    ("eax",16);                     #  3, 2
1459         &or     ("edx",$acc);                   # 14
1460
1461         &movz   ($acc,&BP(-128,$tbl,$key,1));   #  5
1462         &movz   ($key,&HB("ebx"));              # 15
1463         &shr    ("ebx",16);                     #  9, 8
1464         &shl    ($acc,8);                       #  5
1465         &movd   ("mm1","edx");                  # t[1] collected
1466         &movz   ("edx",&BP(-128,$tbl,$key,1));  # 15
1467         &movz   ($key,&HB("ebx"));              #  9
1468         &shl    ("edx",24);                     # 15
1469         &and    ("ebx",0xff);                   #  8
1470         &or     ("edx",$acc);                   # 15
1471
1472         &punpckldq      ("mm0","mm1");          # t[0,1] collected
1473
1474         &movz   ($acc,&BP(-128,$tbl,$key,1));   #  9
1475         &movz   ($key,&LB("eax"));              #  2
1476         &shl    ($acc,8);                       #  9
1477         &movz   ("eax",&HB("eax"));             #  3
1478         &movz   ("ebx",&BP(-128,$tbl,"ebx",1)); #  8
1479         &or     ("ecx",$acc);                   #  9
1480         &movz   ($acc,&BP(-128,$tbl,$key,1));   #  2
1481         &or     ("edx","ebx");                  #  8
1482         &shl    ($acc,16);                      #  2
1483         &movz   ("eax",&BP(-128,$tbl,"eax",1)); #  3
1484         &or     ("edx",$acc);                   #  2
1485         &shl    ("eax",24);                     #  3
1486         &or     ("ecx","eax");                  #  3
1487         &mov    ($key,$__key);
1488         &movd   ("mm4","edx");                  # t[2] collected
1489         &movd   ("mm5","ecx");                  # t[3] collected
1490
1491         &punpckldq      ("mm4","mm5");          # t[2,3] collected
1492 }
1493
1494                                         if (!$x86only) {
1495 &function_begin_B("_sse_AES_decrypt_compact");
1496         &pxor   ("mm0",&QWP(0,$key));   #  7, 6, 5, 4, 3, 2, 1, 0
1497         &pxor   ("mm4",&QWP(8,$key));   # 15,14,13,12,11,10, 9, 8
1498
1499         # note that caller is expected to allocate stack frame for me!
1500         &mov    ($acc,&DWP(240,$key));          # load key->rounds
1501         &lea    ($acc,&DWP(-2,$acc,$acc));
1502         &lea    ($acc,&DWP(0,$key,$acc,8));
1503         &mov    ($__end,$acc);                  # end of key schedule
1504
1505         &mov    ($s0,0x1b1b1b1b);               # magic constant
1506         &mov    (&DWP(8,"esp"),$s0);
1507         &mov    (&DWP(12,"esp"),$s0);
1508
1509         # prefetch Td4
1510         &mov    ($s0,&DWP(0-128,$tbl));
1511         &mov    ($s1,&DWP(32-128,$tbl));
1512         &mov    ($s2,&DWP(64-128,$tbl));
1513         &mov    ($s3,&DWP(96-128,$tbl));
1514         &mov    ($s0,&DWP(128-128,$tbl));
1515         &mov    ($s1,&DWP(160-128,$tbl));
1516         &mov    ($s2,&DWP(192-128,$tbl));
1517         &mov    ($s3,&DWP(224-128,$tbl));
1518
1519         &set_label("loop",16);
1520                 &sse_deccompact();
1521                 &add    ($key,16);
1522                 &cmp    ($key,$__end);
1523                 &ja     (&label("out"));
1524
1525                 # ROTATE(x^y,N) == ROTATE(x,N)^ROTATE(y,N)
1526                 &movq   ("mm3","mm0");          &movq   ("mm7","mm4");
1527                 &movq   ("mm2","mm0",1);        &movq   ("mm6","mm4",1);
1528                 &movq   ("mm1","mm0");          &movq   ("mm5","mm4");
1529                 &pshufw ("mm0","mm0",0xb1);     &pshufw ("mm4","mm4",0xb1);# = ROTATE(tp0,16)
1530                 &pslld  ("mm2",8);              &pslld  ("mm6",8);
1531                 &psrld  ("mm3",8);              &psrld  ("mm7",8);
1532                 &pxor   ("mm0","mm2");          &pxor   ("mm4","mm6");  # ^= tp0<<8
1533                 &pxor   ("mm0","mm3");          &pxor   ("mm4","mm7");  # ^= tp0>>8
1534                 &pslld  ("mm2",16);             &pslld  ("mm6",16);
1535                 &psrld  ("mm3",16);             &psrld  ("mm7",16);
1536                 &pxor   ("mm0","mm2");          &pxor   ("mm4","mm6");  # ^= tp0<<24
1537                 &pxor   ("mm0","mm3");          &pxor   ("mm4","mm7");  # ^= tp0>>24
1538
1539                 &movq   ("mm3",&QWP(8,"esp"));
1540                 &pxor   ("mm2","mm2");          &pxor   ("mm6","mm6");
1541                 &pcmpgtb("mm2","mm1");          &pcmpgtb("mm6","mm5");
1542                 &pand   ("mm2","mm3");          &pand   ("mm6","mm3");
1543                 &paddb  ("mm1","mm1");          &paddb  ("mm5","mm5");
1544                 &pxor   ("mm1","mm2");          &pxor   ("mm5","mm6");  # tp2
1545                 &movq   ("mm3","mm1");          &movq   ("mm7","mm5");
1546                 &movq   ("mm2","mm1");          &movq   ("mm6","mm5");
1547                 &pxor   ("mm0","mm1");          &pxor   ("mm4","mm5");  # ^= tp2
1548                 &pslld  ("mm3",24);             &pslld  ("mm7",24);
1549                 &psrld  ("mm2",8);              &psrld  ("mm6",8);
1550                 &pxor   ("mm0","mm3");          &pxor   ("mm4","mm7");  # ^= tp2<<24
1551                 &pxor   ("mm0","mm2");          &pxor   ("mm4","mm6");  # ^= tp2>>8
1552
1553                 &movq   ("mm2",&QWP(8,"esp"));
1554                 &pxor   ("mm3","mm3");          &pxor   ("mm7","mm7");
1555                 &pcmpgtb("mm3","mm1");          &pcmpgtb("mm7","mm5");
1556                 &pand   ("mm3","mm2");          &pand   ("mm7","mm2");
1557                 &paddb  ("mm1","mm1");          &paddb  ("mm5","mm5");
1558                 &pxor   ("mm1","mm3");          &pxor   ("mm5","mm7");  # tp4
1559                 &pshufw ("mm3","mm1",0xb1);     &pshufw ("mm7","mm5",0xb1);
1560                 &pxor   ("mm0","mm1");          &pxor   ("mm4","mm5");  # ^= tp4
1561                 &pxor   ("mm0","mm3");          &pxor   ("mm4","mm7");  # ^= ROTATE(tp4,16)
1562
1563                 &pxor   ("mm3","mm3");          &pxor   ("mm7","mm7");
1564                 &pcmpgtb("mm3","mm1");          &pcmpgtb("mm7","mm5");
1565                 &pand   ("mm3","mm2");          &pand   ("mm7","mm2");
1566                 &paddb  ("mm1","mm1");          &paddb  ("mm5","mm5");
1567                 &pxor   ("mm1","mm3");          &pxor   ("mm5","mm7");  # tp8
1568                 &pxor   ("mm0","mm1");          &pxor   ("mm4","mm5");  # ^= tp8
1569                 &movq   ("mm3","mm1");          &movq   ("mm7","mm5");
1570                 &pshufw ("mm2","mm1",0xb1);     &pshufw ("mm6","mm5",0xb1);
1571                 &pxor   ("mm0","mm2");          &pxor   ("mm4","mm6");  # ^= ROTATE(tp8,16)
1572                 &pslld  ("mm1",8);              &pslld  ("mm5",8);
1573                 &psrld  ("mm3",8);              &psrld  ("mm7",8);
1574                 &movq   ("mm2",&QWP(0,$key));   &movq   ("mm6",&QWP(8,$key));
1575                 &pxor   ("mm0","mm1");          &pxor   ("mm4","mm5");  # ^= tp8<<8
1576                 &pxor   ("mm0","mm3");          &pxor   ("mm4","mm7");  # ^= tp8>>8
1577                 &mov    ($s0,&DWP(0-128,$tbl));
1578                 &pslld  ("mm1",16);             &pslld  ("mm5",16);
1579                 &mov    ($s1,&DWP(64-128,$tbl));
1580                 &psrld  ("mm3",16);             &psrld  ("mm7",16);
1581                 &mov    ($s2,&DWP(128-128,$tbl));
1582                 &pxor   ("mm0","mm1");          &pxor   ("mm4","mm5");  # ^= tp8<<24
1583                 &mov    ($s3,&DWP(192-128,$tbl));
1584                 &pxor   ("mm0","mm3");          &pxor   ("mm4","mm7");  # ^= tp8>>24
1585
1586                 &pxor   ("mm0","mm2");          &pxor   ("mm4","mm6");
1587         &jmp    (&label("loop"));
1588
1589         &set_label("out",16);
1590         &pxor   ("mm0",&QWP(0,$key));
1591         &pxor   ("mm4",&QWP(8,$key));
1592
1593         &ret    ();
1594 &function_end_B("_sse_AES_decrypt_compact");
1595                                         }
1596
1597 ######################################################################
1598 # Vanilla block function.
1599 ######################################################################
1600
1601 sub decstep()
1602 { my ($i,$td,@s) = @_;
1603   my $tmp = $key;
1604   my $out = $i==3?$s[0]:$acc;
1605
1606         # no instructions are reordered, as performance appears
1607         # optimal... or rather that all attempts to reorder didn't
1608         # result in better performance [which by the way is not a
1609         # bit lower than encryption].
1610         if($i==3)   {   &mov    ($key,$__key);                  }
1611         else        {   &mov    ($out,$s[0]);                   }
1612                         &and    ($out,0xFF);
1613                         &mov    ($out,&DWP(0,$td,$out,8));
1614
1615         if ($i==3)  {   $tmp=$s[1];                             }
1616                         &movz   ($tmp,&HB($s[1]));
1617                         &xor    ($out,&DWP(3,$td,$tmp,8));
1618
1619         if ($i==3)  {   $tmp=$s[2]; &mov ($s[1],$acc);          }
1620         else        {   &mov    ($tmp,$s[2]);                   }
1621                         &shr    ($tmp,16);
1622                         &and    ($tmp,0xFF);
1623                         &xor    ($out,&DWP(2,$td,$tmp,8));
1624
1625         if ($i==3)  {   $tmp=$s[3]; &mov ($s[2],$__s1);         }
1626         else        {   &mov    ($tmp,$s[3]);                   }
1627                         &shr    ($tmp,24);
1628                         &xor    ($out,&DWP(1,$td,$tmp,8));
1629         if ($i<2)   {   &mov    (&DWP(4+4*$i,"esp"),$out);      }
1630         if ($i==3)  {   &mov    ($s[3],$__s0);                  }
1631                         &comment();
1632 }
1633
1634 sub declast()
1635 { my ($i,$td,@s)=@_;
1636   my $tmp = $key;
1637   my $out = $i==3?$s[0]:$acc;
1638
1639         if($i==0)   {   &lea    ($td,&DWP(2048+128,$td));
1640                         &mov    ($tmp,&DWP(0-128,$td));
1641                         &mov    ($acc,&DWP(32-128,$td));
1642                         &mov    ($tmp,&DWP(64-128,$td));
1643                         &mov    ($acc,&DWP(96-128,$td));
1644                         &mov    ($tmp,&DWP(128-128,$td));
1645                         &mov    ($acc,&DWP(160-128,$td));
1646                         &mov    ($tmp,&DWP(192-128,$td));
1647                         &mov    ($acc,&DWP(224-128,$td));
1648                         &lea    ($td,&DWP(-128,$td));           }
1649         if($i==3)   {   &mov    ($key,$__key);                  }
1650         else        {   &mov    ($out,$s[0]);                   }
1651                         &and    ($out,0xFF);
1652                         &movz   ($out,&BP(0,$td,$out,1));
1653
1654         if ($i==3)  {   $tmp=$s[1];                             }
1655                         &movz   ($tmp,&HB($s[1]));
1656                         &movz   ($tmp,&BP(0,$td,$tmp,1));
1657                         &shl    ($tmp,8);
1658                         &xor    ($out,$tmp);
1659
1660         if ($i==3)  {   $tmp=$s[2]; &mov ($s[1],$acc);          }
1661         else        {   mov     ($tmp,$s[2]);                   }
1662                         &shr    ($tmp,16);
1663                         &and    ($tmp,0xFF);
1664                         &movz   ($tmp,&BP(0,$td,$tmp,1));
1665                         &shl    ($tmp,16);
1666                         &xor    ($out,$tmp);
1667
1668         if ($i==3)  {   $tmp=$s[3]; &mov ($s[2],$__s1);         }
1669         else        {   &mov    ($tmp,$s[3]);                   }
1670                         &shr    ($tmp,24);
1671                         &movz   ($tmp,&BP(0,$td,$tmp,1));
1672                         &shl    ($tmp,24);
1673                         &xor    ($out,$tmp);
1674         if ($i<2)   {   &mov    (&DWP(4+4*$i,"esp"),$out);      }
1675         if ($i==3)  {   &mov    ($s[3],$__s0);
1676                         &lea    ($td,&DWP(-2048,$td));          }
1677 }
1678
1679 &function_begin_B("_x86_AES_decrypt");
1680         # note that caller is expected to allocate stack frame for me!
1681         &mov    ($__key,$key);                  # save key
1682
1683         &xor    ($s0,&DWP(0,$key));             # xor with key
1684         &xor    ($s1,&DWP(4,$key));
1685         &xor    ($s2,&DWP(8,$key));
1686         &xor    ($s3,&DWP(12,$key));
1687
1688         &mov    ($acc,&DWP(240,$key));          # load key->rounds
1689
1690         if ($small_footprint) {
1691             &lea        ($acc,&DWP(-2,$acc,$acc));
1692             &lea        ($acc,&DWP(0,$key,$acc,8));
1693             &mov        ($__end,$acc);          # end of key schedule
1694             &set_label("loop",16);
1695                 &decstep(0,$tbl,$s0,$s3,$s2,$s1);
1696                 &decstep(1,$tbl,$s1,$s0,$s3,$s2);
1697                 &decstep(2,$tbl,$s2,$s1,$s0,$s3);
1698                 &decstep(3,$tbl,$s3,$s2,$s1,$s0);
1699                 &add    ($key,16);              # advance rd_key
1700                 &xor    ($s0,&DWP(0,$key));
1701                 &xor    ($s1,&DWP(4,$key));
1702                 &xor    ($s2,&DWP(8,$key));
1703                 &xor    ($s3,&DWP(12,$key));
1704             &cmp        ($key,$__end);
1705             &mov        ($__key,$key);
1706             &jb         (&label("loop"));
1707         }
1708         else {
1709             &cmp        ($acc,10);
1710             &jle        (&label("10rounds"));
1711             &cmp        ($acc,12);
1712             &jle        (&label("12rounds"));
1713
1714         &set_label("14rounds",4);
1715             for ($i=1;$i<3;$i++) {
1716                 &decstep(0,$tbl,$s0,$s3,$s2,$s1);
1717                 &decstep(1,$tbl,$s1,$s0,$s3,$s2);
1718                 &decstep(2,$tbl,$s2,$s1,$s0,$s3);
1719                 &decstep(3,$tbl,$s3,$s2,$s1,$s0);
1720                 &xor    ($s0,&DWP(16*$i+0,$key));
1721                 &xor    ($s1,&DWP(16*$i+4,$key));
1722                 &xor    ($s2,&DWP(16*$i+8,$key));
1723                 &xor    ($s3,&DWP(16*$i+12,$key));
1724             }
1725             &add        ($key,32);
1726             &mov        ($__key,$key);          # advance rd_key
1727         &set_label("12rounds",4);
1728             for ($i=1;$i<3;$i++) {
1729                 &decstep(0,$tbl,$s0,$s3,$s2,$s1);
1730                 &decstep(1,$tbl,$s1,$s0,$s3,$s2);
1731                 &decstep(2,$tbl,$s2,$s1,$s0,$s3);
1732                 &decstep(3,$tbl,$s3,$s2,$s1,$s0);
1733                 &xor    ($s0,&DWP(16*$i+0,$key));
1734                 &xor    ($s1,&DWP(16*$i+4,$key));
1735                 &xor    ($s2,&DWP(16*$i+8,$key));
1736                 &xor    ($s3,&DWP(16*$i+12,$key));
1737             }
1738             &add        ($key,32);
1739             &mov        ($__key,$key);          # advance rd_key
1740         &set_label("10rounds",4);
1741             for ($i=1;$i<10;$i++) {
1742                 &decstep(0,$tbl,$s0,$s3,$s2,$s1);
1743                 &decstep(1,$tbl,$s1,$s0,$s3,$s2);
1744                 &decstep(2,$tbl,$s2,$s1,$s0,$s3);
1745                 &decstep(3,$tbl,$s3,$s2,$s1,$s0);
1746                 &xor    ($s0,&DWP(16*$i+0,$key));
1747                 &xor    ($s1,&DWP(16*$i+4,$key));
1748                 &xor    ($s2,&DWP(16*$i+8,$key));
1749                 &xor    ($s3,&DWP(16*$i+12,$key));
1750             }
1751         }
1752
1753         &declast(0,$tbl,$s0,$s3,$s2,$s1);
1754         &declast(1,$tbl,$s1,$s0,$s3,$s2);
1755         &declast(2,$tbl,$s2,$s1,$s0,$s3);
1756         &declast(3,$tbl,$s3,$s2,$s1,$s0);
1757
1758         &add    ($key,$small_footprint?16:160);
1759         &xor    ($s0,&DWP(0,$key));
1760         &xor    ($s1,&DWP(4,$key));
1761         &xor    ($s2,&DWP(8,$key));
1762         &xor    ($s3,&DWP(12,$key));
1763
1764         &ret    ();
1765
1766 &set_label("AES_Td",64);        # Yes! I keep it in the code segment!
1767         &_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a);
1768         &_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b);
1769         &_data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5);
1770         &_data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5);
1771         &_data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d);
1772         &_data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b);
1773         &_data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295);
1774         &_data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e);
1775         &_data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927);
1776         &_data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d);
1777         &_data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362);
1778         &_data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9);
1779         &_data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52);
1780         &_data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566);
1781         &_data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3);
1782         &_data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed);
1783         &_data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e);
1784         &_data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4);
1785         &_data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4);
1786         &_data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd);
1787         &_data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d);
1788         &_data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060);
1789         &_data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967);
1790         &_data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879);
1791         &_data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000);
1792         &_data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c);
1793         &_data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36);
1794         &_data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624);
1795         &_data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b);
1796         &_data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c);
1797         &_data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12);
1798         &_data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14);
1799         &_data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3);
1800         &_data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b);
1801         &_data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8);
1802         &_data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684);
1803         &_data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7);
1804         &_data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177);
1805         &_data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947);
1806         &_data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322);
1807         &_data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498);
1808         &_data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f);
1809         &_data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54);
1810         &_data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382);
1811         &_data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf);
1812         &_data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb);
1813         &_data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83);
1814         &_data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef);
1815         &_data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029);
1816         &_data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235);
1817         &_data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733);
1818         &_data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117);
1819         &_data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4);
1820         &_data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546);
1821         &_data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb);
1822         &_data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d);
1823         &_data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb);
1824         &_data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a);
1825         &_data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773);
1826         &_data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478);
1827         &_data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2);
1828         &_data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff);
1829         &_data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664);
1830         &_data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0);
1831
1832 #Td4:   # four copies of Td4 to choose from to avoid L1 aliasing
1833         &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
1834         &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
1835         &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
1836         &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
1837         &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
1838         &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
1839         &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
1840         &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
1841         &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
1842         &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
1843         &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
1844         &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
1845         &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
1846         &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
1847         &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
1848         &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
1849         &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
1850         &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
1851         &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
1852         &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
1853         &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
1854         &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
1855         &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
1856         &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
1857         &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
1858         &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
1859         &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
1860         &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
1861         &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
1862         &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
1863         &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
1864         &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
1865
1866         &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
1867         &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
1868         &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
1869         &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
1870         &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
1871         &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
1872         &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
1873         &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
1874         &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
1875         &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
1876         &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
1877         &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
1878         &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
1879         &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
1880         &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
1881         &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
1882         &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
1883         &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
1884         &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
1885         &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
1886         &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
1887         &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
1888         &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
1889         &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
1890         &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
1891         &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
1892         &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
1893         &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
1894         &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
1895         &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
1896         &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
1897         &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
1898
1899         &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
1900         &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
1901         &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
1902         &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
1903         &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
1904         &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
1905         &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
1906         &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
1907         &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
1908         &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
1909         &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
1910         &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
1911         &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
1912         &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
1913         &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
1914         &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
1915         &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
1916         &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
1917         &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
1918         &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
1919         &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
1920         &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
1921         &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
1922         &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
1923         &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
1924         &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
1925         &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
1926         &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
1927         &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
1928         &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
1929         &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
1930         &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
1931
1932         &data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
1933         &data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
1934         &data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
1935         &data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
1936         &data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
1937         &data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
1938         &data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
1939         &data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
1940         &data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
1941         &data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
1942         &data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
1943         &data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
1944         &data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
1945         &data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
1946         &data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
1947         &data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
1948         &data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
1949         &data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
1950         &data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
1951         &data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
1952         &data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
1953         &data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
1954         &data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
1955         &data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
1956         &data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
1957         &data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
1958         &data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
1959         &data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
1960         &data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
1961         &data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
1962         &data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
1963         &data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
1964 &function_end_B("_x86_AES_decrypt");
1965
1966 # void AES_decrypt (const void *inp,void *out,const AES_KEY *key);
1967 &function_begin("AES_decrypt");
1968         &mov    ($acc,&wparam(0));              # load inp
1969         &mov    ($key,&wparam(2));              # load key
1970
1971         &mov    ($s0,"esp");
1972         &sub    ("esp",36);
1973         &and    ("esp",-64);                    # align to cache-line
1974
1975         # place stack frame just "above" the key schedule
1976         &lea    ($s1,&DWP(-64-63,$key));
1977         &sub    ($s1,"esp");
1978         &neg    ($s1);
1979         &and    ($s1,0x3C0);    # modulo 1024, but aligned to cache-line
1980         &sub    ("esp",$s1);
1981         &add    ("esp",4);      # 4 is reserved for caller's return address
1982         &mov    ($_esp,$s0);    # save stack pointer
1983
1984         &call   (&label("pic_point"));          # make it PIC!
1985         &set_label("pic_point");
1986         &blindpop($tbl);
1987         &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only);
1988         &lea    ($tbl,&DWP(&label("AES_Td")."-".&label("pic_point"),$tbl));
1989
1990         # pick Td4 copy which can't "overlap" with stack frame or key schedule
1991         &lea    ($s1,&DWP(768-4,"esp"));
1992         &sub    ($s1,$tbl);
1993         &and    ($s1,0x300);
1994         &lea    ($tbl,&DWP(2048+128,$tbl,$s1));
1995
1996                                         if (!$x86only) {
1997         &bt     (&DWP(0,$s0),25);       # check for SSE bit
1998         &jnc    (&label("x86"));
1999
2000         &movq   ("mm0",&QWP(0,$acc));
2001         &movq   ("mm4",&QWP(8,$acc));
2002         &call   ("_sse_AES_decrypt_compact");
2003         &mov    ("esp",$_esp);                  # restore stack pointer
2004         &mov    ($acc,&wparam(1));              # load out
2005         &movq   (&QWP(0,$acc),"mm0");           # write output data
2006         &movq   (&QWP(8,$acc),"mm4");
2007         &emms   ();
2008         &function_end_A();
2009                                         }
2010         &set_label("x86",16);
2011         &mov    ($_tbl,$tbl);
2012         &mov    ($s0,&DWP(0,$acc));             # load input data
2013         &mov    ($s1,&DWP(4,$acc));
2014         &mov    ($s2,&DWP(8,$acc));
2015         &mov    ($s3,&DWP(12,$acc));
2016         &call   ("_x86_AES_decrypt_compact");
2017         &mov    ("esp",$_esp);                  # restore stack pointer
2018         &mov    ($acc,&wparam(1));              # load out
2019         &mov    (&DWP(0,$acc),$s0);             # write output data
2020         &mov    (&DWP(4,$acc),$s1);
2021         &mov    (&DWP(8,$acc),$s2);
2022         &mov    (&DWP(12,$acc),$s3);
2023 &function_end("AES_decrypt");
2024
2025 # void AES_cbc_encrypt (const void char *inp, unsigned char *out,
2026 #                       size_t length, const AES_KEY *key,
2027 #                       unsigned char *ivp,const int enc);
2028 {
2029 # stack frame layout
2030 #             -4(%esp)          # return address         0(%esp)
2031 #              0(%esp)          # s0 backing store       4(%esp)
2032 #              4(%esp)          # s1 backing store       8(%esp)
2033 #              8(%esp)          # s2 backing store      12(%esp)
2034 #             12(%esp)          # s3 backing store      16(%esp)
2035 #             16(%esp)          # key backup            20(%esp)
2036 #             20(%esp)          # end of key schedule   24(%esp)
2037 #             24(%esp)          # %ebp backup           28(%esp)
2038 #             28(%esp)          # %esp backup
2039 my $_inp=&DWP(32,"esp");        # copy of wparam(0)
2040 my $_out=&DWP(36,"esp");        # copy of wparam(1)
2041 my $_len=&DWP(40,"esp");        # copy of wparam(2)
2042 my $_key=&DWP(44,"esp");        # copy of wparam(3)
2043 my $_ivp=&DWP(48,"esp");        # copy of wparam(4)
2044 my $_tmp=&DWP(52,"esp");        # volatile variable
2045 #
2046 my $ivec=&DWP(60,"esp");        # ivec[16]
2047 my $aes_key=&DWP(76,"esp");     # copy of aes_key
2048 my $mark=&DWP(76+240,"esp");    # copy of aes_key->rounds
2049
2050 &function_begin("AES_cbc_encrypt");
2051         &mov    ($s2 eq "ecx"? $s2 : "",&wparam(2));    # load len
2052         &cmp    ($s2,0);
2053         &je     (&label("drop_out"));
2054
2055         &call   (&label("pic_point"));          # make it PIC!
2056         &set_label("pic_point");
2057         &blindpop($tbl);
2058         &picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only);
2059
2060         &cmp    (&wparam(5),0);
2061         &lea    ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
2062         &jne    (&label("picked_te"));
2063         &lea    ($tbl,&DWP(&label("AES_Td")."-".&label("AES_Te"),$tbl));
2064         &set_label("picked_te");
2065
2066         # one can argue if this is required
2067         &pushf  ();
2068         &cld    ();
2069
2070         &cmp    ($s2,$speed_limit);
2071         &jb     (&label("slow_way"));
2072         &test   ($s2,15);
2073         &jnz    (&label("slow_way"));
2074                                         if (!$x86only) {
2075         &bt     (&DWP(0,$s0),28);       # check for hyper-threading bit
2076         &jc     (&label("slow_way"));
2077                                         }
2078         # pre-allocate aligned stack frame...
2079         &lea    ($acc,&DWP(-80-244,"esp"));
2080         &and    ($acc,-64);
2081
2082         # ... and make sure it doesn't alias with $tbl modulo 4096
2083         &mov    ($s0,$tbl);
2084         &lea    ($s1,&DWP(2048+256,$tbl));
2085         &mov    ($s3,$acc);
2086         &and    ($s0,0xfff);            # s = %ebp&0xfff
2087         &and    ($s1,0xfff);            # e = (%ebp+2048+256)&0xfff
2088         &and    ($s3,0xfff);            # p = %esp&0xfff
2089
2090         &cmp    ($s3,$s1);              # if (p>=e) %esp =- (p-e);
2091         &jb     (&label("tbl_break_out"));
2092         &sub    ($s3,$s1);
2093         &sub    ($acc,$s3);
2094         &jmp    (&label("tbl_ok"));
2095         &set_label("tbl_break_out",4);  # else %esp -= (p-s)&0xfff + framesz;
2096         &sub    ($s3,$s0);
2097         &and    ($s3,0xfff);
2098         &add    ($s3,384);
2099         &sub    ($acc,$s3);
2100         &set_label("tbl_ok",4);
2101
2102         &lea    ($s3,&wparam(0));       # obtain pointer to parameter block
2103         &exch   ("esp",$acc);           # allocate stack frame
2104         &add    ("esp",4);              # reserve for return address!
2105         &mov    ($_tbl,$tbl);           # save %ebp
2106         &mov    ($_esp,$acc);           # save %esp
2107
2108         &mov    ($s0,&DWP(0,$s3));      # load inp
2109         &mov    ($s1,&DWP(4,$s3));      # load out
2110         #&mov   ($s2,&DWP(8,$s3));      # load len
2111         &mov    ($key,&DWP(12,$s3));    # load key
2112         &mov    ($acc,&DWP(16,$s3));    # load ivp
2113         &mov    ($s3,&DWP(20,$s3));     # load enc flag
2114
2115         &mov    ($_inp,$s0);            # save copy of inp
2116         &mov    ($_out,$s1);            # save copy of out
2117         &mov    ($_len,$s2);            # save copy of len
2118         &mov    ($_key,$key);           # save copy of key
2119         &mov    ($_ivp,$acc);           # save copy of ivp
2120
2121         &mov    ($mark,0);              # copy of aes_key->rounds = 0;
2122         # do we copy key schedule to stack?
2123         &mov    ($s1 eq "ebx" ? $s1 : "",$key);
2124         &mov    ($s2 eq "ecx" ? $s2 : "",244/4);
2125         &sub    ($s1,$tbl);
2126         &mov    ("esi",$key);
2127         &and    ($s1,0xfff);
2128         &lea    ("edi",$aes_key);
2129         &cmp    ($s1,2048+256);
2130         &jb     (&label("do_copy"));
2131         &cmp    ($s1,4096-244);
2132         &jb     (&label("skip_copy"));
2133         &set_label("do_copy",4);
2134                 &mov    ($_key,"edi");
2135                 &data_word(0xA5F3F689); # rep movsd
2136         &set_label("skip_copy");
2137
2138         &mov    ($key,16);
2139         &set_label("prefetch_tbl",4);
2140                 &mov    ($s0,&DWP(0,$tbl));
2141                 &mov    ($s1,&DWP(32,$tbl));
2142                 &mov    ($s2,&DWP(64,$tbl));
2143                 &mov    ($acc,&DWP(96,$tbl));
2144                 &lea    ($tbl,&DWP(128,$tbl));
2145                 &sub    ($key,1);
2146         &jnz    (&label("prefetch_tbl"));
2147         &sub    ($tbl,2048);
2148
2149         &mov    ($acc,$_inp);
2150         &mov    ($key,$_ivp);
2151
2152         &cmp    ($s3,0);
2153         &je     (&label("fast_decrypt"));
2154
2155 #----------------------------- ENCRYPT -----------------------------#
2156         &mov    ($s0,&DWP(0,$key));             # load iv
2157         &mov    ($s1,&DWP(4,$key));
2158
2159         &set_label("fast_enc_loop",16);
2160                 &mov    ($s2,&DWP(8,$key));
2161                 &mov    ($s3,&DWP(12,$key));
2162
2163                 &xor    ($s0,&DWP(0,$acc));     # xor input data
2164                 &xor    ($s1,&DWP(4,$acc));
2165                 &xor    ($s2,&DWP(8,$acc));
2166                 &xor    ($s3,&DWP(12,$acc));
2167
2168                 &mov    ($key,$_key);           # load key
2169                 &call   ("_x86_AES_encrypt");
2170
2171                 &mov    ($acc,$_inp);           # load inp
2172                 &mov    ($key,$_out);           # load out
2173
2174                 &mov    (&DWP(0,$key),$s0);     # save output data
2175                 &mov    (&DWP(4,$key),$s1);
2176                 &mov    (&DWP(8,$key),$s2);
2177                 &mov    (&DWP(12,$key),$s3);
2178
2179                 &lea    ($acc,&DWP(16,$acc));   # advance inp
2180                 &mov    ($s2,$_len);            # load len
2181                 &mov    ($_inp,$acc);           # save inp
2182                 &lea    ($s3,&DWP(16,$key));    # advance out
2183                 &mov    ($_out,$s3);            # save out
2184                 &sub    ($s2,16);               # decrease len
2185                 &mov    ($_len,$s2);            # save len
2186         &jnz    (&label("fast_enc_loop"));
2187         &mov    ($acc,$_ivp);           # load ivp
2188         &mov    ($s2,&DWP(8,$key));     # restore last 2 dwords
2189         &mov    ($s3,&DWP(12,$key));
2190         &mov    (&DWP(0,$acc),$s0);     # save ivec
2191         &mov    (&DWP(4,$acc),$s1);
2192         &mov    (&DWP(8,$acc),$s2);
2193         &mov    (&DWP(12,$acc),$s3);
2194
2195         &cmp    ($mark,0);              # was the key schedule copied?
2196         &mov    ("edi",$_key);
2197         &je     (&label("skip_ezero"));
2198         # zero copy of key schedule
2199         &mov    ("ecx",240/4);
2200         &xor    ("eax","eax");
2201         &align  (4);
2202         &data_word(0xABF3F689);         # rep stosd
2203         &set_label("skip_ezero");
2204         &mov    ("esp",$_esp);
2205         &popf   ();
2206     &set_label("drop_out");
2207         &function_end_A();
2208         &pushf  ();                     # kludge, never executed
2209
2210 #----------------------------- DECRYPT -----------------------------#
2211 &set_label("fast_decrypt",16);
2212
2213         &cmp    ($acc,$_out);
2214         &je     (&label("fast_dec_in_place"));  # in-place processing...
2215
2216         &mov    ($_tmp,$key);
2217
2218         &align  (4);
2219         &set_label("fast_dec_loop",16);
2220                 &mov    ($s0,&DWP(0,$acc));     # read input
2221                 &mov    ($s1,&DWP(4,$acc));
2222                 &mov    ($s2,&DWP(8,$acc));
2223                 &mov    ($s3,&DWP(12,$acc));
2224
2225                 &mov    ($key,$_key);           # load key
2226                 &call   ("_x86_AES_decrypt");
2227
2228                 &mov    ($key,$_tmp);           # load ivp
2229                 &mov    ($acc,$_len);           # load len
2230                 &xor    ($s0,&DWP(0,$key));     # xor iv
2231                 &xor    ($s1,&DWP(4,$key));
2232                 &xor    ($s2,&DWP(8,$key));
2233                 &xor    ($s3,&DWP(12,$key));
2234
2235                 &mov    ($key,$_out);           # load out
2236                 &mov    ($acc,$_inp);           # load inp
2237
2238                 &mov    (&DWP(0,$key),$s0);     # write output
2239                 &mov    (&DWP(4,$key),$s1);
2240                 &mov    (&DWP(8,$key),$s2);
2241                 &mov    (&DWP(12,$key),$s3);
2242
2243                 &mov    ($s2,$_len);            # load len
2244                 &mov    ($_tmp,$acc);           # save ivp
2245                 &lea    ($acc,&DWP(16,$acc));   # advance inp
2246                 &mov    ($_inp,$acc);           # save inp
2247                 &lea    ($key,&DWP(16,$key));   # advance out
2248                 &mov    ($_out,$key);           # save out
2249                 &sub    ($s2,16);               # decrease len
2250                 &mov    ($_len,$s2);            # save len
2251         &jnz    (&label("fast_dec_loop"));
2252         &mov    ($key,$_tmp);           # load temp ivp
2253         &mov    ($acc,$_ivp);           # load user ivp
2254         &mov    ($s0,&DWP(0,$key));     # load iv
2255         &mov    ($s1,&DWP(4,$key));
2256         &mov    ($s2,&DWP(8,$key));
2257         &mov    ($s3,&DWP(12,$key));
2258         &mov    (&DWP(0,$acc),$s0);     # copy back to user
2259         &mov    (&DWP(4,$acc),$s1);
2260         &mov    (&DWP(8,$acc),$s2);
2261         &mov    (&DWP(12,$acc),$s3);
2262         &jmp    (&label("fast_dec_out"));
2263
2264     &set_label("fast_dec_in_place",16);
2265         &set_label("fast_dec_in_place_loop");
2266                 &mov    ($s0,&DWP(0,$acc));     # read input
2267                 &mov    ($s1,&DWP(4,$acc));
2268                 &mov    ($s2,&DWP(8,$acc));
2269                 &mov    ($s3,&DWP(12,$acc));
2270
2271                 &lea    ($key,$ivec);
2272                 &mov    (&DWP(0,$key),$s0);     # copy to temp
2273                 &mov    (&DWP(4,$key),$s1);
2274                 &mov    (&DWP(8,$key),$s2);
2275                 &mov    (&DWP(12,$key),$s3);
2276
2277                 &mov    ($key,$_key);           # load key
2278                 &call   ("_x86_AES_decrypt");
2279
2280                 &mov    ($key,$_ivp);           # load ivp
2281                 &mov    ($acc,$_out);           # load out
2282                 &xor    ($s0,&DWP(0,$key));     # xor iv
2283                 &xor    ($s1,&DWP(4,$key));
2284                 &xor    ($s2,&DWP(8,$key));
2285                 &xor    ($s3,&DWP(12,$key));
2286
2287                 &mov    (&DWP(0,$acc),$s0);     # write output
2288                 &mov    (&DWP(4,$acc),$s1);
2289                 &mov    (&DWP(8,$acc),$s2);
2290                 &mov    (&DWP(12,$acc),$s3);
2291
2292                 &lea    ($acc,&DWP(16,$acc));   # advance out
2293                 &mov    ($_out,$acc);           # save out
2294
2295                 &lea    ($acc,$ivec);
2296                 &mov    ($s0,&DWP(0,$acc));     # read temp
2297                 &mov    ($s1,&DWP(4,$acc));
2298                 &mov    ($s2,&DWP(8,$acc));
2299                 &mov    ($s3,&DWP(12,$acc));
2300
2301                 &mov    (&DWP(0,$key),$s0);     # copy iv
2302                 &mov    (&DWP(4,$key),$s1);
2303                 &mov    (&DWP(8,$key),$s2);
2304                 &mov    (&DWP(12,$key),$s3);
2305
2306                 &mov    ($acc,$_inp);           # load inp
2307                 &mov    ($s2,$_len);            # load len
2308                 &lea    ($acc,&DWP(16,$acc));   # advance inp
2309                 &mov    ($_inp,$acc);           # save inp
2310                 &sub    ($s2,16);               # decrease len
2311                 &mov    ($_len,$s2);            # save len
2312         &jnz    (&label("fast_dec_in_place_loop"));
2313
2314     &set_label("fast_dec_out",4);
2315         &cmp    ($mark,0);              # was the key schedule copied?
2316         &mov    ("edi",$_key);
2317         &je     (&label("skip_dzero"));
2318         # zero copy of key schedule
2319         &mov    ("ecx",240/4);
2320         &xor    ("eax","eax");
2321         &align  (4);
2322         &data_word(0xABF3F689);         # rep stosd
2323         &set_label("skip_dzero");
2324         &mov    ("esp",$_esp);
2325         &popf   ();
2326         &function_end_A();
2327         &pushf  ();                     # kludge, never executed
2328
2329 #--------------------------- SLOW ROUTINE ---------------------------#
2330 &set_label("slow_way",16);
2331
2332         &mov    ($s0,&DWP(0,$s0)) if (!$x86only);# load OPENSSL_ia32cap
2333         &mov    ($key,&wparam(3));      # load key
2334
2335         # pre-allocate aligned stack frame...
2336         &lea    ($acc,&DWP(-80,"esp"));
2337         &and    ($acc,-64);
2338
2339         # ... and make sure it doesn't alias with $key modulo 1024
2340         &lea    ($s1,&DWP(-80-63,$key));
2341         &sub    ($s1,$acc);
2342         &neg    ($s1);
2343         &and    ($s1,0x3C0);    # modulo 1024, but aligned to cache-line
2344         &sub    ($acc,$s1);
2345
2346         # pick S-box copy which can't overlap with stack frame or $key
2347         &lea    ($s1,&DWP(768,$acc));
2348         &sub    ($s1,$tbl);
2349         &and    ($s1,0x300);
2350         &lea    ($tbl,&DWP(2048+128,$tbl,$s1));
2351
2352         &lea    ($s3,&wparam(0));       # pointer to parameter block
2353
2354         &exch   ("esp",$acc);
2355         &add    ("esp",4);              # reserve for return address!
2356         &mov    ($_tbl,$tbl);           # save %ebp
2357         &mov    ($_esp,$acc);           # save %esp
2358         &mov    ($_tmp,$s0);            # save OPENSSL_ia32cap
2359
2360         &mov    ($s0,&DWP(0,$s3));      # load inp
2361         &mov    ($s1,&DWP(4,$s3));      # load out
2362         #&mov   ($s2,&DWP(8,$s3));      # load len
2363         #&mov   ($key,&DWP(12,$s3));    # load key
2364         &mov    ($acc,&DWP(16,$s3));    # load ivp
2365         &mov    ($s3,&DWP(20,$s3));     # load enc flag
2366
2367         &mov    ($_inp,$s0);            # save copy of inp
2368         &mov    ($_out,$s1);            # save copy of out
2369         &mov    ($_len,$s2);            # save copy of len
2370         &mov    ($_key,$key);           # save copy of key
2371         &mov    ($_ivp,$acc);           # save copy of ivp
2372
2373         &mov    ($key,$acc);
2374         &mov    ($acc,$s0);
2375
2376         &cmp    ($s3,0);
2377         &je     (&label("slow_decrypt"));
2378
2379 #--------------------------- SLOW ENCRYPT ---------------------------#
2380         &cmp    ($s2,16);
2381         &mov    ($s3,$s1);
2382         &jb     (&label("slow_enc_tail"));
2383
2384                                         if (!$x86only) {
2385         &bt     ($_tmp,25);             # check for SSE bit
2386         &jnc    (&label("slow_enc_x86"));
2387
2388         &movq   ("mm0",&QWP(0,$key));   # load iv
2389         &movq   ("mm4",&QWP(8,$key));
2390
2391         &set_label("slow_enc_loop_sse",16);
2392                 &pxor   ("mm0",&QWP(0,$acc));   # xor input data
2393                 &pxor   ("mm4",&QWP(8,$acc));
2394
2395                 &mov    ($key,$_key);
2396                 &call   ("_sse_AES_encrypt_compact");
2397
2398                 &mov    ($acc,$_inp);           # load inp
2399                 &mov    ($key,$_out);           # load out
2400                 &mov    ($s2,$_len);            # load len
2401
2402                 &movq   (&QWP(0,$key),"mm0");   # save output data
2403                 &movq   (&QWP(8,$key),"mm4");
2404
2405                 &lea    ($acc,&DWP(16,$acc));   # advance inp
2406                 &mov    ($_inp,$acc);           # save inp
2407                 &lea    ($s3,&DWP(16,$key));    # advance out
2408                 &mov    ($_out,$s3);            # save out
2409                 &sub    ($s2,16);               # decrease len
2410                 &cmp    ($s2,16);
2411                 &mov    ($_len,$s2);            # save len
2412         &jae    (&label("slow_enc_loop_sse"));
2413         &test   ($s2,15);
2414         &jnz    (&label("slow_enc_tail"));
2415         &mov    ($acc,$_ivp);           # load ivp
2416         &movq   (&QWP(0,$acc),"mm0");   # save ivec
2417         &movq   (&QWP(8,$acc),"mm4");
2418         &emms   ();
2419         &mov    ("esp",$_esp);
2420         &popf   ();
2421         &function_end_A();
2422         &pushf  ();                     # kludge, never executed
2423                                         }
2424     &set_label("slow_enc_x86",16);
2425         &mov    ($s0,&DWP(0,$key));     # load iv
2426         &mov    ($s1,&DWP(4,$key));
2427
2428         &set_label("slow_enc_loop_x86",4);
2429                 &mov    ($s2,&DWP(8,$key));
2430                 &mov    ($s3,&DWP(12,$key));
2431
2432                 &xor    ($s0,&DWP(0,$acc));     # xor input data
2433                 &xor    ($s1,&DWP(4,$acc));
2434                 &xor    ($s2,&DWP(8,$acc));
2435                 &xor    ($s3,&DWP(12,$acc));
2436
2437                 &mov    ($key,$_key);           # load key
2438                 &call   ("_x86_AES_encrypt_compact");
2439
2440                 &mov    ($acc,$_inp);           # load inp
2441                 &mov    ($key,$_out);           # load out
2442
2443                 &mov    (&DWP(0,$key),$s0);     # save output data
2444                 &mov    (&DWP(4,$key),$s1);
2445                 &mov    (&DWP(8,$key),$s2);
2446                 &mov    (&DWP(12,$key),$s3);
2447
2448                 &mov    ($s2,$_len);            # load len
2449                 &lea    ($acc,&DWP(16,$acc));   # advance inp
2450                 &mov    ($_inp,$acc);           # save inp
2451                 &lea    ($s3,&DWP(16,$key));    # advance out
2452                 &mov    ($_out,$s3);            # save out
2453                 &sub    ($s2,16);               # decrease len
2454                 &cmp    ($s2,16);
2455                 &mov    ($_len,$s2);            # save len
2456         &jae    (&label("slow_enc_loop_x86"));
2457         &test   ($s2,15);
2458         &jnz    (&label("slow_enc_tail"));
2459         &mov    ($acc,$_ivp);           # load ivp
2460         &mov    ($s2,&DWP(8,$key));     # restore last dwords
2461         &mov    ($s3,&DWP(12,$key));
2462         &mov    (&DWP(0,$acc),$s0);     # save ivec
2463         &mov    (&DWP(4,$acc),$s1);
2464         &mov    (&DWP(8,$acc),$s2);
2465         &mov    (&DWP(12,$acc),$s3);
2466
2467         &mov    ("esp",$_esp);
2468         &popf   ();
2469         &function_end_A();
2470         &pushf  ();                     # kludge, never executed
2471
2472     &set_label("slow_enc_tail",16);
2473         &emms   ()      if (!$x86only);
2474         &mov    ($key eq "edi"? $key:"",$s3);   # load out to edi
2475         &mov    ($s1,16);
2476         &sub    ($s1,$s2);
2477         &cmp    ($key,$acc eq "esi"? $acc:"");  # compare with inp
2478         &je     (&label("enc_in_place"));
2479         &align  (4);
2480         &data_word(0xA4F3F689); # rep movsb     # copy input
2481         &jmp    (&label("enc_skip_in_place"));
2482     &set_label("enc_in_place");
2483         &lea    ($key,&DWP(0,$key,$s2));
2484     &set_label("enc_skip_in_place");
2485         &mov    ($s2,$s1);
2486         &xor    ($s0,$s0);
2487         &align  (4);
2488         &data_word(0xAAF3F689); # rep stosb     # zero tail
2489
2490         &mov    ($key,$_ivp);                   # restore ivp
2491         &mov    ($acc,$s3);                     # output as input
2492         &mov    ($s0,&DWP(0,$key));
2493         &mov    ($s1,&DWP(4,$key));
2494         &mov    ($_len,16);                     # len=16
2495         &jmp    (&label("slow_enc_loop_x86"));  # one more spin...
2496
2497 #--------------------------- SLOW DECRYPT ---------------------------#
2498 &set_label("slow_decrypt",16);
2499                                         if (!$x86only) {
2500         &bt     ($_tmp,25);             # check for SSE bit
2501         &jnc    (&label("slow_dec_loop_x86"));
2502
2503         &set_label("slow_dec_loop_sse",4);
2504                 &movq   ("mm0",&QWP(0,$acc));   # read input
2505                 &movq   ("mm4",&QWP(8,$acc));
2506
2507                 &mov    ($key,$_key);
2508                 &call   ("_sse_AES_decrypt_compact");
2509
2510                 &mov    ($acc,$_inp);           # load inp
2511                 &lea    ($s0,$ivec);
2512                 &mov    ($s1,$_out);            # load out
2513                 &mov    ($s2,$_len);            # load len
2514                 &mov    ($key,$_ivp);           # load ivp
2515
2516                 &movq   ("mm1",&QWP(0,$acc));   # re-read input
2517                 &movq   ("mm5",&QWP(8,$acc));
2518
2519                 &pxor   ("mm0",&QWP(0,$key));   # xor iv
2520                 &pxor   ("mm4",&QWP(8,$key));
2521
2522                 &movq   (&QWP(0,$key),"mm1");   # copy input to iv
2523                 &movq   (&QWP(8,$key),"mm5");
2524
2525                 &sub    ($s2,16);               # decrease len
2526                 &jc     (&label("slow_dec_partial_sse"));
2527
2528                 &movq   (&QWP(0,$s1),"mm0");    # write output
2529                 &movq   (&QWP(8,$s1),"mm4");
2530
2531                 &lea    ($s1,&DWP(16,$s1));     # advance out
2532                 &mov    ($_out,$s1);            # save out
2533                 &lea    ($acc,&DWP(16,$acc));   # advance inp
2534                 &mov    ($_inp,$acc);           # save inp
2535                 &mov    ($_len,$s2);            # save len
2536         &jnz    (&label("slow_dec_loop_sse"));
2537         &emms   ();
2538         &mov    ("esp",$_esp);
2539         &popf   ();
2540         &function_end_A();
2541         &pushf  ();                     # kludge, never executed
2542
2543     &set_label("slow_dec_partial_sse",16);
2544         &movq   (&QWP(0,$s0),"mm0");    # save output to temp
2545         &movq   (&QWP(8,$s0),"mm4");
2546         &emms   ();
2547
2548         &add    ($s2 eq "ecx" ? "ecx":"",16);
2549         &mov    ("edi",$s1);            # out
2550         &mov    ("esi",$s0);            # temp
2551         &align  (4);
2552         &data_word(0xA4F3F689);         # rep movsb # copy partial output
2553
2554         &mov    ("esp",$_esp);
2555         &popf   ();
2556         &function_end_A();
2557         &pushf  ();                     # kludge, never executed
2558                                         }
2559         &set_label("slow_dec_loop_x86",16);
2560                 &mov    ($s0,&DWP(0,$acc));     # read input
2561                 &mov    ($s1,&DWP(4,$acc));
2562                 &mov    ($s2,&DWP(8,$acc));
2563                 &mov    ($s3,&DWP(12,$acc));
2564
2565                 &lea    ($key,$ivec);
2566                 &mov    (&DWP(0,$key),$s0);     # copy to temp
2567                 &mov    (&DWP(4,$key),$s1);
2568                 &mov    (&DWP(8,$key),$s2);
2569                 &mov    (&DWP(12,$key),$s3);
2570
2571                 &mov    ($key,$_key);           # load key
2572                 &call   ("_x86_AES_decrypt_compact");
2573
2574                 &mov    ($key,$_ivp);           # load ivp
2575                 &mov    ($acc,$_len);           # load len
2576                 &xor    ($s0,&DWP(0,$key));     # xor iv
2577                 &xor    ($s1,&DWP(4,$key));
2578                 &xor    ($s2,&DWP(8,$key));
2579                 &xor    ($s3,&DWP(12,$key));
2580
2581                 &sub    ($acc,16);
2582                 &jc     (&label("slow_dec_partial_x86"));
2583
2584                 &mov    ($_len,$acc);           # save len
2585                 &mov    ($acc,$_out);           # load out
2586
2587                 &mov    (&DWP(0,$acc),$s0);     # write output
2588                 &mov    (&DWP(4,$acc),$s1);
2589                 &mov    (&DWP(8,$acc),$s2);
2590                 &mov    (&DWP(12,$acc),$s3);
2591
2592                 &lea    ($acc,&DWP(16,$acc));   # advance out
2593                 &mov    ($_out,$acc);           # save out
2594
2595                 &lea    ($acc,$ivec);
2596                 &mov    ($s0,&DWP(0,$acc));     # read temp
2597                 &mov    ($s1,&DWP(4,$acc));
2598                 &mov    ($s2,&DWP(8,$acc));
2599                 &mov    ($s3,&DWP(12,$acc));
2600
2601                 &mov    (&DWP(0,$key),$s0);     # copy it to iv
2602                 &mov    (&DWP(4,$key),$s1);
2603                 &mov    (&DWP(8,$key),$s2);
2604                 &mov    (&DWP(12,$key),$s3);
2605
2606                 &mov    ($acc,$_inp);           # load inp
2607                 &lea    ($acc,&DWP(16,$acc));   # advance inp
2608                 &mov    ($_inp,$acc);           # save inp
2609         &jnz    (&label("slow_dec_loop_x86"));
2610         &mov    ("esp",$_esp);
2611         &popf   ();
2612         &function_end_A();
2613         &pushf  ();                     # kludge, never executed
2614
2615     &set_label("slow_dec_partial_x86",16);
2616         &lea    ($acc,$ivec);
2617         &mov    (&DWP(0,$acc),$s0);     # save output to temp
2618         &mov    (&DWP(4,$acc),$s1);
2619         &mov    (&DWP(8,$acc),$s2);
2620         &mov    (&DWP(12,$acc),$s3);
2621
2622         &mov    ($acc,$_inp);
2623         &mov    ($s0,&DWP(0,$acc));     # re-read input
2624         &mov    ($s1,&DWP(4,$acc));
2625         &mov    ($s2,&DWP(8,$acc));
2626         &mov    ($s3,&DWP(12,$acc));
2627
2628         &mov    (&DWP(0,$key),$s0);     # copy it to iv
2629         &mov    (&DWP(4,$key),$s1);
2630         &mov    (&DWP(8,$key),$s2);
2631         &mov    (&DWP(12,$key),$s3);
2632
2633         &mov    ("ecx",$_len);
2634         &mov    ("edi",$_out);
2635         &lea    ("esi",$ivec);
2636         &align  (4);
2637         &data_word(0xA4F3F689);         # rep movsb # copy partial output
2638
2639         &mov    ("esp",$_esp);
2640         &popf   ();
2641 &function_end("AES_cbc_encrypt");
2642 }
2643
2644 #------------------------------------------------------------------#
2645
2646 sub enckey()
2647 {
2648         &movz   ("esi",&LB("edx"));             # rk[i]>>0
2649         &movz   ("ebx",&BP(-128,$tbl,"esi",1));
2650         &movz   ("esi",&HB("edx"));             # rk[i]>>8
2651         &shl    ("ebx",24);
2652         &xor    ("eax","ebx");
2653
2654         &movz   ("ebx",&BP(-128,$tbl,"esi",1));
2655         &shr    ("edx",16);
2656         &movz   ("esi",&LB("edx"));             # rk[i]>>16
2657         &xor    ("eax","ebx");
2658
2659         &movz   ("ebx",&BP(-128,$tbl,"esi",1));
2660         &movz   ("esi",&HB("edx"));             # rk[i]>>24
2661         &shl    ("ebx",8);
2662         &xor    ("eax","ebx");
2663
2664         &movz   ("ebx",&BP(-128,$tbl,"esi",1));
2665         &shl    ("ebx",16);
2666         &xor    ("eax","ebx");
2667
2668         &xor    ("eax",&DWP(1024-128,$tbl,"ecx",4));    # rcon
2669 }
2670
2671 &function_begin("_x86_AES_set_encrypt_key");
2672         &mov    ("esi",&wparam(1));             # user supplied key
2673         &mov    ("edi",&wparam(3));             # private key schedule
2674
2675         &test   ("esi",-1);
2676         &jz     (&label("badpointer"));
2677         &test   ("edi",-1);
2678         &jz     (&label("badpointer"));
2679
2680         &call   (&label("pic_point"));
2681         &set_label("pic_point");
2682         &blindpop($tbl);
2683         &lea    ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
2684         &lea    ($tbl,&DWP(2048+128,$tbl));
2685
2686         # prefetch Te4
2687         &mov    ("eax",&DWP(0-128,$tbl));
2688         &mov    ("ebx",&DWP(32-128,$tbl));
2689         &mov    ("ecx",&DWP(64-128,$tbl));
2690         &mov    ("edx",&DWP(96-128,$tbl));
2691         &mov    ("eax",&DWP(128-128,$tbl));
2692         &mov    ("ebx",&DWP(160-128,$tbl));
2693         &mov    ("ecx",&DWP(192-128,$tbl));
2694         &mov    ("edx",&DWP(224-128,$tbl));
2695
2696         &mov    ("ecx",&wparam(2));             # number of bits in key
2697         &cmp    ("ecx",128);
2698         &je     (&label("10rounds"));
2699         &cmp    ("ecx",192);
2700         &je     (&label("12rounds"));
2701         &cmp    ("ecx",256);
2702         &je     (&label("14rounds"));
2703         &mov    ("eax",-2);                     # invalid number of bits
2704         &jmp    (&label("exit"));
2705
2706     &set_label("10rounds");
2707         &mov    ("eax",&DWP(0,"esi"));          # copy first 4 dwords
2708         &mov    ("ebx",&DWP(4,"esi"));
2709         &mov    ("ecx",&DWP(8,"esi"));
2710         &mov    ("edx",&DWP(12,"esi"));
2711         &mov    (&DWP(0,"edi"),"eax");
2712         &mov    (&DWP(4,"edi"),"ebx");
2713         &mov    (&DWP(8,"edi"),"ecx");
2714         &mov    (&DWP(12,"edi"),"edx");
2715
2716         &xor    ("ecx","ecx");
2717         &jmp    (&label("10shortcut"));
2718
2719         &align  (4);
2720         &set_label("10loop");
2721                 &mov    ("eax",&DWP(0,"edi"));          # rk[0]
2722                 &mov    ("edx",&DWP(12,"edi"));         # rk[3]
2723         &set_label("10shortcut");
2724                 &enckey ();
2725
2726                 &mov    (&DWP(16,"edi"),"eax");         # rk[4]
2727                 &xor    ("eax",&DWP(4,"edi"));
2728                 &mov    (&DWP(20,"edi"),"eax");         # rk[5]
2729                 &xor    ("eax",&DWP(8,"edi"));
2730                 &mov    (&DWP(24,"edi"),"eax");         # rk[6]
2731                 &xor    ("eax",&DWP(12,"edi"));
2732                 &mov    (&DWP(28,"edi"),"eax");         # rk[7]
2733                 &inc    ("ecx");
2734                 &add    ("edi",16);
2735                 &cmp    ("ecx",10);
2736         &jl     (&label("10loop"));
2737
2738         &mov    (&DWP(80,"edi"),10);            # setup number of rounds
2739         &xor    ("eax","eax");
2740         &jmp    (&label("exit"));
2741
2742     &set_label("12rounds");
2743         &mov    ("eax",&DWP(0,"esi"));          # copy first 6 dwords
2744         &mov    ("ebx",&DWP(4,"esi"));
2745         &mov    ("ecx",&DWP(8,"esi"));
2746         &mov    ("edx",&DWP(12,"esi"));
2747         &mov    (&DWP(0,"edi"),"eax");
2748         &mov    (&DWP(4,"edi"),"ebx");
2749         &mov    (&DWP(8,"edi"),"ecx");
2750         &mov    (&DWP(12,"edi"),"edx");
2751         &mov    ("ecx",&DWP(16,"esi"));
2752         &mov    ("edx",&DWP(20,"esi"));
2753         &mov    (&DWP(16,"edi"),"ecx");
2754         &mov    (&DWP(20,"edi"),"edx");
2755
2756         &xor    ("ecx","ecx");
2757         &jmp    (&label("12shortcut"));
2758
2759         &align  (4);
2760         &set_label("12loop");
2761                 &mov    ("eax",&DWP(0,"edi"));          # rk[0]
2762                 &mov    ("edx",&DWP(20,"edi"));         # rk[5]
2763         &set_label("12shortcut");
2764                 &enckey ();
2765
2766                 &mov    (&DWP(24,"edi"),"eax");         # rk[6]
2767                 &xor    ("eax",&DWP(4,"edi"));
2768                 &mov    (&DWP(28,"edi"),"eax");         # rk[7]
2769                 &xor    ("eax",&DWP(8,"edi"));
2770                 &mov    (&DWP(32,"edi"),"eax");         # rk[8]
2771                 &xor    ("eax",&DWP(12,"edi"));
2772                 &mov    (&DWP(36,"edi"),"eax");         # rk[9]
2773
2774                 &cmp    ("ecx",7);
2775                 &je     (&label("12break"));
2776                 &inc    ("ecx");
2777
2778                 &xor    ("eax",&DWP(16,"edi"));
2779                 &mov    (&DWP(40,"edi"),"eax");         # rk[10]
2780                 &xor    ("eax",&DWP(20,"edi"));
2781                 &mov    (&DWP(44,"edi"),"eax");         # rk[11]
2782
2783                 &add    ("edi",24);
2784         &jmp    (&label("12loop"));
2785
2786         &set_label("12break");
2787         &mov    (&DWP(72,"edi"),12);            # setup number of rounds
2788         &xor    ("eax","eax");
2789         &jmp    (&label("exit"));
2790
2791     &set_label("14rounds");
2792         &mov    ("eax",&DWP(0,"esi"));          # copy first 8 dwords
2793         &mov    ("ebx",&DWP(4,"esi"));
2794         &mov    ("ecx",&DWP(8,"esi"));
2795         &mov    ("edx",&DWP(12,"esi"));
2796         &mov    (&DWP(0,"edi"),"eax");
2797         &mov    (&DWP(4,"edi"),"ebx");
2798         &mov    (&DWP(8,"edi"),"ecx");
2799         &mov    (&DWP(12,"edi"),"edx");
2800         &mov    ("eax",&DWP(16,"esi"));
2801         &mov    ("ebx",&DWP(20,"esi"));
2802         &mov    ("ecx",&DWP(24,"esi"));
2803         &mov    ("edx",&DWP(28,"esi"));
2804         &mov    (&DWP(16,"edi"),"eax");
2805         &mov    (&DWP(20,"edi"),"ebx");
2806         &mov    (&DWP(24,"edi"),"ecx");
2807         &mov    (&DWP(28,"edi"),"edx");
2808
2809         &xor    ("ecx","ecx");
2810         &jmp    (&label("14shortcut"));
2811
2812         &align  (4);
2813         &set_label("14loop");
2814                 &mov    ("edx",&DWP(28,"edi"));         # rk[7]
2815         &set_label("14shortcut");
2816                 &mov    ("eax",&DWP(0,"edi"));          # rk[0]
2817
2818                 &enckey ();
2819
2820                 &mov    (&DWP(32,"edi"),"eax");         # rk[8]
2821                 &xor    ("eax",&DWP(4,"edi"));
2822                 &mov    (&DWP(36,"edi"),"eax");         # rk[9]
2823                 &xor    ("eax",&DWP(8,"edi"));
2824                 &mov    (&DWP(40,"edi"),"eax");         # rk[10]
2825                 &xor    ("eax",&DWP(12,"edi"));
2826                 &mov    (&DWP(44,"edi"),"eax");         # rk[11]
2827
2828                 &cmp    ("ecx",6);
2829                 &je     (&label("14break"));
2830                 &inc    ("ecx");
2831
2832                 &mov    ("edx","eax");
2833                 &mov    ("eax",&DWP(16,"edi"));         # rk[4]
2834                 &movz   ("esi",&LB("edx"));             # rk[11]>>0
2835                 &movz   ("ebx",&BP(-128,$tbl,"esi",1));
2836                 &movz   ("esi",&HB("edx"));             # rk[11]>>8
2837                 &xor    ("eax","ebx");
2838
2839                 &movz   ("ebx",&BP(-128,$tbl,"esi",1));
2840                 &shr    ("edx",16);
2841                 &shl    ("ebx",8);
2842                 &movz   ("esi",&LB("edx"));             # rk[11]>>16
2843                 &xor    ("eax","ebx");
2844
2845                 &movz   ("ebx",&BP(-128,$tbl,"esi",1));
2846                 &movz   ("esi",&HB("edx"));             # rk[11]>>24
2847                 &shl    ("ebx",16);
2848                 &xor    ("eax","ebx");
2849
2850                 &movz   ("ebx",&BP(-128,$tbl,"esi",1));
2851                 &shl    ("ebx",24);
2852                 &xor    ("eax","ebx");
2853
2854                 &mov    (&DWP(48,"edi"),"eax");         # rk[12]
2855                 &xor    ("eax",&DWP(20,"edi"));
2856                 &mov    (&DWP(52,"edi"),"eax");         # rk[13]
2857                 &xor    ("eax",&DWP(24,"edi"));
2858                 &mov    (&DWP(56,"edi"),"eax");         # rk[14]
2859                 &xor    ("eax",&DWP(28,"edi"));
2860                 &mov    (&DWP(60,"edi"),"eax");         # rk[15]
2861
2862                 &add    ("edi",32);
2863         &jmp    (&label("14loop"));
2864
2865         &set_label("14break");
2866         &mov    (&DWP(48,"edi"),14);            # setup number of rounds
2867         &xor    ("eax","eax");
2868         &jmp    (&label("exit"));
2869
2870     &set_label("badpointer");
2871         &mov    ("eax",-1);
2872     &set_label("exit");
2873 &function_end("_x86_AES_set_encrypt_key");
2874
2875 # int AES_set_encrypt_key(const unsigned char *userKey, const int bits,
2876 #                        AES_KEY *key)
2877 &function_begin_B("AES_set_encrypt_key");
2878         &call   ("_x86_AES_set_encrypt_key");
2879         &ret    ();
2880 &function_end_B("AES_set_encrypt_key");
2881
2882 sub deckey()
2883 { my ($i,$key,$tp1,$tp2,$tp4,$tp8) = @_;
2884   my $tmp = $tbl;
2885
2886         &mov    ($tmp,0x80808080);
2887         &and    ($tmp,$tp1);
2888         &lea    ($tp2,&DWP(0,$tp1,$tp1));
2889         &mov    ($acc,$tmp);
2890         &shr    ($tmp,7);
2891         &sub    ($acc,$tmp);
2892         &and    ($tp2,0xfefefefe);
2893         &and    ($acc,0x1b1b1b1b);
2894         &xor    ($tp2,$acc);
2895         &mov    ($tmp,0x80808080);
2896
2897         &and    ($tmp,$tp2);
2898         &lea    ($tp4,&DWP(0,$tp2,$tp2));
2899         &mov    ($acc,$tmp);
2900         &shr    ($tmp,7);
2901         &sub    ($acc,$tmp);
2902         &and    ($tp4,0xfefefefe);
2903         &and    ($acc,0x1b1b1b1b);
2904          &xor   ($tp2,$tp1);    # tp2^tp1
2905         &xor    ($tp4,$acc);
2906         &mov    ($tmp,0x80808080);
2907
2908         &and    ($tmp,$tp4);
2909         &lea    ($tp8,&DWP(0,$tp4,$tp4));
2910         &mov    ($acc,$tmp);
2911         &shr    ($tmp,7);
2912          &xor   ($tp4,$tp1);    # tp4^tp1
2913         &sub    ($acc,$tmp);
2914         &and    ($tp8,0xfefefefe);
2915         &and    ($acc,0x1b1b1b1b);
2916          &rotl  ($tp1,8);       # = ROTATE(tp1,8)
2917         &xor    ($tp8,$acc);
2918
2919         &mov    ($tmp,&DWP(4*($i+1),$key));     # modulo-scheduled load
2920
2921         &xor    ($tp1,$tp2);
2922         &xor    ($tp2,$tp8);
2923         &xor    ($tp1,$tp4);
2924         &rotl   ($tp2,24);
2925         &xor    ($tp4,$tp8);
2926         &xor    ($tp1,$tp8);    # ^= tp8^(tp4^tp1)^(tp2^tp1)
2927         &rotl   ($tp4,16);
2928         &xor    ($tp1,$tp2);    # ^= ROTATE(tp8^tp2^tp1,24)
2929         &rotl   ($tp8,8);
2930         &xor    ($tp1,$tp4);    # ^= ROTATE(tp8^tp4^tp1,16)
2931         &mov    ($tp2,$tmp);
2932         &xor    ($tp1,$tp8);    # ^= ROTATE(tp8,8)
2933
2934         &mov    (&DWP(4*$i,$key),$tp1);
2935 }
2936
2937 # int AES_set_decrypt_key(const unsigned char *userKey, const int bits,
2938 #                        AES_KEY *key)
2939 &function_begin_B("AES_set_decrypt_key");
2940         &call   ("_x86_AES_set_encrypt_key");
2941         &cmp    ("eax",0);
2942         &je     (&label("proceed"));
2943         &ret    ();
2944
2945     &set_label("proceed");
2946         &push   ("ebp");
2947         &push   ("ebx");
2948         &push   ("esi");
2949         &push   ("edi");
2950
2951         &mov    ("esi",&wparam(2));
2952         &mov    ("ecx",&DWP(240,"esi"));        # pull number of rounds
2953         &lea    ("ecx",&DWP(0,"","ecx",4));
2954         &lea    ("edi",&DWP(0,"esi","ecx",4));  # pointer to last chunk
2955
2956         &set_label("invert",4);                 # invert order of chunks
2957                 &mov    ("eax",&DWP(0,"esi"));
2958                 &mov    ("ebx",&DWP(4,"esi"));
2959                 &mov    ("ecx",&DWP(0,"edi"));
2960                 &mov    ("edx",&DWP(4,"edi"));
2961                 &mov    (&DWP(0,"edi"),"eax");
2962                 &mov    (&DWP(4,"edi"),"ebx");
2963                 &mov    (&DWP(0,"esi"),"ecx");
2964                 &mov    (&DWP(4,"esi"),"edx");
2965                 &mov    ("eax",&DWP(8,"esi"));
2966                 &mov    ("ebx",&DWP(12,"esi"));
2967                 &mov    ("ecx",&DWP(8,"edi"));
2968                 &mov    ("edx",&DWP(12,"edi"));
2969                 &mov    (&DWP(8,"edi"),"eax");
2970                 &mov    (&DWP(12,"edi"),"ebx");
2971                 &mov    (&DWP(8,"esi"),"ecx");
2972                 &mov    (&DWP(12,"esi"),"edx");
2973                 &add    ("esi",16);
2974                 &sub    ("edi",16);
2975                 &cmp    ("esi","edi");
2976         &jne    (&label("invert"));
2977
2978         &mov    ($key,&wparam(2));
2979         &mov    ($acc,&DWP(240,$key));          # pull number of rounds
2980         &lea    ($acc,&DWP(-2,$acc,$acc));
2981         &lea    ($acc,&DWP(0,$key,$acc,8));
2982         &mov    (&wparam(2),$acc);
2983
2984         &mov    ($s0,&DWP(16,$key));            # modulo-scheduled load
2985         &set_label("permute",4);                # permute the key schedule
2986                 &add    ($key,16);
2987                 &deckey (0,$key,$s0,$s1,$s2,$s3);
2988                 &deckey (1,$key,$s1,$s2,$s3,$s0);
2989                 &deckey (2,$key,$s2,$s3,$s0,$s1);
2990                 &deckey (3,$key,$s3,$s0,$s1,$s2);
2991                 &cmp    ($key,&wparam(2));
2992         &jb     (&label("permute"));
2993
2994         &xor    ("eax","eax");                  # return success
2995 &function_end("AES_set_decrypt_key");
2996 &asciz("AES for x86, CRYPTOGAMS by <appro\@openssl.org>");
2997
2998 &asm_finish();
2999
3000 close STDOUT;