aes-586.pl revision 160814 
1#!/usr/bin/env perl
2#
3# ====================================================================
4# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
5# project. Rights for redistribution and usage in source and binary
6# forms are granted according to the OpenSSL license.
7# ====================================================================
8#
9# Version 3.4.
10#
11# You might fail to appreciate this module performance from the first
12# try. If compared to "vanilla" linux-ia32-icc target, i.e. considered
13# to be *the* best Intel C compiler without -KPIC, performance appears
14# to be virtually identical... But try to re-configure with shared
15# library support... Aha! Intel compiler "suddenly" lags behind by 30%
16# [on P4, more on others]:-) And if compared to position-independent
17# code generated by GNU C, this code performs *more* than *twice* as
18# fast! Yes, all this buzz about PIC means that unlike other hand-
19# coded implementations, this one was explicitly designed to be safe
20# to use even in shared library context... This also means that this
21# code isn't necessarily absolutely fastest "ever," because in order
22# to achieve position independence an extra register has to be
23# off-loaded to stack, which affects the benchmark result.
24#
25# Special note about instruction choice. Do you recall RC4_INT code
26# performing poorly on P4? It might be the time to figure out why.
27# RC4_INT code implies effective address calculations in base+offset*4
28# form. Trouble is that it seems that offset scaling turned to be
29# critical path... At least eliminating scaling resulted in 2.8x RC4
30# performance improvement [as you might recall]. As AES code is hungry
31# for scaling too, I [try to] avoid the latter by favoring off-by-2
32# shifts and masking the result with 0xFF<<2 instead of "boring" 0xFF.
33#
34# As was shown by Dean Gaudet <dean@arctic.org>, the above note turned
35# void. Performance improvement with off-by-2 shifts was observed on
36# intermediate implementation, which was spilling yet another register
37# to stack... Final offset*4 code below runs just a tad faster on P4,
38# but exhibits up to 10% improvement on other cores.
39#
40# Second version is "monolithic" replacement for aes_core.c, which in
41# addition to AES_[de|en]crypt implements AES_set_[de|en]cryption_key.
42# This made it possible to implement little-endian variant of the
43# algorithm without modifying the base C code. Motivating factor for
44# the undertaken effort was that it appeared that in tight IA-32
45# register window little-endian flavor could achieve slightly higher
46# Instruction Level Parallelism, and it indeed resulted in up to 15%
47# better performance on most recent �-archs...
48#
49# Third version adds AES_cbc_encrypt implementation, which resulted in
50# up to 40% performance imrovement of CBC benchmark results. 40% was
51# observed on P4 core, where "overall" imrovement coefficient, i.e. if
52# compared to PIC generated by GCC and in CBC mode, was observed to be
53# as large as 4x:-) CBC performance is virtually identical to ECB now
54# and on some platforms even better, e.g. 17.6 "small" cycles/byte on
55# Opteron, because certain function prologues and epilogues are
56# effectively taken out of the loop...
57#
58# Version 3.2 implements compressed tables and prefetch of these tables
59# in CBC[!] mode. Former means that 3/4 of table references are now
60# misaligned, which unfortunately has negative impact on elder IA-32
61# implementations, Pentium suffered 30% penalty, PIII - 10%.
62#
63# Version 3.3 avoids L1 cache aliasing between stack frame and
64# S-boxes, and 3.4 - L1 cache aliasing even between key schedule. The
65# latter is achieved by copying the key schedule to controlled place in
66# stack. This unfortunately has rather strong impact on small block CBC
67# performance, ~2x deterioration on 16-byte block if compared to 3.3.
68#
69# Current ECB performance numbers for 128-bit key in CPU cycles per
70# processed byte [measure commonly used by AES benchmarkers] are:
71#
72#		small footprint		fully unrolled
73# P4		24			22
74# AMD K8	20			19
75# PIII		25			23
76# Pentium	81			78
77
78push(@INC,"perlasm","../../perlasm");
79require "x86asm.pl";
80
81&asm_init($ARGV[0],"aes-586.pl",$ARGV[$#ARGV] eq "386");
82
83$s0="eax";
84$s1="ebx";
85$s2="ecx";
86$s3="edx";
87$key="edi";
88$acc="esi";
89
90$compromise=0;		# $compromise=128 abstains from copying key
91			# schedule to stack when encrypting inputs
92			# shorter than 128 bytes at the cost of
93			# risksing aliasing with S-boxes. In return
94			# you get way better, up to +70%, small block
95			# performance.
96$small_footprint=1;	# $small_footprint=1 code is ~5% slower [on
97			# recent �-archs], but ~5 times smaller!
98			# I favor compact code to minimize cache
99			# contention and in hope to "collect" 5% back
100			# in real-life applications...
101$vertical_spin=0;	# shift "verticaly" defaults to 0, because of
102			# its proof-of-concept status...
103
104# Note that there is no decvert(), as well as last encryption round is
105# performed with "horizontal" shifts. This is because this "vertical"
106# implementation [one which groups shifts on a given $s[i] to form a
107# "column," unlike "horizontal" one, which groups shifts on different
108# $s[i] to form a "row"] is work in progress. It was observed to run
109# few percents faster on Intel cores, but not AMD. On AMD K8 core it's
110# whole 12% slower:-( So we face a trade-off... Shall it be resolved
111# some day? Till then the code is considered experimental and by
112# default remains dormant...
113
114sub encvert()
115{ my ($te,@s) = @_;
116  my $v0 = $acc, $v1 = $key;
117
118	&mov	($v0,$s[3]);				# copy s3
119	&mov	(&DWP(4,"esp"),$s[2]);			# save s2
120	&mov	($v1,$s[0]);				# copy s0
121	&mov	(&DWP(8,"esp"),$s[1]);			# save s1
122
123	&movz	($s[2],&HB($s[0]));
124	&and	($s[0],0xFF);
125	&mov	($s[0],&DWP(0,$te,$s[0],8));		# s0>>0
126	&shr	($v1,16);
127	&mov	($s[3],&DWP(3,$te,$s[2],8));		# s0>>8
128	&movz	($s[1],&HB($v1));
129	&and	($v1,0xFF);
130	&mov	($s[2],&DWP(2,$te,$v1,8));		# s0>>16
131	 &mov	($v1,$v0);
132	&mov	($s[1],&DWP(1,$te,$s[1],8));		# s0>>24
133
134	&and	($v0,0xFF);
135	&xor	($s[3],&DWP(0,$te,$v0,8));		# s3>>0
136	&movz	($v0,&HB($v1));
137	&shr	($v1,16);
138	&xor	($s[2],&DWP(3,$te,$v0,8));		# s3>>8
139	&movz	($v0,&HB($v1));
140	&and	($v1,0xFF);
141	&xor	($s[1],&DWP(2,$te,$v1,8));		# s3>>16
142	 &mov	($v1,&DWP(4,"esp"));			# restore s2
143	&xor	($s[0],&DWP(1,$te,$v0,8));		# s3>>24
144
145	&mov	($v0,$v1);
146	&and	($v1,0xFF);
147	&xor	($s[2],&DWP(0,$te,$v1,8));		# s2>>0
148	&movz	($v1,&HB($v0));
149	&shr	($v0,16);
150	&xor	($s[1],&DWP(3,$te,$v1,8));		# s2>>8
151	&movz	($v1,&HB($v0));
152	&and	($v0,0xFF);
153	&xor	($s[0],&DWP(2,$te,$v0,8));		# s2>>16
154	 &mov	($v0,&DWP(8,"esp"));			# restore s1
155	&xor	($s[3],&DWP(1,$te,$v1,8));		# s2>>24
156
157	&mov	($v1,$v0);
158	&and	($v0,0xFF);
159	&xor	($s[1],&DWP(0,$te,$v0,8));		# s1>>0
160	&movz	($v0,&HB($v1));
161	&shr	($v1,16);
162	&xor	($s[0],&DWP(3,$te,$v0,8));		# s1>>8
163	&movz	($v0,&HB($v1));
164	&and	($v1,0xFF);
165	&xor	($s[3],&DWP(2,$te,$v1,8));		# s1>>16
166	 &mov	($key,&DWP(12,"esp"));			# reincarnate v1 as key
167	&xor	($s[2],&DWP(1,$te,$v0,8));		# s1>>24
168}
169
170sub encstep()
171{ my ($i,$te,@s) = @_;
172  my $tmp = $key;
173  my $out = $i==3?$s[0]:$acc;
174
175	# lines marked with #%e?x[i] denote "reordered" instructions...
176	if ($i==3)  {	&mov	($key,&DWP(12,"esp"));		}##%edx
177	else        {	&mov	($out,$s[0]);
178			&and	($out,0xFF);			}
179	if ($i==1)  {	&shr	($s[0],16);			}#%ebx[1]
180	if ($i==2)  {	&shr	($s[0],24);			}#%ecx[2]
181			&mov	($out,&DWP(0,$te,$out,8));
182
183	if ($i==3)  {	$tmp=$s[1];				}##%eax
184			&movz	($tmp,&HB($s[1]));
185			&xor	($out,&DWP(3,$te,$tmp,8));
186
187	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],&DWP(4,"esp"));	}##%ebx
188	else        {	&mov	($tmp,$s[2]);
189			&shr	($tmp,16);			}
190	if ($i==2)  {	&and	($s[1],0xFF);			}#%edx[2]
191			&and	($tmp,0xFF);
192			&xor	($out,&DWP(2,$te,$tmp,8));
193
194	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],&DWP(8,"esp"));	}##%ecx
195	elsif($i==2){	&movz	($tmp,&HB($s[3]));		}#%ebx[2]
196	else        {	&mov	($tmp,$s[3]);
197			&shr	($tmp,24)			}
198			&xor	($out,&DWP(1,$te,$tmp,8));
199	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
200	if ($i==3)  {	&mov	($s[3],$acc);			}
201			&comment();
202}
203
204sub enclast()
205{ my ($i,$te,@s)=@_;
206  my $tmp = $key;
207  my $out = $i==3?$s[0]:$acc;
208
209	if ($i==3)  {	&mov	($key,&DWP(12,"esp"));		}##%edx
210	else        {	&mov	($out,$s[0]);			}
211			&and	($out,0xFF);
212	if ($i==1)  {	&shr	($s[0],16);			}#%ebx[1]
213	if ($i==2)  {	&shr	($s[0],24);			}#%ecx[2]
214			&mov	($out,&DWP(2,$te,$out,8));
215			&and	($out,0x000000ff);
216
217	if ($i==3)  {	$tmp=$s[1];				}##%eax
218			&movz	($tmp,&HB($s[1]));
219			&mov	($tmp,&DWP(0,$te,$tmp,8));
220			&and	($tmp,0x0000ff00);
221			&xor	($out,$tmp);
222
223	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],&DWP(4,"esp"));	}##%ebx
224	else        {	mov	($tmp,$s[2]);
225			&shr	($tmp,16);			}
226	if ($i==2)  {	&and	($s[1],0xFF);			}#%edx[2]
227			&and	($tmp,0xFF);
228			&mov	($tmp,&DWP(0,$te,$tmp,8));
229			&and	($tmp,0x00ff0000);
230			&xor	($out,$tmp);
231
232	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],&DWP(8,"esp"));	}##%ecx
233	elsif($i==2){	&movz	($tmp,&HB($s[3]));		}#%ebx[2]
234	else        {	&mov	($tmp,$s[3]);
235			&shr	($tmp,24);			}
236			&mov	($tmp,&DWP(2,$te,$tmp,8));
237			&and	($tmp,0xff000000);
238			&xor	($out,$tmp);
239	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
240	if ($i==3)  {	&mov	($s[3],$acc);			}
241}
242
243sub _data_word() { my $i; while(defined($i=shift)) { &data_word($i,$i); } }
244
245&public_label("AES_Te");
246&function_begin_B("_x86_AES_encrypt");
247	if ($vertical_spin) {
248		# I need high parts of volatile registers to be accessible...
249		&exch	($s1="edi",$key="ebx");
250		&mov	($s2="esi",$acc="ecx");
251	}
252
253	# note that caller is expected to allocate stack frame for me!
254	&mov	(&DWP(12,"esp"),$key);		# save key
255
256	&xor	($s0,&DWP(0,$key));		# xor with key
257	&xor	($s1,&DWP(4,$key));
258	&xor	($s2,&DWP(8,$key));
259	&xor	($s3,&DWP(12,$key));
260
261	&mov	($acc,&DWP(240,$key));		# load key->rounds
262
263	if ($small_footprint) {
264	    &lea	($acc,&DWP(-2,$acc,$acc));
265	    &lea	($acc,&DWP(0,$key,$acc,8));
266	    &mov	(&DWP(16,"esp"),$acc);	# end of key schedule
267	    &align	(4);
268	    &set_label("loop");
269		if ($vertical_spin) {
270		    &encvert("ebp",$s0,$s1,$s2,$s3);
271		} else {
272		    &encstep(0,"ebp",$s0,$s1,$s2,$s3);
273		    &encstep(1,"ebp",$s1,$s2,$s3,$s0);
274		    &encstep(2,"ebp",$s2,$s3,$s0,$s1);
275		    &encstep(3,"ebp",$s3,$s0,$s1,$s2);
276		}
277		&add	($key,16);		# advance rd_key
278		&xor	($s0,&DWP(0,$key));
279		&xor	($s1,&DWP(4,$key));
280		&xor	($s2,&DWP(8,$key));
281		&xor	($s3,&DWP(12,$key));
282	    &cmp	($key,&DWP(16,"esp"));
283	    &mov	(&DWP(12,"esp"),$key);
284	    &jb		(&label("loop"));
285	}
286	else {
287	    &cmp	($acc,10);
288	    &jle	(&label("10rounds"));
289	    &cmp	($acc,12);
290	    &jle	(&label("12rounds"));
291
292	&set_label("14rounds");
293	    for ($i=1;$i<3;$i++) {
294		if ($vertical_spin) {
295		    &encvert("ebp",$s0,$s1,$s2,$s3);
296		} else {
297		    &encstep(0,"ebp",$s0,$s1,$s2,$s3);
298		    &encstep(1,"ebp",$s1,$s2,$s3,$s0);
299		    &encstep(2,"ebp",$s2,$s3,$s0,$s1);
300		    &encstep(3,"ebp",$s3,$s0,$s1,$s2);
301		}
302		&xor	($s0,&DWP(16*$i+0,$key));
303		&xor	($s1,&DWP(16*$i+4,$key));
304		&xor	($s2,&DWP(16*$i+8,$key));
305		&xor	($s3,&DWP(16*$i+12,$key));
306	    }
307	    &add	($key,32);
308	    &mov	(&DWP(12,"esp"),$key);	# advance rd_key
309	&set_label("12rounds");
310	    for ($i=1;$i<3;$i++) {
311		if ($vertical_spin) {
312		    &encvert("ebp",$s0,$s1,$s2,$s3);
313		} else {
314		    &encstep(0,"ebp",$s0,$s1,$s2,$s3);
315		    &encstep(1,"ebp",$s1,$s2,$s3,$s0);
316		    &encstep(2,"ebp",$s2,$s3,$s0,$s1);
317		    &encstep(3,"ebp",$s3,$s0,$s1,$s2);
318		}
319		&xor	($s0,&DWP(16*$i+0,$key));
320		&xor	($s1,&DWP(16*$i+4,$key));
321		&xor	($s2,&DWP(16*$i+8,$key));
322		&xor	($s3,&DWP(16*$i+12,$key));
323	    }
324	    &add	($key,32);
325	    &mov	(&DWP(12,"esp"),$key);	# advance rd_key
326	&set_label("10rounds");
327	    for ($i=1;$i<10;$i++) {
328		if ($vertical_spin) {
329		    &encvert("ebp",$s0,$s1,$s2,$s3);
330		} else {
331		    &encstep(0,"ebp",$s0,$s1,$s2,$s3);
332		    &encstep(1,"ebp",$s1,$s2,$s3,$s0);
333		    &encstep(2,"ebp",$s2,$s3,$s0,$s1);
334		    &encstep(3,"ebp",$s3,$s0,$s1,$s2);
335		}
336		&xor	($s0,&DWP(16*$i+0,$key));
337		&xor	($s1,&DWP(16*$i+4,$key));
338		&xor	($s2,&DWP(16*$i+8,$key));
339		&xor	($s3,&DWP(16*$i+12,$key));
340	    }
341	}
342
343	if ($vertical_spin) {
344	    # "reincarnate" some registers for "horizontal" spin...
345	    &mov	($s1="ebx",$key="edi");
346	    &mov	($s2="ecx",$acc="esi");
347	}
348	&enclast(0,"ebp",$s0,$s1,$s2,$s3);
349	&enclast(1,"ebp",$s1,$s2,$s3,$s0);
350	&enclast(2,"ebp",$s2,$s3,$s0,$s1);
351	&enclast(3,"ebp",$s3,$s0,$s1,$s2);
352
353	&add	($key,$small_footprint?16:160);
354	&xor	($s0,&DWP(0,$key));
355	&xor	($s1,&DWP(4,$key));
356	&xor	($s2,&DWP(8,$key));
357	&xor	($s3,&DWP(12,$key));
358
359	&ret	();
360
361&set_label("AES_Te",64);	# Yes! I keep it in the code segment!
362	&_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6);
363	&_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591);
364	&_data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56);
365	&_data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec);
366	&_data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa);
367	&_data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb);
368	&_data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45);
369	&_data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b);
370	&_data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c);
371	&_data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83);
372	&_data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9);
373	&_data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a);
374	&_data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d);
375	&_data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f);
376	&_data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df);
377	&_data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea);
378	&_data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34);
379	&_data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b);
380	&_data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d);
381	&_data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413);
382	&_data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1);
383	&_data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6);
384	&_data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972);
385	&_data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85);
386	&_data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed);
387	&_data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511);
388	&_data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe);
389	&_data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b);
390	&_data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05);
391	&_data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1);
392	&_data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142);
393	&_data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf);
394	&_data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3);
395	&_data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e);
396	&_data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a);
397	&_data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6);
398	&_data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3);
399	&_data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b);
400	&_data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428);
401	&_data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad);
402	&_data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14);
403	&_data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8);
404	&_data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4);
405	&_data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2);
406	&_data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda);
407	&_data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949);
408	&_data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf);
409	&_data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810);
410	&_data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c);
411	&_data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697);
412	&_data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e);
413	&_data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f);
414	&_data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc);
415	&_data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c);
416	&_data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969);
417	&_data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27);
418	&_data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122);
419	&_data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433);
420	&_data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9);
421	&_data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5);
422	&_data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a);
423	&_data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0);
424	&_data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e);
425	&_data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c);
426#rcon:
427	&data_word(0x00000001, 0x00000002, 0x00000004, 0x00000008);
428	&data_word(0x00000010, 0x00000020, 0x00000040, 0x00000080);
429	&data_word(0x0000001b, 0x00000036, 0, 0, 0, 0, 0, 0);
430&function_end_B("_x86_AES_encrypt");
431
432# void AES_encrypt (const void *inp,void *out,const AES_KEY *key);
433&public_label("AES_Te");
434&function_begin("AES_encrypt");
435	&mov	($acc,&wparam(0));		# load inp
436	&mov	($key,&wparam(2));		# load key
437
438	&mov	($s0,"esp");
439	&sub	("esp",24);
440	&and	("esp",-64);
441	&add	("esp",4);
442	&mov	(&DWP(16,"esp"),$s0);
443
444	&call   (&label("pic_point"));          # make it PIC!
445	&set_label("pic_point");
446	&blindpop("ebp");
447	&lea    ("ebp",&DWP(&label("AES_Te")."-".&label("pic_point"),"ebp"));
448
449	&mov	($s0,&DWP(0,$acc));		# load input data
450	&mov	($s1,&DWP(4,$acc));
451	&mov	($s2,&DWP(8,$acc));
452	&mov	($s3,&DWP(12,$acc));
453
454	&call	("_x86_AES_encrypt");
455
456	&mov	("esp",&DWP(16,"esp"));
457
458	&mov	($acc,&wparam(1));		# load out
459	&mov	(&DWP(0,$acc),$s0);		# write output data
460	&mov	(&DWP(4,$acc),$s1);
461	&mov	(&DWP(8,$acc),$s2);
462	&mov	(&DWP(12,$acc),$s3);
463&function_end("AES_encrypt");
464
465#------------------------------------------------------------------#
466
467sub decstep()
468{ my ($i,$td,@s) = @_;
469  my $tmp = $key;
470  my $out = $i==3?$s[0]:$acc;
471
472	# no instructions are reordered, as performance appears
473	# optimal... or rather that all attempts to reorder didn't
474	# result in better performance [which by the way is not a
475	# bit lower than ecryption].
476	if($i==3)   {	&mov	($key,&DWP(12,"esp"));		}
477	else        {	&mov	($out,$s[0]);			}
478			&and	($out,0xFF);
479			&mov	($out,&DWP(0,$td,$out,8));
480
481	if ($i==3)  {	$tmp=$s[1];				}
482			&movz	($tmp,&HB($s[1]));
483			&xor	($out,&DWP(3,$td,$tmp,8));
484
485	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],$acc);		}
486	else        {	&mov	($tmp,$s[2]);			}
487			&shr	($tmp,16);
488			&and	($tmp,0xFF);
489			&xor	($out,&DWP(2,$td,$tmp,8));
490
491	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],&DWP(8,"esp"));	}
492	else        {	&mov	($tmp,$s[3]);			}
493			&shr	($tmp,24);
494			&xor	($out,&DWP(1,$td,$tmp,8));
495	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
496	if ($i==3)  {	&mov	($s[3],&DWP(4,"esp"));		}
497			&comment();
498}
499
500sub declast()
501{ my ($i,$td,@s)=@_;
502  my $tmp = $key;
503  my $out = $i==3?$s[0]:$acc;
504
505	if($i==3)   {	&mov	($key,&DWP(12,"esp"));		}
506	else        {	&mov	($out,$s[0]);			}
507			&and	($out,0xFF);
508			&mov	($out,&DWP(2048,$td,$out,4));
509			&and	($out,0x000000ff);
510
511	if ($i==3)  {	$tmp=$s[1];				}
512			&movz	($tmp,&HB($s[1]));
513			&mov	($tmp,&DWP(2048,$td,$tmp,4));
514			&and	($tmp,0x0000ff00);
515			&xor	($out,$tmp);
516
517	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],$acc);		}
518	else        {	mov	($tmp,$s[2]);			}
519			&shr	($tmp,16);
520			&and	($tmp,0xFF);
521			&mov	($tmp,&DWP(2048,$td,$tmp,4));
522			&and	($tmp,0x00ff0000);
523			&xor	($out,$tmp);
524
525	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],&DWP(8,"esp"));	}
526	else        {	&mov	($tmp,$s[3]);			}
527			&shr	($tmp,24);
528			&mov	($tmp,&DWP(2048,$td,$tmp,4));
529			&and	($tmp,0xff000000);
530			&xor	($out,$tmp);
531	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
532	if ($i==3)  {	&mov	($s[3],&DWP(4,"esp"));		}
533}
534
535&public_label("AES_Td");
536&function_begin_B("_x86_AES_decrypt");
537	# note that caller is expected to allocate stack frame for me!
538	&mov	(&DWP(12,"esp"),$key);		# save key
539
540	&xor	($s0,&DWP(0,$key));		# xor with key
541	&xor	($s1,&DWP(4,$key));
542	&xor	($s2,&DWP(8,$key));
543	&xor	($s3,&DWP(12,$key));
544
545	&mov	($acc,&DWP(240,$key));		# load key->rounds
546
547	if ($small_footprint) {
548	    &lea	($acc,&DWP(-2,$acc,$acc));
549	    &lea	($acc,&DWP(0,$key,$acc,8));
550	    &mov	(&DWP(16,"esp"),$acc);	# end of key schedule
551	    &align	(4);
552	    &set_label("loop");
553		&decstep(0,"ebp",$s0,$s3,$s2,$s1);
554		&decstep(1,"ebp",$s1,$s0,$s3,$s2);
555		&decstep(2,"ebp",$s2,$s1,$s0,$s3);
556		&decstep(3,"ebp",$s3,$s2,$s1,$s0);
557		&add	($key,16);		# advance rd_key
558		&xor	($s0,&DWP(0,$key));
559		&xor	($s1,&DWP(4,$key));
560		&xor	($s2,&DWP(8,$key));
561		&xor	($s3,&DWP(12,$key));
562	    &cmp	($key,&DWP(16,"esp"));
563	    &mov	(&DWP(12,"esp"),$key);
564	    &jb		(&label("loop"));
565	}
566	else {
567	    &cmp	($acc,10);
568	    &jle	(&label("10rounds"));
569	    &cmp	($acc,12);
570	    &jle	(&label("12rounds"));
571
572	&set_label("14rounds");
573	    for ($i=1;$i<3;$i++) {
574		&decstep(0,"ebp",$s0,$s3,$s2,$s1);
575		&decstep(1,"ebp",$s1,$s0,$s3,$s2);
576		&decstep(2,"ebp",$s2,$s1,$s0,$s3);
577		&decstep(3,"ebp",$s3,$s2,$s1,$s0);
578		&xor	($s0,&DWP(16*$i+0,$key));
579		&xor	($s1,&DWP(16*$i+4,$key));
580		&xor	($s2,&DWP(16*$i+8,$key));
581		&xor	($s3,&DWP(16*$i+12,$key));
582	    }
583	    &add	($key,32);
584	    &mov	(&DWP(12,"esp"),$key);	# advance rd_key
585	&set_label("12rounds");
586	    for ($i=1;$i<3;$i++) {
587		&decstep(0,"ebp",$s0,$s3,$s2,$s1);
588		&decstep(1,"ebp",$s1,$s0,$s3,$s2);
589		&decstep(2,"ebp",$s2,$s1,$s0,$s3);
590		&decstep(3,"ebp",$s3,$s2,$s1,$s0);
591		&xor	($s0,&DWP(16*$i+0,$key));
592		&xor	($s1,&DWP(16*$i+4,$key));
593		&xor	($s2,&DWP(16*$i+8,$key));
594		&xor	($s3,&DWP(16*$i+12,$key));
595	    }
596	    &add	($key,32);
597	    &mov	(&DWP(12,"esp"),$key);	# advance rd_key
598	&set_label("10rounds");
599	    for ($i=1;$i<10;$i++) {
600		&decstep(0,"ebp",$s0,$s3,$s2,$s1);
601		&decstep(1,"ebp",$s1,$s0,$s3,$s2);
602		&decstep(2,"ebp",$s2,$s1,$s0,$s3);
603		&decstep(3,"ebp",$s3,$s2,$s1,$s0);
604		&xor	($s0,&DWP(16*$i+0,$key));
605		&xor	($s1,&DWP(16*$i+4,$key));
606		&xor	($s2,&DWP(16*$i+8,$key));
607		&xor	($s3,&DWP(16*$i+12,$key));
608	    }
609	}
610
611	&declast(0,"ebp",$s0,$s3,$s2,$s1);
612	&declast(1,"ebp",$s1,$s0,$s3,$s2);
613	&declast(2,"ebp",$s2,$s1,$s0,$s3);
614	&declast(3,"ebp",$s3,$s2,$s1,$s0);
615
616	&add	($key,$small_footprint?16:160);
617	&xor	($s0,&DWP(0,$key));
618	&xor	($s1,&DWP(4,$key));
619	&xor	($s2,&DWP(8,$key));
620	&xor	($s3,&DWP(12,$key));
621
622	&ret	();
623
624&set_label("AES_Td",64);	# Yes! I keep it in the code segment!
625	&_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a);
626	&_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b);
627	&_data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5);
628	&_data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5);
629	&_data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d);
630	&_data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b);
631	&_data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295);
632	&_data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e);
633	&_data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927);
634	&_data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d);
635	&_data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362);
636	&_data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9);
637	&_data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52);
638	&_data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566);
639	&_data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3);
640	&_data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed);
641	&_data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e);
642	&_data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4);
643	&_data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4);
644	&_data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd);
645	&_data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d);
646	&_data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060);
647	&_data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967);
648	&_data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879);
649	&_data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000);
650	&_data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c);
651	&_data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36);
652	&_data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624);
653	&_data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b);
654	&_data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c);
655	&_data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12);
656	&_data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14);
657	&_data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3);
658	&_data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b);
659	&_data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8);
660	&_data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684);
661	&_data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7);
662	&_data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177);
663	&_data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947);
664	&_data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322);
665	&_data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498);
666	&_data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f);
667	&_data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54);
668	&_data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382);
669	&_data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf);
670	&_data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb);
671	&_data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83);
672	&_data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef);
673	&_data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029);
674	&_data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235);
675	&_data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733);
676	&_data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117);
677	&_data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4);
678	&_data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546);
679	&_data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb);
680	&_data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d);
681	&_data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb);
682	&_data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a);
683	&_data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773);
684	&_data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478);
685	&_data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2);
686	&_data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff);
687	&_data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664);
688	&_data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0);
689#Td4:
690	&data_word(0x52525252, 0x09090909, 0x6a6a6a6a, 0xd5d5d5d5);
691	&data_word(0x30303030, 0x36363636, 0xa5a5a5a5, 0x38383838);
692	&data_word(0xbfbfbfbf, 0x40404040, 0xa3a3a3a3, 0x9e9e9e9e);
693	&data_word(0x81818181, 0xf3f3f3f3, 0xd7d7d7d7, 0xfbfbfbfb);
694	&data_word(0x7c7c7c7c, 0xe3e3e3e3, 0x39393939, 0x82828282);
695	&data_word(0x9b9b9b9b, 0x2f2f2f2f, 0xffffffff, 0x87878787);
696	&data_word(0x34343434, 0x8e8e8e8e, 0x43434343, 0x44444444);
697	&data_word(0xc4c4c4c4, 0xdededede, 0xe9e9e9e9, 0xcbcbcbcb);
698	&data_word(0x54545454, 0x7b7b7b7b, 0x94949494, 0x32323232);
699	&data_word(0xa6a6a6a6, 0xc2c2c2c2, 0x23232323, 0x3d3d3d3d);
700	&data_word(0xeeeeeeee, 0x4c4c4c4c, 0x95959595, 0x0b0b0b0b);
701	&data_word(0x42424242, 0xfafafafa, 0xc3c3c3c3, 0x4e4e4e4e);
702	&data_word(0x08080808, 0x2e2e2e2e, 0xa1a1a1a1, 0x66666666);
703	&data_word(0x28282828, 0xd9d9d9d9, 0x24242424, 0xb2b2b2b2);
704	&data_word(0x76767676, 0x5b5b5b5b, 0xa2a2a2a2, 0x49494949);
705	&data_word(0x6d6d6d6d, 0x8b8b8b8b, 0xd1d1d1d1, 0x25252525);
706	&data_word(0x72727272, 0xf8f8f8f8, 0xf6f6f6f6, 0x64646464);
707	&data_word(0x86868686, 0x68686868, 0x98989898, 0x16161616);
708	&data_word(0xd4d4d4d4, 0xa4a4a4a4, 0x5c5c5c5c, 0xcccccccc);
709	&data_word(0x5d5d5d5d, 0x65656565, 0xb6b6b6b6, 0x92929292);
710	&data_word(0x6c6c6c6c, 0x70707070, 0x48484848, 0x50505050);
711	&data_word(0xfdfdfdfd, 0xedededed, 0xb9b9b9b9, 0xdadadada);
712	&data_word(0x5e5e5e5e, 0x15151515, 0x46464646, 0x57575757);
713	&data_word(0xa7a7a7a7, 0x8d8d8d8d, 0x9d9d9d9d, 0x84848484);
714	&data_word(0x90909090, 0xd8d8d8d8, 0xabababab, 0x00000000);
715	&data_word(0x8c8c8c8c, 0xbcbcbcbc, 0xd3d3d3d3, 0x0a0a0a0a);
716	&data_word(0xf7f7f7f7, 0xe4e4e4e4, 0x58585858, 0x05050505);
717	&data_word(0xb8b8b8b8, 0xb3b3b3b3, 0x45454545, 0x06060606);
718	&data_word(0xd0d0d0d0, 0x2c2c2c2c, 0x1e1e1e1e, 0x8f8f8f8f);
719	&data_word(0xcacacaca, 0x3f3f3f3f, 0x0f0f0f0f, 0x02020202);
720	&data_word(0xc1c1c1c1, 0xafafafaf, 0xbdbdbdbd, 0x03030303);
721	&data_word(0x01010101, 0x13131313, 0x8a8a8a8a, 0x6b6b6b6b);
722	&data_word(0x3a3a3a3a, 0x91919191, 0x11111111, 0x41414141);
723	&data_word(0x4f4f4f4f, 0x67676767, 0xdcdcdcdc, 0xeaeaeaea);
724	&data_word(0x97979797, 0xf2f2f2f2, 0xcfcfcfcf, 0xcececece);
725	&data_word(0xf0f0f0f0, 0xb4b4b4b4, 0xe6e6e6e6, 0x73737373);
726	&data_word(0x96969696, 0xacacacac, 0x74747474, 0x22222222);
727	&data_word(0xe7e7e7e7, 0xadadadad, 0x35353535, 0x85858585);
728	&data_word(0xe2e2e2e2, 0xf9f9f9f9, 0x37373737, 0xe8e8e8e8);
729	&data_word(0x1c1c1c1c, 0x75757575, 0xdfdfdfdf, 0x6e6e6e6e);
730	&data_word(0x47474747, 0xf1f1f1f1, 0x1a1a1a1a, 0x71717171);
731	&data_word(0x1d1d1d1d, 0x29292929, 0xc5c5c5c5, 0x89898989);
732	&data_word(0x6f6f6f6f, 0xb7b7b7b7, 0x62626262, 0x0e0e0e0e);
733	&data_word(0xaaaaaaaa, 0x18181818, 0xbebebebe, 0x1b1b1b1b);
734	&data_word(0xfcfcfcfc, 0x56565656, 0x3e3e3e3e, 0x4b4b4b4b);
735	&data_word(0xc6c6c6c6, 0xd2d2d2d2, 0x79797979, 0x20202020);
736	&data_word(0x9a9a9a9a, 0xdbdbdbdb, 0xc0c0c0c0, 0xfefefefe);
737	&data_word(0x78787878, 0xcdcdcdcd, 0x5a5a5a5a, 0xf4f4f4f4);
738	&data_word(0x1f1f1f1f, 0xdddddddd, 0xa8a8a8a8, 0x33333333);
739	&data_word(0x88888888, 0x07070707, 0xc7c7c7c7, 0x31313131);
740	&data_word(0xb1b1b1b1, 0x12121212, 0x10101010, 0x59595959);
741	&data_word(0x27272727, 0x80808080, 0xecececec, 0x5f5f5f5f);
742	&data_word(0x60606060, 0x51515151, 0x7f7f7f7f, 0xa9a9a9a9);
743	&data_word(0x19191919, 0xb5b5b5b5, 0x4a4a4a4a, 0x0d0d0d0d);
744	&data_word(0x2d2d2d2d, 0xe5e5e5e5, 0x7a7a7a7a, 0x9f9f9f9f);
745	&data_word(0x93939393, 0xc9c9c9c9, 0x9c9c9c9c, 0xefefefef);
746	&data_word(0xa0a0a0a0, 0xe0e0e0e0, 0x3b3b3b3b, 0x4d4d4d4d);
747	&data_word(0xaeaeaeae, 0x2a2a2a2a, 0xf5f5f5f5, 0xb0b0b0b0);
748	&data_word(0xc8c8c8c8, 0xebebebeb, 0xbbbbbbbb, 0x3c3c3c3c);
749	&data_word(0x83838383, 0x53535353, 0x99999999, 0x61616161);
750	&data_word(0x17171717, 0x2b2b2b2b, 0x04040404, 0x7e7e7e7e);
751	&data_word(0xbabababa, 0x77777777, 0xd6d6d6d6, 0x26262626);
752	&data_word(0xe1e1e1e1, 0x69696969, 0x14141414, 0x63636363);
753	&data_word(0x55555555, 0x21212121, 0x0c0c0c0c, 0x7d7d7d7d);
754&function_end_B("_x86_AES_decrypt");
755
756# void AES_decrypt (const void *inp,void *out,const AES_KEY *key);
757&public_label("AES_Td");
758&function_begin("AES_decrypt");
759	&mov	($acc,&wparam(0));		# load inp
760	&mov	($key,&wparam(2));		# load key
761
762	&mov	($s0,"esp");
763	&sub	("esp",24);
764	&and	("esp",-64);
765	&add	("esp",4);
766	&mov	(&DWP(16,"esp"),$s0);
767
768	&call   (&label("pic_point"));          # make it PIC!
769	&set_label("pic_point");
770	&blindpop("ebp");
771	&lea    ("ebp",&DWP(&label("AES_Td")."-".&label("pic_point"),"ebp"));
772
773	&mov	($s0,&DWP(0,$acc));		# load input data
774	&mov	($s1,&DWP(4,$acc));
775	&mov	($s2,&DWP(8,$acc));
776	&mov	($s3,&DWP(12,$acc));
777
778	&call	("_x86_AES_decrypt");
779
780	&mov	("esp",&DWP(16,"esp"));
781
782	&mov	($acc,&wparam(1));		# load out
783	&mov	(&DWP(0,$acc),$s0);		# write output data
784	&mov	(&DWP(4,$acc),$s1);
785	&mov	(&DWP(8,$acc),$s2);
786	&mov	(&DWP(12,$acc),$s3);
787&function_end("AES_decrypt");
788
789# void AES_cbc_encrypt (const void char *inp, unsigned char *out,
790#			size_t length, const AES_KEY *key,
791#			unsigned char *ivp,const int enc);
792{
793# stack frame layout
794# -4(%esp)	0(%esp)		return address
795# 0(%esp)	4(%esp)		tmp1
796# 4(%esp)	8(%esp)		tmp2
797# 8(%esp)	12(%esp)	key
798# 12(%esp)	16(%esp)	end of key schedule
799my $_esp=&DWP(16,"esp");	#saved %esp
800my $_inp=&DWP(20,"esp");	#copy of wparam(0)
801my $_out=&DWP(24,"esp");	#copy of wparam(1)
802my $_len=&DWP(28,"esp");	#copy of wparam(2)
803my $_key=&DWP(32,"esp");	#copy of wparam(3)
804my $_ivp=&DWP(36,"esp");	#copy of wparam(4)
805my $_tmp=&DWP(40,"esp");	#volatile variable
806my $ivec=&DWP(44,"esp");	#ivec[16]
807my $aes_key=&DWP(60,"esp");	#copy of aes_key
808
809&public_label("AES_Te");
810&public_label("AES_Td");
811&function_begin("AES_cbc_encrypt");
812	&mov	($s2 eq "ecx"? $s2 : "",&wparam(2));	# load len
813	&cmp	($s2,0);
814	&je	(&label("enc_out"));
815
816	&call   (&label("pic_point"));		# make it PIC!
817	&set_label("pic_point");
818	&blindpop("ebp");
819
820	&pushf	();
821	&cld	();
822
823	&cmp	(&wparam(5),0);
824	&je	(&label("DECRYPT"));
825
826	&lea    ("ebp",&DWP(&label("AES_Te")."-".&label("pic_point"),"ebp"));
827
828	# allocate aligned stack frame...
829	&lea	($key,&DWP(-64-244,"esp"));
830	&and	($key,-64);
831
832	# ... and make sure it doesn't alias with AES_Te modulo 4096
833	&mov	($s0,"ebp");
834	&lea	($s1,&DWP(2048,"ebp"));
835	&mov	($s3,$key);
836	&and	($s0,0xfff);		# s = %ebp&0xfff
837	&and	($s1,0xfff);		# e = (%ebp+2048)&0xfff
838	&and	($s3,0xfff);		# p = %esp&0xfff
839
840	&cmp	($s3,$s1);		# if (p>=e) %esp =- (p-e);
841	&jb	(&label("te_break_out"));
842	&sub	($s3,$s1);
843	&sub	($key,$s3);
844	&jmp	(&label("te_ok"));
845	&set_label("te_break_out");	# else %esp -= (p-s)&0xfff + framesz;
846	&sub	($s3,$s0);
847	&and	($s3,0xfff);
848	&add	($s3,64+256);
849	&sub	($key,$s3);
850	&align	(4);
851	&set_label("te_ok");
852
853	&mov	($s0,&wparam(0));	# load inp
854	&mov	($s1,&wparam(1));	# load out
855	&mov	($s3,&wparam(3));	# load key
856	&mov	($acc,&wparam(4));	# load ivp
857
858	&exch	("esp",$key);
859	&add	("esp",4);		# reserve for return address!
860	&mov	($_esp,$key);		# save %esp
861
862	&mov	($_inp,$s0);		# save copy of inp
863	&mov	($_out,$s1);		# save copy of out
864	&mov	($_len,$s2);		# save copy of len
865	&mov	($_key,$s3);		# save copy of key
866	&mov	($_ivp,$acc);		# save copy of ivp
867
868	if ($compromise) {
869		&cmp	($s2,$compromise);
870		&jb	(&label("skip_ecopy"));
871	}
872	# copy key schedule to stack
873	&mov	("ecx",244/4);
874	&mov	("esi",$s3);
875	&lea	("edi",$aes_key);
876	&mov	($_key,"edi");
877	&align	(4);
878	&data_word(0xF689A5F3);	# rep movsd
879	&set_label("skip_ecopy") if ($compromise);
880
881	&mov	($acc,$s0);
882	&mov	($key,16);
883	&align	(4);
884	&set_label("prefetch_te");
885		&mov	($s0,&DWP(0,"ebp"));
886		&mov	($s1,&DWP(32,"ebp"));
887		&mov	($s2,&DWP(64,"ebp"));
888		&mov	($s3,&DWP(96,"ebp"));
889		&lea	("ebp",&DWP(128,"ebp"));
890		&dec	($key);
891	&jnz	(&label("prefetch_te"));
892	&sub	("ebp",2048);
893
894	&mov	($s2,$_len);
895	&mov	($key,$_ivp);
896	&test	($s2,0xFFFFFFF0);
897	&jz	(&label("enc_tail"));		# short input...
898
899	&mov	($s0,&DWP(0,$key));		# load iv
900	&mov	($s1,&DWP(4,$key));
901
902	&align	(4);
903	&set_label("enc_loop");
904		&mov	($s2,&DWP(8,$key));
905		&mov	($s3,&DWP(12,$key));
906
907		&xor	($s0,&DWP(0,$acc));	# xor input data
908		&xor	($s1,&DWP(4,$acc));
909		&xor	($s2,&DWP(8,$acc));
910		&xor	($s3,&DWP(12,$acc));
911
912		&mov	($key,$_key);		# load key
913		&call	("_x86_AES_encrypt");
914
915		&mov	($acc,$_inp);		# load inp
916		&mov	($key,$_out);		# load out
917
918		&mov	(&DWP(0,$key),$s0);	# save output data
919		&mov	(&DWP(4,$key),$s1);
920		&mov	(&DWP(8,$key),$s2);
921		&mov	(&DWP(12,$key),$s3);
922
923		&mov	($s2,$_len);		# load len
924
925		&lea	($acc,&DWP(16,$acc));
926		&mov	($_inp,$acc);		# save inp
927
928		&lea	($s3,&DWP(16,$key));
929		&mov	($_out,$s3);		# save out
930
931		&sub	($s2,16);
932		&test	($s2,0xFFFFFFF0);
933		&mov	($_len,$s2);		# save len
934	&jnz	(&label("enc_loop"));
935	&test	($s2,15);
936	&jnz	(&label("enc_tail"));
937	&mov	($acc,$_ivp);		# load ivp
938	&mov	($s2,&DWP(8,$key));	# restore last dwords
939	&mov	($s3,&DWP(12,$key));
940	&mov	(&DWP(0,$acc),$s0);	# save ivec
941	&mov	(&DWP(4,$acc),$s1);
942	&mov	(&DWP(8,$acc),$s2);
943	&mov	(&DWP(12,$acc),$s3);
944
945	&mov	("edi",$_key);
946	&mov	("esp",$_esp);
947	if ($compromise) {
948		&cmp	(&wparam(2),$compromise);
949		&jb	(&label("skip_ezero"));
950	}
951	# zero copy of key schedule
952	&mov	("ecx",240/4);
953	&xor	("eax","eax");
954	&align	(4);
955	&data_word(0xF689ABF3);	# rep stosd
956	&set_label("skip_ezero") if ($compromise);
957	&popf	();
958    &set_label("enc_out");
959	&function_end_A();
960	&pushf	();			# kludge, never executed
961
962    &align	(4);
963    &set_label("enc_tail");
964	&push	($key eq "edi" ? $key : "");	# push ivp
965	&mov	($key,$_out);			# load out
966	&mov	($s1,16);
967	&sub	($s1,$s2);
968	&cmp	($key,$acc);			# compare with inp
969	&je	(&label("enc_in_place"));
970	&align	(4);
971	&data_word(0xF689A4F3);	# rep movsb	# copy input
972	&jmp	(&label("enc_skip_in_place"));
973    &set_label("enc_in_place");
974	&lea	($key,&DWP(0,$key,$s2));
975    &set_label("enc_skip_in_place");
976	&mov	($s2,$s1);
977	&xor	($s0,$s0);
978	&align	(4);
979	&data_word(0xF689AAF3);	# rep stosb	# zero tail
980	&pop	($key);				# pop ivp
981
982	&mov	($acc,$_out);			# output as input
983	&mov	($s0,&DWP(0,$key));
984	&mov	($s1,&DWP(4,$key));
985	&mov	($_len,16);			# len=16
986	&jmp	(&label("enc_loop"));		# one more spin...
987
988#----------------------------- DECRYPT -----------------------------#
989&align	(4);
990&set_label("DECRYPT");
991	&lea    ("ebp",&DWP(&label("AES_Td")."-".&label("pic_point"),"ebp"));
992
993	# allocate aligned stack frame...
994	&lea	($key,&DWP(-64-244,"esp"));
995	&and	($key,-64);
996
997	# ... and make sure it doesn't alias with AES_Td modulo 4096
998	&mov	($s0,"ebp");
999	&lea	($s1,&DWP(3072,"ebp"));
1000	&mov	($s3,$key);
1001	&and	($s0,0xfff);		# s = %ebp&0xfff
1002	&and	($s1,0xfff);		# e = (%ebp+3072)&0xfff
1003	&and	($s3,0xfff);		# p = %esp&0xfff
1004
1005	&cmp	($s3,$s1);		# if (p>=e) %esp =- (p-e);
1006	&jb	(&label("td_break_out"));
1007	&sub	($s3,$s1);
1008	&sub	($key,$s3);
1009	&jmp	(&label("td_ok"));
1010	&set_label("td_break_out");	# else %esp -= (p-s)&0xfff + framesz;
1011	&sub	($s3,$s0);
1012	&and	($s3,0xfff);
1013	&add	($s3,64+256);
1014	&sub	($key,$s3);
1015	&align	(4);
1016	&set_label("td_ok");
1017
1018	&mov	($s0,&wparam(0));	# load inp
1019	&mov	($s1,&wparam(1));	# load out
1020	&mov	($s3,&wparam(3));	# load key
1021	&mov	($acc,&wparam(4));	# load ivp
1022
1023	&exch	("esp",$key);
1024	&add	("esp",4);		# reserve for return address!
1025	&mov	($_esp,$key);		# save %esp
1026
1027	&mov	($_inp,$s0);		# save copy of inp
1028	&mov	($_out,$s1);		# save copy of out
1029	&mov	($_len,$s2);		# save copy of len
1030	&mov	($_key,$s3);		# save copy of key
1031	&mov	($_ivp,$acc);		# save copy of ivp
1032
1033	if ($compromise) {
1034		&cmp	($s2,$compromise);
1035		&jb	(&label("skip_dcopy"));
1036	}
1037	# copy key schedule to stack
1038	&mov	("ecx",244/4);
1039	&mov	("esi",$s3);
1040	&lea	("edi",$aes_key);
1041	&mov	($_key,"edi");
1042	&align	(4);
1043	&data_word(0xF689A5F3);	# rep movsd
1044	&set_label("skip_dcopy") if ($compromise);
1045
1046	&mov	($acc,$s0);
1047	&mov	($key,24);
1048	&align	(4);
1049	&set_label("prefetch_td");
1050		&mov	($s0,&DWP(0,"ebp"));
1051		&mov	($s1,&DWP(32,"ebp"));
1052		&mov	($s2,&DWP(64,"ebp"));
1053		&mov	($s3,&DWP(96,"ebp"));
1054		&lea	("ebp",&DWP(128,"ebp"));
1055		&dec	($key);
1056	&jnz	(&label("prefetch_td"));
1057	&sub	("ebp",3072);
1058
1059	&cmp	($acc,$_out);
1060	&je	(&label("dec_in_place"));	# in-place processing...
1061
1062	&mov	($key,$_ivp);		# load ivp
1063	&mov	($_tmp,$key);
1064
1065	&align	(4);
1066	&set_label("dec_loop");
1067		&mov	($s0,&DWP(0,$acc));	# read input
1068		&mov	($s1,&DWP(4,$acc));
1069		&mov	($s2,&DWP(8,$acc));
1070		&mov	($s3,&DWP(12,$acc));
1071
1072		&mov	($key,$_key);		# load key
1073		&call	("_x86_AES_decrypt");
1074
1075		&mov	($key,$_tmp);		# load ivp
1076		&mov	($acc,$_len);		# load len
1077		&xor	($s0,&DWP(0,$key));	# xor iv
1078		&xor	($s1,&DWP(4,$key));
1079		&xor	($s2,&DWP(8,$key));
1080		&xor	($s3,&DWP(12,$key));
1081
1082		&sub	($acc,16);
1083		&jc	(&label("dec_partial"));
1084		&mov	($_len,$acc);		# save len
1085		&mov	($acc,$_inp);		# load inp
1086		&mov	($key,$_out);		# load out
1087
1088		&mov	(&DWP(0,$key),$s0);	# write output
1089		&mov	(&DWP(4,$key),$s1);
1090		&mov	(&DWP(8,$key),$s2);
1091		&mov	(&DWP(12,$key),$s3);
1092
1093		&mov	($_tmp,$acc);		# save ivp
1094		&lea	($acc,&DWP(16,$acc));
1095		&mov	($_inp,$acc);		# save inp
1096
1097		&lea	($key,&DWP(16,$key));
1098		&mov	($_out,$key);		# save out
1099
1100	&jnz	(&label("dec_loop"));
1101	&mov	($key,$_tmp);		# load temp ivp
1102    &set_label("dec_end");
1103	&mov	($acc,$_ivp);		# load user ivp
1104	&mov	($s0,&DWP(0,$key));	# load iv
1105	&mov	($s1,&DWP(4,$key));
1106	&mov	($s2,&DWP(8,$key));
1107	&mov	($s3,&DWP(12,$key));
1108	&mov	(&DWP(0,$acc),$s0);	# copy back to user
1109	&mov	(&DWP(4,$acc),$s1);
1110	&mov	(&DWP(8,$acc),$s2);
1111	&mov	(&DWP(12,$acc),$s3);
1112	&jmp	(&label("dec_out"));
1113
1114    &align	(4);
1115    &set_label("dec_partial");
1116	&lea	($key,$ivec);
1117	&mov	(&DWP(0,$key),$s0);	# dump output to stack
1118	&mov	(&DWP(4,$key),$s1);
1119	&mov	(&DWP(8,$key),$s2);
1120	&mov	(&DWP(12,$key),$s3);
1121	&lea	($s2 eq "ecx" ? $s2 : "",&DWP(16,$acc));
1122	&mov	($acc eq "esi" ? $acc : "",$key);
1123	&mov	($key eq "edi" ? $key : "",$_out);	# load out
1124	&data_word(0xF689A4F3);	# rep movsb		# copy output
1125	&mov	($key,$_inp);				# use inp as temp ivp
1126	&jmp	(&label("dec_end"));
1127
1128    &align	(4);
1129    &set_label("dec_in_place");
1130	&set_label("dec_in_place_loop");
1131		&lea	($key,$ivec);
1132		&mov	($s0,&DWP(0,$acc));	# read input
1133		&mov	($s1,&DWP(4,$acc));
1134		&mov	($s2,&DWP(8,$acc));
1135		&mov	($s3,&DWP(12,$acc));
1136
1137		&mov	(&DWP(0,$key),$s0);	# copy to temp
1138		&mov	(&DWP(4,$key),$s1);
1139		&mov	(&DWP(8,$key),$s2);
1140		&mov	(&DWP(12,$key),$s3);
1141
1142		&mov	($key,$_key);		# load key
1143		&call	("_x86_AES_decrypt");
1144
1145		&mov	($key,$_ivp);		# load ivp
1146		&mov	($acc,$_out);		# load out
1147		&xor	($s0,&DWP(0,$key));	# xor iv
1148		&xor	($s1,&DWP(4,$key));
1149		&xor	($s2,&DWP(8,$key));
1150		&xor	($s3,&DWP(12,$key));
1151
1152		&mov	(&DWP(0,$acc),$s0);	# write output
1153		&mov	(&DWP(4,$acc),$s1);
1154		&mov	(&DWP(8,$acc),$s2);
1155		&mov	(&DWP(12,$acc),$s3);
1156
1157		&lea	($acc,&DWP(16,$acc));
1158		&mov	($_out,$acc);		# save out
1159
1160		&lea	($acc,$ivec);
1161		&mov	($s0,&DWP(0,$acc));	# read temp
1162		&mov	($s1,&DWP(4,$acc));
1163		&mov	($s2,&DWP(8,$acc));
1164		&mov	($s3,&DWP(12,$acc));
1165
1166		&mov	(&DWP(0,$key),$s0);	# copy iv
1167		&mov	(&DWP(4,$key),$s1);
1168		&mov	(&DWP(8,$key),$s2);
1169		&mov	(&DWP(12,$key),$s3);
1170
1171		&mov	($acc,$_inp);		# load inp
1172
1173		&lea	($acc,&DWP(16,$acc));
1174		&mov	($_inp,$acc);		# save inp
1175
1176		&mov	($s2,$_len);		# load len
1177		&sub	($s2,16);
1178		&jc	(&label("dec_in_place_partial"));
1179		&mov	($_len,$s2);		# save len
1180	&jnz	(&label("dec_in_place_loop"));
1181	&jmp	(&label("dec_out"));
1182
1183    &align	(4);
1184    &set_label("dec_in_place_partial");
1185	# one can argue if this is actually required...
1186	&mov	($key eq "edi" ? $key : "",$_out);
1187	&lea	($acc eq "esi" ? $acc : "",$ivec);
1188	&lea	($key,&DWP(0,$key,$s2));
1189	&lea	($acc,&DWP(16,$acc,$s2));
1190	&neg	($s2 eq "ecx" ? $s2 : "");
1191	&data_word(0xF689A4F3);	# rep movsb	# restore tail
1192
1193    &align	(4);
1194    &set_label("dec_out");
1195    &mov	("edi",$_key);
1196    &mov	("esp",$_esp);
1197    if ($compromise) {
1198	&cmp	(&wparam(2),$compromise);
1199	&jb	(&label("skip_dzero"));
1200    }
1201    # zero copy of key schedule
1202    &mov	("ecx",240/4);
1203    &xor	("eax","eax");
1204    &align	(4);
1205    &data_word(0xF689ABF3);	# rep stosd
1206    &set_label("skip_dzero") if ($compromise);
1207    &popf	();
1208&function_end("AES_cbc_encrypt");
1209}
1210
1211#------------------------------------------------------------------#
1212
1213sub enckey()
1214{
1215	&movz	("esi",&LB("edx"));		# rk[i]>>0
1216	&mov	("ebx",&DWP(2,"ebp","esi",8));
1217	&movz	("esi",&HB("edx"));		# rk[i]>>8
1218	&and	("ebx",0xFF000000);
1219	&xor	("eax","ebx");
1220
1221	&mov	("ebx",&DWP(2,"ebp","esi",8));
1222	&shr	("edx",16);
1223	&and	("ebx",0x000000FF);
1224	&movz	("esi",&LB("edx"));		# rk[i]>>16
1225	&xor	("eax","ebx");
1226
1227	&mov	("ebx",&DWP(0,"ebp","esi",8));
1228	&movz	("esi",&HB("edx"));		# rk[i]>>24
1229	&and	("ebx",0x0000FF00);
1230	&xor	("eax","ebx");
1231
1232	&mov	("ebx",&DWP(0,"ebp","esi",8));
1233	&and	("ebx",0x00FF0000);
1234	&xor	("eax","ebx");
1235
1236	&xor	("eax",&DWP(2048,"ebp","ecx",4));	# rcon
1237}
1238
1239# int AES_set_encrypt_key(const unsigned char *userKey, const int bits,
1240#                        AES_KEY *key)
1241&public_label("AES_Te");
1242&function_begin("AES_set_encrypt_key");
1243	&mov	("esi",&wparam(0));		# user supplied key
1244	&mov	("edi",&wparam(2));		# private key schedule
1245
1246	&test	("esi",-1);
1247	&jz	(&label("badpointer"));
1248	&test	("edi",-1);
1249	&jz	(&label("badpointer"));
1250
1251	&call	(&label("pic_point"));
1252	&set_label("pic_point");
1253	&blindpop("ebp");
1254	&lea	("ebp",&DWP(&label("AES_Te")."-".&label("pic_point"),"ebp"));
1255
1256	&mov	("ecx",&wparam(1));		# number of bits in key
1257	&cmp	("ecx",128);
1258	&je	(&label("10rounds"));
1259	&cmp	("ecx",192);
1260	&je	(&label("12rounds"));
1261	&cmp	("ecx",256);
1262	&je	(&label("14rounds"));
1263	&mov	("eax",-2);			# invalid number of bits
1264	&jmp	(&label("exit"));
1265
1266    &set_label("10rounds");
1267	&mov	("eax",&DWP(0,"esi"));		# copy first 4 dwords
1268	&mov	("ebx",&DWP(4,"esi"));
1269	&mov	("ecx",&DWP(8,"esi"));
1270	&mov	("edx",&DWP(12,"esi"));
1271	&mov	(&DWP(0,"edi"),"eax");
1272	&mov	(&DWP(4,"edi"),"ebx");
1273	&mov	(&DWP(8,"edi"),"ecx");
1274	&mov	(&DWP(12,"edi"),"edx");
1275
1276	&xor	("ecx","ecx");
1277	&jmp	(&label("10shortcut"));
1278
1279	&align	(4);
1280	&set_label("10loop");
1281		&mov	("eax",&DWP(0,"edi"));		# rk[0]
1282		&mov	("edx",&DWP(12,"edi"));		# rk[3]
1283	&set_label("10shortcut");
1284		&enckey	();
1285
1286		&mov	(&DWP(16,"edi"),"eax");		# rk[4]
1287		&xor	("eax",&DWP(4,"edi"));
1288		&mov	(&DWP(20,"edi"),"eax");		# rk[5]
1289		&xor	("eax",&DWP(8,"edi"));
1290		&mov	(&DWP(24,"edi"),"eax");		# rk[6]
1291		&xor	("eax",&DWP(12,"edi"));
1292		&mov	(&DWP(28,"edi"),"eax");		# rk[7]
1293		&inc	("ecx");
1294		&add	("edi",16);
1295		&cmp	("ecx",10);
1296	&jl	(&label("10loop"));
1297
1298	&mov	(&DWP(80,"edi"),10);		# setup number of rounds
1299	&xor	("eax","eax");
1300	&jmp	(&label("exit"));
1301
1302    &set_label("12rounds");
1303	&mov	("eax",&DWP(0,"esi"));		# copy first 6 dwords
1304	&mov	("ebx",&DWP(4,"esi"));
1305	&mov	("ecx",&DWP(8,"esi"));
1306	&mov	("edx",&DWP(12,"esi"));
1307	&mov	(&DWP(0,"edi"),"eax");
1308	&mov	(&DWP(4,"edi"),"ebx");
1309	&mov	(&DWP(8,"edi"),"ecx");
1310	&mov	(&DWP(12,"edi"),"edx");
1311	&mov	("ecx",&DWP(16,"esi"));
1312	&mov	("edx",&DWP(20,"esi"));
1313	&mov	(&DWP(16,"edi"),"ecx");
1314	&mov	(&DWP(20,"edi"),"edx");
1315
1316	&xor	("ecx","ecx");
1317	&jmp	(&label("12shortcut"));
1318
1319	&align	(4);
1320	&set_label("12loop");
1321		&mov	("eax",&DWP(0,"edi"));		# rk[0]
1322		&mov	("edx",&DWP(20,"edi"));		# rk[5]
1323	&set_label("12shortcut");
1324		&enckey	();
1325
1326		&mov	(&DWP(24,"edi"),"eax");		# rk[6]
1327		&xor	("eax",&DWP(4,"edi"));
1328		&mov	(&DWP(28,"edi"),"eax");		# rk[7]
1329		&xor	("eax",&DWP(8,"edi"));
1330		&mov	(&DWP(32,"edi"),"eax");		# rk[8]
1331		&xor	("eax",&DWP(12,"edi"));
1332		&mov	(&DWP(36,"edi"),"eax");		# rk[9]
1333
1334		&cmp	("ecx",7);
1335		&je	(&label("12break"));
1336		&inc	("ecx");
1337
1338		&xor	("eax",&DWP(16,"edi"));
1339		&mov	(&DWP(40,"edi"),"eax");		# rk[10]
1340		&xor	("eax",&DWP(20,"edi"));
1341		&mov	(&DWP(44,"edi"),"eax");		# rk[11]
1342
1343		&add	("edi",24);
1344	&jmp	(&label("12loop"));
1345
1346	&set_label("12break");
1347	&mov	(&DWP(72,"edi"),12);		# setup number of rounds
1348	&xor	("eax","eax");
1349	&jmp	(&label("exit"));
1350
1351    &set_label("14rounds");
1352	&mov	("eax",&DWP(0,"esi"));		# copy first 8 dwords
1353	&mov	("ebx",&DWP(4,"esi"));
1354	&mov	("ecx",&DWP(8,"esi"));
1355	&mov	("edx",&DWP(12,"esi"));
1356	&mov	(&DWP(0,"edi"),"eax");
1357	&mov	(&DWP(4,"edi"),"ebx");
1358	&mov	(&DWP(8,"edi"),"ecx");
1359	&mov	(&DWP(12,"edi"),"edx");
1360	&mov	("eax",&DWP(16,"esi"));
1361	&mov	("ebx",&DWP(20,"esi"));
1362	&mov	("ecx",&DWP(24,"esi"));
1363	&mov	("edx",&DWP(28,"esi"));
1364	&mov	(&DWP(16,"edi"),"eax");
1365	&mov	(&DWP(20,"edi"),"ebx");
1366	&mov	(&DWP(24,"edi"),"ecx");
1367	&mov	(&DWP(28,"edi"),"edx");
1368
1369	&xor	("ecx","ecx");
1370	&jmp	(&label("14shortcut"));
1371
1372	&align	(4);
1373	&set_label("14loop");
1374		&mov	("edx",&DWP(28,"edi"));		# rk[7]
1375	&set_label("14shortcut");
1376		&mov	("eax",&DWP(0,"edi"));		# rk[0]
1377
1378		&enckey	();
1379
1380		&mov	(&DWP(32,"edi"),"eax");		# rk[8]
1381		&xor	("eax",&DWP(4,"edi"));
1382		&mov	(&DWP(36,"edi"),"eax");		# rk[9]
1383		&xor	("eax",&DWP(8,"edi"));
1384		&mov	(&DWP(40,"edi"),"eax");		# rk[10]
1385		&xor	("eax",&DWP(12,"edi"));
1386		&mov	(&DWP(44,"edi"),"eax");		# rk[11]
1387
1388		&cmp	("ecx",6);
1389		&je	(&label("14break"));
1390		&inc	("ecx");
1391
1392		&mov	("edx","eax");
1393		&mov	("eax",&DWP(16,"edi"));		# rk[4]
1394		&movz	("esi",&LB("edx"));		# rk[11]>>0
1395		&mov	("ebx",&DWP(2,"ebp","esi",8));
1396		&movz	("esi",&HB("edx"));		# rk[11]>>8
1397		&and	("ebx",0x000000FF);
1398		&xor	("eax","ebx");
1399
1400		&mov	("ebx",&DWP(0,"ebp","esi",8));
1401		&shr	("edx",16);
1402		&and	("ebx",0x0000FF00);
1403		&movz	("esi",&LB("edx"));		# rk[11]>>16
1404		&xor	("eax","ebx");
1405
1406		&mov	("ebx",&DWP(0,"ebp","esi",8));
1407		&movz	("esi",&HB("edx"));		# rk[11]>>24
1408		&and	("ebx",0x00FF0000);
1409		&xor	("eax","ebx");
1410
1411		&mov	("ebx",&DWP(2,"ebp","esi",8));
1412		&and	("ebx",0xFF000000);
1413		&xor	("eax","ebx");
1414
1415		&mov	(&DWP(48,"edi"),"eax");		# rk[12]
1416		&xor	("eax",&DWP(20,"edi"));
1417		&mov	(&DWP(52,"edi"),"eax");		# rk[13]
1418		&xor	("eax",&DWP(24,"edi"));
1419		&mov	(&DWP(56,"edi"),"eax");		# rk[14]
1420		&xor	("eax",&DWP(28,"edi"));
1421		&mov	(&DWP(60,"edi"),"eax");		# rk[15]
1422
1423		&add	("edi",32);
1424	&jmp	(&label("14loop"));
1425
1426	&set_label("14break");
1427	&mov	(&DWP(48,"edi"),14);		# setup number of rounds
1428	&xor	("eax","eax");
1429	&jmp	(&label("exit"));
1430
1431    &set_label("badpointer");
1432	&mov	("eax",-1);
1433    &set_label("exit");
1434&function_end("AES_set_encrypt_key");
1435
1436sub deckey()
1437{ my ($i,$ptr,$te,$td) = @_;
1438
1439	&mov	("eax",&DWP($i,$ptr));
1440	&mov	("edx","eax");
1441	&movz	("ebx",&HB("eax"));
1442	&shr	("edx",16);
1443	&and	("eax",0xFF);
1444	&movz	("eax",&BP(2,$te,"eax",8));
1445	&movz	("ebx",&BP(2,$te,"ebx",8));
1446	&mov	("eax",&DWP(0,$td,"eax",8));
1447	&xor	("eax",&DWP(3,$td,"ebx",8));
1448	&movz	("ebx",&HB("edx"));
1449	&and	("edx",0xFF);
1450	&movz	("edx",&BP(2,$te,"edx",8));
1451	&movz	("ebx",&BP(2,$te,"ebx",8));
1452	&xor	("eax",&DWP(2,$td,"edx",8));
1453	&xor	("eax",&DWP(1,$td,"ebx",8));
1454	&mov	(&DWP($i,$ptr),"eax");
1455}
1456
1457# int AES_set_decrypt_key(const unsigned char *userKey, const int bits,
1458#                        AES_KEY *key)
1459&public_label("AES_Td");
1460&public_label("AES_Te");
1461&function_begin_B("AES_set_decrypt_key");
1462	&mov	("eax",&wparam(0));
1463	&mov	("ecx",&wparam(1));
1464	&mov	("edx",&wparam(2));
1465	&sub	("esp",12);
1466	&mov	(&DWP(0,"esp"),"eax");
1467	&mov	(&DWP(4,"esp"),"ecx");
1468	&mov	(&DWP(8,"esp"),"edx");
1469	&call	("AES_set_encrypt_key");
1470	&add	("esp",12);
1471	&cmp	("eax",0);
1472	&je	(&label("proceed"));
1473	&ret	();
1474
1475    &set_label("proceed");
1476	&push	("ebp");
1477	&push	("ebx");
1478	&push	("esi");
1479	&push	("edi");
1480
1481	&mov	("esi",&wparam(2));
1482	&mov	("ecx",&DWP(240,"esi"));	# pull number of rounds
1483	&lea	("ecx",&DWP(0,"","ecx",4));
1484	&lea	("edi",&DWP(0,"esi","ecx",4));	# pointer to last chunk
1485
1486	&align	(4);
1487	&set_label("invert");			# invert order of chunks
1488		&mov	("eax",&DWP(0,"esi"));
1489		&mov	("ebx",&DWP(4,"esi"));
1490		&mov	("ecx",&DWP(0,"edi"));
1491		&mov	("edx",&DWP(4,"edi"));
1492		&mov	(&DWP(0,"edi"),"eax");
1493		&mov	(&DWP(4,"edi"),"ebx");
1494		&mov	(&DWP(0,"esi"),"ecx");
1495		&mov	(&DWP(4,"esi"),"edx");
1496		&mov	("eax",&DWP(8,"esi"));
1497		&mov	("ebx",&DWP(12,"esi"));
1498		&mov	("ecx",&DWP(8,"edi"));
1499		&mov	("edx",&DWP(12,"edi"));
1500		&mov	(&DWP(8,"edi"),"eax");
1501		&mov	(&DWP(12,"edi"),"ebx");
1502		&mov	(&DWP(8,"esi"),"ecx");
1503		&mov	(&DWP(12,"esi"),"edx");
1504		&add	("esi",16);
1505		&sub	("edi",16);
1506		&cmp	("esi","edi");
1507	&jne	(&label("invert"));
1508
1509	&call	(&label("pic_point"));
1510	&set_label("pic_point");
1511	blindpop("ebp");
1512	&lea	("edi",&DWP(&label("AES_Td")."-".&label("pic_point"),"ebp"));
1513	&lea	("ebp",&DWP(&label("AES_Te")."-".&label("pic_point"),"ebp"));
1514
1515	&mov	("esi",&wparam(2));
1516	&mov	("ecx",&DWP(240,"esi"));	# pull number of rounds
1517	&dec	("ecx");
1518	&align	(4);
1519	&set_label("permute");			# permute the key schedule
1520		&add	("esi",16);
1521		&deckey	(0,"esi","ebp","edi");
1522		&deckey	(4,"esi","ebp","edi");
1523		&deckey	(8,"esi","ebp","edi");
1524		&deckey	(12,"esi","ebp","edi");
1525		&dec	("ecx");
1526	&jnz	(&label("permute"));
1527
1528	&xor	("eax","eax");			# return success
1529&function_end("AES_set_decrypt_key");
1530
1531&asm_finish();
1532