aes-x86_64.pl revision 221304ee937bc0910948a8be1320cb8cc4eb6d36 
1#!/usr/bin/env perl
2#
3# ====================================================================
4# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
5# project. The module is, however, dual licensed under OpenSSL and
6# CRYPTOGAMS licenses depending on where you obtain it. For further
7# details see http://www.openssl.org/~appro/cryptogams/.
8# ====================================================================
9#
10# Version 2.1.
11#
12# aes-*-cbc benchmarks are improved by >70% [compared to gcc 3.3.2 on
13# Opteron 240 CPU] plus all the bells-n-whistles from 32-bit version
14# [you'll notice a lot of resemblance], such as compressed S-boxes
15# in little-endian byte order, prefetch of these tables in CBC mode,
16# as well as avoiding L1 cache aliasing between stack frame and key
17# schedule and already mentioned tables, compressed Td4...
18#
19# Performance in number of cycles per processed byte for 128-bit key:
20#
21#		ECB encrypt	ECB decrypt	CBC large chunk
22# AMD64		33		41		13.0
23# EM64T		38		59		18.6(*)
24# Core 2	30		43		14.5(*)
25#
26# (*) with hyper-threading off
27
28$flavour = shift;
29$output  = shift;
30if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
31
32$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
33
34$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
35( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
36( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
37die "can't locate x86_64-xlate.pl";
38
39open STDOUT,"| $^X $xlate $flavour $output";
40
41$verticalspin=1;	# unlike 32-bit version $verticalspin performs
42			# ~15% better on both AMD and Intel cores
43$speed_limit=512;	# see aes-586.pl for details
44
45$code=".text\n";
46
47$s0="%eax";
48$s1="%ebx";
49$s2="%ecx";
50$s3="%edx";
51$acc0="%esi";	$mask80="%rsi";
52$acc1="%edi";	$maskfe="%rdi";
53$acc2="%ebp";	$mask1b="%rbp";
54$inp="%r8";
55$out="%r9";
56$t0="%r10d";
57$t1="%r11d";
58$t2="%r12d";
59$rnds="%r13d";
60$sbox="%r14";
61$key="%r15";
62
63sub hi() { my $r=shift;	$r =~ s/%[er]([a-d])x/%\1h/;	$r; }
64sub lo() { my $r=shift;	$r =~ s/%[er]([a-d])x/%\1l/;
65			$r =~ s/%[er]([sd]i)/%\1l/;
66			$r =~ s/%(r[0-9]+)[d]?/%\1b/;	$r; }
67sub LO() { my $r=shift; $r =~ s/%r([a-z]+)/%e\1/;
68			$r =~ s/%r([0-9]+)/%r\1d/;	$r; }
69sub _data_word()
70{ my $i;
71    while(defined($i=shift)) { $code.=sprintf".long\t0x%08x,0x%08x\n",$i,$i; }
72}
73sub data_word()
74{ my $i;
75  my $last=pop(@_);
76    $code.=".long\t";
77    while(defined($i=shift)) { $code.=sprintf"0x%08x,",$i; }
78    $code.=sprintf"0x%08x\n",$last;
79}
80
81sub data_byte()
82{ my $i;
83  my $last=pop(@_);
84    $code.=".byte\t";
85    while(defined($i=shift)) { $code.=sprintf"0x%02x,",$i&0xff; }
86    $code.=sprintf"0x%02x\n",$last&0xff;
87}
88
89sub encvert()
90{ my $t3="%r8d";	# zaps $inp!
91
92$code.=<<___;
93	# favor 3-way issue Opteron pipeline...
94	movzb	`&lo("$s0")`,$acc0
95	movzb	`&lo("$s1")`,$acc1
96	movzb	`&lo("$s2")`,$acc2
97	mov	0($sbox,$acc0,8),$t0
98	mov	0($sbox,$acc1,8),$t1
99	mov	0($sbox,$acc2,8),$t2
100
101	movzb	`&hi("$s1")`,$acc0
102	movzb	`&hi("$s2")`,$acc1
103	movzb	`&lo("$s3")`,$acc2
104	xor	3($sbox,$acc0,8),$t0
105	xor	3($sbox,$acc1,8),$t1
106	mov	0($sbox,$acc2,8),$t3
107
108	movzb	`&hi("$s3")`,$acc0
109	shr	\$16,$s2
110	movzb	`&hi("$s0")`,$acc2
111	xor	3($sbox,$acc0,8),$t2
112	shr	\$16,$s3
113	xor	3($sbox,$acc2,8),$t3
114
115	shr	\$16,$s1
116	lea	16($key),$key
117	shr	\$16,$s0
118
119	movzb	`&lo("$s2")`,$acc0
120	movzb	`&lo("$s3")`,$acc1
121	movzb	`&lo("$s0")`,$acc2
122	xor	2($sbox,$acc0,8),$t0
123	xor	2($sbox,$acc1,8),$t1
124	xor	2($sbox,$acc2,8),$t2
125
126	movzb	`&hi("$s3")`,$acc0
127	movzb	`&hi("$s0")`,$acc1
128	movzb	`&lo("$s1")`,$acc2
129	xor	1($sbox,$acc0,8),$t0
130	xor	1($sbox,$acc1,8),$t1
131	xor	2($sbox,$acc2,8),$t3
132
133	mov	12($key),$s3
134	movzb	`&hi("$s1")`,$acc1
135	movzb	`&hi("$s2")`,$acc2
136	mov	0($key),$s0
137	xor	1($sbox,$acc1,8),$t2
138	xor	1($sbox,$acc2,8),$t3
139
140	mov	4($key),$s1
141	mov	8($key),$s2
142	xor	$t0,$s0
143	xor	$t1,$s1
144	xor	$t2,$s2
145	xor	$t3,$s3
146___
147}
148
149sub enclastvert()
150{ my $t3="%r8d";	# zaps $inp!
151
152$code.=<<___;
153	movzb	`&lo("$s0")`,$acc0
154	movzb	`&lo("$s1")`,$acc1
155	movzb	`&lo("$s2")`,$acc2
156	movzb	2($sbox,$acc0,8),$t0
157	movzb	2($sbox,$acc1,8),$t1
158	movzb	2($sbox,$acc2,8),$t2
159
160	movzb	`&lo("$s3")`,$acc0
161	movzb	`&hi("$s1")`,$acc1
162	movzb	`&hi("$s2")`,$acc2
163	movzb	2($sbox,$acc0,8),$t3
164	mov	0($sbox,$acc1,8),$acc1	#$t0
165	mov	0($sbox,$acc2,8),$acc2	#$t1
166
167	and	\$0x0000ff00,$acc1
168	and	\$0x0000ff00,$acc2
169
170	xor	$acc1,$t0
171	xor	$acc2,$t1
172	shr	\$16,$s2
173
174	movzb	`&hi("$s3")`,$acc0
175	movzb	`&hi("$s0")`,$acc1
176	shr	\$16,$s3
177	mov	0($sbox,$acc0,8),$acc0	#$t2
178	mov	0($sbox,$acc1,8),$acc1	#$t3
179
180	and	\$0x0000ff00,$acc0
181	and	\$0x0000ff00,$acc1
182	shr	\$16,$s1
183	xor	$acc0,$t2
184	xor	$acc1,$t3
185	shr	\$16,$s0
186
187	movzb	`&lo("$s2")`,$acc0
188	movzb	`&lo("$s3")`,$acc1
189	movzb	`&lo("$s0")`,$acc2
190	mov	0($sbox,$acc0,8),$acc0	#$t0
191	mov	0($sbox,$acc1,8),$acc1	#$t1
192	mov	0($sbox,$acc2,8),$acc2	#$t2
193
194	and	\$0x00ff0000,$acc0
195	and	\$0x00ff0000,$acc1
196	and	\$0x00ff0000,$acc2
197
198	xor	$acc0,$t0
199	xor	$acc1,$t1
200	xor	$acc2,$t2
201
202	movzb	`&lo("$s1")`,$acc0
203	movzb	`&hi("$s3")`,$acc1
204	movzb	`&hi("$s0")`,$acc2
205	mov	0($sbox,$acc0,8),$acc0	#$t3
206	mov	2($sbox,$acc1,8),$acc1	#$t0
207	mov	2($sbox,$acc2,8),$acc2	#$t1
208
209	and	\$0x00ff0000,$acc0
210	and	\$0xff000000,$acc1
211	and	\$0xff000000,$acc2
212
213	xor	$acc0,$t3
214	xor	$acc1,$t0
215	xor	$acc2,$t1
216
217	movzb	`&hi("$s1")`,$acc0
218	movzb	`&hi("$s2")`,$acc1
219	mov	16+12($key),$s3
220	mov	2($sbox,$acc0,8),$acc0	#$t2
221	mov	2($sbox,$acc1,8),$acc1	#$t3
222	mov	16+0($key),$s0
223
224	and	\$0xff000000,$acc0
225	and	\$0xff000000,$acc1
226
227	xor	$acc0,$t2
228	xor	$acc1,$t3
229
230	mov	16+4($key),$s1
231	mov	16+8($key),$s2
232	xor	$t0,$s0
233	xor	$t1,$s1
234	xor	$t2,$s2
235	xor	$t3,$s3
236___
237}
238
239sub encstep()
240{ my ($i,@s) = @_;
241  my $tmp0=$acc0;
242  my $tmp1=$acc1;
243  my $tmp2=$acc2;
244  my $out=($t0,$t1,$t2,$s[0])[$i];
245
246	if ($i==3) {
247		$tmp0=$s[1];
248		$tmp1=$s[2];
249		$tmp2=$s[3];
250	}
251	$code.="	movzb	".&lo($s[0]).",$out\n";
252	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
253	$code.="	lea	16($key),$key\n"	if ($i==0);
254
255	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
256	$code.="	mov	0($sbox,$out,8),$out\n";
257
258	$code.="	shr	\$16,$tmp1\n";
259	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
260	$code.="	xor	3($sbox,$tmp0,8),$out\n";
261
262	$code.="	movzb	".&lo($tmp1).",$tmp1\n";
263	$code.="	shr	\$24,$tmp2\n";
264	$code.="	xor	4*$i($key),$out\n";
265
266	$code.="	xor	2($sbox,$tmp1,8),$out\n";
267	$code.="	xor	1($sbox,$tmp2,8),$out\n";
268
269	$code.="	mov	$t0,$s[1]\n"		if ($i==3);
270	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
271	$code.="	mov	$t2,$s[3]\n"		if ($i==3);
272	$code.="\n";
273}
274
275sub enclast()
276{ my ($i,@s)=@_;
277  my $tmp0=$acc0;
278  my $tmp1=$acc1;
279  my $tmp2=$acc2;
280  my $out=($t0,$t1,$t2,$s[0])[$i];
281
282	if ($i==3) {
283		$tmp0=$s[1];
284		$tmp1=$s[2];
285		$tmp2=$s[3];
286	}
287	$code.="	movzb	".&lo($s[0]).",$out\n";
288	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
289
290	$code.="	mov	2($sbox,$out,8),$out\n";
291	$code.="	shr	\$16,$tmp1\n";
292	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
293
294	$code.="	and	\$0x000000ff,$out\n";
295	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
296	$code.="	movzb	".&lo($tmp1).",$tmp1\n";
297	$code.="	shr	\$24,$tmp2\n";
298
299	$code.="	mov	0($sbox,$tmp0,8),$tmp0\n";
300	$code.="	mov	0($sbox,$tmp1,8),$tmp1\n";
301	$code.="	mov	2($sbox,$tmp2,8),$tmp2\n";
302
303	$code.="	and	\$0x0000ff00,$tmp0\n";
304	$code.="	and	\$0x00ff0000,$tmp1\n";
305	$code.="	and	\$0xff000000,$tmp2\n";
306
307	$code.="	xor	$tmp0,$out\n";
308	$code.="	mov	$t0,$s[1]\n"		if ($i==3);
309	$code.="	xor	$tmp1,$out\n";
310	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
311	$code.="	xor	$tmp2,$out\n";
312	$code.="	mov	$t2,$s[3]\n"		if ($i==3);
313	$code.="\n";
314}
315
316$code.=<<___;
317.type	_x86_64_AES_encrypt,\@abi-omnipotent
318.align	16
319_x86_64_AES_encrypt:
320	xor	0($key),$s0			# xor with key
321	xor	4($key),$s1
322	xor	8($key),$s2
323	xor	12($key),$s3
324
325	mov	240($key),$rnds			# load key->rounds
326	sub	\$1,$rnds
327	jmp	.Lenc_loop
328.align	16
329.Lenc_loop:
330___
331	if ($verticalspin) { &encvert(); }
332	else {	&encstep(0,$s0,$s1,$s2,$s3);
333		&encstep(1,$s1,$s2,$s3,$s0);
334		&encstep(2,$s2,$s3,$s0,$s1);
335		&encstep(3,$s3,$s0,$s1,$s2);
336	}
337$code.=<<___;
338	sub	\$1,$rnds
339	jnz	.Lenc_loop
340___
341	if ($verticalspin) { &enclastvert(); }
342	else {	&enclast(0,$s0,$s1,$s2,$s3);
343		&enclast(1,$s1,$s2,$s3,$s0);
344		&enclast(2,$s2,$s3,$s0,$s1);
345		&enclast(3,$s3,$s0,$s1,$s2);
346		$code.=<<___;
347		xor	16+0($key),$s0		# xor with key
348		xor	16+4($key),$s1
349		xor	16+8($key),$s2
350		xor	16+12($key),$s3
351___
352	}
353$code.=<<___;
354	.byte	0xf3,0xc3			# rep ret
355.size	_x86_64_AES_encrypt,.-_x86_64_AES_encrypt
356___
357
358# it's possible to implement this by shifting tN by 8, filling least
359# significant byte with byte load and finally bswap-ing at the end,
360# but such partial register load kills Core 2...
361sub enccompactvert()
362{ my ($t3,$t4,$t5)=("%r8d","%r9d","%r13d");
363
364$code.=<<___;
365	movzb	`&lo("$s0")`,$t0
366	movzb	`&lo("$s1")`,$t1
367	movzb	`&lo("$s2")`,$t2
368	movzb	($sbox,$t0,1),$t0
369	movzb	($sbox,$t1,1),$t1
370	movzb	($sbox,$t2,1),$t2
371
372	movzb	`&lo("$s3")`,$t3
373	movzb	`&hi("$s1")`,$acc0
374	movzb	`&hi("$s2")`,$acc1
375	movzb	($sbox,$t3,1),$t3
376	movzb	($sbox,$acc0,1),$t4	#$t0
377	movzb	($sbox,$acc1,1),$t5	#$t1
378
379	movzb	`&hi("$s3")`,$acc2
380	movzb	`&hi("$s0")`,$acc0
381	shr	\$16,$s2
382	movzb	($sbox,$acc2,1),$acc2	#$t2
383	movzb	($sbox,$acc0,1),$acc0	#$t3
384	shr	\$16,$s3
385
386	movzb	`&lo("$s2")`,$acc1
387	shl	\$8,$t4
388	shl	\$8,$t5
389	movzb	($sbox,$acc1,1),$acc1	#$t0
390	xor	$t4,$t0
391	xor	$t5,$t1
392
393	movzb	`&lo("$s3")`,$t4
394	shr	\$16,$s0
395	shr	\$16,$s1
396	movzb	`&lo("$s0")`,$t5
397	shl	\$8,$acc2
398	shl	\$8,$acc0
399	movzb	($sbox,$t4,1),$t4	#$t1
400	movzb	($sbox,$t5,1),$t5	#$t2
401	xor	$acc2,$t2
402	xor	$acc0,$t3
403
404	movzb	`&lo("$s1")`,$acc2
405	movzb	`&hi("$s3")`,$acc0
406	shl	\$16,$acc1
407	movzb	($sbox,$acc2,1),$acc2	#$t3
408	movzb	($sbox,$acc0,1),$acc0	#$t0
409	xor	$acc1,$t0
410
411	movzb	`&hi("$s0")`,$acc1
412	shr	\$8,$s2
413	shr	\$8,$s1
414	movzb	($sbox,$acc1,1),$acc1	#$t1
415	movzb	($sbox,$s2,1),$s3	#$t3
416	movzb	($sbox,$s1,1),$s2	#$t2
417	shl	\$16,$t4
418	shl	\$16,$t5
419	shl	\$16,$acc2
420	xor	$t4,$t1
421	xor	$t5,$t2
422	xor	$acc2,$t3
423
424	shl	\$24,$acc0
425	shl	\$24,$acc1
426	shl	\$24,$s3
427	xor	$acc0,$t0
428	shl	\$24,$s2
429	xor	$acc1,$t1
430	mov	$t0,$s0
431	mov	$t1,$s1
432	xor	$t2,$s2
433	xor	$t3,$s3
434___
435}
436
437sub enctransform_ref()
438{ my $sn = shift;
439  my ($acc,$r2,$tmp)=("%r8d","%r9d","%r13d");
440
441$code.=<<___;
442	mov	$sn,$acc
443	and	\$0x80808080,$acc
444	mov	$acc,$tmp
445	shr	\$7,$tmp
446	lea	($sn,$sn),$r2
447	sub	$tmp,$acc
448	and	\$0xfefefefe,$r2
449	and	\$0x1b1b1b1b,$acc
450	mov	$sn,$tmp
451	xor	$acc,$r2
452
453	xor	$r2,$sn
454	rol	\$24,$sn
455	xor	$r2,$sn
456	ror	\$16,$tmp
457	xor	$tmp,$sn
458	ror	\$8,$tmp
459	xor	$tmp,$sn
460___
461}
462
463# unlike decrypt case it does not pay off to parallelize enctransform
464sub enctransform()
465{ my ($t3,$r20,$r21)=($acc2,"%r8d","%r9d");
466
467$code.=<<___;
468	mov	$s0,$acc0
469	mov	$s1,$acc1
470	and	\$0x80808080,$acc0
471	and	\$0x80808080,$acc1
472	mov	$acc0,$t0
473	mov	$acc1,$t1
474	shr	\$7,$t0
475	lea	($s0,$s0),$r20
476	shr	\$7,$t1
477	lea	($s1,$s1),$r21
478	sub	$t0,$acc0
479	sub	$t1,$acc1
480	and	\$0xfefefefe,$r20
481	and	\$0xfefefefe,$r21
482	and	\$0x1b1b1b1b,$acc0
483	and	\$0x1b1b1b1b,$acc1
484	mov	$s0,$t0
485	mov	$s1,$t1
486	xor	$acc0,$r20
487	xor	$acc1,$r21
488
489	xor	$r20,$s0
490	xor	$r21,$s1
491	 mov	$s2,$acc0
492	 mov	$s3,$acc1
493	rol	\$24,$s0
494	rol	\$24,$s1
495	 and	\$0x80808080,$acc0
496	 and	\$0x80808080,$acc1
497	xor	$r20,$s0
498	xor	$r21,$s1
499	 mov	$acc0,$t2
500	 mov	$acc1,$t3
501	ror	\$16,$t0
502	ror	\$16,$t1
503	 shr	\$7,$t2
504	 lea	($s2,$s2),$r20
505	xor	$t0,$s0
506	xor	$t1,$s1
507	 shr	\$7,$t3
508	 lea	($s3,$s3),$r21
509	ror	\$8,$t0
510	ror	\$8,$t1
511	 sub	$t2,$acc0
512	 sub	$t3,$acc1
513	xor	$t0,$s0
514	xor	$t1,$s1
515
516	and	\$0xfefefefe,$r20
517	and	\$0xfefefefe,$r21
518	and	\$0x1b1b1b1b,$acc0
519	and	\$0x1b1b1b1b,$acc1
520	mov	$s2,$t2
521	mov	$s3,$t3
522	xor	$acc0,$r20
523	xor	$acc1,$r21
524
525	xor	$r20,$s2
526	xor	$r21,$s3
527	rol	\$24,$s2
528	rol	\$24,$s3
529	xor	$r20,$s2
530	xor	$r21,$s3
531	mov	0($sbox),$acc0			# prefetch Te4
532	ror	\$16,$t2
533	ror	\$16,$t3
534	mov	64($sbox),$acc1
535	xor	$t2,$s2
536	xor	$t3,$s3
537	mov	128($sbox),$r20
538	ror	\$8,$t2
539	ror	\$8,$t3
540	mov	192($sbox),$r21
541	xor	$t2,$s2
542	xor	$t3,$s3
543___
544}
545
546$code.=<<___;
547.type	_x86_64_AES_encrypt_compact,\@abi-omnipotent
548.align	16
549_x86_64_AES_encrypt_compact:
550	lea	128($sbox),$inp			# size optimization
551	mov	0-128($inp),$acc1		# prefetch Te4
552	mov	32-128($inp),$acc2
553	mov	64-128($inp),$t0
554	mov	96-128($inp),$t1
555	mov	128-128($inp),$acc1
556	mov	160-128($inp),$acc2
557	mov	192-128($inp),$t0
558	mov	224-128($inp),$t1
559	jmp	.Lenc_loop_compact
560.align	16
561.Lenc_loop_compact:
562		xor	0($key),$s0		# xor with key
563		xor	4($key),$s1
564		xor	8($key),$s2
565		xor	12($key),$s3
566		lea	16($key),$key
567___
568		&enccompactvert();
569$code.=<<___;
570		cmp	16(%rsp),$key
571		je	.Lenc_compact_done
572___
573		&enctransform();
574$code.=<<___;
575	jmp	.Lenc_loop_compact
576.align	16
577.Lenc_compact_done:
578	xor	0($key),$s0
579	xor	4($key),$s1
580	xor	8($key),$s2
581	xor	12($key),$s3
582	.byte	0xf3,0xc3			# rep ret
583.size	_x86_64_AES_encrypt_compact,.-_x86_64_AES_encrypt_compact
584___
585
586# void AES_encrypt (const void *inp,void *out,const AES_KEY *key);
587$code.=<<___;
588.globl	AES_encrypt
589.type	AES_encrypt,\@function,3
590.align	16
591AES_encrypt:
592	push	%rbx
593	push	%rbp
594	push	%r12
595	push	%r13
596	push	%r14
597	push	%r15
598
599	# allocate frame "above" key schedule
600	mov	%rsp,%r10
601	lea	-63(%rdx),%rcx	# %rdx is key argument
602	and	\$-64,%rsp
603	sub	%rsp,%rcx
604	neg	%rcx
605	and	\$0x3c0,%rcx
606	sub	%rcx,%rsp
607	sub	\$32,%rsp
608
609	mov	%rsi,16(%rsp)	# save out
610	mov	%r10,24(%rsp)	# save real stack pointer
611.Lenc_prologue:
612
613	mov	%rdx,$key
614	mov	240($key),$rnds	# load rounds
615
616	mov	0(%rdi),$s0	# load input vector
617	mov	4(%rdi),$s1
618	mov	8(%rdi),$s2
619	mov	12(%rdi),$s3
620
621	shl	\$4,$rnds
622	lea	($key,$rnds),%rbp
623	mov	$key,(%rsp)	# key schedule
624	mov	%rbp,8(%rsp)	# end of key schedule
625
626	# pick Te4 copy which can't "overlap" with stack frame or key schedule
627	lea	.LAES_Te+2048(%rip),$sbox
628	lea	768(%rsp),%rbp
629	sub	$sbox,%rbp
630	and	\$0x300,%rbp
631	lea	($sbox,%rbp),$sbox
632
633	call	_x86_64_AES_encrypt_compact
634
635	mov	16(%rsp),$out	# restore out
636	mov	24(%rsp),%rsi	# restore saved stack pointer
637	mov	$s0,0($out)	# write output vector
638	mov	$s1,4($out)
639	mov	$s2,8($out)
640	mov	$s3,12($out)
641
642	mov	(%rsi),%r15
643	mov	8(%rsi),%r14
644	mov	16(%rsi),%r13
645	mov	24(%rsi),%r12
646	mov	32(%rsi),%rbp
647	mov	40(%rsi),%rbx
648	lea	48(%rsi),%rsp
649.Lenc_epilogue:
650	ret
651.size	AES_encrypt,.-AES_encrypt
652___
653
654#------------------------------------------------------------------#
655
656sub decvert()
657{ my $t3="%r8d";	# zaps $inp!
658
659$code.=<<___;
660	# favor 3-way issue Opteron pipeline...
661	movzb	`&lo("$s0")`,$acc0
662	movzb	`&lo("$s1")`,$acc1
663	movzb	`&lo("$s2")`,$acc2
664	mov	0($sbox,$acc0,8),$t0
665	mov	0($sbox,$acc1,8),$t1
666	mov	0($sbox,$acc2,8),$t2
667
668	movzb	`&hi("$s3")`,$acc0
669	movzb	`&hi("$s0")`,$acc1
670	movzb	`&lo("$s3")`,$acc2
671	xor	3($sbox,$acc0,8),$t0
672	xor	3($sbox,$acc1,8),$t1
673	mov	0($sbox,$acc2,8),$t3
674
675	movzb	`&hi("$s1")`,$acc0
676	shr	\$16,$s0
677	movzb	`&hi("$s2")`,$acc2
678	xor	3($sbox,$acc0,8),$t2
679	shr	\$16,$s3
680	xor	3($sbox,$acc2,8),$t3
681
682	shr	\$16,$s1
683	lea	16($key),$key
684	shr	\$16,$s2
685
686	movzb	`&lo("$s2")`,$acc0
687	movzb	`&lo("$s3")`,$acc1
688	movzb	`&lo("$s0")`,$acc2
689	xor	2($sbox,$acc0,8),$t0
690	xor	2($sbox,$acc1,8),$t1
691	xor	2($sbox,$acc2,8),$t2
692
693	movzb	`&hi("$s1")`,$acc0
694	movzb	`&hi("$s2")`,$acc1
695	movzb	`&lo("$s1")`,$acc2
696	xor	1($sbox,$acc0,8),$t0
697	xor	1($sbox,$acc1,8),$t1
698	xor	2($sbox,$acc2,8),$t3
699
700	movzb	`&hi("$s3")`,$acc0
701	mov	12($key),$s3
702	movzb	`&hi("$s0")`,$acc2
703	xor	1($sbox,$acc0,8),$t2
704	mov	0($key),$s0
705	xor	1($sbox,$acc2,8),$t3
706
707	xor	$t0,$s0
708	mov	4($key),$s1
709	mov	8($key),$s2
710	xor	$t2,$s2
711	xor	$t1,$s1
712	xor	$t3,$s3
713___
714}
715
716sub declastvert()
717{ my $t3="%r8d";	# zaps $inp!
718
719$code.=<<___;
720	lea	2048($sbox),$sbox	# size optimization
721	movzb	`&lo("$s0")`,$acc0
722	movzb	`&lo("$s1")`,$acc1
723	movzb	`&lo("$s2")`,$acc2
724	movzb	($sbox,$acc0,1),$t0
725	movzb	($sbox,$acc1,1),$t1
726	movzb	($sbox,$acc2,1),$t2
727
728	movzb	`&lo("$s3")`,$acc0
729	movzb	`&hi("$s3")`,$acc1
730	movzb	`&hi("$s0")`,$acc2
731	movzb	($sbox,$acc0,1),$t3
732	movzb	($sbox,$acc1,1),$acc1	#$t0
733	movzb	($sbox,$acc2,1),$acc2	#$t1
734
735	shl	\$8,$acc1
736	shl	\$8,$acc2
737
738	xor	$acc1,$t0
739	xor	$acc2,$t1
740	shr	\$16,$s3
741
742	movzb	`&hi("$s1")`,$acc0
743	movzb	`&hi("$s2")`,$acc1
744	shr	\$16,$s0
745	movzb	($sbox,$acc0,1),$acc0	#$t2
746	movzb	($sbox,$acc1,1),$acc1	#$t3
747
748	shl	\$8,$acc0
749	shl	\$8,$acc1
750	shr	\$16,$s1
751	xor	$acc0,$t2
752	xor	$acc1,$t3
753	shr	\$16,$s2
754
755	movzb	`&lo("$s2")`,$acc0
756	movzb	`&lo("$s3")`,$acc1
757	movzb	`&lo("$s0")`,$acc2
758	movzb	($sbox,$acc0,1),$acc0	#$t0
759	movzb	($sbox,$acc1,1),$acc1	#$t1
760	movzb	($sbox,$acc2,1),$acc2	#$t2
761
762	shl	\$16,$acc0
763	shl	\$16,$acc1
764	shl	\$16,$acc2
765
766	xor	$acc0,$t0
767	xor	$acc1,$t1
768	xor	$acc2,$t2
769
770	movzb	`&lo("$s1")`,$acc0
771	movzb	`&hi("$s1")`,$acc1
772	movzb	`&hi("$s2")`,$acc2
773	movzb	($sbox,$acc0,1),$acc0	#$t3
774	movzb	($sbox,$acc1,1),$acc1	#$t0
775	movzb	($sbox,$acc2,1),$acc2	#$t1
776
777	shl	\$16,$acc0
778	shl	\$24,$acc1
779	shl	\$24,$acc2
780
781	xor	$acc0,$t3
782	xor	$acc1,$t0
783	xor	$acc2,$t1
784
785	movzb	`&hi("$s3")`,$acc0
786	movzb	`&hi("$s0")`,$acc1
787	mov	16+12($key),$s3
788	movzb	($sbox,$acc0,1),$acc0	#$t2
789	movzb	($sbox,$acc1,1),$acc1	#$t3
790	mov	16+0($key),$s0
791
792	shl	\$24,$acc0
793	shl	\$24,$acc1
794
795	xor	$acc0,$t2
796	xor	$acc1,$t3
797
798	mov	16+4($key),$s1
799	mov	16+8($key),$s2
800	lea	-2048($sbox),$sbox
801	xor	$t0,$s0
802	xor	$t1,$s1
803	xor	$t2,$s2
804	xor	$t3,$s3
805___
806}
807
808sub decstep()
809{ my ($i,@s) = @_;
810  my $tmp0=$acc0;
811  my $tmp1=$acc1;
812  my $tmp2=$acc2;
813  my $out=($t0,$t1,$t2,$s[0])[$i];
814
815	$code.="	mov	$s[0],$out\n"		if ($i!=3);
816			$tmp1=$s[2]			if ($i==3);
817	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
818	$code.="	and	\$0xFF,$out\n";
819
820	$code.="	mov	0($sbox,$out,8),$out\n";
821	$code.="	shr	\$16,$tmp1\n";
822			$tmp2=$s[3]			if ($i==3);
823	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
824
825			$tmp0=$s[1]			if ($i==3);
826	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
827	$code.="	and	\$0xFF,$tmp1\n";
828	$code.="	shr	\$24,$tmp2\n";
829
830	$code.="	xor	3($sbox,$tmp0,8),$out\n";
831	$code.="	xor	2($sbox,$tmp1,8),$out\n";
832	$code.="	xor	1($sbox,$tmp2,8),$out\n";
833
834	$code.="	mov	$t2,$s[1]\n"		if ($i==3);
835	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
836	$code.="	mov	$t0,$s[3]\n"		if ($i==3);
837	$code.="\n";
838}
839
840sub declast()
841{ my ($i,@s)=@_;
842  my $tmp0=$acc0;
843  my $tmp1=$acc1;
844  my $tmp2=$acc2;
845  my $out=($t0,$t1,$t2,$s[0])[$i];
846
847	$code.="	mov	$s[0],$out\n"		if ($i!=3);
848			$tmp1=$s[2]			if ($i==3);
849	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
850	$code.="	and	\$0xFF,$out\n";
851
852	$code.="	movzb	2048($sbox,$out,1),$out\n";
853	$code.="	shr	\$16,$tmp1\n";
854			$tmp2=$s[3]			if ($i==3);
855	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
856
857			$tmp0=$s[1]			if ($i==3);
858	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
859	$code.="	and	\$0xFF,$tmp1\n";
860	$code.="	shr	\$24,$tmp2\n";
861
862	$code.="	movzb	2048($sbox,$tmp0,1),$tmp0\n";
863	$code.="	movzb	2048($sbox,$tmp1,1),$tmp1\n";
864	$code.="	movzb	2048($sbox,$tmp2,1),$tmp2\n";
865
866	$code.="	shl	\$8,$tmp0\n";
867	$code.="	shl	\$16,$tmp1\n";
868	$code.="	shl	\$24,$tmp2\n";
869
870	$code.="	xor	$tmp0,$out\n";
871	$code.="	mov	$t2,$s[1]\n"		if ($i==3);
872	$code.="	xor	$tmp1,$out\n";
873	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
874	$code.="	xor	$tmp2,$out\n";
875	$code.="	mov	$t0,$s[3]\n"		if ($i==3);
876	$code.="\n";
877}
878
879$code.=<<___;
880.type	_x86_64_AES_decrypt,\@abi-omnipotent
881.align	16
882_x86_64_AES_decrypt:
883	xor	0($key),$s0			# xor with key
884	xor	4($key),$s1
885	xor	8($key),$s2
886	xor	12($key),$s3
887
888	mov	240($key),$rnds			# load key->rounds
889	sub	\$1,$rnds
890	jmp	.Ldec_loop
891.align	16
892.Ldec_loop:
893___
894	if ($verticalspin) { &decvert(); }
895	else {	&decstep(0,$s0,$s3,$s2,$s1);
896		&decstep(1,$s1,$s0,$s3,$s2);
897		&decstep(2,$s2,$s1,$s0,$s3);
898		&decstep(3,$s3,$s2,$s1,$s0);
899		$code.=<<___;
900		lea	16($key),$key
901		xor	0($key),$s0			# xor with key
902		xor	4($key),$s1
903		xor	8($key),$s2
904		xor	12($key),$s3
905___
906	}
907$code.=<<___;
908	sub	\$1,$rnds
909	jnz	.Ldec_loop
910___
911	if ($verticalspin) { &declastvert(); }
912	else {	&declast(0,$s0,$s3,$s2,$s1);
913		&declast(1,$s1,$s0,$s3,$s2);
914		&declast(2,$s2,$s1,$s0,$s3);
915		&declast(3,$s3,$s2,$s1,$s0);
916		$code.=<<___;
917		xor	16+0($key),$s0			# xor with key
918		xor	16+4($key),$s1
919		xor	16+8($key),$s2
920		xor	16+12($key),$s3
921___
922	}
923$code.=<<___;
924	.byte	0xf3,0xc3			# rep ret
925.size	_x86_64_AES_decrypt,.-_x86_64_AES_decrypt
926___
927
928sub deccompactvert()
929{ my ($t3,$t4,$t5)=("%r8d","%r9d","%r13d");
930
931$code.=<<___;
932	movzb	`&lo("$s0")`,$t0
933	movzb	`&lo("$s1")`,$t1
934	movzb	`&lo("$s2")`,$t2
935	movzb	($sbox,$t0,1),$t0
936	movzb	($sbox,$t1,1),$t1
937	movzb	($sbox,$t2,1),$t2
938
939	movzb	`&lo("$s3")`,$t3
940	movzb	`&hi("$s3")`,$acc0
941	movzb	`&hi("$s0")`,$acc1
942	movzb	($sbox,$t3,1),$t3
943	movzb	($sbox,$acc0,1),$t4	#$t0
944	movzb	($sbox,$acc1,1),$t5	#$t1
945
946	movzb	`&hi("$s1")`,$acc2
947	movzb	`&hi("$s2")`,$acc0
948	shr	\$16,$s2
949	movzb	($sbox,$acc2,1),$acc2	#$t2
950	movzb	($sbox,$acc0,1),$acc0	#$t3
951	shr	\$16,$s3
952
953	movzb	`&lo("$s2")`,$acc1
954	shl	\$8,$t4
955	shl	\$8,$t5
956	movzb	($sbox,$acc1,1),$acc1	#$t0
957	xor	$t4,$t0
958	xor	$t5,$t1
959
960	movzb	`&lo("$s3")`,$t4
961	shr	\$16,$s0
962	shr	\$16,$s1
963	movzb	`&lo("$s0")`,$t5
964	shl	\$8,$acc2
965	shl	\$8,$acc0
966	movzb	($sbox,$t4,1),$t4	#$t1
967	movzb	($sbox,$t5,1),$t5	#$t2
968	xor	$acc2,$t2
969	xor	$acc0,$t3
970
971	movzb	`&lo("$s1")`,$acc2
972	movzb	`&hi("$s1")`,$acc0
973	shl	\$16,$acc1
974	movzb	($sbox,$acc2,1),$acc2	#$t3
975	movzb	($sbox,$acc0,1),$acc0	#$t0
976	xor	$acc1,$t0
977
978	movzb	`&hi("$s2")`,$acc1
979	shl	\$16,$t4
980	shl	\$16,$t5
981	movzb	($sbox,$acc1,1),$s1	#$t1
982	xor	$t4,$t1
983	xor	$t5,$t2
984
985	movzb	`&hi("$s3")`,$acc1
986	shr	\$8,$s0
987	shl	\$16,$acc2
988	movzb	($sbox,$acc1,1),$s2	#$t2
989	movzb	($sbox,$s0,1),$s3	#$t3
990	xor	$acc2,$t3
991
992	shl	\$24,$acc0
993	shl	\$24,$s1
994	shl	\$24,$s2
995	xor	$acc0,$t0
996	shl	\$24,$s3
997	xor	$t1,$s1
998	mov	$t0,$s0
999	xor	$t2,$s2
1000	xor	$t3,$s3
1001___
1002}
1003
1004# parallelized version! input is pair of 64-bit values: %rax=s1.s0
1005# and %rcx=s3.s2, output is four 32-bit values in %eax=s0, %ebx=s1,
1006# %ecx=s2 and %edx=s3.
1007sub dectransform()
1008{ my ($tp10,$tp20,$tp40,$tp80,$acc0)=("%rax","%r8", "%r9", "%r10","%rbx");
1009  my ($tp18,$tp28,$tp48,$tp88,$acc8)=("%rcx","%r11","%r12","%r13","%rdx");
1010  my $prefetch = shift;
1011
1012$code.=<<___;
1013	mov	$tp10,$acc0
1014	mov	$tp18,$acc8
1015	and	$mask80,$acc0
1016	and	$mask80,$acc8
1017	mov	$acc0,$tp40
1018	mov	$acc8,$tp48
1019	shr	\$7,$tp40
1020	lea	($tp10,$tp10),$tp20
1021	shr	\$7,$tp48
1022	lea	($tp18,$tp18),$tp28
1023	sub	$tp40,$acc0
1024	sub	$tp48,$acc8
1025	and	$maskfe,$tp20
1026	and	$maskfe,$tp28
1027	and	$mask1b,$acc0
1028	and	$mask1b,$acc8
1029	xor	$tp20,$acc0
1030	xor	$tp28,$acc8
1031	mov	$acc0,$tp20
1032	mov	$acc8,$tp28
1033
1034	and	$mask80,$acc0
1035	and	$mask80,$acc8
1036	mov	$acc0,$tp80
1037	mov	$acc8,$tp88
1038	shr	\$7,$tp80
1039	lea	($tp20,$tp20),$tp40
1040	shr	\$7,$tp88
1041	lea	($tp28,$tp28),$tp48
1042	sub	$tp80,$acc0
1043	sub	$tp88,$acc8
1044	and	$maskfe,$tp40
1045	and	$maskfe,$tp48
1046	and	$mask1b,$acc0
1047	and	$mask1b,$acc8
1048	xor	$tp40,$acc0
1049	xor	$tp48,$acc8
1050	mov	$acc0,$tp40
1051	mov	$acc8,$tp48
1052
1053	and	$mask80,$acc0
1054	and	$mask80,$acc8
1055	mov	$acc0,$tp80
1056	mov	$acc8,$tp88
1057	shr	\$7,$tp80
1058	 xor	$tp10,$tp20		# tp2^=tp1
1059	shr	\$7,$tp88
1060	 xor	$tp18,$tp28		# tp2^=tp1
1061	sub	$tp80,$acc0
1062	sub	$tp88,$acc8
1063	lea	($tp40,$tp40),$tp80
1064	lea	($tp48,$tp48),$tp88
1065	 xor	$tp10,$tp40		# tp4^=tp1
1066	 xor	$tp18,$tp48		# tp4^=tp1
1067	and	$maskfe,$tp80
1068	and	$maskfe,$tp88
1069	and	$mask1b,$acc0
1070	and	$mask1b,$acc8
1071	xor	$acc0,$tp80
1072	xor	$acc8,$tp88
1073
1074	xor	$tp80,$tp10		# tp1^=tp8
1075	xor	$tp88,$tp18		# tp1^=tp8
1076	xor	$tp80,$tp20		# tp2^tp1^=tp8
1077	xor	$tp88,$tp28		# tp2^tp1^=tp8
1078	mov	$tp10,$acc0
1079	mov	$tp18,$acc8
1080	xor	$tp80,$tp40		# tp4^tp1^=tp8
1081	xor	$tp88,$tp48		# tp4^tp1^=tp8
1082	shr	\$32,$acc0
1083	shr	\$32,$acc8
1084	xor	$tp20,$tp80		# tp8^=tp8^tp2^tp1=tp2^tp1
1085	xor	$tp28,$tp88		# tp8^=tp8^tp2^tp1=tp2^tp1
1086	rol	\$8,`&LO("$tp10")`	# ROTATE(tp1^tp8,8)
1087	rol	\$8,`&LO("$tp18")`	# ROTATE(tp1^tp8,8)
1088	xor	$tp40,$tp80		# tp2^tp1^=tp8^tp4^tp1=tp8^tp4^tp2
1089	xor	$tp48,$tp88		# tp2^tp1^=tp8^tp4^tp1=tp8^tp4^tp2
1090
1091	rol	\$8,`&LO("$acc0")`	# ROTATE(tp1^tp8,8)
1092	rol	\$8,`&LO("$acc8")`	# ROTATE(tp1^tp8,8)
1093	xor	`&LO("$tp80")`,`&LO("$tp10")`
1094	xor	`&LO("$tp88")`,`&LO("$tp18")`
1095	shr	\$32,$tp80
1096	shr	\$32,$tp88
1097	xor	`&LO("$tp80")`,`&LO("$acc0")`
1098	xor	`&LO("$tp88")`,`&LO("$acc8")`
1099
1100	mov	$tp20,$tp80
1101	mov	$tp28,$tp88
1102	shr	\$32,$tp80
1103	shr	\$32,$tp88
1104	rol	\$24,`&LO("$tp20")`	# ROTATE(tp2^tp1^tp8,24)
1105	rol	\$24,`&LO("$tp28")`	# ROTATE(tp2^tp1^tp8,24)
1106	rol	\$24,`&LO("$tp80")`	# ROTATE(tp2^tp1^tp8,24)
1107	rol	\$24,`&LO("$tp88")`	# ROTATE(tp2^tp1^tp8,24)
1108	xor	`&LO("$tp20")`,`&LO("$tp10")`
1109	xor	`&LO("$tp28")`,`&LO("$tp18")`
1110	mov	$tp40,$tp20
1111	mov	$tp48,$tp28
1112	xor	`&LO("$tp80")`,`&LO("$acc0")`
1113	xor	`&LO("$tp88")`,`&LO("$acc8")`
1114
1115	`"mov	0($sbox),$mask80"	if ($prefetch)`
1116	shr	\$32,$tp20
1117	shr	\$32,$tp28
1118	`"mov	64($sbox),$maskfe"	if ($prefetch)`
1119	rol	\$16,`&LO("$tp40")`	# ROTATE(tp4^tp1^tp8,16)
1120	rol	\$16,`&LO("$tp48")`	# ROTATE(tp4^tp1^tp8,16)
1121	`"mov	128($sbox),$mask1b"	if ($prefetch)`
1122	rol	\$16,`&LO("$tp20")`	# ROTATE(tp4^tp1^tp8,16)
1123	rol	\$16,`&LO("$tp28")`	# ROTATE(tp4^tp1^tp8,16)
1124	`"mov	192($sbox),$tp80"	if ($prefetch)`
1125	xor	`&LO("$tp40")`,`&LO("$tp10")`
1126	xor	`&LO("$tp48")`,`&LO("$tp18")`
1127	`"mov	256($sbox),$tp88"	if ($prefetch)`
1128	xor	`&LO("$tp20")`,`&LO("$acc0")`
1129	xor	`&LO("$tp28")`,`&LO("$acc8")`
1130___
1131}
1132
1133$code.=<<___;
1134.type	_x86_64_AES_decrypt_compact,\@abi-omnipotent
1135.align	16
1136_x86_64_AES_decrypt_compact:
1137	lea	128($sbox),$inp			# size optimization
1138	mov	0-128($inp),$acc1		# prefetch Td4
1139	mov	32-128($inp),$acc2
1140	mov	64-128($inp),$t0
1141	mov	96-128($inp),$t1
1142	mov	128-128($inp),$acc1
1143	mov	160-128($inp),$acc2
1144	mov	192-128($inp),$t0
1145	mov	224-128($inp),$t1
1146	jmp	.Ldec_loop_compact
1147
1148.align	16
1149.Ldec_loop_compact:
1150		xor	0($key),$s0		# xor with key
1151		xor	4($key),$s1
1152		xor	8($key),$s2
1153		xor	12($key),$s3
1154		lea	16($key),$key
1155___
1156		&deccompactvert();
1157$code.=<<___;
1158		cmp	16(%rsp),$key
1159		je	.Ldec_compact_done
1160
1161		mov	256+0($sbox),$mask80
1162		shl	\$32,%rbx
1163		shl	\$32,%rdx
1164		mov	256+8($sbox),$maskfe
1165		or	%rbx,%rax
1166		or	%rdx,%rcx
1167		mov	256+16($sbox),$mask1b
1168___
1169		&dectransform(1);
1170$code.=<<___;
1171	jmp	.Ldec_loop_compact
1172.align	16
1173.Ldec_compact_done:
1174	xor	0($key),$s0
1175	xor	4($key),$s1
1176	xor	8($key),$s2
1177	xor	12($key),$s3
1178	.byte	0xf3,0xc3			# rep ret
1179.size	_x86_64_AES_decrypt_compact,.-_x86_64_AES_decrypt_compact
1180___
1181
1182# void AES_decrypt (const void *inp,void *out,const AES_KEY *key);
1183$code.=<<___;
1184.globl	AES_decrypt
1185.type	AES_decrypt,\@function,3
1186.align	16
1187AES_decrypt:
1188	push	%rbx
1189	push	%rbp
1190	push	%r12
1191	push	%r13
1192	push	%r14
1193	push	%r15
1194
1195	# allocate frame "above" key schedule
1196	mov	%rsp,%r10
1197	lea	-63(%rdx),%rcx	# %rdx is key argument
1198	and	\$-64,%rsp
1199	sub	%rsp,%rcx
1200	neg	%rcx
1201	and	\$0x3c0,%rcx
1202	sub	%rcx,%rsp
1203	sub	\$32,%rsp
1204
1205	mov	%rsi,16(%rsp)	# save out
1206	mov	%r10,24(%rsp)	# save real stack pointer
1207.Ldec_prologue:
1208
1209	mov	%rdx,$key
1210	mov	240($key),$rnds	# load rounds
1211
1212	mov	0(%rdi),$s0	# load input vector
1213	mov	4(%rdi),$s1
1214	mov	8(%rdi),$s2
1215	mov	12(%rdi),$s3
1216
1217	shl	\$4,$rnds
1218	lea	($key,$rnds),%rbp
1219	mov	$key,(%rsp)	# key schedule
1220	mov	%rbp,8(%rsp)	# end of key schedule
1221
1222	# pick Td4 copy which can't "overlap" with stack frame or key schedule
1223	lea	.LAES_Td+2048(%rip),$sbox
1224	lea	768(%rsp),%rbp
1225	sub	$sbox,%rbp
1226	and	\$0x300,%rbp
1227	lea	($sbox,%rbp),$sbox
1228	shr	\$3,%rbp	# recall "magic" constants!
1229	add	%rbp,$sbox
1230
1231	call	_x86_64_AES_decrypt_compact
1232
1233	mov	16(%rsp),$out	# restore out
1234	mov	24(%rsp),%rsi	# restore saved stack pointer
1235	mov	$s0,0($out)	# write output vector
1236	mov	$s1,4($out)
1237	mov	$s2,8($out)
1238	mov	$s3,12($out)
1239
1240	mov	(%rsi),%r15
1241	mov	8(%rsi),%r14
1242	mov	16(%rsi),%r13
1243	mov	24(%rsi),%r12
1244	mov	32(%rsi),%rbp
1245	mov	40(%rsi),%rbx
1246	lea	48(%rsi),%rsp
1247.Ldec_epilogue:
1248	ret
1249.size	AES_decrypt,.-AES_decrypt
1250___
1251#------------------------------------------------------------------#
1252
1253sub enckey()
1254{
1255$code.=<<___;
1256	movz	%dl,%esi		# rk[i]>>0
1257	movzb	-128(%rbp,%rsi),%ebx
1258	movz	%dh,%esi		# rk[i]>>8
1259	shl	\$24,%ebx
1260	xor	%ebx,%eax
1261
1262	movzb	-128(%rbp,%rsi),%ebx
1263	shr	\$16,%edx
1264	movz	%dl,%esi		# rk[i]>>16
1265	xor	%ebx,%eax
1266
1267	movzb	-128(%rbp,%rsi),%ebx
1268	movz	%dh,%esi		# rk[i]>>24
1269	shl	\$8,%ebx
1270	xor	%ebx,%eax
1271
1272	movzb	-128(%rbp,%rsi),%ebx
1273	shl	\$16,%ebx
1274	xor	%ebx,%eax
1275
1276	xor	1024-128(%rbp,%rcx,4),%eax		# rcon
1277___
1278}
1279
1280# int AES_set_encrypt_key(const unsigned char *userKey, const int bits,
1281#                        AES_KEY *key)
1282$code.=<<___;
1283.globl	AES_set_encrypt_key
1284.type	AES_set_encrypt_key,\@function,3
1285.align	16
1286AES_set_encrypt_key:
1287	push	%rbx
1288	push	%rbp
1289	push	%r12			# redundant, but allows to share
1290	push	%r13			# exception handler...
1291	push	%r14
1292	push	%r15
1293	sub	\$8,%rsp
1294.Lenc_key_prologue:
1295
1296	call	_x86_64_AES_set_encrypt_key
1297
1298	mov	8(%rsp),%r15
1299	mov	16(%rsp),%r14
1300	mov	24(%rsp),%r13
1301	mov	32(%rsp),%r12
1302	mov	40(%rsp),%rbp
1303	mov	48(%rsp),%rbx
1304	add	\$56,%rsp
1305.Lenc_key_epilogue:
1306	ret
1307.size	AES_set_encrypt_key,.-AES_set_encrypt_key
1308
1309.type	_x86_64_AES_set_encrypt_key,\@abi-omnipotent
1310.align	16
1311_x86_64_AES_set_encrypt_key:
1312	mov	%esi,%ecx			# %ecx=bits
1313	mov	%rdi,%rsi			# %rsi=userKey
1314	mov	%rdx,%rdi			# %rdi=key
1315
1316	test	\$-1,%rsi
1317	jz	.Lbadpointer
1318	test	\$-1,%rdi
1319	jz	.Lbadpointer
1320
1321	lea	.LAES_Te(%rip),%rbp
1322	lea	2048+128(%rbp),%rbp
1323
1324	# prefetch Te4
1325	mov	0-128(%rbp),%eax
1326	mov	32-128(%rbp),%ebx
1327	mov	64-128(%rbp),%r8d
1328	mov	96-128(%rbp),%edx
1329	mov	128-128(%rbp),%eax
1330	mov	160-128(%rbp),%ebx
1331	mov	192-128(%rbp),%r8d
1332	mov	224-128(%rbp),%edx
1333
1334	cmp	\$128,%ecx
1335	je	.L10rounds
1336	cmp	\$192,%ecx
1337	je	.L12rounds
1338	cmp	\$256,%ecx
1339	je	.L14rounds
1340	mov	\$-2,%rax			# invalid number of bits
1341	jmp	.Lexit
1342
1343.L10rounds:
1344	mov	0(%rsi),%rax			# copy first 4 dwords
1345	mov	8(%rsi),%rdx
1346	mov	%rax,0(%rdi)
1347	mov	%rdx,8(%rdi)
1348
1349	shr	\$32,%rdx
1350	xor	%ecx,%ecx
1351	jmp	.L10shortcut
1352.align	4
1353.L10loop:
1354		mov	0(%rdi),%eax			# rk[0]
1355		mov	12(%rdi),%edx			# rk[3]
1356.L10shortcut:
1357___
1358		&enckey	();
1359$code.=<<___;
1360		mov	%eax,16(%rdi)			# rk[4]
1361		xor	4(%rdi),%eax
1362		mov	%eax,20(%rdi)			# rk[5]
1363		xor	8(%rdi),%eax
1364		mov	%eax,24(%rdi)			# rk[6]
1365		xor	12(%rdi),%eax
1366		mov	%eax,28(%rdi)			# rk[7]
1367		add	\$1,%ecx
1368		lea	16(%rdi),%rdi
1369		cmp	\$10,%ecx
1370	jl	.L10loop
1371
1372	movl	\$10,80(%rdi)			# setup number of rounds
1373	xor	%rax,%rax
1374	jmp	.Lexit
1375
1376.L12rounds:
1377	mov	0(%rsi),%rax			# copy first 6 dwords
1378	mov	8(%rsi),%rbx
1379	mov	16(%rsi),%rdx
1380	mov	%rax,0(%rdi)
1381	mov	%rbx,8(%rdi)
1382	mov	%rdx,16(%rdi)
1383
1384	shr	\$32,%rdx
1385	xor	%ecx,%ecx
1386	jmp	.L12shortcut
1387.align	4
1388.L12loop:
1389		mov	0(%rdi),%eax			# rk[0]
1390		mov	20(%rdi),%edx			# rk[5]
1391.L12shortcut:
1392___
1393		&enckey	();
1394$code.=<<___;
1395		mov	%eax,24(%rdi)			# rk[6]
1396		xor	4(%rdi),%eax
1397		mov	%eax,28(%rdi)			# rk[7]
1398		xor	8(%rdi),%eax
1399		mov	%eax,32(%rdi)			# rk[8]
1400		xor	12(%rdi),%eax
1401		mov	%eax,36(%rdi)			# rk[9]
1402
1403		cmp	\$7,%ecx
1404		je	.L12break
1405		add	\$1,%ecx
1406
1407		xor	16(%rdi),%eax
1408		mov	%eax,40(%rdi)			# rk[10]
1409		xor	20(%rdi),%eax
1410		mov	%eax,44(%rdi)			# rk[11]
1411
1412		lea	24(%rdi),%rdi
1413	jmp	.L12loop
1414.L12break:
1415	movl	\$12,72(%rdi)		# setup number of rounds
1416	xor	%rax,%rax
1417	jmp	.Lexit
1418
1419.L14rounds:
1420	mov	0(%rsi),%rax			# copy first 8 dwords
1421	mov	8(%rsi),%rbx
1422	mov	16(%rsi),%rcx
1423	mov	24(%rsi),%rdx
1424	mov	%rax,0(%rdi)
1425	mov	%rbx,8(%rdi)
1426	mov	%rcx,16(%rdi)
1427	mov	%rdx,24(%rdi)
1428
1429	shr	\$32,%rdx
1430	xor	%ecx,%ecx
1431	jmp	.L14shortcut
1432.align	4
1433.L14loop:
1434		mov	0(%rdi),%eax			# rk[0]
1435		mov	28(%rdi),%edx			# rk[4]
1436.L14shortcut:
1437___
1438		&enckey	();
1439$code.=<<___;
1440		mov	%eax,32(%rdi)			# rk[8]
1441		xor	4(%rdi),%eax
1442		mov	%eax,36(%rdi)			# rk[9]
1443		xor	8(%rdi),%eax
1444		mov	%eax,40(%rdi)			# rk[10]
1445		xor	12(%rdi),%eax
1446		mov	%eax,44(%rdi)			# rk[11]
1447
1448		cmp	\$6,%ecx
1449		je	.L14break
1450		add	\$1,%ecx
1451
1452		mov	%eax,%edx
1453		mov	16(%rdi),%eax			# rk[4]
1454		movz	%dl,%esi			# rk[11]>>0
1455		movzb	-128(%rbp,%rsi),%ebx
1456		movz	%dh,%esi			# rk[11]>>8
1457		xor	%ebx,%eax
1458
1459		movzb	-128(%rbp,%rsi),%ebx
1460		shr	\$16,%edx
1461		shl	\$8,%ebx
1462		movz	%dl,%esi			# rk[11]>>16
1463		xor	%ebx,%eax
1464
1465		movzb	-128(%rbp,%rsi),%ebx
1466		movz	%dh,%esi			# rk[11]>>24
1467		shl	\$16,%ebx
1468		xor	%ebx,%eax
1469
1470		movzb	-128(%rbp,%rsi),%ebx
1471		shl	\$24,%ebx
1472		xor	%ebx,%eax
1473
1474		mov	%eax,48(%rdi)			# rk[12]
1475		xor	20(%rdi),%eax
1476		mov	%eax,52(%rdi)			# rk[13]
1477		xor	24(%rdi),%eax
1478		mov	%eax,56(%rdi)			# rk[14]
1479		xor	28(%rdi),%eax
1480		mov	%eax,60(%rdi)			# rk[15]
1481
1482		lea	32(%rdi),%rdi
1483	jmp	.L14loop
1484.L14break:
1485	movl	\$14,48(%rdi)		# setup number of rounds
1486	xor	%rax,%rax
1487	jmp	.Lexit
1488
1489.Lbadpointer:
1490	mov	\$-1,%rax
1491.Lexit:
1492	.byte	0xf3,0xc3			# rep ret
1493.size	_x86_64_AES_set_encrypt_key,.-_x86_64_AES_set_encrypt_key
1494___
1495
1496sub deckey_ref()
1497{ my ($i,$ptr,$te,$td) = @_;
1498  my ($tp1,$tp2,$tp4,$tp8,$acc)=("%eax","%ebx","%edi","%edx","%r8d");
1499$code.=<<___;
1500	mov	$i($ptr),$tp1
1501	mov	$tp1,$acc
1502	and	\$0x80808080,$acc
1503	mov	$acc,$tp4
1504	shr	\$7,$tp4
1505	lea	0($tp1,$tp1),$tp2
1506	sub	$tp4,$acc
1507	and	\$0xfefefefe,$tp2
1508	and	\$0x1b1b1b1b,$acc
1509	xor	$tp2,$acc
1510	mov	$acc,$tp2
1511
1512	and	\$0x80808080,$acc
1513	mov	$acc,$tp8
1514	shr	\$7,$tp8
1515	lea	0($tp2,$tp2),$tp4
1516	sub	$tp8,$acc
1517	and	\$0xfefefefe,$tp4
1518	and	\$0x1b1b1b1b,$acc
1519	 xor	$tp1,$tp2		# tp2^tp1
1520	xor	$tp4,$acc
1521	mov	$acc,$tp4
1522
1523	and	\$0x80808080,$acc
1524	mov	$acc,$tp8
1525	shr	\$7,$tp8
1526	sub	$tp8,$acc
1527	lea	0($tp4,$tp4),$tp8
1528	 xor	$tp1,$tp4		# tp4^tp1
1529	and	\$0xfefefefe,$tp8
1530	and	\$0x1b1b1b1b,$acc
1531	xor	$acc,$tp8
1532
1533	xor	$tp8,$tp1		# tp1^tp8
1534	rol	\$8,$tp1		# ROTATE(tp1^tp8,8)
1535	xor	$tp8,$tp2		# tp2^tp1^tp8
1536	xor	$tp8,$tp4		# tp4^tp1^tp8
1537	xor	$tp2,$tp8
1538	xor	$tp4,$tp8		# tp8^(tp8^tp4^tp1)^(tp8^tp2^tp1)=tp8^tp4^tp2
1539
1540	xor	$tp8,$tp1
1541	rol	\$24,$tp2		# ROTATE(tp2^tp1^tp8,24)
1542	xor	$tp2,$tp1
1543	rol	\$16,$tp4		# ROTATE(tp4^tp1^tp8,16)
1544	xor	$tp4,$tp1
1545
1546	mov	$tp1,$i($ptr)
1547___
1548}
1549
1550# int AES_set_decrypt_key(const unsigned char *userKey, const int bits,
1551#                        AES_KEY *key)
1552$code.=<<___;
1553.globl	AES_set_decrypt_key
1554.type	AES_set_decrypt_key,\@function,3
1555.align	16
1556AES_set_decrypt_key:
1557	push	%rbx
1558	push	%rbp
1559	push	%r12
1560	push	%r13
1561	push	%r14
1562	push	%r15
1563	push	%rdx			# save key schedule
1564.Ldec_key_prologue:
1565
1566	call	_x86_64_AES_set_encrypt_key
1567	mov	(%rsp),%r8		# restore key schedule
1568	cmp	\$0,%eax
1569	jne	.Labort
1570
1571	mov	240(%r8),%r14d		# pull number of rounds
1572	xor	%rdi,%rdi
1573	lea	(%rdi,%r14d,4),%rcx
1574	mov	%r8,%rsi
1575	lea	(%r8,%rcx,4),%rdi	# pointer to last chunk
1576.align	4
1577.Linvert:
1578		mov	0(%rsi),%rax
1579		mov	8(%rsi),%rbx
1580		mov	0(%rdi),%rcx
1581		mov	8(%rdi),%rdx
1582		mov	%rax,0(%rdi)
1583		mov	%rbx,8(%rdi)
1584		mov	%rcx,0(%rsi)
1585		mov	%rdx,8(%rsi)
1586		lea	16(%rsi),%rsi
1587		lea	-16(%rdi),%rdi
1588		cmp	%rsi,%rdi
1589	jne	.Linvert
1590
1591	lea	.LAES_Te+2048+1024(%rip),%rax	# rcon
1592
1593	mov	40(%rax),$mask80
1594	mov	48(%rax),$maskfe
1595	mov	56(%rax),$mask1b
1596
1597	mov	%r8,$key
1598	sub	\$1,%r14d
1599.align	4
1600.Lpermute:
1601		lea	16($key),$key
1602		mov	0($key),%rax
1603		mov	8($key),%rcx
1604___
1605		&dectransform ();
1606$code.=<<___;
1607		mov	%eax,0($key)
1608		mov	%ebx,4($key)
1609		mov	%ecx,8($key)
1610		mov	%edx,12($key)
1611		sub	\$1,%r14d
1612	jnz	.Lpermute
1613
1614	xor	%rax,%rax
1615.Labort:
1616	mov	8(%rsp),%r15
1617	mov	16(%rsp),%r14
1618	mov	24(%rsp),%r13
1619	mov	32(%rsp),%r12
1620	mov	40(%rsp),%rbp
1621	mov	48(%rsp),%rbx
1622	add	\$56,%rsp
1623.Ldec_key_epilogue:
1624	ret
1625.size	AES_set_decrypt_key,.-AES_set_decrypt_key
1626___
1627
1628# void AES_cbc_encrypt (const void char *inp, unsigned char *out,
1629#			size_t length, const AES_KEY *key,
1630#			unsigned char *ivp,const int enc);
1631{
1632# stack frame layout
1633# -8(%rsp)		return address
1634my $keyp="0(%rsp)";		# one to pass as $key
1635my $keyend="8(%rsp)";		# &(keyp->rd_key[4*keyp->rounds])
1636my $_rsp="16(%rsp)";		# saved %rsp
1637my $_inp="24(%rsp)";		# copy of 1st parameter, inp
1638my $_out="32(%rsp)";		# copy of 2nd parameter, out
1639my $_len="40(%rsp)";		# copy of 3rd parameter, length
1640my $_key="48(%rsp)";		# copy of 4th parameter, key
1641my $_ivp="56(%rsp)";		# copy of 5th parameter, ivp
1642my $ivec="64(%rsp)";		# ivec[16]
1643my $aes_key="80(%rsp)";		# copy of aes_key
1644my $mark="80+240(%rsp)";	# copy of aes_key->rounds
1645
1646$code.=<<___;
1647.globl	AES_cbc_encrypt
1648.type	AES_cbc_encrypt,\@function,6
1649.align	16
1650.extern	OPENSSL_ia32cap_P
1651AES_cbc_encrypt:
1652	cmp	\$0,%rdx	# check length
1653	je	.Lcbc_epilogue
1654	pushfq
1655	push	%rbx
1656	push	%rbp
1657	push	%r12
1658	push	%r13
1659	push	%r14
1660	push	%r15
1661.Lcbc_prologue:
1662
1663	cld
1664	mov	%r9d,%r9d	# clear upper half of enc
1665
1666	lea	.LAES_Te(%rip),$sbox
1667	cmp	\$0,%r9
1668	jne	.Lcbc_picked_te
1669	lea	.LAES_Td(%rip),$sbox
1670.Lcbc_picked_te:
1671
1672	mov	OPENSSL_ia32cap_P(%rip),%r10d
1673	cmp	\$$speed_limit,%rdx
1674	jb	.Lcbc_slow_prologue
1675	test	\$15,%rdx
1676	jnz	.Lcbc_slow_prologue
1677	bt	\$28,%r10d
1678	jc	.Lcbc_slow_prologue
1679
1680	# allocate aligned stack frame...
1681	lea	-88-248(%rsp),$key
1682	and	\$-64,$key
1683
1684	# ... and make sure it doesn't alias with AES_T[ed] modulo 4096
1685	mov	$sbox,%r10
1686	lea	2304($sbox),%r11
1687	mov	$key,%r12
1688	and	\$0xFFF,%r10	# s = $sbox&0xfff
1689	and	\$0xFFF,%r11	# e = ($sbox+2048)&0xfff
1690	and	\$0xFFF,%r12	# p = %rsp&0xfff
1691
1692	cmp	%r11,%r12	# if (p=>e) %rsp =- (p-e);
1693	jb	.Lcbc_te_break_out
1694	sub	%r11,%r12
1695	sub	%r12,$key
1696	jmp	.Lcbc_te_ok
1697.Lcbc_te_break_out:		# else %rsp -= (p-s)&0xfff + framesz
1698	sub	%r10,%r12
1699	and	\$0xFFF,%r12
1700	add	\$320,%r12
1701	sub	%r12,$key
1702.align	4
1703.Lcbc_te_ok:
1704
1705	xchg	%rsp,$key
1706	#add	\$8,%rsp	# reserve for return address!
1707	mov	$key,$_rsp	# save %rsp
1708.Lcbc_fast_body:
1709	mov	%rdi,$_inp	# save copy of inp
1710	mov	%rsi,$_out	# save copy of out
1711	mov	%rdx,$_len	# save copy of len
1712	mov	%rcx,$_key	# save copy of key
1713	mov	%r8,$_ivp	# save copy of ivp
1714	movl	\$0,$mark	# copy of aes_key->rounds = 0;
1715	mov	%r8,%rbp	# rearrange input arguments
1716	mov	%r9,%rbx
1717	mov	%rsi,$out
1718	mov	%rdi,$inp
1719	mov	%rcx,$key
1720
1721	mov	240($key),%eax		# key->rounds
1722	# do we copy key schedule to stack?
1723	mov	$key,%r10
1724	sub	$sbox,%r10
1725	and	\$0xfff,%r10
1726	cmp	\$2304,%r10
1727	jb	.Lcbc_do_ecopy
1728	cmp	\$4096-248,%r10
1729	jb	.Lcbc_skip_ecopy
1730.align	4
1731.Lcbc_do_ecopy:
1732		mov	$key,%rsi
1733		lea	$aes_key,%rdi
1734		lea	$aes_key,$key
1735		mov	\$240/8,%ecx
1736		.long	0x90A548F3	# rep movsq
1737		mov	%eax,(%rdi)	# copy aes_key->rounds
1738.Lcbc_skip_ecopy:
1739	mov	$key,$keyp	# save key pointer
1740
1741	mov	\$18,%ecx
1742.align	4
1743.Lcbc_prefetch_te:
1744		mov	0($sbox),%r10
1745		mov	32($sbox),%r11
1746		mov	64($sbox),%r12
1747		mov	96($sbox),%r13
1748		lea	128($sbox),$sbox
1749		sub	\$1,%ecx
1750	jnz	.Lcbc_prefetch_te
1751	lea	-2304($sbox),$sbox
1752
1753	cmp	\$0,%rbx
1754	je	.LFAST_DECRYPT
1755
1756#----------------------------- ENCRYPT -----------------------------#
1757	mov	0(%rbp),$s0		# load iv
1758	mov	4(%rbp),$s1
1759	mov	8(%rbp),$s2
1760	mov	12(%rbp),$s3
1761
1762.align	4
1763.Lcbc_fast_enc_loop:
1764		xor	0($inp),$s0
1765		xor	4($inp),$s1
1766		xor	8($inp),$s2
1767		xor	12($inp),$s3
1768		mov	$keyp,$key	# restore key
1769		mov	$inp,$_inp	# if ($verticalspin) save inp
1770
1771		call	_x86_64_AES_encrypt
1772
1773		mov	$_inp,$inp	# if ($verticalspin) restore inp
1774		mov	$_len,%r10
1775		mov	$s0,0($out)
1776		mov	$s1,4($out)
1777		mov	$s2,8($out)
1778		mov	$s3,12($out)
1779
1780		lea	16($inp),$inp
1781		lea	16($out),$out
1782		sub	\$16,%r10
1783		test	\$-16,%r10
1784		mov	%r10,$_len
1785	jnz	.Lcbc_fast_enc_loop
1786	mov	$_ivp,%rbp	# restore ivp
1787	mov	$s0,0(%rbp)	# save ivec
1788	mov	$s1,4(%rbp)
1789	mov	$s2,8(%rbp)
1790	mov	$s3,12(%rbp)
1791
1792	jmp	.Lcbc_fast_cleanup
1793
1794#----------------------------- DECRYPT -----------------------------#
1795.align	16
1796.LFAST_DECRYPT:
1797	cmp	$inp,$out
1798	je	.Lcbc_fast_dec_in_place
1799
1800	mov	%rbp,$ivec
1801.align	4
1802.Lcbc_fast_dec_loop:
1803		mov	0($inp),$s0	# read input
1804		mov	4($inp),$s1
1805		mov	8($inp),$s2
1806		mov	12($inp),$s3
1807		mov	$keyp,$key	# restore key
1808		mov	$inp,$_inp	# if ($verticalspin) save inp
1809
1810		call	_x86_64_AES_decrypt
1811
1812		mov	$ivec,%rbp	# load ivp
1813		mov	$_inp,$inp	# if ($verticalspin) restore inp
1814		mov	$_len,%r10	# load len
1815		xor	0(%rbp),$s0	# xor iv
1816		xor	4(%rbp),$s1
1817		xor	8(%rbp),$s2
1818		xor	12(%rbp),$s3
1819		mov	$inp,%rbp	# current input, next iv
1820
1821		sub	\$16,%r10
1822		mov	%r10,$_len	# update len
1823		mov	%rbp,$ivec	# update ivp
1824
1825		mov	$s0,0($out)	# write output
1826		mov	$s1,4($out)
1827		mov	$s2,8($out)
1828		mov	$s3,12($out)
1829
1830		lea	16($inp),$inp
1831		lea	16($out),$out
1832	jnz	.Lcbc_fast_dec_loop
1833	mov	$_ivp,%r12		# load user ivp
1834	mov	0(%rbp),%r10		# load iv
1835	mov	8(%rbp),%r11
1836	mov	%r10,0(%r12)		# copy back to user
1837	mov	%r11,8(%r12)
1838	jmp	.Lcbc_fast_cleanup
1839
1840.align	16
1841.Lcbc_fast_dec_in_place:
1842	mov	0(%rbp),%r10		# copy iv to stack
1843	mov	8(%rbp),%r11
1844	mov	%r10,0+$ivec
1845	mov	%r11,8+$ivec
1846.align	4
1847.Lcbc_fast_dec_in_place_loop:
1848		mov	0($inp),$s0	# load input
1849		mov	4($inp),$s1
1850		mov	8($inp),$s2
1851		mov	12($inp),$s3
1852		mov	$keyp,$key	# restore key
1853		mov	$inp,$_inp	# if ($verticalspin) save inp
1854
1855		call	_x86_64_AES_decrypt
1856
1857		mov	$_inp,$inp	# if ($verticalspin) restore inp
1858		mov	$_len,%r10
1859		xor	0+$ivec,$s0
1860		xor	4+$ivec,$s1
1861		xor	8+$ivec,$s2
1862		xor	12+$ivec,$s3
1863
1864		mov	0($inp),%r11	# load input
1865		mov	8($inp),%r12
1866		sub	\$16,%r10
1867		jz	.Lcbc_fast_dec_in_place_done
1868
1869		mov	%r11,0+$ivec	# copy input to iv
1870		mov	%r12,8+$ivec
1871
1872		mov	$s0,0($out)	# save output [zaps input]
1873		mov	$s1,4($out)
1874		mov	$s2,8($out)
1875		mov	$s3,12($out)
1876
1877		lea	16($inp),$inp
1878		lea	16($out),$out
1879		mov	%r10,$_len
1880	jmp	.Lcbc_fast_dec_in_place_loop
1881.Lcbc_fast_dec_in_place_done:
1882	mov	$_ivp,%rdi
1883	mov	%r11,0(%rdi)	# copy iv back to user
1884	mov	%r12,8(%rdi)
1885
1886	mov	$s0,0($out)	# save output [zaps input]
1887	mov	$s1,4($out)
1888	mov	$s2,8($out)
1889	mov	$s3,12($out)
1890
1891.align	4
1892.Lcbc_fast_cleanup:
1893	cmpl	\$0,$mark	# was the key schedule copied?
1894	lea	$aes_key,%rdi
1895	je	.Lcbc_exit
1896		mov	\$240/8,%ecx
1897		xor	%rax,%rax
1898		.long	0x90AB48F3	# rep stosq
1899
1900	jmp	.Lcbc_exit
1901
1902#--------------------------- SLOW ROUTINE ---------------------------#
1903.align	16
1904.Lcbc_slow_prologue:
1905	# allocate aligned stack frame...
1906	lea	-88(%rsp),%rbp
1907	and	\$-64,%rbp
1908	# ... just "above" key schedule
1909	lea	-88-63(%rcx),%r10
1910	sub	%rbp,%r10
1911	neg	%r10
1912	and	\$0x3c0,%r10
1913	sub	%r10,%rbp
1914
1915	xchg	%rsp,%rbp
1916	#add	\$8,%rsp	# reserve for return address!
1917	mov	%rbp,$_rsp	# save %rsp
1918.Lcbc_slow_body:
1919	#mov	%rdi,$_inp	# save copy of inp
1920	#mov	%rsi,$_out	# save copy of out
1921	#mov	%rdx,$_len	# save copy of len
1922	#mov	%rcx,$_key	# save copy of key
1923	mov	%r8,$_ivp	# save copy of ivp
1924	mov	%r8,%rbp	# rearrange input arguments
1925	mov	%r9,%rbx
1926	mov	%rsi,$out
1927	mov	%rdi,$inp
1928	mov	%rcx,$key
1929	mov	%rdx,%r10
1930
1931	mov	240($key),%eax
1932	mov	$key,$keyp	# save key pointer
1933	shl	\$4,%eax
1934	lea	($key,%rax),%rax
1935	mov	%rax,$keyend
1936
1937	# pick Te4 copy which can't "overlap" with stack frame or key scdedule
1938	lea	2048($sbox),$sbox
1939	lea	768-8(%rsp),%rax
1940	sub	$sbox,%rax
1941	and	\$0x300,%rax
1942	lea	($sbox,%rax),$sbox
1943
1944	cmp	\$0,%rbx
1945	je	.LSLOW_DECRYPT
1946
1947#--------------------------- SLOW ENCRYPT ---------------------------#
1948	test	\$-16,%r10		# check upon length
1949	mov	0(%rbp),$s0		# load iv
1950	mov	4(%rbp),$s1
1951	mov	8(%rbp),$s2
1952	mov	12(%rbp),$s3
1953	jz	.Lcbc_slow_enc_tail	# short input...
1954
1955.align	4
1956.Lcbc_slow_enc_loop:
1957		xor	0($inp),$s0
1958		xor	4($inp),$s1
1959		xor	8($inp),$s2
1960		xor	12($inp),$s3
1961		mov	$keyp,$key	# restore key
1962		mov	$inp,$_inp	# save inp
1963		mov	$out,$_out	# save out
1964		mov	%r10,$_len	# save len
1965
1966		call	_x86_64_AES_encrypt_compact
1967
1968		mov	$_inp,$inp	# restore inp
1969		mov	$_out,$out	# restore out
1970		mov	$_len,%r10	# restore len
1971		mov	$s0,0($out)
1972		mov	$s1,4($out)
1973		mov	$s2,8($out)
1974		mov	$s3,12($out)
1975
1976		lea	16($inp),$inp
1977		lea	16($out),$out
1978		sub	\$16,%r10
1979		test	\$-16,%r10
1980	jnz	.Lcbc_slow_enc_loop
1981	test	\$15,%r10
1982	jnz	.Lcbc_slow_enc_tail
1983	mov	$_ivp,%rbp	# restore ivp
1984	mov	$s0,0(%rbp)	# save ivec
1985	mov	$s1,4(%rbp)
1986	mov	$s2,8(%rbp)
1987	mov	$s3,12(%rbp)
1988
1989	jmp	.Lcbc_exit
1990
1991.align	4
1992.Lcbc_slow_enc_tail:
1993	mov	%rax,%r11
1994	mov	%rcx,%r12
1995	mov	%r10,%rcx
1996	mov	$inp,%rsi
1997	mov	$out,%rdi
1998	.long	0x9066A4F3		# rep movsb
1999	mov	\$16,%rcx		# zero tail
2000	sub	%r10,%rcx
2001	xor	%rax,%rax
2002	.long	0x9066AAF3		# rep stosb
2003	mov	$out,$inp		# this is not a mistake!
2004	mov	\$16,%r10		# len=16
2005	mov	%r11,%rax
2006	mov	%r12,%rcx
2007	jmp	.Lcbc_slow_enc_loop	# one more spin...
2008#--------------------------- SLOW DECRYPT ---------------------------#
2009.align	16
2010.LSLOW_DECRYPT:
2011	shr	\$3,%rax
2012	add	%rax,$sbox		# recall "magic" constants!
2013
2014	mov	0(%rbp),%r11		# copy iv to stack
2015	mov	8(%rbp),%r12
2016	mov	%r11,0+$ivec
2017	mov	%r12,8+$ivec
2018
2019.align	4
2020.Lcbc_slow_dec_loop:
2021		mov	0($inp),$s0	# load input
2022		mov	4($inp),$s1
2023		mov	8($inp),$s2
2024		mov	12($inp),$s3
2025		mov	$keyp,$key	# restore key
2026		mov	$inp,$_inp	# save inp
2027		mov	$out,$_out	# save out
2028		mov	%r10,$_len	# save len
2029
2030		call	_x86_64_AES_decrypt_compact
2031
2032		mov	$_inp,$inp	# restore inp
2033		mov	$_out,$out	# restore out
2034		mov	$_len,%r10
2035		xor	0+$ivec,$s0
2036		xor	4+$ivec,$s1
2037		xor	8+$ivec,$s2
2038		xor	12+$ivec,$s3
2039
2040		mov	0($inp),%r11	# load input
2041		mov	8($inp),%r12
2042		sub	\$16,%r10
2043		jc	.Lcbc_slow_dec_partial
2044		jz	.Lcbc_slow_dec_done
2045
2046		mov	%r11,0+$ivec	# copy input to iv
2047		mov	%r12,8+$ivec
2048
2049		mov	$s0,0($out)	# save output [can zap input]
2050		mov	$s1,4($out)
2051		mov	$s2,8($out)
2052		mov	$s3,12($out)
2053
2054		lea	16($inp),$inp
2055		lea	16($out),$out
2056	jmp	.Lcbc_slow_dec_loop
2057.Lcbc_slow_dec_done:
2058	mov	$_ivp,%rdi
2059	mov	%r11,0(%rdi)		# copy iv back to user
2060	mov	%r12,8(%rdi)
2061
2062	mov	$s0,0($out)		# save output [can zap input]
2063	mov	$s1,4($out)
2064	mov	$s2,8($out)
2065	mov	$s3,12($out)
2066
2067	jmp	.Lcbc_exit
2068
2069.align	4
2070.Lcbc_slow_dec_partial:
2071	mov	$_ivp,%rdi
2072	mov	%r11,0(%rdi)		# copy iv back to user
2073	mov	%r12,8(%rdi)
2074
2075	mov	$s0,0+$ivec		# save output to stack
2076	mov	$s1,4+$ivec
2077	mov	$s2,8+$ivec
2078	mov	$s3,12+$ivec
2079
2080	mov	$out,%rdi
2081	lea	$ivec,%rsi
2082	lea	16(%r10),%rcx
2083	.long	0x9066A4F3	# rep movsb
2084	jmp	.Lcbc_exit
2085
2086.align	16
2087.Lcbc_exit:
2088	mov	$_rsp,%rsi
2089	mov	(%rsi),%r15
2090	mov	8(%rsi),%r14
2091	mov	16(%rsi),%r13
2092	mov	24(%rsi),%r12
2093	mov	32(%rsi),%rbp
2094	mov	40(%rsi),%rbx
2095	lea	48(%rsi),%rsp
2096.Lcbc_popfq:
2097	popfq
2098.Lcbc_epilogue:
2099	ret
2100.size	AES_cbc_encrypt,.-AES_cbc_encrypt
2101___
2102}
2103
2104$code.=<<___;
2105.align	64
2106.LAES_Te:
2107___
2108	&_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6);
2109	&_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591);
2110	&_data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56);
2111	&_data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec);
2112	&_data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa);
2113	&_data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb);
2114	&_data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45);
2115	&_data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b);
2116	&_data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c);
2117	&_data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83);
2118	&_data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9);
2119	&_data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a);
2120	&_data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d);
2121	&_data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f);
2122	&_data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df);
2123	&_data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea);
2124	&_data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34);
2125	&_data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b);
2126	&_data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d);
2127	&_data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413);
2128	&_data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1);
2129	&_data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6);
2130	&_data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972);
2131	&_data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85);
2132	&_data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed);
2133	&_data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511);
2134	&_data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe);
2135	&_data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b);
2136	&_data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05);
2137	&_data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1);
2138	&_data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142);
2139	&_data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf);
2140	&_data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3);
2141	&_data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e);
2142	&_data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a);
2143	&_data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6);
2144	&_data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3);
2145	&_data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b);
2146	&_data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428);
2147	&_data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad);
2148	&_data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14);
2149	&_data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8);
2150	&_data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4);
2151	&_data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2);
2152	&_data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda);
2153	&_data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949);
2154	&_data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf);
2155	&_data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810);
2156	&_data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c);
2157	&_data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697);
2158	&_data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e);
2159	&_data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f);
2160	&_data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc);
2161	&_data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c);
2162	&_data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969);
2163	&_data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27);
2164	&_data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122);
2165	&_data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433);
2166	&_data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9);
2167	&_data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5);
2168	&_data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a);
2169	&_data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0);
2170	&_data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e);
2171	&_data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c);
2172
2173#Te4	# four copies of Te4 to choose from to avoid L1 aliasing
2174	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2175	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2176	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2177	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2178	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2179	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2180	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2181	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2182	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2183	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2184	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2185	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2186	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2187	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2188	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2189	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2190	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2191	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2192	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2193	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2194	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2195	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2196	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2197	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2198	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2199	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2200	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2201	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2202	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2203	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2204	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2205	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2206
2207	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2208	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2209	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2210	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2211	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2212	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2213	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2214	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2215	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2216	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2217	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2218	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2219	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2220	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2221	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2222	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2223	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2224	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2225	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2226	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2227	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2228	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2229	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2230	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2231	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2232	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2233	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2234	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2235	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2236	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2237	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2238	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2239
2240	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2241	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2242	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2243	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2244	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2245	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2246	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2247	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2248	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2249	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2250	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2251	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2252	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2253	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2254	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2255	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2256	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2257	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2258	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2259	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2260	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2261	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2262	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2263	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2264	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2265	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2266	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2267	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2268	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2269	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2270	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2271	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2272
2273	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2274	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2275	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2276	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2277	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2278	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2279	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2280	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2281	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2282	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2283	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2284	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2285	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2286	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2287	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2288	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2289	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2290	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2291	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2292	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2293	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2294	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2295	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2296	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2297	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2298	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2299	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2300	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2301	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2302	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2303	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2304	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2305#rcon:
2306$code.=<<___;
2307	.long	0x00000001, 0x00000002, 0x00000004, 0x00000008
2308	.long	0x00000010, 0x00000020, 0x00000040, 0x00000080
2309	.long	0x0000001b, 0x00000036, 0x80808080, 0x80808080
2310	.long	0xfefefefe, 0xfefefefe, 0x1b1b1b1b, 0x1b1b1b1b
2311___
2312$code.=<<___;
2313.align	64
2314.LAES_Td:
2315___
2316	&_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a);
2317	&_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b);
2318	&_data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5);
2319	&_data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5);
2320	&_data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d);
2321	&_data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b);
2322	&_data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295);
2323	&_data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e);
2324	&_data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927);
2325	&_data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d);
2326	&_data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362);
2327	&_data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9);
2328	&_data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52);
2329	&_data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566);
2330	&_data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3);
2331	&_data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed);
2332	&_data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e);
2333	&_data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4);
2334	&_data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4);
2335	&_data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd);
2336	&_data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d);
2337	&_data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060);
2338	&_data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967);
2339	&_data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879);
2340	&_data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000);
2341	&_data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c);
2342	&_data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36);
2343	&_data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624);
2344	&_data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b);
2345	&_data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c);
2346	&_data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12);
2347	&_data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14);
2348	&_data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3);
2349	&_data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b);
2350	&_data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8);
2351	&_data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684);
2352	&_data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7);
2353	&_data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177);
2354	&_data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947);
2355	&_data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322);
2356	&_data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498);
2357	&_data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f);
2358	&_data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54);
2359	&_data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382);
2360	&_data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf);
2361	&_data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb);
2362	&_data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83);
2363	&_data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef);
2364	&_data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029);
2365	&_data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235);
2366	&_data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733);
2367	&_data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117);
2368	&_data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4);
2369	&_data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546);
2370	&_data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb);
2371	&_data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d);
2372	&_data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb);
2373	&_data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a);
2374	&_data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773);
2375	&_data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478);
2376	&_data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2);
2377	&_data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff);
2378	&_data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664);
2379	&_data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0);
2380
2381#Td4:	# four copies of Td4 to choose from to avoid L1 aliasing
2382	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2383	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2384	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2385	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2386	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2387	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2388	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2389	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2390	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2391	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2392	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2393	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2394	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2395	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2396	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2397	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2398	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2399	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2400	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2401	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2402	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2403	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2404	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2405	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2406	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2407	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2408	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2409	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2410	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2411	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2412	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2413	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2414$code.=<<___;
2415	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2416	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2417___
2418	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2419	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2420	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2421	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2422	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2423	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2424	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2425	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2426	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2427	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2428	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2429	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2430	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2431	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2432	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2433	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2434	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2435	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2436	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2437	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2438	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2439	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2440	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2441	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2442	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2443	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2444	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2445	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2446	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2447	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2448	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2449	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2450$code.=<<___;
2451	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2452	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2453___
2454	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2455	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2456	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2457	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2458	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2459	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2460	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2461	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2462	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2463	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2464	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2465	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2466	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2467	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2468	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2469	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2470	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2471	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2472	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2473	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2474	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2475	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2476	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2477	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2478	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2479	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2480	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2481	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2482	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2483	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2484	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2485	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2486$code.=<<___;
2487	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2488	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2489___
2490	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2491	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2492	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2493	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2494	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2495	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2496	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2497	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2498	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2499	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2500	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2501	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2502	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2503	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2504	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2505	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2506	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2507	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2508	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2509	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2510	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2511	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2512	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2513	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2514	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2515	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2516	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2517	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2518	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2519	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2520	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2521	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2522$code.=<<___;
2523	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2524	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2525.asciz  "AES for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
2526.align	64
2527___
2528
2529# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
2530#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
2531if ($win64) {
2532$rec="%rcx";
2533$frame="%rdx";
2534$context="%r8";
2535$disp="%r9";
2536
2537$code.=<<___;
2538.extern	__imp_RtlVirtualUnwind
2539.type	block_se_handler,\@abi-omnipotent
2540.align	16
2541block_se_handler:
2542	push	%rsi
2543	push	%rdi
2544	push	%rbx
2545	push	%rbp
2546	push	%r12
2547	push	%r13
2548	push	%r14
2549	push	%r15
2550	pushfq
2551	sub	\$64,%rsp
2552
2553	mov	120($context),%rax	# pull context->Rax
2554	mov	248($context),%rbx	# pull context->Rip
2555
2556	mov	8($disp),%rsi		# disp->ImageBase
2557	mov	56($disp),%r11		# disp->HandlerData
2558
2559	mov	0(%r11),%r10d		# HandlerData[0]
2560	lea	(%rsi,%r10),%r10	# prologue label
2561	cmp	%r10,%rbx		# context->Rip<prologue label
2562	jb	.Lin_block_prologue
2563
2564	mov	152($context),%rax	# pull context->Rsp
2565
2566	mov	4(%r11),%r10d		# HandlerData[1]
2567	lea	(%rsi,%r10),%r10	# epilogue label
2568	cmp	%r10,%rbx		# context->Rip>=epilogue label
2569	jae	.Lin_block_prologue
2570
2571	mov	24(%rax),%rax		# pull saved real stack pointer
2572	lea	48(%rax),%rax		# adjust...
2573
2574	mov	-8(%rax),%rbx
2575	mov	-16(%rax),%rbp
2576	mov	-24(%rax),%r12
2577	mov	-32(%rax),%r13
2578	mov	-40(%rax),%r14
2579	mov	-48(%rax),%r15
2580	mov	%rbx,144($context)	# restore context->Rbx
2581	mov	%rbp,160($context)	# restore context->Rbp
2582	mov	%r12,216($context)	# restore context->R12
2583	mov	%r13,224($context)	# restore context->R13
2584	mov	%r14,232($context)	# restore context->R14
2585	mov	%r15,240($context)	# restore context->R15
2586
2587.Lin_block_prologue:
2588	mov	8(%rax),%rdi
2589	mov	16(%rax),%rsi
2590	mov	%rax,152($context)	# restore context->Rsp
2591	mov	%rsi,168($context)	# restore context->Rsi
2592	mov	%rdi,176($context)	# restore context->Rdi
2593
2594	jmp	.Lcommon_seh_exit
2595.size	block_se_handler,.-block_se_handler
2596
2597.type	key_se_handler,\@abi-omnipotent
2598.align	16
2599key_se_handler:
2600	push	%rsi
2601	push	%rdi
2602	push	%rbx
2603	push	%rbp
2604	push	%r12
2605	push	%r13
2606	push	%r14
2607	push	%r15
2608	pushfq
2609	sub	\$64,%rsp
2610
2611	mov	120($context),%rax	# pull context->Rax
2612	mov	248($context),%rbx	# pull context->Rip
2613
2614	mov	8($disp),%rsi		# disp->ImageBase
2615	mov	56($disp),%r11		# disp->HandlerData
2616
2617	mov	0(%r11),%r10d		# HandlerData[0]
2618	lea	(%rsi,%r10),%r10	# prologue label
2619	cmp	%r10,%rbx		# context->Rip<prologue label
2620	jb	.Lin_key_prologue
2621
2622	mov	152($context),%rax	# pull context->Rsp
2623
2624	mov	4(%r11),%r10d		# HandlerData[1]
2625	lea	(%rsi,%r10),%r10	# epilogue label
2626	cmp	%r10,%rbx		# context->Rip>=epilogue label
2627	jae	.Lin_key_prologue
2628
2629	lea	56(%rax),%rax
2630
2631	mov	-8(%rax),%rbx
2632	mov	-16(%rax),%rbp
2633	mov	-24(%rax),%r12
2634	mov	-32(%rax),%r13
2635	mov	-40(%rax),%r14
2636	mov	-48(%rax),%r15
2637	mov	%rbx,144($context)	# restore context->Rbx
2638	mov	%rbp,160($context)	# restore context->Rbp
2639	mov	%r12,216($context)	# restore context->R12
2640	mov	%r13,224($context)	# restore context->R13
2641	mov	%r14,232($context)	# restore context->R14
2642	mov	%r15,240($context)	# restore context->R15
2643
2644.Lin_key_prologue:
2645	mov	8(%rax),%rdi
2646	mov	16(%rax),%rsi
2647	mov	%rax,152($context)	# restore context->Rsp
2648	mov	%rsi,168($context)	# restore context->Rsi
2649	mov	%rdi,176($context)	# restore context->Rdi
2650
2651	jmp	.Lcommon_seh_exit
2652.size	key_se_handler,.-key_se_handler
2653
2654.type	cbc_se_handler,\@abi-omnipotent
2655.align	16
2656cbc_se_handler:
2657	push	%rsi
2658	push	%rdi
2659	push	%rbx
2660	push	%rbp
2661	push	%r12
2662	push	%r13
2663	push	%r14
2664	push	%r15
2665	pushfq
2666	sub	\$64,%rsp
2667
2668	mov	120($context),%rax	# pull context->Rax
2669	mov	248($context),%rbx	# pull context->Rip
2670
2671	lea	.Lcbc_prologue(%rip),%r10
2672	cmp	%r10,%rbx		# context->Rip<.Lcbc_prologue
2673	jb	.Lin_cbc_prologue
2674
2675	lea	.Lcbc_fast_body(%rip),%r10
2676	cmp	%r10,%rbx		# context->Rip<.Lcbc_fast_body
2677	jb	.Lin_cbc_frame_setup
2678
2679	lea	.Lcbc_slow_prologue(%rip),%r10
2680	cmp	%r10,%rbx		# context->Rip<.Lcbc_slow_prologue
2681	jb	.Lin_cbc_body
2682
2683	lea	.Lcbc_slow_body(%rip),%r10
2684	cmp	%r10,%rbx		# context->Rip<.Lcbc_slow_body
2685	jb	.Lin_cbc_frame_setup
2686
2687.Lin_cbc_body:
2688	mov	152($context),%rax	# pull context->Rsp
2689
2690	lea	.Lcbc_epilogue(%rip),%r10
2691	cmp	%r10,%rbx		# context->Rip>=.Lcbc_epilogue
2692	jae	.Lin_cbc_prologue
2693
2694	lea	8(%rax),%rax
2695
2696	lea	.Lcbc_popfq(%rip),%r10
2697	cmp	%r10,%rbx		# context->Rip>=.Lcbc_popfq
2698	jae	.Lin_cbc_prologue
2699
2700	mov	`16-8`(%rax),%rax	# biased $_rsp
2701	lea	56(%rax),%rax
2702
2703.Lin_cbc_frame_setup:
2704	mov	-16(%rax),%rbx
2705	mov	-24(%rax),%rbp
2706	mov	-32(%rax),%r12
2707	mov	-40(%rax),%r13
2708	mov	-48(%rax),%r14
2709	mov	-56(%rax),%r15
2710	mov	%rbx,144($context)	# restore context->Rbx
2711	mov	%rbp,160($context)	# restore context->Rbp
2712	mov	%r12,216($context)	# restore context->R12
2713	mov	%r13,224($context)	# restore context->R13
2714	mov	%r14,232($context)	# restore context->R14
2715	mov	%r15,240($context)	# restore context->R15
2716
2717.Lin_cbc_prologue:
2718	mov	8(%rax),%rdi
2719	mov	16(%rax),%rsi
2720	mov	%rax,152($context)	# restore context->Rsp
2721	mov	%rsi,168($context)	# restore context->Rsi
2722	mov	%rdi,176($context)	# restore context->Rdi
2723
2724.Lcommon_seh_exit:
2725
2726	mov	40($disp),%rdi		# disp->ContextRecord
2727	mov	$context,%rsi		# context
2728	mov	\$`1232/8`,%ecx		# sizeof(CONTEXT)
2729	.long	0xa548f3fc		# cld; rep movsq
2730
2731	mov	$disp,%rsi
2732	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
2733	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
2734	mov	0(%rsi),%r8		# arg3, disp->ControlPc
2735	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
2736	mov	40(%rsi),%r10		# disp->ContextRecord
2737	lea	56(%rsi),%r11		# &disp->HandlerData
2738	lea	24(%rsi),%r12		# &disp->EstablisherFrame
2739	mov	%r10,32(%rsp)		# arg5
2740	mov	%r11,40(%rsp)		# arg6
2741	mov	%r12,48(%rsp)		# arg7
2742	mov	%rcx,56(%rsp)		# arg8, (NULL)
2743	call	*__imp_RtlVirtualUnwind(%rip)
2744
2745	mov	\$1,%eax		# ExceptionContinueSearch
2746	add	\$64,%rsp
2747	popfq
2748	pop	%r15
2749	pop	%r14
2750	pop	%r13
2751	pop	%r12
2752	pop	%rbp
2753	pop	%rbx
2754	pop	%rdi
2755	pop	%rsi
2756	ret
2757.size	cbc_se_handler,.-cbc_se_handler
2758
2759.section	.pdata
2760.align	4
2761	.rva	.LSEH_begin_AES_encrypt
2762	.rva	.LSEH_end_AES_encrypt
2763	.rva	.LSEH_info_AES_encrypt
2764
2765	.rva	.LSEH_begin_AES_decrypt
2766	.rva	.LSEH_end_AES_decrypt
2767	.rva	.LSEH_info_AES_decrypt
2768
2769	.rva	.LSEH_begin_AES_set_encrypt_key
2770	.rva	.LSEH_end_AES_set_encrypt_key
2771	.rva	.LSEH_info_AES_set_encrypt_key
2772
2773	.rva	.LSEH_begin_AES_set_decrypt_key
2774	.rva	.LSEH_end_AES_set_decrypt_key
2775	.rva	.LSEH_info_AES_set_decrypt_key
2776
2777	.rva	.LSEH_begin_AES_cbc_encrypt
2778	.rva	.LSEH_end_AES_cbc_encrypt
2779	.rva	.LSEH_info_AES_cbc_encrypt
2780
2781.section	.xdata
2782.align	8
2783.LSEH_info_AES_encrypt:
2784	.byte	9,0,0,0
2785	.rva	block_se_handler
2786	.rva	.Lenc_prologue,.Lenc_epilogue	# HandlerData[]
2787.LSEH_info_AES_decrypt:
2788	.byte	9,0,0,0
2789	.rva	block_se_handler
2790	.rva	.Ldec_prologue,.Ldec_epilogue	# HandlerData[]
2791.LSEH_info_AES_set_encrypt_key:
2792	.byte	9,0,0,0
2793	.rva	key_se_handler
2794	.rva	.Lenc_key_prologue,.Lenc_key_epilogue	# HandlerData[]
2795.LSEH_info_AES_set_decrypt_key:
2796	.byte	9,0,0,0
2797	.rva	key_se_handler
2798	.rva	.Ldec_key_prologue,.Ldec_key_epilogue	# HandlerData[]
2799.LSEH_info_AES_cbc_encrypt:
2800	.byte	9,0,0,0
2801	.rva	cbc_se_handler
2802___
2803}
2804
2805$code =~ s/\`([^\`]*)\`/eval($1)/gem;
2806
2807print $code;
2808
2809close STDOUT;
2810