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		43		13.0
23# EM64T		38		56		18.6(*)
24# Core 2	30		42		14.5(*)
25# Atom		65		86		32.1(*)
26#
27# (*) with hyper-threading off
28
29$flavour = shift;
30$output  = shift;
31if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
32
33$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
34
35$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
36( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
37( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
38die "can't locate x86_64-xlate.pl";
39
40open OUT,"| \"$^X\" $xlate $flavour $output";
41*STDOUT=*OUT;
42
43$verticalspin=1;	# unlike 32-bit version $verticalspin performs
44			# ~15% better on both AMD and Intel cores
45$speed_limit=512;	# see aes-586.pl for details
46
47$code=".text\n";
48
49$s0="%eax";
50$s1="%ebx";
51$s2="%ecx";
52$s3="%edx";
53$acc0="%esi";	$mask80="%rsi";
54$acc1="%edi";	$maskfe="%rdi";
55$acc2="%ebp";	$mask1b="%rbp";
56$inp="%r8";
57$out="%r9";
58$t0="%r10d";
59$t1="%r11d";
60$t2="%r12d";
61$rnds="%r13d";
62$sbox="%r14";
63$key="%r15";
64
65sub hi() { my $r=shift;	$r =~ s/%[er]([a-d])x/%\1h/;	$r; }
66sub lo() { my $r=shift;	$r =~ s/%[er]([a-d])x/%\1l/;
67			$r =~ s/%[er]([sd]i)/%\1l/;
68			$r =~ s/%(r[0-9]+)[d]?/%\1b/;	$r; }
69sub LO() { my $r=shift; $r =~ s/%r([a-z]+)/%e\1/;
70			$r =~ s/%r([0-9]+)/%r\1d/;	$r; }
71sub _data_word()
72{ my $i;
73    while(defined($i=shift)) { $code.=sprintf".long\t0x%08x,0x%08x\n",$i,$i; }
74}
75sub data_word()
76{ my $i;
77  my $last=pop(@_);
78    $code.=".long\t";
79    while(defined($i=shift)) { $code.=sprintf"0x%08x,",$i; }
80    $code.=sprintf"0x%08x\n",$last;
81}
82
83sub data_byte()
84{ my $i;
85  my $last=pop(@_);
86    $code.=".byte\t";
87    while(defined($i=shift)) { $code.=sprintf"0x%02x,",$i&0xff; }
88    $code.=sprintf"0x%02x\n",$last&0xff;
89}
90
91sub encvert()
92{ my $t3="%r8d";	# zaps $inp!
93
94$code.=<<___;
95	# favor 3-way issue Opteron pipeline...
96	movzb	`&lo("$s0")`,$acc0
97	movzb	`&lo("$s1")`,$acc1
98	movzb	`&lo("$s2")`,$acc2
99	mov	0($sbox,$acc0,8),$t0
100	mov	0($sbox,$acc1,8),$t1
101	mov	0($sbox,$acc2,8),$t2
102
103	movzb	`&hi("$s1")`,$acc0
104	movzb	`&hi("$s2")`,$acc1
105	movzb	`&lo("$s3")`,$acc2
106	xor	3($sbox,$acc0,8),$t0
107	xor	3($sbox,$acc1,8),$t1
108	mov	0($sbox,$acc2,8),$t3
109
110	movzb	`&hi("$s3")`,$acc0
111	shr	\$16,$s2
112	movzb	`&hi("$s0")`,$acc2
113	xor	3($sbox,$acc0,8),$t2
114	shr	\$16,$s3
115	xor	3($sbox,$acc2,8),$t3
116
117	shr	\$16,$s1
118	lea	16($key),$key
119	shr	\$16,$s0
120
121	movzb	`&lo("$s2")`,$acc0
122	movzb	`&lo("$s3")`,$acc1
123	movzb	`&lo("$s0")`,$acc2
124	xor	2($sbox,$acc0,8),$t0
125	xor	2($sbox,$acc1,8),$t1
126	xor	2($sbox,$acc2,8),$t2
127
128	movzb	`&hi("$s3")`,$acc0
129	movzb	`&hi("$s0")`,$acc1
130	movzb	`&lo("$s1")`,$acc2
131	xor	1($sbox,$acc0,8),$t0
132	xor	1($sbox,$acc1,8),$t1
133	xor	2($sbox,$acc2,8),$t3
134
135	mov	12($key),$s3
136	movzb	`&hi("$s1")`,$acc1
137	movzb	`&hi("$s2")`,$acc2
138	mov	0($key),$s0
139	xor	1($sbox,$acc1,8),$t2
140	xor	1($sbox,$acc2,8),$t3
141
142	mov	4($key),$s1
143	mov	8($key),$s2
144	xor	$t0,$s0
145	xor	$t1,$s1
146	xor	$t2,$s2
147	xor	$t3,$s3
148___
149}
150
151sub enclastvert()
152{ my $t3="%r8d";	# zaps $inp!
153
154$code.=<<___;
155	movzb	`&lo("$s0")`,$acc0
156	movzb	`&lo("$s1")`,$acc1
157	movzb	`&lo("$s2")`,$acc2
158	movzb	2($sbox,$acc0,8),$t0
159	movzb	2($sbox,$acc1,8),$t1
160	movzb	2($sbox,$acc2,8),$t2
161
162	movzb	`&lo("$s3")`,$acc0
163	movzb	`&hi("$s1")`,$acc1
164	movzb	`&hi("$s2")`,$acc2
165	movzb	2($sbox,$acc0,8),$t3
166	mov	0($sbox,$acc1,8),$acc1	#$t0
167	mov	0($sbox,$acc2,8),$acc2	#$t1
168
169	and	\$0x0000ff00,$acc1
170	and	\$0x0000ff00,$acc2
171
172	xor	$acc1,$t0
173	xor	$acc2,$t1
174	shr	\$16,$s2
175
176	movzb	`&hi("$s3")`,$acc0
177	movzb	`&hi("$s0")`,$acc1
178	shr	\$16,$s3
179	mov	0($sbox,$acc0,8),$acc0	#$t2
180	mov	0($sbox,$acc1,8),$acc1	#$t3
181
182	and	\$0x0000ff00,$acc0
183	and	\$0x0000ff00,$acc1
184	shr	\$16,$s1
185	xor	$acc0,$t2
186	xor	$acc1,$t3
187	shr	\$16,$s0
188
189	movzb	`&lo("$s2")`,$acc0
190	movzb	`&lo("$s3")`,$acc1
191	movzb	`&lo("$s0")`,$acc2
192	mov	0($sbox,$acc0,8),$acc0	#$t0
193	mov	0($sbox,$acc1,8),$acc1	#$t1
194	mov	0($sbox,$acc2,8),$acc2	#$t2
195
196	and	\$0x00ff0000,$acc0
197	and	\$0x00ff0000,$acc1
198	and	\$0x00ff0000,$acc2
199
200	xor	$acc0,$t0
201	xor	$acc1,$t1
202	xor	$acc2,$t2
203
204	movzb	`&lo("$s1")`,$acc0
205	movzb	`&hi("$s3")`,$acc1
206	movzb	`&hi("$s0")`,$acc2
207	mov	0($sbox,$acc0,8),$acc0	#$t3
208	mov	2($sbox,$acc1,8),$acc1	#$t0
209	mov	2($sbox,$acc2,8),$acc2	#$t1
210
211	and	\$0x00ff0000,$acc0
212	and	\$0xff000000,$acc1
213	and	\$0xff000000,$acc2
214
215	xor	$acc0,$t3
216	xor	$acc1,$t0
217	xor	$acc2,$t1
218
219	movzb	`&hi("$s1")`,$acc0
220	movzb	`&hi("$s2")`,$acc1
221	mov	16+12($key),$s3
222	mov	2($sbox,$acc0,8),$acc0	#$t2
223	mov	2($sbox,$acc1,8),$acc1	#$t3
224	mov	16+0($key),$s0
225
226	and	\$0xff000000,$acc0
227	and	\$0xff000000,$acc1
228
229	xor	$acc0,$t2
230	xor	$acc1,$t3
231
232	mov	16+4($key),$s1
233	mov	16+8($key),$s2
234	xor	$t0,$s0
235	xor	$t1,$s1
236	xor	$t2,$s2
237	xor	$t3,$s3
238___
239}
240
241sub encstep()
242{ my ($i,@s) = @_;
243  my $tmp0=$acc0;
244  my $tmp1=$acc1;
245  my $tmp2=$acc2;
246  my $out=($t0,$t1,$t2,$s[0])[$i];
247
248	if ($i==3) {
249		$tmp0=$s[1];
250		$tmp1=$s[2];
251		$tmp2=$s[3];
252	}
253	$code.="	movzb	".&lo($s[0]).",$out\n";
254	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
255	$code.="	lea	16($key),$key\n"	if ($i==0);
256
257	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
258	$code.="	mov	0($sbox,$out,8),$out\n";
259
260	$code.="	shr	\$16,$tmp1\n";
261	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
262	$code.="	xor	3($sbox,$tmp0,8),$out\n";
263
264	$code.="	movzb	".&lo($tmp1).",$tmp1\n";
265	$code.="	shr	\$24,$tmp2\n";
266	$code.="	xor	4*$i($key),$out\n";
267
268	$code.="	xor	2($sbox,$tmp1,8),$out\n";
269	$code.="	xor	1($sbox,$tmp2,8),$out\n";
270
271	$code.="	mov	$t0,$s[1]\n"		if ($i==3);
272	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
273	$code.="	mov	$t2,$s[3]\n"		if ($i==3);
274	$code.="\n";
275}
276
277sub enclast()
278{ my ($i,@s)=@_;
279  my $tmp0=$acc0;
280  my $tmp1=$acc1;
281  my $tmp2=$acc2;
282  my $out=($t0,$t1,$t2,$s[0])[$i];
283
284	if ($i==3) {
285		$tmp0=$s[1];
286		$tmp1=$s[2];
287		$tmp2=$s[3];
288	}
289	$code.="	movzb	".&lo($s[0]).",$out\n";
290	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
291
292	$code.="	mov	2($sbox,$out,8),$out\n";
293	$code.="	shr	\$16,$tmp1\n";
294	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
295
296	$code.="	and	\$0x000000ff,$out\n";
297	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
298	$code.="	movzb	".&lo($tmp1).",$tmp1\n";
299	$code.="	shr	\$24,$tmp2\n";
300
301	$code.="	mov	0($sbox,$tmp0,8),$tmp0\n";
302	$code.="	mov	0($sbox,$tmp1,8),$tmp1\n";
303	$code.="	mov	2($sbox,$tmp2,8),$tmp2\n";
304
305	$code.="	and	\$0x0000ff00,$tmp0\n";
306	$code.="	and	\$0x00ff0000,$tmp1\n";
307	$code.="	and	\$0xff000000,$tmp2\n";
308
309	$code.="	xor	$tmp0,$out\n";
310	$code.="	mov	$t0,$s[1]\n"		if ($i==3);
311	$code.="	xor	$tmp1,$out\n";
312	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
313	$code.="	xor	$tmp2,$out\n";
314	$code.="	mov	$t2,$s[3]\n"		if ($i==3);
315	$code.="\n";
316}
317
318$code.=<<___;
319.type	_x86_64_AES_encrypt,\@abi-omnipotent
320.align	16
321_x86_64_AES_encrypt:
322	xor	0($key),$s0			# xor with key
323	xor	4($key),$s1
324	xor	8($key),$s2
325	xor	12($key),$s3
326
327	mov	240($key),$rnds			# load key->rounds
328	sub	\$1,$rnds
329	jmp	.Lenc_loop
330.align	16
331.Lenc_loop:
332___
333	if ($verticalspin) { &encvert(); }
334	else {	&encstep(0,$s0,$s1,$s2,$s3);
335		&encstep(1,$s1,$s2,$s3,$s0);
336		&encstep(2,$s2,$s3,$s0,$s1);
337		&encstep(3,$s3,$s0,$s1,$s2);
338	}
339$code.=<<___;
340	sub	\$1,$rnds
341	jnz	.Lenc_loop
342___
343	if ($verticalspin) { &enclastvert(); }
344	else {	&enclast(0,$s0,$s1,$s2,$s3);
345		&enclast(1,$s1,$s2,$s3,$s0);
346		&enclast(2,$s2,$s3,$s0,$s1);
347		&enclast(3,$s3,$s0,$s1,$s2);
348		$code.=<<___;
349		xor	16+0($key),$s0		# xor with key
350		xor	16+4($key),$s1
351		xor	16+8($key),$s2
352		xor	16+12($key),$s3
353___
354	}
355$code.=<<___;
356	.byte	0xf3,0xc3			# rep ret
357.size	_x86_64_AES_encrypt,.-_x86_64_AES_encrypt
358___
359
360# it's possible to implement this by shifting tN by 8, filling least
361# significant byte with byte load and finally bswap-ing at the end,
362# but such partial register load kills Core 2...
363sub enccompactvert()
364{ my ($t3,$t4,$t5)=("%r8d","%r9d","%r13d");
365
366$code.=<<___;
367	movzb	`&lo("$s0")`,$t0
368	movzb	`&lo("$s1")`,$t1
369	movzb	`&lo("$s2")`,$t2
370	movzb	`&lo("$s3")`,$t3
371	movzb	`&hi("$s1")`,$acc0
372	movzb	`&hi("$s2")`,$acc1
373	shr	\$16,$s2
374	movzb	`&hi("$s3")`,$acc2
375	movzb	($sbox,$t0,1),$t0
376	movzb	($sbox,$t1,1),$t1
377	movzb	($sbox,$t2,1),$t2
378	movzb	($sbox,$t3,1),$t3
379
380	movzb	($sbox,$acc0,1),$t4	#$t0
381	movzb	`&hi("$s0")`,$acc0
382	movzb	($sbox,$acc1,1),$t5	#$t1
383	movzb	`&lo("$s2")`,$acc1
384	movzb	($sbox,$acc2,1),$acc2	#$t2
385	movzb	($sbox,$acc0,1),$acc0	#$t3
386
387	shl	\$8,$t4
388	shr	\$16,$s3
389	shl	\$8,$t5
390	xor	$t4,$t0
391	shr	\$16,$s0
392	movzb	`&lo("$s3")`,$t4
393	shr	\$16,$s1
394	xor	$t5,$t1
395	shl	\$8,$acc2
396	movzb	`&lo("$s0")`,$t5
397	movzb	($sbox,$acc1,1),$acc1	#$t0
398	xor	$acc2,$t2
399
400	shl	\$8,$acc0
401	movzb	`&lo("$s1")`,$acc2
402	shl	\$16,$acc1
403	xor	$acc0,$t3
404	movzb	($sbox,$t4,1),$t4	#$t1
405	movzb	`&hi("$s3")`,$acc0
406	movzb	($sbox,$t5,1),$t5	#$t2
407	xor	$acc1,$t0
408
409	shr	\$8,$s2
410	movzb	`&hi("$s0")`,$acc1
411	shl	\$16,$t4
412	shr	\$8,$s1
413	shl	\$16,$t5
414	xor	$t4,$t1
415	movzb	($sbox,$acc2,1),$acc2	#$t3
416	movzb	($sbox,$acc0,1),$acc0	#$t0
417	movzb	($sbox,$acc1,1),$acc1	#$t1
418	movzb	($sbox,$s2,1),$s3	#$t3
419	movzb	($sbox,$s1,1),$s2	#$t2
420
421	shl	\$16,$acc2
422	xor	$t5,$t2
423	shl	\$24,$acc0
424	xor	$acc2,$t3
425	shl	\$24,$acc1
426	xor	$acc0,$t0
427	shl	\$24,$s3
428	xor	$acc1,$t1
429	shl	\$24,$s2
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$code.=<<___;
468	mov	\$0x80808080,$t0
469	mov	\$0x80808080,$t1
470	and	$s0,$t0
471	and	$s1,$t1
472	mov	$t0,$acc0
473	mov	$t1,$acc1
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	\$0x80808080,$t2
492	rol	\$24,$s0
493	 mov	\$0x80808080,$t3
494	rol	\$24,$s1
495	 and	$s2,$t2
496	 and	$s3,$t3
497	xor	$r20,$s0
498	xor	$r21,$s1
499	 mov	$t2,$acc0
500	ror	\$16,$t0
501	 mov	$t3,$acc1
502	ror	\$16,$t1
503	 lea	($s2,$s2),$r20
504	 shr	\$7,$t2
505	xor	$t0,$s0
506	 shr	\$7,$t3
507	xor	$t1,$s1
508	ror	\$8,$t0
509	 lea	($s3,$s3),$r21
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	ror	\$16,$t2
526	xor	$r20,$s2
527	ror	\$16,$t3
528	xor	$r21,$s3
529	rol	\$24,$s2
530	mov	0($sbox),$acc0			# prefetch Te4
531	rol	\$24,$s3
532	xor	$r20,$s2
533	mov	64($sbox),$acc1
534	xor	$r21,$s3
535	mov	128($sbox),$r20
536	xor	$t2,$s2
537	ror	\$8,$t2
538	xor	$t3,$s3
539	ror	\$8,$t3
540	xor	$t2,$s2
541	mov	192($sbox),$r21
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$code.=<<___;
588.globl	AES_encrypt
589.type	AES_encrypt,\@function,3
590.align	16
591.globl	asm_AES_encrypt
592.hidden	asm_AES_encrypt
593asm_AES_encrypt:
594AES_encrypt:
595	push	%rbx
596	push	%rbp
597	push	%r12
598	push	%r13
599	push	%r14
600	push	%r15
601
602	# allocate frame "above" key schedule
603	mov	%rsp,%r10
604	lea	-63(%rdx),%rcx	# %rdx is key argument
605	and	\$-64,%rsp
606	sub	%rsp,%rcx
607	neg	%rcx
608	and	\$0x3c0,%rcx
609	sub	%rcx,%rsp
610	sub	\$32,%rsp
611
612	mov	%rsi,16(%rsp)	# save out
613	mov	%r10,24(%rsp)	# save real stack pointer
614.Lenc_prologue:
615
616	mov	%rdx,$key
617	mov	240($key),$rnds	# load rounds
618
619	mov	0(%rdi),$s0	# load input vector
620	mov	4(%rdi),$s1
621	mov	8(%rdi),$s2
622	mov	12(%rdi),$s3
623
624	shl	\$4,$rnds
625	lea	($key,$rnds),%rbp
626	mov	$key,(%rsp)	# key schedule
627	mov	%rbp,8(%rsp)	# end of key schedule
628
629	# pick Te4 copy which can't "overlap" with stack frame or key schedule
630	lea	.LAES_Te+2048(%rip),$sbox
631	lea	768(%rsp),%rbp
632	sub	$sbox,%rbp
633	and	\$0x300,%rbp
634	lea	($sbox,%rbp),$sbox
635
636	call	_x86_64_AES_encrypt_compact
637
638	mov	16(%rsp),$out	# restore out
639	mov	24(%rsp),%rsi	# restore saved stack pointer
640	mov	$s0,0($out)	# write output vector
641	mov	$s1,4($out)
642	mov	$s2,8($out)
643	mov	$s3,12($out)
644
645	mov	(%rsi),%r15
646	mov	8(%rsi),%r14
647	mov	16(%rsi),%r13
648	mov	24(%rsi),%r12
649	mov	32(%rsi),%rbp
650	mov	40(%rsi),%rbx
651	lea	48(%rsi),%rsp
652.Lenc_epilogue:
653	ret
654.size	AES_encrypt,.-AES_encrypt
655___
656
657#------------------------------------------------------------------#
658
659sub decvert()
660{ my $t3="%r8d";	# zaps $inp!
661
662$code.=<<___;
663	# favor 3-way issue Opteron pipeline...
664	movzb	`&lo("$s0")`,$acc0
665	movzb	`&lo("$s1")`,$acc1
666	movzb	`&lo("$s2")`,$acc2
667	mov	0($sbox,$acc0,8),$t0
668	mov	0($sbox,$acc1,8),$t1
669	mov	0($sbox,$acc2,8),$t2
670
671	movzb	`&hi("$s3")`,$acc0
672	movzb	`&hi("$s0")`,$acc1
673	movzb	`&lo("$s3")`,$acc2
674	xor	3($sbox,$acc0,8),$t0
675	xor	3($sbox,$acc1,8),$t1
676	mov	0($sbox,$acc2,8),$t3
677
678	movzb	`&hi("$s1")`,$acc0
679	shr	\$16,$s0
680	movzb	`&hi("$s2")`,$acc2
681	xor	3($sbox,$acc0,8),$t2
682	shr	\$16,$s3
683	xor	3($sbox,$acc2,8),$t3
684
685	shr	\$16,$s1
686	lea	16($key),$key
687	shr	\$16,$s2
688
689	movzb	`&lo("$s2")`,$acc0
690	movzb	`&lo("$s3")`,$acc1
691	movzb	`&lo("$s0")`,$acc2
692	xor	2($sbox,$acc0,8),$t0
693	xor	2($sbox,$acc1,8),$t1
694	xor	2($sbox,$acc2,8),$t2
695
696	movzb	`&hi("$s1")`,$acc0
697	movzb	`&hi("$s2")`,$acc1
698	movzb	`&lo("$s1")`,$acc2
699	xor	1($sbox,$acc0,8),$t0
700	xor	1($sbox,$acc1,8),$t1
701	xor	2($sbox,$acc2,8),$t3
702
703	movzb	`&hi("$s3")`,$acc0
704	mov	12($key),$s3
705	movzb	`&hi("$s0")`,$acc2
706	xor	1($sbox,$acc0,8),$t2
707	mov	0($key),$s0
708	xor	1($sbox,$acc2,8),$t3
709
710	xor	$t0,$s0
711	mov	4($key),$s1
712	mov	8($key),$s2
713	xor	$t2,$s2
714	xor	$t1,$s1
715	xor	$t3,$s3
716___
717}
718
719sub declastvert()
720{ my $t3="%r8d";	# zaps $inp!
721
722$code.=<<___;
723	lea	2048($sbox),$sbox	# size optimization
724	movzb	`&lo("$s0")`,$acc0
725	movzb	`&lo("$s1")`,$acc1
726	movzb	`&lo("$s2")`,$acc2
727	movzb	($sbox,$acc0,1),$t0
728	movzb	($sbox,$acc1,1),$t1
729	movzb	($sbox,$acc2,1),$t2
730
731	movzb	`&lo("$s3")`,$acc0
732	movzb	`&hi("$s3")`,$acc1
733	movzb	`&hi("$s0")`,$acc2
734	movzb	($sbox,$acc0,1),$t3
735	movzb	($sbox,$acc1,1),$acc1	#$t0
736	movzb	($sbox,$acc2,1),$acc2	#$t1
737
738	shl	\$8,$acc1
739	shl	\$8,$acc2
740
741	xor	$acc1,$t0
742	xor	$acc2,$t1
743	shr	\$16,$s3
744
745	movzb	`&hi("$s1")`,$acc0
746	movzb	`&hi("$s2")`,$acc1
747	shr	\$16,$s0
748	movzb	($sbox,$acc0,1),$acc0	#$t2
749	movzb	($sbox,$acc1,1),$acc1	#$t3
750
751	shl	\$8,$acc0
752	shl	\$8,$acc1
753	shr	\$16,$s1
754	xor	$acc0,$t2
755	xor	$acc1,$t3
756	shr	\$16,$s2
757
758	movzb	`&lo("$s2")`,$acc0
759	movzb	`&lo("$s3")`,$acc1
760	movzb	`&lo("$s0")`,$acc2
761	movzb	($sbox,$acc0,1),$acc0	#$t0
762	movzb	($sbox,$acc1,1),$acc1	#$t1
763	movzb	($sbox,$acc2,1),$acc2	#$t2
764
765	shl	\$16,$acc0
766	shl	\$16,$acc1
767	shl	\$16,$acc2
768
769	xor	$acc0,$t0
770	xor	$acc1,$t1
771	xor	$acc2,$t2
772
773	movzb	`&lo("$s1")`,$acc0
774	movzb	`&hi("$s1")`,$acc1
775	movzb	`&hi("$s2")`,$acc2
776	movzb	($sbox,$acc0,1),$acc0	#$t3
777	movzb	($sbox,$acc1,1),$acc1	#$t0
778	movzb	($sbox,$acc2,1),$acc2	#$t1
779
780	shl	\$16,$acc0
781	shl	\$24,$acc1
782	shl	\$24,$acc2
783
784	xor	$acc0,$t3
785	xor	$acc1,$t0
786	xor	$acc2,$t1
787
788	movzb	`&hi("$s3")`,$acc0
789	movzb	`&hi("$s0")`,$acc1
790	mov	16+12($key),$s3
791	movzb	($sbox,$acc0,1),$acc0	#$t2
792	movzb	($sbox,$acc1,1),$acc1	#$t3
793	mov	16+0($key),$s0
794
795	shl	\$24,$acc0
796	shl	\$24,$acc1
797
798	xor	$acc0,$t2
799	xor	$acc1,$t3
800
801	mov	16+4($key),$s1
802	mov	16+8($key),$s2
803	lea	-2048($sbox),$sbox
804	xor	$t0,$s0
805	xor	$t1,$s1
806	xor	$t2,$s2
807	xor	$t3,$s3
808___
809}
810
811sub decstep()
812{ my ($i,@s) = @_;
813  my $tmp0=$acc0;
814  my $tmp1=$acc1;
815  my $tmp2=$acc2;
816  my $out=($t0,$t1,$t2,$s[0])[$i];
817
818	$code.="	mov	$s[0],$out\n"		if ($i!=3);
819			$tmp1=$s[2]			if ($i==3);
820	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
821	$code.="	and	\$0xFF,$out\n";
822
823	$code.="	mov	0($sbox,$out,8),$out\n";
824	$code.="	shr	\$16,$tmp1\n";
825			$tmp2=$s[3]			if ($i==3);
826	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
827
828			$tmp0=$s[1]			if ($i==3);
829	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
830	$code.="	and	\$0xFF,$tmp1\n";
831	$code.="	shr	\$24,$tmp2\n";
832
833	$code.="	xor	3($sbox,$tmp0,8),$out\n";
834	$code.="	xor	2($sbox,$tmp1,8),$out\n";
835	$code.="	xor	1($sbox,$tmp2,8),$out\n";
836
837	$code.="	mov	$t2,$s[1]\n"		if ($i==3);
838	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
839	$code.="	mov	$t0,$s[3]\n"		if ($i==3);
840	$code.="\n";
841}
842
843sub declast()
844{ my ($i,@s)=@_;
845  my $tmp0=$acc0;
846  my $tmp1=$acc1;
847  my $tmp2=$acc2;
848  my $out=($t0,$t1,$t2,$s[0])[$i];
849
850	$code.="	mov	$s[0],$out\n"		if ($i!=3);
851			$tmp1=$s[2]			if ($i==3);
852	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
853	$code.="	and	\$0xFF,$out\n";
854
855	$code.="	movzb	2048($sbox,$out,1),$out\n";
856	$code.="	shr	\$16,$tmp1\n";
857			$tmp2=$s[3]			if ($i==3);
858	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
859
860			$tmp0=$s[1]			if ($i==3);
861	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
862	$code.="	and	\$0xFF,$tmp1\n";
863	$code.="	shr	\$24,$tmp2\n";
864
865	$code.="	movzb	2048($sbox,$tmp0,1),$tmp0\n";
866	$code.="	movzb	2048($sbox,$tmp1,1),$tmp1\n";
867	$code.="	movzb	2048($sbox,$tmp2,1),$tmp2\n";
868
869	$code.="	shl	\$8,$tmp0\n";
870	$code.="	shl	\$16,$tmp1\n";
871	$code.="	shl	\$24,$tmp2\n";
872
873	$code.="	xor	$tmp0,$out\n";
874	$code.="	mov	$t2,$s[1]\n"		if ($i==3);
875	$code.="	xor	$tmp1,$out\n";
876	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
877	$code.="	xor	$tmp2,$out\n";
878	$code.="	mov	$t0,$s[3]\n"		if ($i==3);
879	$code.="\n";
880}
881
882$code.=<<___;
883.type	_x86_64_AES_decrypt,\@abi-omnipotent
884.align	16
885_x86_64_AES_decrypt:
886	xor	0($key),$s0			# xor with key
887	xor	4($key),$s1
888	xor	8($key),$s2
889	xor	12($key),$s3
890
891	mov	240($key),$rnds			# load key->rounds
892	sub	\$1,$rnds
893	jmp	.Ldec_loop
894.align	16
895.Ldec_loop:
896___
897	if ($verticalspin) { &decvert(); }
898	else {	&decstep(0,$s0,$s3,$s2,$s1);
899		&decstep(1,$s1,$s0,$s3,$s2);
900		&decstep(2,$s2,$s1,$s0,$s3);
901		&decstep(3,$s3,$s2,$s1,$s0);
902		$code.=<<___;
903		lea	16($key),$key
904		xor	0($key),$s0			# xor with key
905		xor	4($key),$s1
906		xor	8($key),$s2
907		xor	12($key),$s3
908___
909	}
910$code.=<<___;
911	sub	\$1,$rnds
912	jnz	.Ldec_loop
913___
914	if ($verticalspin) { &declastvert(); }
915	else {	&declast(0,$s0,$s3,$s2,$s1);
916		&declast(1,$s1,$s0,$s3,$s2);
917		&declast(2,$s2,$s1,$s0,$s3);
918		&declast(3,$s3,$s2,$s1,$s0);
919		$code.=<<___;
920		xor	16+0($key),$s0			# xor with key
921		xor	16+4($key),$s1
922		xor	16+8($key),$s2
923		xor	16+12($key),$s3
924___
925	}
926$code.=<<___;
927	.byte	0xf3,0xc3			# rep ret
928.size	_x86_64_AES_decrypt,.-_x86_64_AES_decrypt
929___
930
931sub deccompactvert()
932{ my ($t3,$t4,$t5)=("%r8d","%r9d","%r13d");
933
934$code.=<<___;
935	movzb	`&lo("$s0")`,$t0
936	movzb	`&lo("$s1")`,$t1
937	movzb	`&lo("$s2")`,$t2
938	movzb	`&lo("$s3")`,$t3
939	movzb	`&hi("$s3")`,$acc0
940	movzb	`&hi("$s0")`,$acc1
941	shr	\$16,$s3
942	movzb	`&hi("$s1")`,$acc2
943	movzb	($sbox,$t0,1),$t0
944	movzb	($sbox,$t1,1),$t1
945	movzb	($sbox,$t2,1),$t2
946	movzb	($sbox,$t3,1),$t3
947
948	movzb	($sbox,$acc0,1),$t4	#$t0
949	movzb	`&hi("$s2")`,$acc0
950	movzb	($sbox,$acc1,1),$t5	#$t1
951	movzb	($sbox,$acc2,1),$acc2	#$t2
952	movzb	($sbox,$acc0,1),$acc0	#$t3
953
954	shr	\$16,$s2
955	shl	\$8,$t5
956	shl	\$8,$t4
957	movzb	`&lo("$s2")`,$acc1
958	shr	\$16,$s0
959	xor	$t4,$t0
960	shr	\$16,$s1
961	movzb	`&lo("$s3")`,$t4
962
963	shl	\$8,$acc2
964	xor	$t5,$t1
965	shl	\$8,$acc0
966	movzb	`&lo("$s0")`,$t5
967	movzb	($sbox,$acc1,1),$acc1	#$t0
968	xor	$acc2,$t2
969	movzb	`&lo("$s1")`,$acc2
970
971	shl	\$16,$acc1
972	xor	$acc0,$t3
973	movzb	($sbox,$t4,1),$t4	#$t1
974	movzb	`&hi("$s1")`,$acc0
975	movzb	($sbox,$acc2,1),$acc2	#$t3
976	xor	$acc1,$t0
977	movzb	($sbox,$t5,1),$t5	#$t2
978	movzb	`&hi("$s2")`,$acc1
979
980	shl	\$16,$acc2
981	shl	\$16,$t4
982	shl	\$16,$t5
983	xor	$acc2,$t3
984	movzb	`&hi("$s3")`,$acc2
985	xor	$t4,$t1
986	shr	\$8,$s0
987	xor	$t5,$t2
988
989	movzb	($sbox,$acc0,1),$acc0	#$t0
990	movzb	($sbox,$acc1,1),$s1	#$t1
991	movzb	($sbox,$acc2,1),$s2	#$t2
992	movzb	($sbox,$s0,1),$s3	#$t3
993
994	mov	$t0,$s0
995	shl	\$24,$acc0
996	shl	\$24,$s1
997	shl	\$24,$s2
998	xor	$acc0,$s0
999	shl	\$24,$s3
1000	xor	$t1,$s1
1001	xor	$t2,$s2
1002	xor	$t3,$s3
1003___
1004}
1005
1006# parallelized version! input is pair of 64-bit values: %rax=s1.s0
1007# and %rcx=s3.s2, output is four 32-bit values in %eax=s0, %ebx=s1,
1008# %ecx=s2 and %edx=s3.
1009sub dectransform()
1010{ my ($tp10,$tp20,$tp40,$tp80,$acc0)=("%rax","%r8", "%r9", "%r10","%rbx");
1011  my ($tp18,$tp28,$tp48,$tp88,$acc8)=("%rcx","%r11","%r12","%r13","%rdx");
1012  my $prefetch = shift;
1013
1014$code.=<<___;
1015	mov	$mask80,$tp40
1016	mov	$mask80,$tp48
1017	and	$tp10,$tp40
1018	and	$tp18,$tp48
1019	mov	$tp40,$acc0
1020	mov	$tp48,$acc8
1021	shr	\$7,$tp40
1022	lea	($tp10,$tp10),$tp20
1023	shr	\$7,$tp48
1024	lea	($tp18,$tp18),$tp28
1025	sub	$tp40,$acc0
1026	sub	$tp48,$acc8
1027	and	$maskfe,$tp20
1028	and	$maskfe,$tp28
1029	and	$mask1b,$acc0
1030	and	$mask1b,$acc8
1031	xor	$acc0,$tp20
1032	xor	$acc8,$tp28
1033	mov	$mask80,$tp80
1034	mov	$mask80,$tp88
1035
1036	and	$tp20,$tp80
1037	and	$tp28,$tp88
1038	mov	$tp80,$acc0
1039	mov	$tp88,$acc8
1040	shr	\$7,$tp80
1041	lea	($tp20,$tp20),$tp40
1042	shr	\$7,$tp88
1043	lea	($tp28,$tp28),$tp48
1044	sub	$tp80,$acc0
1045	sub	$tp88,$acc8
1046	and	$maskfe,$tp40
1047	and	$maskfe,$tp48
1048	and	$mask1b,$acc0
1049	and	$mask1b,$acc8
1050	xor	$acc0,$tp40
1051	xor	$acc8,$tp48
1052	mov	$mask80,$tp80
1053	mov	$mask80,$tp88
1054
1055	and	$tp40,$tp80
1056	and	$tp48,$tp88
1057	mov	$tp80,$acc0
1058	mov	$tp88,$acc8
1059	shr	\$7,$tp80
1060	 xor	$tp10,$tp20		# tp2^=tp1
1061	shr	\$7,$tp88
1062	 xor	$tp18,$tp28		# tp2^=tp1
1063	sub	$tp80,$acc0
1064	sub	$tp88,$acc8
1065	lea	($tp40,$tp40),$tp80
1066	lea	($tp48,$tp48),$tp88
1067	 xor	$tp10,$tp40		# tp4^=tp1
1068	 xor	$tp18,$tp48		# tp4^=tp1
1069	and	$maskfe,$tp80
1070	and	$maskfe,$tp88
1071	and	$mask1b,$acc0
1072	and	$mask1b,$acc8
1073	xor	$acc0,$tp80
1074	xor	$acc8,$tp88
1075
1076	xor	$tp80,$tp10		# tp1^=tp8
1077	xor	$tp88,$tp18		# tp1^=tp8
1078	xor	$tp80,$tp20		# tp2^tp1^=tp8
1079	xor	$tp88,$tp28		# tp2^tp1^=tp8
1080	mov	$tp10,$acc0
1081	mov	$tp18,$acc8
1082	xor	$tp80,$tp40		# tp4^tp1^=tp8
1083	shr	\$32,$acc0
1084	xor	$tp88,$tp48		# tp4^tp1^=tp8
1085	shr	\$32,$acc8
1086	xor	$tp20,$tp80		# tp8^=tp8^tp2^tp1=tp2^tp1
1087	rol	\$8,`&LO("$tp10")`	# ROTATE(tp1^tp8,8)
1088	xor	$tp28,$tp88		# tp8^=tp8^tp2^tp1=tp2^tp1
1089	rol	\$8,`&LO("$tp18")`	# ROTATE(tp1^tp8,8)
1090	xor	$tp40,$tp80		# tp2^tp1^=tp8^tp4^tp1=tp8^tp4^tp2
1091	rol	\$8,`&LO("$acc0")`	# ROTATE(tp1^tp8,8)
1092	xor	$tp48,$tp88		# tp2^tp1^=tp8^tp4^tp1=tp8^tp4^tp2
1093
1094	rol	\$8,`&LO("$acc8")`	# ROTATE(tp1^tp8,8)
1095	xor	`&LO("$tp80")`,`&LO("$tp10")`
1096	shr	\$32,$tp80
1097	xor	`&LO("$tp88")`,`&LO("$tp18")`
1098	shr	\$32,$tp88
1099	xor	`&LO("$tp80")`,`&LO("$acc0")`
1100	xor	`&LO("$tp88")`,`&LO("$acc8")`
1101
1102	mov	$tp20,$tp80
1103	rol	\$24,`&LO("$tp20")`	# ROTATE(tp2^tp1^tp8,24)
1104	mov	$tp28,$tp88
1105	rol	\$24,`&LO("$tp28")`	# ROTATE(tp2^tp1^tp8,24)
1106	shr	\$32,$tp80
1107	xor	`&LO("$tp20")`,`&LO("$tp10")`
1108	shr	\$32,$tp88
1109	xor	`&LO("$tp28")`,`&LO("$tp18")`
1110	rol	\$24,`&LO("$tp80")`	# ROTATE(tp2^tp1^tp8,24)
1111	mov	$tp40,$tp20
1112	rol	\$24,`&LO("$tp88")`	# ROTATE(tp2^tp1^tp8,24)
1113	mov	$tp48,$tp28
1114	shr	\$32,$tp20
1115	xor	`&LO("$tp80")`,`&LO("$acc0")`
1116	shr	\$32,$tp28
1117	xor	`&LO("$tp88")`,`&LO("$acc8")`
1118
1119	`"mov	0($sbox),$mask80"	if ($prefetch)`
1120	rol	\$16,`&LO("$tp40")`	# ROTATE(tp4^tp1^tp8,16)
1121	`"mov	64($sbox),$maskfe"	if ($prefetch)`
1122	rol	\$16,`&LO("$tp48")`	# ROTATE(tp4^tp1^tp8,16)
1123	`"mov	128($sbox),$mask1b"	if ($prefetch)`
1124	rol	\$16,`&LO("$tp20")`	# ROTATE(tp4^tp1^tp8,16)
1125	`"mov	192($sbox),$tp80"	if ($prefetch)`
1126	xor	`&LO("$tp40")`,`&LO("$tp10")`
1127	rol	\$16,`&LO("$tp28")`	# ROTATE(tp4^tp1^tp8,16)
1128	xor	`&LO("$tp48")`,`&LO("$tp18")`
1129	`"mov	256($sbox),$tp88"	if ($prefetch)`
1130	xor	`&LO("$tp20")`,`&LO("$acc0")`
1131	xor	`&LO("$tp28")`,`&LO("$acc8")`
1132___
1133}
1134
1135$code.=<<___;
1136.type	_x86_64_AES_decrypt_compact,\@abi-omnipotent
1137.align	16
1138_x86_64_AES_decrypt_compact:
1139	lea	128($sbox),$inp			# size optimization
1140	mov	0-128($inp),$acc1		# prefetch Td4
1141	mov	32-128($inp),$acc2
1142	mov	64-128($inp),$t0
1143	mov	96-128($inp),$t1
1144	mov	128-128($inp),$acc1
1145	mov	160-128($inp),$acc2
1146	mov	192-128($inp),$t0
1147	mov	224-128($inp),$t1
1148	jmp	.Ldec_loop_compact
1149
1150.align	16
1151.Ldec_loop_compact:
1152		xor	0($key),$s0		# xor with key
1153		xor	4($key),$s1
1154		xor	8($key),$s2
1155		xor	12($key),$s3
1156		lea	16($key),$key
1157___
1158		&deccompactvert();
1159$code.=<<___;
1160		cmp	16(%rsp),$key
1161		je	.Ldec_compact_done
1162
1163		mov	256+0($sbox),$mask80
1164		shl	\$32,%rbx
1165		shl	\$32,%rdx
1166		mov	256+8($sbox),$maskfe
1167		or	%rbx,%rax
1168		or	%rdx,%rcx
1169		mov	256+16($sbox),$mask1b
1170___
1171		&dectransform(1);
1172$code.=<<___;
1173	jmp	.Ldec_loop_compact
1174.align	16
1175.Ldec_compact_done:
1176	xor	0($key),$s0
1177	xor	4($key),$s1
1178	xor	8($key),$s2
1179	xor	12($key),$s3
1180	.byte	0xf3,0xc3			# rep ret
1181.size	_x86_64_AES_decrypt_compact,.-_x86_64_AES_decrypt_compact
1182___
1183
1184# void AES_decrypt (const void *inp,void *out,const AES_KEY *key);
1185$code.=<<___;
1186.globl	AES_decrypt
1187.type	AES_decrypt,\@function,3
1188.align	16
1189.globl	asm_AES_decrypt
1190.hidden	asm_AES_decrypt
1191asm_AES_decrypt:
1192AES_decrypt:
1193	push	%rbx
1194	push	%rbp
1195	push	%r12
1196	push	%r13
1197	push	%r14
1198	push	%r15
1199
1200	# allocate frame "above" key schedule
1201	mov	%rsp,%r10
1202	lea	-63(%rdx),%rcx	# %rdx is key argument
1203	and	\$-64,%rsp
1204	sub	%rsp,%rcx
1205	neg	%rcx
1206	and	\$0x3c0,%rcx
1207	sub	%rcx,%rsp
1208	sub	\$32,%rsp
1209
1210	mov	%rsi,16(%rsp)	# save out
1211	mov	%r10,24(%rsp)	# save real stack pointer
1212.Ldec_prologue:
1213
1214	mov	%rdx,$key
1215	mov	240($key),$rnds	# load rounds
1216
1217	mov	0(%rdi),$s0	# load input vector
1218	mov	4(%rdi),$s1
1219	mov	8(%rdi),$s2
1220	mov	12(%rdi),$s3
1221
1222	shl	\$4,$rnds
1223	lea	($key,$rnds),%rbp
1224	mov	$key,(%rsp)	# key schedule
1225	mov	%rbp,8(%rsp)	# end of key schedule
1226
1227	# pick Td4 copy which can't "overlap" with stack frame or key schedule
1228	lea	.LAES_Td+2048(%rip),$sbox
1229	lea	768(%rsp),%rbp
1230	sub	$sbox,%rbp
1231	and	\$0x300,%rbp
1232	lea	($sbox,%rbp),$sbox
1233	shr	\$3,%rbp	# recall "magic" constants!
1234	add	%rbp,$sbox
1235
1236	call	_x86_64_AES_decrypt_compact
1237
1238	mov	16(%rsp),$out	# restore out
1239	mov	24(%rsp),%rsi	# restore saved stack pointer
1240	mov	$s0,0($out)	# write output vector
1241	mov	$s1,4($out)
1242	mov	$s2,8($out)
1243	mov	$s3,12($out)
1244
1245	mov	(%rsi),%r15
1246	mov	8(%rsi),%r14
1247	mov	16(%rsi),%r13
1248	mov	24(%rsi),%r12
1249	mov	32(%rsi),%rbp
1250	mov	40(%rsi),%rbx
1251	lea	48(%rsi),%rsp
1252.Ldec_epilogue:
1253	ret
1254.size	AES_decrypt,.-AES_decrypt
1255___
1256#------------------------------------------------------------------#
1257
1258sub enckey()
1259{
1260$code.=<<___;
1261	movz	%dl,%esi		# rk[i]>>0
1262	movzb	-128(%rbp,%rsi),%ebx
1263	movz	%dh,%esi		# rk[i]>>8
1264	shl	\$24,%ebx
1265	xor	%ebx,%eax
1266
1267	movzb	-128(%rbp,%rsi),%ebx
1268	shr	\$16,%edx
1269	movz	%dl,%esi		# rk[i]>>16
1270	xor	%ebx,%eax
1271
1272	movzb	-128(%rbp,%rsi),%ebx
1273	movz	%dh,%esi		# rk[i]>>24
1274	shl	\$8,%ebx
1275	xor	%ebx,%eax
1276
1277	movzb	-128(%rbp,%rsi),%ebx
1278	shl	\$16,%ebx
1279	xor	%ebx,%eax
1280
1281	xor	1024-128(%rbp,%rcx,4),%eax		# rcon
1282___
1283}
1284
1285# int private_AES_set_encrypt_key(const unsigned char *userKey, const int bits,
1286#                        AES_KEY *key)
1287$code.=<<___;
1288.globl	private_AES_set_encrypt_key
1289.type	private_AES_set_encrypt_key,\@function,3
1290.align	16
1291private_AES_set_encrypt_key:
1292	push	%rbx
1293	push	%rbp
1294	push	%r12			# redundant, but allows to share
1295	push	%r13			# exception handler...
1296	push	%r14
1297	push	%r15
1298	sub	\$8,%rsp
1299.Lenc_key_prologue:
1300
1301	call	_x86_64_AES_set_encrypt_key
1302
1303	mov	40(%rsp),%rbp
1304	mov	48(%rsp),%rbx
1305	add	\$56,%rsp
1306.Lenc_key_epilogue:
1307	ret
1308.size	private_AES_set_encrypt_key,.-private_AES_set_encrypt_key
1309
1310.type	_x86_64_AES_set_encrypt_key,\@abi-omnipotent
1311.align	16
1312_x86_64_AES_set_encrypt_key:
1313	mov	%esi,%ecx			# %ecx=bits
1314	mov	%rdi,%rsi			# %rsi=userKey
1315	mov	%rdx,%rdi			# %rdi=key
1316
1317	test	\$-1,%rsi
1318	jz	.Lbadpointer
1319	test	\$-1,%rdi
1320	jz	.Lbadpointer
1321
1322	lea	.LAES_Te(%rip),%rbp
1323	lea	2048+128(%rbp),%rbp
1324
1325	# prefetch Te4
1326	mov	0-128(%rbp),%eax
1327	mov	32-128(%rbp),%ebx
1328	mov	64-128(%rbp),%r8d
1329	mov	96-128(%rbp),%edx
1330	mov	128-128(%rbp),%eax
1331	mov	160-128(%rbp),%ebx
1332	mov	192-128(%rbp),%r8d
1333	mov	224-128(%rbp),%edx
1334
1335	cmp	\$128,%ecx
1336	je	.L10rounds
1337	cmp	\$192,%ecx
1338	je	.L12rounds
1339	cmp	\$256,%ecx
1340	je	.L14rounds
1341	mov	\$-2,%rax			# invalid number of bits
1342	jmp	.Lexit
1343
1344.L10rounds:
1345	mov	0(%rsi),%rax			# copy first 4 dwords
1346	mov	8(%rsi),%rdx
1347	mov	%rax,0(%rdi)
1348	mov	%rdx,8(%rdi)
1349
1350	shr	\$32,%rdx
1351	xor	%ecx,%ecx
1352	jmp	.L10shortcut
1353.align	4
1354.L10loop:
1355		mov	0(%rdi),%eax			# rk[0]
1356		mov	12(%rdi),%edx			# rk[3]
1357.L10shortcut:
1358___
1359		&enckey	();
1360$code.=<<___;
1361		mov	%eax,16(%rdi)			# rk[4]
1362		xor	4(%rdi),%eax
1363		mov	%eax,20(%rdi)			# rk[5]
1364		xor	8(%rdi),%eax
1365		mov	%eax,24(%rdi)			# rk[6]
1366		xor	12(%rdi),%eax
1367		mov	%eax,28(%rdi)			# rk[7]
1368		add	\$1,%ecx
1369		lea	16(%rdi),%rdi
1370		cmp	\$10,%ecx
1371	jl	.L10loop
1372
1373	movl	\$10,80(%rdi)			# setup number of rounds
1374	xor	%rax,%rax
1375	jmp	.Lexit
1376
1377.L12rounds:
1378	mov	0(%rsi),%rax			# copy first 6 dwords
1379	mov	8(%rsi),%rbx
1380	mov	16(%rsi),%rdx
1381	mov	%rax,0(%rdi)
1382	mov	%rbx,8(%rdi)
1383	mov	%rdx,16(%rdi)
1384
1385	shr	\$32,%rdx
1386	xor	%ecx,%ecx
1387	jmp	.L12shortcut
1388.align	4
1389.L12loop:
1390		mov	0(%rdi),%eax			# rk[0]
1391		mov	20(%rdi),%edx			# rk[5]
1392.L12shortcut:
1393___
1394		&enckey	();
1395$code.=<<___;
1396		mov	%eax,24(%rdi)			# rk[6]
1397		xor	4(%rdi),%eax
1398		mov	%eax,28(%rdi)			# rk[7]
1399		xor	8(%rdi),%eax
1400		mov	%eax,32(%rdi)			# rk[8]
1401		xor	12(%rdi),%eax
1402		mov	%eax,36(%rdi)			# rk[9]
1403
1404		cmp	\$7,%ecx
1405		je	.L12break
1406		add	\$1,%ecx
1407
1408		xor	16(%rdi),%eax
1409		mov	%eax,40(%rdi)			# rk[10]
1410		xor	20(%rdi),%eax
1411		mov	%eax,44(%rdi)			# rk[11]
1412
1413		lea	24(%rdi),%rdi
1414	jmp	.L12loop
1415.L12break:
1416	movl	\$12,72(%rdi)		# setup number of rounds
1417	xor	%rax,%rax
1418	jmp	.Lexit
1419
1420.L14rounds:
1421	mov	0(%rsi),%rax			# copy first 8 dwords
1422	mov	8(%rsi),%rbx
1423	mov	16(%rsi),%rcx
1424	mov	24(%rsi),%rdx
1425	mov	%rax,0(%rdi)
1426	mov	%rbx,8(%rdi)
1427	mov	%rcx,16(%rdi)
1428	mov	%rdx,24(%rdi)
1429
1430	shr	\$32,%rdx
1431	xor	%ecx,%ecx
1432	jmp	.L14shortcut
1433.align	4
1434.L14loop:
1435		mov	0(%rdi),%eax			# rk[0]
1436		mov	28(%rdi),%edx			# rk[4]
1437.L14shortcut:
1438___
1439		&enckey	();
1440$code.=<<___;
1441		mov	%eax,32(%rdi)			# rk[8]
1442		xor	4(%rdi),%eax
1443		mov	%eax,36(%rdi)			# rk[9]
1444		xor	8(%rdi),%eax
1445		mov	%eax,40(%rdi)			# rk[10]
1446		xor	12(%rdi),%eax
1447		mov	%eax,44(%rdi)			# rk[11]
1448
1449		cmp	\$6,%ecx
1450		je	.L14break
1451		add	\$1,%ecx
1452
1453		mov	%eax,%edx
1454		mov	16(%rdi),%eax			# rk[4]
1455		movz	%dl,%esi			# rk[11]>>0
1456		movzb	-128(%rbp,%rsi),%ebx
1457		movz	%dh,%esi			# rk[11]>>8
1458		xor	%ebx,%eax
1459
1460		movzb	-128(%rbp,%rsi),%ebx
1461		shr	\$16,%edx
1462		shl	\$8,%ebx
1463		movz	%dl,%esi			# rk[11]>>16
1464		xor	%ebx,%eax
1465
1466		movzb	-128(%rbp,%rsi),%ebx
1467		movz	%dh,%esi			# rk[11]>>24
1468		shl	\$16,%ebx
1469		xor	%ebx,%eax
1470
1471		movzb	-128(%rbp,%rsi),%ebx
1472		shl	\$24,%ebx
1473		xor	%ebx,%eax
1474
1475		mov	%eax,48(%rdi)			# rk[12]
1476		xor	20(%rdi),%eax
1477		mov	%eax,52(%rdi)			# rk[13]
1478		xor	24(%rdi),%eax
1479		mov	%eax,56(%rdi)			# rk[14]
1480		xor	28(%rdi),%eax
1481		mov	%eax,60(%rdi)			# rk[15]
1482
1483		lea	32(%rdi),%rdi
1484	jmp	.L14loop
1485.L14break:
1486	movl	\$14,48(%rdi)		# setup number of rounds
1487	xor	%rax,%rax
1488	jmp	.Lexit
1489
1490.Lbadpointer:
1491	mov	\$-1,%rax
1492.Lexit:
1493	.byte	0xf3,0xc3			# rep ret
1494.size	_x86_64_AES_set_encrypt_key,.-_x86_64_AES_set_encrypt_key
1495___
1496
1497sub deckey_ref()
1498{ my ($i,$ptr,$te,$td) = @_;
1499  my ($tp1,$tp2,$tp4,$tp8,$acc)=("%eax","%ebx","%edi","%edx","%r8d");
1500$code.=<<___;
1501	mov	$i($ptr),$tp1
1502	mov	$tp1,$acc
1503	and	\$0x80808080,$acc
1504	mov	$acc,$tp4
1505	shr	\$7,$tp4
1506	lea	0($tp1,$tp1),$tp2
1507	sub	$tp4,$acc
1508	and	\$0xfefefefe,$tp2
1509	and	\$0x1b1b1b1b,$acc
1510	xor	$tp2,$acc
1511	mov	$acc,$tp2
1512
1513	and	\$0x80808080,$acc
1514	mov	$acc,$tp8
1515	shr	\$7,$tp8
1516	lea	0($tp2,$tp2),$tp4
1517	sub	$tp8,$acc
1518	and	\$0xfefefefe,$tp4
1519	and	\$0x1b1b1b1b,$acc
1520	 xor	$tp1,$tp2		# tp2^tp1
1521	xor	$tp4,$acc
1522	mov	$acc,$tp4
1523
1524	and	\$0x80808080,$acc
1525	mov	$acc,$tp8
1526	shr	\$7,$tp8
1527	sub	$tp8,$acc
1528	lea	0($tp4,$tp4),$tp8
1529	 xor	$tp1,$tp4		# tp4^tp1
1530	and	\$0xfefefefe,$tp8
1531	and	\$0x1b1b1b1b,$acc
1532	xor	$acc,$tp8
1533
1534	xor	$tp8,$tp1		# tp1^tp8
1535	rol	\$8,$tp1		# ROTATE(tp1^tp8,8)
1536	xor	$tp8,$tp2		# tp2^tp1^tp8
1537	xor	$tp8,$tp4		# tp4^tp1^tp8
1538	xor	$tp2,$tp8
1539	xor	$tp4,$tp8		# tp8^(tp8^tp4^tp1)^(tp8^tp2^tp1)=tp8^tp4^tp2
1540
1541	xor	$tp8,$tp1
1542	rol	\$24,$tp2		# ROTATE(tp2^tp1^tp8,24)
1543	xor	$tp2,$tp1
1544	rol	\$16,$tp4		# ROTATE(tp4^tp1^tp8,16)
1545	xor	$tp4,$tp1
1546
1547	mov	$tp1,$i($ptr)
1548___
1549}
1550
1551# int private_AES_set_decrypt_key(const unsigned char *userKey, const int bits,
1552#                        AES_KEY *key)
1553$code.=<<___;
1554.globl	private_AES_set_decrypt_key
1555.type	private_AES_set_decrypt_key,\@function,3
1556.align	16
1557private_AES_set_decrypt_key:
1558	push	%rbx
1559	push	%rbp
1560	push	%r12
1561	push	%r13
1562	push	%r14
1563	push	%r15
1564	push	%rdx			# save key schedule
1565.Ldec_key_prologue:
1566
1567	call	_x86_64_AES_set_encrypt_key
1568	mov	(%rsp),%r8		# restore key schedule
1569	cmp	\$0,%eax
1570	jne	.Labort
1571
1572	mov	240(%r8),%r14d		# pull number of rounds
1573	xor	%rdi,%rdi
1574	lea	(%rdi,%r14d,4),%rcx
1575	mov	%r8,%rsi
1576	lea	(%r8,%rcx,4),%rdi	# pointer to last chunk
1577.align	4
1578.Linvert:
1579		mov	0(%rsi),%rax
1580		mov	8(%rsi),%rbx
1581		mov	0(%rdi),%rcx
1582		mov	8(%rdi),%rdx
1583		mov	%rax,0(%rdi)
1584		mov	%rbx,8(%rdi)
1585		mov	%rcx,0(%rsi)
1586		mov	%rdx,8(%rsi)
1587		lea	16(%rsi),%rsi
1588		lea	-16(%rdi),%rdi
1589		cmp	%rsi,%rdi
1590	jne	.Linvert
1591
1592	lea	.LAES_Te+2048+1024(%rip),%rax	# rcon
1593
1594	mov	40(%rax),$mask80
1595	mov	48(%rax),$maskfe
1596	mov	56(%rax),$mask1b
1597
1598	mov	%r8,$key
1599	sub	\$1,%r14d
1600.align	4
1601.Lpermute:
1602		lea	16($key),$key
1603		mov	0($key),%rax
1604		mov	8($key),%rcx
1605___
1606		&dectransform ();
1607$code.=<<___;
1608		mov	%eax,0($key)
1609		mov	%ebx,4($key)
1610		mov	%ecx,8($key)
1611		mov	%edx,12($key)
1612		sub	\$1,%r14d
1613	jnz	.Lpermute
1614
1615	xor	%rax,%rax
1616.Labort:
1617	mov	8(%rsp),%r15
1618	mov	16(%rsp),%r14
1619	mov	24(%rsp),%r13
1620	mov	32(%rsp),%r12
1621	mov	40(%rsp),%rbp
1622	mov	48(%rsp),%rbx
1623	add	\$56,%rsp
1624.Ldec_key_epilogue:
1625	ret
1626.size	private_AES_set_decrypt_key,.-private_AES_set_decrypt_key
1627___
1628
1629# void AES_cbc_encrypt (const void char *inp, unsigned char *out,
1630#			size_t length, const AES_KEY *key,
1631#			unsigned char *ivp,const int enc);
1632{
1633# stack frame layout
1634# -8(%rsp)		return address
1635my $keyp="0(%rsp)";		# one to pass as $key
1636my $keyend="8(%rsp)";		# &(keyp->rd_key[4*keyp->rounds])
1637my $_rsp="16(%rsp)";		# saved %rsp
1638my $_inp="24(%rsp)";		# copy of 1st parameter, inp
1639my $_out="32(%rsp)";		# copy of 2nd parameter, out
1640my $_len="40(%rsp)";		# copy of 3rd parameter, length
1641my $_key="48(%rsp)";		# copy of 4th parameter, key
1642my $_ivp="56(%rsp)";		# copy of 5th parameter, ivp
1643my $ivec="64(%rsp)";		# ivec[16]
1644my $aes_key="80(%rsp)";		# copy of aes_key
1645my $mark="80+240(%rsp)";	# copy of aes_key->rounds
1646
1647$code.=<<___;
1648.globl	AES_cbc_encrypt
1649.type	AES_cbc_encrypt,\@function,6
1650.align	16
1651.extern	OPENSSL_ia32cap_P
1652.globl	asm_AES_cbc_encrypt
1653.hidden	asm_AES_cbc_encrypt
1654asm_AES_cbc_encrypt:
1655AES_cbc_encrypt:
1656	cmp	\$0,%rdx	# check length
1657	je	.Lcbc_epilogue
1658	pushfq
1659	push	%rbx
1660	push	%rbp
1661	push	%r12
1662	push	%r13
1663	push	%r14
1664	push	%r15
1665.Lcbc_prologue:
1666
1667	cld
1668	mov	%r9d,%r9d	# clear upper half of enc
1669
1670	lea	.LAES_Te(%rip),$sbox
1671	cmp	\$0,%r9
1672	jne	.Lcbc_picked_te
1673	lea	.LAES_Td(%rip),$sbox
1674.Lcbc_picked_te:
1675
1676	mov	OPENSSL_ia32cap_P(%rip),%r10d
1677	cmp	\$$speed_limit,%rdx
1678	jb	.Lcbc_slow_prologue
1679	test	\$15,%rdx
1680	jnz	.Lcbc_slow_prologue
1681	bt	\$28,%r10d
1682	jc	.Lcbc_slow_prologue
1683
1684	# allocate aligned stack frame...
1685	lea	-88-248(%rsp),$key
1686	and	\$-64,$key
1687
1688	# ... and make sure it doesn't alias with AES_T[ed] modulo 4096
1689	mov	$sbox,%r10
1690	lea	2304($sbox),%r11
1691	mov	$key,%r12
1692	and	\$0xFFF,%r10	# s = $sbox&0xfff
1693	and	\$0xFFF,%r11	# e = ($sbox+2048)&0xfff
1694	and	\$0xFFF,%r12	# p = %rsp&0xfff
1695
1696	cmp	%r11,%r12	# if (p=>e) %rsp =- (p-e);
1697	jb	.Lcbc_te_break_out
1698	sub	%r11,%r12
1699	sub	%r12,$key
1700	jmp	.Lcbc_te_ok
1701.Lcbc_te_break_out:		# else %rsp -= (p-s)&0xfff + framesz
1702	sub	%r10,%r12
1703	and	\$0xFFF,%r12
1704	add	\$320,%r12
1705	sub	%r12,$key
1706.align	4
1707.Lcbc_te_ok:
1708
1709	xchg	%rsp,$key
1710	#add	\$8,%rsp	# reserve for return address!
1711	mov	$key,$_rsp	# save %rsp
1712.Lcbc_fast_body:
1713	mov	%rdi,$_inp	# save copy of inp
1714	mov	%rsi,$_out	# save copy of out
1715	mov	%rdx,$_len	# save copy of len
1716	mov	%rcx,$_key	# save copy of key
1717	mov	%r8,$_ivp	# save copy of ivp
1718	movl	\$0,$mark	# copy of aes_key->rounds = 0;
1719	mov	%r8,%rbp	# rearrange input arguments
1720	mov	%r9,%rbx
1721	mov	%rsi,$out
1722	mov	%rdi,$inp
1723	mov	%rcx,$key
1724
1725	mov	240($key),%eax		# key->rounds
1726	# do we copy key schedule to stack?
1727	mov	$key,%r10
1728	sub	$sbox,%r10
1729	and	\$0xfff,%r10
1730	cmp	\$2304,%r10
1731	jb	.Lcbc_do_ecopy
1732	cmp	\$4096-248,%r10
1733	jb	.Lcbc_skip_ecopy
1734.align	4
1735.Lcbc_do_ecopy:
1736		mov	$key,%rsi
1737		lea	$aes_key,%rdi
1738		lea	$aes_key,$key
1739		mov	\$240/8,%ecx
1740		.long	0x90A548F3	# rep movsq
1741		mov	%eax,(%rdi)	# copy aes_key->rounds
1742.Lcbc_skip_ecopy:
1743	mov	$key,$keyp	# save key pointer
1744
1745	mov	\$18,%ecx
1746.align	4
1747.Lcbc_prefetch_te:
1748		mov	0($sbox),%r10
1749		mov	32($sbox),%r11
1750		mov	64($sbox),%r12
1751		mov	96($sbox),%r13
1752		lea	128($sbox),$sbox
1753		sub	\$1,%ecx
1754	jnz	.Lcbc_prefetch_te
1755	lea	-2304($sbox),$sbox
1756
1757	cmp	\$0,%rbx
1758	je	.LFAST_DECRYPT
1759
1760#----------------------------- ENCRYPT -----------------------------#
1761	mov	0(%rbp),$s0		# load iv
1762	mov	4(%rbp),$s1
1763	mov	8(%rbp),$s2
1764	mov	12(%rbp),$s3
1765
1766.align	4
1767.Lcbc_fast_enc_loop:
1768		xor	0($inp),$s0
1769		xor	4($inp),$s1
1770		xor	8($inp),$s2
1771		xor	12($inp),$s3
1772		mov	$keyp,$key	# restore key
1773		mov	$inp,$_inp	# if ($verticalspin) save inp
1774
1775		call	_x86_64_AES_encrypt
1776
1777		mov	$_inp,$inp	# if ($verticalspin) restore inp
1778		mov	$_len,%r10
1779		mov	$s0,0($out)
1780		mov	$s1,4($out)
1781		mov	$s2,8($out)
1782		mov	$s3,12($out)
1783
1784		lea	16($inp),$inp
1785		lea	16($out),$out
1786		sub	\$16,%r10
1787		test	\$-16,%r10
1788		mov	%r10,$_len
1789	jnz	.Lcbc_fast_enc_loop
1790	mov	$_ivp,%rbp	# restore ivp
1791	mov	$s0,0(%rbp)	# save ivec
1792	mov	$s1,4(%rbp)
1793	mov	$s2,8(%rbp)
1794	mov	$s3,12(%rbp)
1795
1796	jmp	.Lcbc_fast_cleanup
1797
1798#----------------------------- DECRYPT -----------------------------#
1799.align	16
1800.LFAST_DECRYPT:
1801	cmp	$inp,$out
1802	je	.Lcbc_fast_dec_in_place
1803
1804	mov	%rbp,$ivec
1805.align	4
1806.Lcbc_fast_dec_loop:
1807		mov	0($inp),$s0	# read input
1808		mov	4($inp),$s1
1809		mov	8($inp),$s2
1810		mov	12($inp),$s3
1811		mov	$keyp,$key	# restore key
1812		mov	$inp,$_inp	# if ($verticalspin) save inp
1813
1814		call	_x86_64_AES_decrypt
1815
1816		mov	$ivec,%rbp	# load ivp
1817		mov	$_inp,$inp	# if ($verticalspin) restore inp
1818		mov	$_len,%r10	# load len
1819		xor	0(%rbp),$s0	# xor iv
1820		xor	4(%rbp),$s1
1821		xor	8(%rbp),$s2
1822		xor	12(%rbp),$s3
1823		mov	$inp,%rbp	# current input, next iv
1824
1825		sub	\$16,%r10
1826		mov	%r10,$_len	# update len
1827		mov	%rbp,$ivec	# update ivp
1828
1829		mov	$s0,0($out)	# write output
1830		mov	$s1,4($out)
1831		mov	$s2,8($out)
1832		mov	$s3,12($out)
1833
1834		lea	16($inp),$inp
1835		lea	16($out),$out
1836	jnz	.Lcbc_fast_dec_loop
1837	mov	$_ivp,%r12		# load user ivp
1838	mov	0(%rbp),%r10		# load iv
1839	mov	8(%rbp),%r11
1840	mov	%r10,0(%r12)		# copy back to user
1841	mov	%r11,8(%r12)
1842	jmp	.Lcbc_fast_cleanup
1843
1844.align	16
1845.Lcbc_fast_dec_in_place:
1846	mov	0(%rbp),%r10		# copy iv to stack
1847	mov	8(%rbp),%r11
1848	mov	%r10,0+$ivec
1849	mov	%r11,8+$ivec
1850.align	4
1851.Lcbc_fast_dec_in_place_loop:
1852		mov	0($inp),$s0	# load input
1853		mov	4($inp),$s1
1854		mov	8($inp),$s2
1855		mov	12($inp),$s3
1856		mov	$keyp,$key	# restore key
1857		mov	$inp,$_inp	# if ($verticalspin) save inp
1858
1859		call	_x86_64_AES_decrypt
1860
1861		mov	$_inp,$inp	# if ($verticalspin) restore inp
1862		mov	$_len,%r10
1863		xor	0+$ivec,$s0
1864		xor	4+$ivec,$s1
1865		xor	8+$ivec,$s2
1866		xor	12+$ivec,$s3
1867
1868		mov	0($inp),%r11	# load input
1869		mov	8($inp),%r12
1870		sub	\$16,%r10
1871		jz	.Lcbc_fast_dec_in_place_done
1872
1873		mov	%r11,0+$ivec	# copy input to iv
1874		mov	%r12,8+$ivec
1875
1876		mov	$s0,0($out)	# save output [zaps input]
1877		mov	$s1,4($out)
1878		mov	$s2,8($out)
1879		mov	$s3,12($out)
1880
1881		lea	16($inp),$inp
1882		lea	16($out),$out
1883		mov	%r10,$_len
1884	jmp	.Lcbc_fast_dec_in_place_loop
1885.Lcbc_fast_dec_in_place_done:
1886	mov	$_ivp,%rdi
1887	mov	%r11,0(%rdi)	# copy iv back to user
1888	mov	%r12,8(%rdi)
1889
1890	mov	$s0,0($out)	# save output [zaps input]
1891	mov	$s1,4($out)
1892	mov	$s2,8($out)
1893	mov	$s3,12($out)
1894
1895.align	4
1896.Lcbc_fast_cleanup:
1897	cmpl	\$0,$mark	# was the key schedule copied?
1898	lea	$aes_key,%rdi
1899	je	.Lcbc_exit
1900		mov	\$240/8,%ecx
1901		xor	%rax,%rax
1902		.long	0x90AB48F3	# rep stosq
1903
1904	jmp	.Lcbc_exit
1905
1906#--------------------------- SLOW ROUTINE ---------------------------#
1907.align	16
1908.Lcbc_slow_prologue:
1909	# allocate aligned stack frame...
1910	lea	-88(%rsp),%rbp
1911	and	\$-64,%rbp
1912	# ... just "above" key schedule
1913	lea	-88-63(%rcx),%r10
1914	sub	%rbp,%r10
1915	neg	%r10
1916	and	\$0x3c0,%r10
1917	sub	%r10,%rbp
1918
1919	xchg	%rsp,%rbp
1920	#add	\$8,%rsp	# reserve for return address!
1921	mov	%rbp,$_rsp	# save %rsp
1922.Lcbc_slow_body:
1923	#mov	%rdi,$_inp	# save copy of inp
1924	#mov	%rsi,$_out	# save copy of out
1925	#mov	%rdx,$_len	# save copy of len
1926	#mov	%rcx,$_key	# save copy of key
1927	mov	%r8,$_ivp	# save copy of ivp
1928	mov	%r8,%rbp	# rearrange input arguments
1929	mov	%r9,%rbx
1930	mov	%rsi,$out
1931	mov	%rdi,$inp
1932	mov	%rcx,$key
1933	mov	%rdx,%r10
1934
1935	mov	240($key),%eax
1936	mov	$key,$keyp	# save key pointer
1937	shl	\$4,%eax
1938	lea	($key,%rax),%rax
1939	mov	%rax,$keyend
1940
1941	# pick Te4 copy which can't "overlap" with stack frame or key scdedule
1942	lea	2048($sbox),$sbox
1943	lea	768-8(%rsp),%rax
1944	sub	$sbox,%rax
1945	and	\$0x300,%rax
1946	lea	($sbox,%rax),$sbox
1947
1948	cmp	\$0,%rbx
1949	je	.LSLOW_DECRYPT
1950
1951#--------------------------- SLOW ENCRYPT ---------------------------#
1952	test	\$-16,%r10		# check upon length
1953	mov	0(%rbp),$s0		# load iv
1954	mov	4(%rbp),$s1
1955	mov	8(%rbp),$s2
1956	mov	12(%rbp),$s3
1957	jz	.Lcbc_slow_enc_tail	# short input...
1958
1959.align	4
1960.Lcbc_slow_enc_loop:
1961		xor	0($inp),$s0
1962		xor	4($inp),$s1
1963		xor	8($inp),$s2
1964		xor	12($inp),$s3
1965		mov	$keyp,$key	# restore key
1966		mov	$inp,$_inp	# save inp
1967		mov	$out,$_out	# save out
1968		mov	%r10,$_len	# save len
1969
1970		call	_x86_64_AES_encrypt_compact
1971
1972		mov	$_inp,$inp	# restore inp
1973		mov	$_out,$out	# restore out
1974		mov	$_len,%r10	# restore len
1975		mov	$s0,0($out)
1976		mov	$s1,4($out)
1977		mov	$s2,8($out)
1978		mov	$s3,12($out)
1979
1980		lea	16($inp),$inp
1981		lea	16($out),$out
1982		sub	\$16,%r10
1983		test	\$-16,%r10
1984	jnz	.Lcbc_slow_enc_loop
1985	test	\$15,%r10
1986	jnz	.Lcbc_slow_enc_tail
1987	mov	$_ivp,%rbp	# restore ivp
1988	mov	$s0,0(%rbp)	# save ivec
1989	mov	$s1,4(%rbp)
1990	mov	$s2,8(%rbp)
1991	mov	$s3,12(%rbp)
1992
1993	jmp	.Lcbc_exit
1994
1995.align	4
1996.Lcbc_slow_enc_tail:
1997	mov	%rax,%r11
1998	mov	%rcx,%r12
1999	mov	%r10,%rcx
2000	mov	$inp,%rsi
2001	mov	$out,%rdi
2002	.long	0x9066A4F3		# rep movsb
2003	mov	\$16,%rcx		# zero tail
2004	sub	%r10,%rcx
2005	xor	%rax,%rax
2006	.long	0x9066AAF3		# rep stosb
2007	mov	$out,$inp		# this is not a mistake!
2008	mov	\$16,%r10		# len=16
2009	mov	%r11,%rax
2010	mov	%r12,%rcx
2011	jmp	.Lcbc_slow_enc_loop	# one more spin...
2012#--------------------------- SLOW DECRYPT ---------------------------#
2013.align	16
2014.LSLOW_DECRYPT:
2015	shr	\$3,%rax
2016	add	%rax,$sbox		# recall "magic" constants!
2017
2018	mov	0(%rbp),%r11		# copy iv to stack
2019	mov	8(%rbp),%r12
2020	mov	%r11,0+$ivec
2021	mov	%r12,8+$ivec
2022
2023.align	4
2024.Lcbc_slow_dec_loop:
2025		mov	0($inp),$s0	# load input
2026		mov	4($inp),$s1
2027		mov	8($inp),$s2
2028		mov	12($inp),$s3
2029		mov	$keyp,$key	# restore key
2030		mov	$inp,$_inp	# save inp
2031		mov	$out,$_out	# save out
2032		mov	%r10,$_len	# save len
2033
2034		call	_x86_64_AES_decrypt_compact
2035
2036		mov	$_inp,$inp	# restore inp
2037		mov	$_out,$out	# restore out
2038		mov	$_len,%r10
2039		xor	0+$ivec,$s0
2040		xor	4+$ivec,$s1
2041		xor	8+$ivec,$s2
2042		xor	12+$ivec,$s3
2043
2044		mov	0($inp),%r11	# load input
2045		mov	8($inp),%r12
2046		sub	\$16,%r10
2047		jc	.Lcbc_slow_dec_partial
2048		jz	.Lcbc_slow_dec_done
2049
2050		mov	%r11,0+$ivec	# copy input to iv
2051		mov	%r12,8+$ivec
2052
2053		mov	$s0,0($out)	# save output [can zap input]
2054		mov	$s1,4($out)
2055		mov	$s2,8($out)
2056		mov	$s3,12($out)
2057
2058		lea	16($inp),$inp
2059		lea	16($out),$out
2060	jmp	.Lcbc_slow_dec_loop
2061.Lcbc_slow_dec_done:
2062	mov	$_ivp,%rdi
2063	mov	%r11,0(%rdi)		# copy iv back to user
2064	mov	%r12,8(%rdi)
2065
2066	mov	$s0,0($out)		# save output [can zap input]
2067	mov	$s1,4($out)
2068	mov	$s2,8($out)
2069	mov	$s3,12($out)
2070
2071	jmp	.Lcbc_exit
2072
2073.align	4
2074.Lcbc_slow_dec_partial:
2075	mov	$_ivp,%rdi
2076	mov	%r11,0(%rdi)		# copy iv back to user
2077	mov	%r12,8(%rdi)
2078
2079	mov	$s0,0+$ivec		# save output to stack
2080	mov	$s1,4+$ivec
2081	mov	$s2,8+$ivec
2082	mov	$s3,12+$ivec
2083
2084	mov	$out,%rdi
2085	lea	$ivec,%rsi
2086	lea	16(%r10),%rcx
2087	.long	0x9066A4F3	# rep movsb
2088	jmp	.Lcbc_exit
2089
2090.align	16
2091.Lcbc_exit:
2092	mov	$_rsp,%rsi
2093	mov	(%rsi),%r15
2094	mov	8(%rsi),%r14
2095	mov	16(%rsi),%r13
2096	mov	24(%rsi),%r12
2097	mov	32(%rsi),%rbp
2098	mov	40(%rsi),%rbx
2099	lea	48(%rsi),%rsp
2100.Lcbc_popfq:
2101	popfq
2102.Lcbc_epilogue:
2103	ret
2104.size	AES_cbc_encrypt,.-AES_cbc_encrypt
2105___
2106}
2107
2108$code.=<<___;
2109.align	64
2110.LAES_Te:
2111___
2112	&_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6);
2113	&_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591);
2114	&_data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56);
2115	&_data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec);
2116	&_data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa);
2117	&_data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb);
2118	&_data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45);
2119	&_data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b);
2120	&_data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c);
2121	&_data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83);
2122	&_data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9);
2123	&_data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a);
2124	&_data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d);
2125	&_data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f);
2126	&_data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df);
2127	&_data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea);
2128	&_data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34);
2129	&_data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b);
2130	&_data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d);
2131	&_data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413);
2132	&_data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1);
2133	&_data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6);
2134	&_data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972);
2135	&_data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85);
2136	&_data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed);
2137	&_data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511);
2138	&_data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe);
2139	&_data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b);
2140	&_data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05);
2141	&_data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1);
2142	&_data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142);
2143	&_data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf);
2144	&_data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3);
2145	&_data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e);
2146	&_data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a);
2147	&_data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6);
2148	&_data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3);
2149	&_data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b);
2150	&_data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428);
2151	&_data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad);
2152	&_data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14);
2153	&_data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8);
2154	&_data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4);
2155	&_data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2);
2156	&_data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda);
2157	&_data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949);
2158	&_data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf);
2159	&_data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810);
2160	&_data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c);
2161	&_data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697);
2162	&_data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e);
2163	&_data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f);
2164	&_data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc);
2165	&_data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c);
2166	&_data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969);
2167	&_data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27);
2168	&_data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122);
2169	&_data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433);
2170	&_data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9);
2171	&_data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5);
2172	&_data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a);
2173	&_data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0);
2174	&_data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e);
2175	&_data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c);
2176
2177#Te4	# four copies of Te4 to choose from to avoid L1 aliasing
2178	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2179	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2180	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2181	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2182	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2183	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2184	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2185	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2186	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2187	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2188	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2189	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2190	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2191	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2192	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2193	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2194	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2195	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2196	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2197	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2198	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2199	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2200	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2201	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2202	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2203	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2204	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2205	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2206	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2207	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2208	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2209	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2210
2211	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2212	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2213	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2214	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2215	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2216	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2217	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2218	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2219	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2220	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2221	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2222	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2223	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2224	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2225	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2226	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2227	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2228	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2229	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2230	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2231	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2232	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2233	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2234	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2235	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2236	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2237	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2238	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2239	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2240	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2241	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2242	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2243
2244	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2245	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2246	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2247	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2248	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2249	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2250	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2251	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2252	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2253	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2254	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2255	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2256	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2257	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2258	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2259	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2260	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2261	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2262	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2263	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2264	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2265	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2266	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2267	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2268	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2269	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2270	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2271	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2272	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2273	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2274	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2275	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2276
2277	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2278	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2279	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2280	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2281	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2282	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2283	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2284	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2285	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2286	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2287	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2288	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2289	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2290	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2291	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2292	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2293	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2294	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2295	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2296	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2297	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2298	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2299	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2300	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2301	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2302	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2303	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2304	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2305	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2306	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2307	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2308	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2309#rcon:
2310$code.=<<___;
2311	.long	0x00000001, 0x00000002, 0x00000004, 0x00000008
2312	.long	0x00000010, 0x00000020, 0x00000040, 0x00000080
2313	.long	0x0000001b, 0x00000036, 0x80808080, 0x80808080
2314	.long	0xfefefefe, 0xfefefefe, 0x1b1b1b1b, 0x1b1b1b1b
2315___
2316$code.=<<___;
2317.align	64
2318.LAES_Td:
2319___
2320	&_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a);
2321	&_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b);
2322	&_data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5);
2323	&_data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5);
2324	&_data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d);
2325	&_data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b);
2326	&_data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295);
2327	&_data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e);
2328	&_data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927);
2329	&_data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d);
2330	&_data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362);
2331	&_data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9);
2332	&_data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52);
2333	&_data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566);
2334	&_data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3);
2335	&_data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed);
2336	&_data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e);
2337	&_data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4);
2338	&_data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4);
2339	&_data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd);
2340	&_data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d);
2341	&_data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060);
2342	&_data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967);
2343	&_data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879);
2344	&_data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000);
2345	&_data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c);
2346	&_data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36);
2347	&_data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624);
2348	&_data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b);
2349	&_data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c);
2350	&_data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12);
2351	&_data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14);
2352	&_data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3);
2353	&_data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b);
2354	&_data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8);
2355	&_data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684);
2356	&_data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7);
2357	&_data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177);
2358	&_data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947);
2359	&_data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322);
2360	&_data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498);
2361	&_data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f);
2362	&_data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54);
2363	&_data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382);
2364	&_data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf);
2365	&_data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb);
2366	&_data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83);
2367	&_data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef);
2368	&_data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029);
2369	&_data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235);
2370	&_data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733);
2371	&_data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117);
2372	&_data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4);
2373	&_data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546);
2374	&_data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb);
2375	&_data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d);
2376	&_data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb);
2377	&_data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a);
2378	&_data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773);
2379	&_data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478);
2380	&_data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2);
2381	&_data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff);
2382	&_data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664);
2383	&_data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0);
2384
2385#Td4:	# four copies of Td4 to choose from to avoid L1 aliasing
2386	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2387	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2388	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2389	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2390	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2391	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2392	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2393	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2394	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2395	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2396	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2397	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2398	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2399	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2400	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2401	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2402	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2403	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2404	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2405	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2406	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2407	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2408	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2409	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2410	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2411	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2412	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2413	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2414	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2415	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2416	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2417	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2418$code.=<<___;
2419	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2420	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2421___
2422	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2423	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2424	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2425	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2426	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2427	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2428	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2429	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2430	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2431	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2432	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2433	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2434	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2435	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2436	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2437	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2438	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2439	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2440	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2441	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2442	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2443	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2444	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2445	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2446	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2447	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2448	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2449	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2450	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2451	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2452	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2453	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2454$code.=<<___;
2455	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2456	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2457___
2458	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2459	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2460	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2461	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2462	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2463	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2464	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2465	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2466	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2467	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2468	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2469	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2470	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2471	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2472	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2473	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2474	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2475	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2476	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2477	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2478	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2479	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2480	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2481	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2482	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2483	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2484	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2485	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2486	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2487	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2488	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2489	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2490$code.=<<___;
2491	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2492	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2493___
2494	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2495	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2496	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2497	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2498	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2499	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2500	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2501	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2502	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2503	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2504	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2505	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2506	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2507	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2508	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2509	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2510	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2511	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2512	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2513	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2514	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2515	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2516	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2517	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2518	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2519	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2520	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2521	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2522	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2523	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2524	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2525	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2526$code.=<<___;
2527	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2528	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2529.asciz  "AES for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
2530.align	64
2531___
2532
2533# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
2534#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
2535if ($win64) {
2536$rec="%rcx";
2537$frame="%rdx";
2538$context="%r8";
2539$disp="%r9";
2540
2541$code.=<<___;
2542.extern	__imp_RtlVirtualUnwind
2543.type	block_se_handler,\@abi-omnipotent
2544.align	16
2545block_se_handler:
2546	push	%rsi
2547	push	%rdi
2548	push	%rbx
2549	push	%rbp
2550	push	%r12
2551	push	%r13
2552	push	%r14
2553	push	%r15
2554	pushfq
2555	sub	\$64,%rsp
2556
2557	mov	120($context),%rax	# pull context->Rax
2558	mov	248($context),%rbx	# pull context->Rip
2559
2560	mov	8($disp),%rsi		# disp->ImageBase
2561	mov	56($disp),%r11		# disp->HandlerData
2562
2563	mov	0(%r11),%r10d		# HandlerData[0]
2564	lea	(%rsi,%r10),%r10	# prologue label
2565	cmp	%r10,%rbx		# context->Rip<prologue label
2566	jb	.Lin_block_prologue
2567
2568	mov	152($context),%rax	# pull context->Rsp
2569
2570	mov	4(%r11),%r10d		# HandlerData[1]
2571	lea	(%rsi,%r10),%r10	# epilogue label
2572	cmp	%r10,%rbx		# context->Rip>=epilogue label
2573	jae	.Lin_block_prologue
2574
2575	mov	24(%rax),%rax		# pull saved real stack pointer
2576	lea	48(%rax),%rax		# adjust...
2577
2578	mov	-8(%rax),%rbx
2579	mov	-16(%rax),%rbp
2580	mov	-24(%rax),%r12
2581	mov	-32(%rax),%r13
2582	mov	-40(%rax),%r14
2583	mov	-48(%rax),%r15
2584	mov	%rbx,144($context)	# restore context->Rbx
2585	mov	%rbp,160($context)	# restore context->Rbp
2586	mov	%r12,216($context)	# restore context->R12
2587	mov	%r13,224($context)	# restore context->R13
2588	mov	%r14,232($context)	# restore context->R14
2589	mov	%r15,240($context)	# restore context->R15
2590
2591.Lin_block_prologue:
2592	mov	8(%rax),%rdi
2593	mov	16(%rax),%rsi
2594	mov	%rax,152($context)	# restore context->Rsp
2595	mov	%rsi,168($context)	# restore context->Rsi
2596	mov	%rdi,176($context)	# restore context->Rdi
2597
2598	jmp	.Lcommon_seh_exit
2599.size	block_se_handler,.-block_se_handler
2600
2601.type	key_se_handler,\@abi-omnipotent
2602.align	16
2603key_se_handler:
2604	push	%rsi
2605	push	%rdi
2606	push	%rbx
2607	push	%rbp
2608	push	%r12
2609	push	%r13
2610	push	%r14
2611	push	%r15
2612	pushfq
2613	sub	\$64,%rsp
2614
2615	mov	120($context),%rax	# pull context->Rax
2616	mov	248($context),%rbx	# pull context->Rip
2617
2618	mov	8($disp),%rsi		# disp->ImageBase
2619	mov	56($disp),%r11		# disp->HandlerData
2620
2621	mov	0(%r11),%r10d		# HandlerData[0]
2622	lea	(%rsi,%r10),%r10	# prologue label
2623	cmp	%r10,%rbx		# context->Rip<prologue label
2624	jb	.Lin_key_prologue
2625
2626	mov	152($context),%rax	# pull context->Rsp
2627
2628	mov	4(%r11),%r10d		# HandlerData[1]
2629	lea	(%rsi,%r10),%r10	# epilogue label
2630	cmp	%r10,%rbx		# context->Rip>=epilogue label
2631	jae	.Lin_key_prologue
2632
2633	lea	56(%rax),%rax
2634
2635	mov	-8(%rax),%rbx
2636	mov	-16(%rax),%rbp
2637	mov	-24(%rax),%r12
2638	mov	-32(%rax),%r13
2639	mov	-40(%rax),%r14
2640	mov	-48(%rax),%r15
2641	mov	%rbx,144($context)	# restore context->Rbx
2642	mov	%rbp,160($context)	# restore context->Rbp
2643	mov	%r12,216($context)	# restore context->R12
2644	mov	%r13,224($context)	# restore context->R13
2645	mov	%r14,232($context)	# restore context->R14
2646	mov	%r15,240($context)	# restore context->R15
2647
2648.Lin_key_prologue:
2649	mov	8(%rax),%rdi
2650	mov	16(%rax),%rsi
2651	mov	%rax,152($context)	# restore context->Rsp
2652	mov	%rsi,168($context)	# restore context->Rsi
2653	mov	%rdi,176($context)	# restore context->Rdi
2654
2655	jmp	.Lcommon_seh_exit
2656.size	key_se_handler,.-key_se_handler
2657
2658.type	cbc_se_handler,\@abi-omnipotent
2659.align	16
2660cbc_se_handler:
2661	push	%rsi
2662	push	%rdi
2663	push	%rbx
2664	push	%rbp
2665	push	%r12
2666	push	%r13
2667	push	%r14
2668	push	%r15
2669	pushfq
2670	sub	\$64,%rsp
2671
2672	mov	120($context),%rax	# pull context->Rax
2673	mov	248($context),%rbx	# pull context->Rip
2674
2675	lea	.Lcbc_prologue(%rip),%r10
2676	cmp	%r10,%rbx		# context->Rip<.Lcbc_prologue
2677	jb	.Lin_cbc_prologue
2678
2679	lea	.Lcbc_fast_body(%rip),%r10
2680	cmp	%r10,%rbx		# context->Rip<.Lcbc_fast_body
2681	jb	.Lin_cbc_frame_setup
2682
2683	lea	.Lcbc_slow_prologue(%rip),%r10
2684	cmp	%r10,%rbx		# context->Rip<.Lcbc_slow_prologue
2685	jb	.Lin_cbc_body
2686
2687	lea	.Lcbc_slow_body(%rip),%r10
2688	cmp	%r10,%rbx		# context->Rip<.Lcbc_slow_body
2689	jb	.Lin_cbc_frame_setup
2690
2691.Lin_cbc_body:
2692	mov	152($context),%rax	# pull context->Rsp
2693
2694	lea	.Lcbc_epilogue(%rip),%r10
2695	cmp	%r10,%rbx		# context->Rip>=.Lcbc_epilogue
2696	jae	.Lin_cbc_prologue
2697
2698	lea	8(%rax),%rax
2699
2700	lea	.Lcbc_popfq(%rip),%r10
2701	cmp	%r10,%rbx		# context->Rip>=.Lcbc_popfq
2702	jae	.Lin_cbc_prologue
2703
2704	mov	`16-8`(%rax),%rax	# biased $_rsp
2705	lea	56(%rax),%rax
2706
2707.Lin_cbc_frame_setup:
2708	mov	-16(%rax),%rbx
2709	mov	-24(%rax),%rbp
2710	mov	-32(%rax),%r12
2711	mov	-40(%rax),%r13
2712	mov	-48(%rax),%r14
2713	mov	-56(%rax),%r15
2714	mov	%rbx,144($context)	# restore context->Rbx
2715	mov	%rbp,160($context)	# restore context->Rbp
2716	mov	%r12,216($context)	# restore context->R12
2717	mov	%r13,224($context)	# restore context->R13
2718	mov	%r14,232($context)	# restore context->R14
2719	mov	%r15,240($context)	# restore context->R15
2720
2721.Lin_cbc_prologue:
2722	mov	8(%rax),%rdi
2723	mov	16(%rax),%rsi
2724	mov	%rax,152($context)	# restore context->Rsp
2725	mov	%rsi,168($context)	# restore context->Rsi
2726	mov	%rdi,176($context)	# restore context->Rdi
2727
2728.Lcommon_seh_exit:
2729
2730	mov	40($disp),%rdi		# disp->ContextRecord
2731	mov	$context,%rsi		# context
2732	mov	\$`1232/8`,%ecx		# sizeof(CONTEXT)
2733	.long	0xa548f3fc		# cld; rep movsq
2734
2735	mov	$disp,%rsi
2736	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
2737	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
2738	mov	0(%rsi),%r8		# arg3, disp->ControlPc
2739	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
2740	mov	40(%rsi),%r10		# disp->ContextRecord
2741	lea	56(%rsi),%r11		# &disp->HandlerData
2742	lea	24(%rsi),%r12		# &disp->EstablisherFrame
2743	mov	%r10,32(%rsp)		# arg5
2744	mov	%r11,40(%rsp)		# arg6
2745	mov	%r12,48(%rsp)		# arg7
2746	mov	%rcx,56(%rsp)		# arg8, (NULL)
2747	call	*__imp_RtlVirtualUnwind(%rip)
2748
2749	mov	\$1,%eax		# ExceptionContinueSearch
2750	add	\$64,%rsp
2751	popfq
2752	pop	%r15
2753	pop	%r14
2754	pop	%r13
2755	pop	%r12
2756	pop	%rbp
2757	pop	%rbx
2758	pop	%rdi
2759	pop	%rsi
2760	ret
2761.size	cbc_se_handler,.-cbc_se_handler
2762
2763.section	.pdata
2764.align	4
2765	.rva	.LSEH_begin_AES_encrypt
2766	.rva	.LSEH_end_AES_encrypt
2767	.rva	.LSEH_info_AES_encrypt
2768
2769	.rva	.LSEH_begin_AES_decrypt
2770	.rva	.LSEH_end_AES_decrypt
2771	.rva	.LSEH_info_AES_decrypt
2772
2773	.rva	.LSEH_begin_private_AES_set_encrypt_key
2774	.rva	.LSEH_end_private_AES_set_encrypt_key
2775	.rva	.LSEH_info_private_AES_set_encrypt_key
2776
2777	.rva	.LSEH_begin_private_AES_set_decrypt_key
2778	.rva	.LSEH_end_private_AES_set_decrypt_key
2779	.rva	.LSEH_info_private_AES_set_decrypt_key
2780
2781	.rva	.LSEH_begin_AES_cbc_encrypt
2782	.rva	.LSEH_end_AES_cbc_encrypt
2783	.rva	.LSEH_info_AES_cbc_encrypt
2784
2785.section	.xdata
2786.align	8
2787.LSEH_info_AES_encrypt:
2788	.byte	9,0,0,0
2789	.rva	block_se_handler
2790	.rva	.Lenc_prologue,.Lenc_epilogue	# HandlerData[]
2791.LSEH_info_AES_decrypt:
2792	.byte	9,0,0,0
2793	.rva	block_se_handler
2794	.rva	.Ldec_prologue,.Ldec_epilogue	# HandlerData[]
2795.LSEH_info_private_AES_set_encrypt_key:
2796	.byte	9,0,0,0
2797	.rva	key_se_handler
2798	.rva	.Lenc_key_prologue,.Lenc_key_epilogue	# HandlerData[]
2799.LSEH_info_private_AES_set_decrypt_key:
2800	.byte	9,0,0,0
2801	.rva	key_se_handler
2802	.rva	.Ldec_key_prologue,.Ldec_key_epilogue	# HandlerData[]
2803.LSEH_info_AES_cbc_encrypt:
2804	.byte	9,0,0,0
2805	.rva	cbc_se_handler
2806___
2807}
2808
2809$code =~ s/\`([^\`]*)\`/eval($1)/gem;
2810
2811print $code;
2812
2813close STDOUT;
2814