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 $xlate) or
37die "can't locate x86_64-xlate.pl";
38
39open OUT,"| \"$^X\" $xlate $flavour $output";
40*STDOUT=*OUT;
41
42$verticalspin=1;	# unlike 32-bit version $verticalspin performs
43			# ~15% better on both AMD and Intel cores
44$speed_limit=512;	# see aes-586.pl for details
45
46$code=".text\n";
47
48$s0="%eax";
49$s1="%ebx";
50$s2="%ecx";
51$s3="%edx";
52$acc0="%esi";	$mask80="%rsi";
53$acc1="%edi";	$maskfe="%rdi";
54$acc2="%ebp";	$mask1b="%rbp";
55$inp="%r8";
56$out="%r9";
57$t0="%r10d";
58$t1="%r11d";
59$t2="%r12d";
60$rnds="%r13d";
61$sbox="%r14";
62$key="%r15";
63
64sub hi() { my $r=shift;	$r =~ s/%[er]([a-d])x/%\1h/;	$r; }
65sub lo() { my $r=shift;	$r =~ s/%[er]([a-d])x/%\1l/;
66			$r =~ s/%[er]([sd]i)/%\1l/;
67			$r =~ s/%(r[0-9]+)[d]?/%\1b/;	$r; }
68sub LO() { my $r=shift; $r =~ s/%r([a-z]+)/%e\1/;
69			$r =~ s/%r([0-9]+)/%r\1d/;	$r; }
70sub _data_word()
71{ my $i;
72    while(defined($i=shift)) { $code.=sprintf".long\t0x%08x,0x%08x\n",$i,$i; }
73}
74sub data_word()
75{ my $i;
76  my $last=pop(@_);
77    $code.=".long\t";
78    while(defined($i=shift)) { $code.=sprintf"0x%08x,",$i; }
79    $code.=sprintf"0x%08x\n",$last;
80}
81
82sub data_byte()
83{ my $i;
84  my $last=pop(@_);
85    $code.=".byte\t";
86    while(defined($i=shift)) { $code.=sprintf"0x%02x,",$i&0xff; }
87    $code.=sprintf"0x%02x\n",$last&0xff;
88}
89
90sub encvert()
91{ my $t3="%r8d";	# zaps $inp!
92
93$code.=<<___;
94	# favor 3-way issue Opteron pipeline...
95	movzb	`&lo("$s0")`,$acc0
96	movzb	`&lo("$s1")`,$acc1
97	movzb	`&lo("$s2")`,$acc2
98	mov	0($sbox,$acc0,8),$t0
99	mov	0($sbox,$acc1,8),$t1
100	mov	0($sbox,$acc2,8),$t2
101
102	movzb	`&hi("$s1")`,$acc0
103	movzb	`&hi("$s2")`,$acc1
104	movzb	`&lo("$s3")`,$acc2
105	xor	3($sbox,$acc0,8),$t0
106	xor	3($sbox,$acc1,8),$t1
107	mov	0($sbox,$acc2,8),$t3
108
109	movzb	`&hi("$s3")`,$acc0
110	shr	\$16,$s2
111	movzb	`&hi("$s0")`,$acc2
112	xor	3($sbox,$acc0,8),$t2
113	shr	\$16,$s3
114	xor	3($sbox,$acc2,8),$t3
115
116	shr	\$16,$s1
117	lea	16($key),$key
118	shr	\$16,$s0
119
120	movzb	`&lo("$s2")`,$acc0
121	movzb	`&lo("$s3")`,$acc1
122	movzb	`&lo("$s0")`,$acc2
123	xor	2($sbox,$acc0,8),$t0
124	xor	2($sbox,$acc1,8),$t1
125	xor	2($sbox,$acc2,8),$t2
126
127	movzb	`&hi("$s3")`,$acc0
128	movzb	`&hi("$s0")`,$acc1
129	movzb	`&lo("$s1")`,$acc2
130	xor	1($sbox,$acc0,8),$t0
131	xor	1($sbox,$acc1,8),$t1
132	xor	2($sbox,$acc2,8),$t3
133
134	mov	12($key),$s3
135	movzb	`&hi("$s1")`,$acc1
136	movzb	`&hi("$s2")`,$acc2
137	mov	0($key),$s0
138	xor	1($sbox,$acc1,8),$t2
139	xor	1($sbox,$acc2,8),$t3
140
141	mov	4($key),$s1
142	mov	8($key),$s2
143	xor	$t0,$s0
144	xor	$t1,$s1
145	xor	$t2,$s2
146	xor	$t3,$s3
147___
148}
149
150sub enclastvert()
151{ my $t3="%r8d";	# zaps $inp!
152
153$code.=<<___;
154	movzb	`&lo("$s0")`,$acc0
155	movzb	`&lo("$s1")`,$acc1
156	movzb	`&lo("$s2")`,$acc2
157	movzb	2($sbox,$acc0,8),$t0
158	movzb	2($sbox,$acc1,8),$t1
159	movzb	2($sbox,$acc2,8),$t2
160
161	movzb	`&lo("$s3")`,$acc0
162	movzb	`&hi("$s1")`,$acc1
163	movzb	`&hi("$s2")`,$acc2
164	movzb	2($sbox,$acc0,8),$t3
165	mov	0($sbox,$acc1,8),$acc1	#$t0
166	mov	0($sbox,$acc2,8),$acc2	#$t1
167
168	and	\$0x0000ff00,$acc1
169	and	\$0x0000ff00,$acc2
170
171	xor	$acc1,$t0
172	xor	$acc2,$t1
173	shr	\$16,$s2
174
175	movzb	`&hi("$s3")`,$acc0
176	movzb	`&hi("$s0")`,$acc1
177	shr	\$16,$s3
178	mov	0($sbox,$acc0,8),$acc0	#$t2
179	mov	0($sbox,$acc1,8),$acc1	#$t3
180
181	and	\$0x0000ff00,$acc0
182	and	\$0x0000ff00,$acc1
183	shr	\$16,$s1
184	xor	$acc0,$t2
185	xor	$acc1,$t3
186	shr	\$16,$s0
187
188	movzb	`&lo("$s2")`,$acc0
189	movzb	`&lo("$s3")`,$acc1
190	movzb	`&lo("$s0")`,$acc2
191	mov	0($sbox,$acc0,8),$acc0	#$t0
192	mov	0($sbox,$acc1,8),$acc1	#$t1
193	mov	0($sbox,$acc2,8),$acc2	#$t2
194
195	and	\$0x00ff0000,$acc0
196	and	\$0x00ff0000,$acc1
197	and	\$0x00ff0000,$acc2
198
199	xor	$acc0,$t0
200	xor	$acc1,$t1
201	xor	$acc2,$t2
202
203	movzb	`&lo("$s1")`,$acc0
204	movzb	`&hi("$s3")`,$acc1
205	movzb	`&hi("$s0")`,$acc2
206	mov	0($sbox,$acc0,8),$acc0	#$t3
207	mov	2($sbox,$acc1,8),$acc1	#$t0
208	mov	2($sbox,$acc2,8),$acc2	#$t1
209
210	and	\$0x00ff0000,$acc0
211	and	\$0xff000000,$acc1
212	and	\$0xff000000,$acc2
213
214	xor	$acc0,$t3
215	xor	$acc1,$t0
216	xor	$acc2,$t1
217
218	movzb	`&hi("$s1")`,$acc0
219	movzb	`&hi("$s2")`,$acc1
220	mov	16+12($key),$s3
221	mov	2($sbox,$acc0,8),$acc0	#$t2
222	mov	2($sbox,$acc1,8),$acc1	#$t3
223	mov	16+0($key),$s0
224
225	and	\$0xff000000,$acc0
226	and	\$0xff000000,$acc1
227
228	xor	$acc0,$t2
229	xor	$acc1,$t3
230
231	mov	16+4($key),$s1
232	mov	16+8($key),$s2
233	xor	$t0,$s0
234	xor	$t1,$s1
235	xor	$t2,$s2
236	xor	$t3,$s3
237___
238}
239
240sub encstep()
241{ my ($i,@s) = @_;
242  my $tmp0=$acc0;
243  my $tmp1=$acc1;
244  my $tmp2=$acc2;
245  my $out=($t0,$t1,$t2,$s[0])[$i];
246
247	if ($i==3) {
248		$tmp0=$s[1];
249		$tmp1=$s[2];
250		$tmp2=$s[3];
251	}
252	$code.="	movzb	".&lo($s[0]).",$out\n";
253	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
254	$code.="	lea	16($key),$key\n"	if ($i==0);
255
256	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
257	$code.="	mov	0($sbox,$out,8),$out\n";
258
259	$code.="	shr	\$16,$tmp1\n";
260	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
261	$code.="	xor	3($sbox,$tmp0,8),$out\n";
262
263	$code.="	movzb	".&lo($tmp1).",$tmp1\n";
264	$code.="	shr	\$24,$tmp2\n";
265	$code.="	xor	4*$i($key),$out\n";
266
267	$code.="	xor	2($sbox,$tmp1,8),$out\n";
268	$code.="	xor	1($sbox,$tmp2,8),$out\n";
269
270	$code.="	mov	$t0,$s[1]\n"		if ($i==3);
271	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
272	$code.="	mov	$t2,$s[3]\n"		if ($i==3);
273	$code.="\n";
274}
275
276sub enclast()
277{ my ($i,@s)=@_;
278  my $tmp0=$acc0;
279  my $tmp1=$acc1;
280  my $tmp2=$acc2;
281  my $out=($t0,$t1,$t2,$s[0])[$i];
282
283	if ($i==3) {
284		$tmp0=$s[1];
285		$tmp1=$s[2];
286		$tmp2=$s[3];
287	}
288	$code.="	movzb	".&lo($s[0]).",$out\n";
289	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
290
291	$code.="	mov	2($sbox,$out,8),$out\n";
292	$code.="	shr	\$16,$tmp1\n";
293	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
294
295	$code.="	and	\$0x000000ff,$out\n";
296	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
297	$code.="	movzb	".&lo($tmp1).",$tmp1\n";
298	$code.="	shr	\$24,$tmp2\n";
299
300	$code.="	mov	0($sbox,$tmp0,8),$tmp0\n";
301	$code.="	mov	0($sbox,$tmp1,8),$tmp1\n";
302	$code.="	mov	2($sbox,$tmp2,8),$tmp2\n";
303
304	$code.="	and	\$0x0000ff00,$tmp0\n";
305	$code.="	and	\$0x00ff0000,$tmp1\n";
306	$code.="	and	\$0xff000000,$tmp2\n";
307
308	$code.="	xor	$tmp0,$out\n";
309	$code.="	mov	$t0,$s[1]\n"		if ($i==3);
310	$code.="	xor	$tmp1,$out\n";
311	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
312	$code.="	xor	$tmp2,$out\n";
313	$code.="	mov	$t2,$s[3]\n"		if ($i==3);
314	$code.="\n";
315}
316
317$code.=<<___;
318.type	_x86_64_AES_encrypt,\@abi-omnipotent
319.align	16
320_x86_64_AES_encrypt:
321	xor	0($key),$s0			# xor with key
322	xor	4($key),$s1
323	xor	8($key),$s2
324	xor	12($key),$s3
325
326	mov	240($key),$rnds			# load key->rounds
327	sub	\$1,$rnds
328	jmp	.Lenc_loop
329.align	16
330.Lenc_loop:
331___
332	if ($verticalspin) { &encvert(); }
333	else {	&encstep(0,$s0,$s1,$s2,$s3);
334		&encstep(1,$s1,$s2,$s3,$s0);
335		&encstep(2,$s2,$s3,$s0,$s1);
336		&encstep(3,$s3,$s0,$s1,$s2);
337	}
338$code.=<<___;
339	sub	\$1,$rnds
340	jnz	.Lenc_loop
341___
342	if ($verticalspin) { &enclastvert(); }
343	else {	&enclast(0,$s0,$s1,$s2,$s3);
344		&enclast(1,$s1,$s2,$s3,$s0);
345		&enclast(2,$s2,$s3,$s0,$s1);
346		&enclast(3,$s3,$s0,$s1,$s2);
347		$code.=<<___;
348		xor	16+0($key),$s0		# xor with key
349		xor	16+4($key),$s1
350		xor	16+8($key),$s2
351		xor	16+12($key),$s3
352___
353	}
354$code.=<<___;
355	.byte	0xf3,0xc3			# rep ret
356.size	_x86_64_AES_encrypt,.-_x86_64_AES_encrypt
357___
358
359# it's possible to implement this by shifting tN by 8, filling least
360# significant byte with byte load and finally bswap-ing at the end,
361# but such partial register load kills Core 2...
362sub enccompactvert()
363{ my ($t3,$t4,$t5)=("%r8d","%r9d","%r13d");
364
365$code.=<<___;
366	movzb	`&lo("$s0")`,$t0
367	movzb	`&lo("$s1")`,$t1
368	movzb	`&lo("$s2")`,$t2
369	movzb	($sbox,$t0,1),$t0
370	movzb	($sbox,$t1,1),$t1
371	movzb	($sbox,$t2,1),$t2
372
373	movzb	`&lo("$s3")`,$t3
374	movzb	`&hi("$s1")`,$acc0
375	movzb	`&hi("$s2")`,$acc1
376	movzb	($sbox,$t3,1),$t3
377	movzb	($sbox,$acc0,1),$t4	#$t0
378	movzb	($sbox,$acc1,1),$t5	#$t1
379
380	movzb	`&hi("$s3")`,$acc2
381	movzb	`&hi("$s0")`,$acc0
382	shr	\$16,$s2
383	movzb	($sbox,$acc2,1),$acc2	#$t2
384	movzb	($sbox,$acc0,1),$acc0	#$t3
385	shr	\$16,$s3
386
387	movzb	`&lo("$s2")`,$acc1
388	shl	\$8,$t4
389	shl	\$8,$t5
390	movzb	($sbox,$acc1,1),$acc1	#$t0
391	xor	$t4,$t0
392	xor	$t5,$t1
393
394	movzb	`&lo("$s3")`,$t4
395	shr	\$16,$s0
396	shr	\$16,$s1
397	movzb	`&lo("$s0")`,$t5
398	shl	\$8,$acc2
399	shl	\$8,$acc0
400	movzb	($sbox,$t4,1),$t4	#$t1
401	movzb	($sbox,$t5,1),$t5	#$t2
402	xor	$acc2,$t2
403	xor	$acc0,$t3
404
405	movzb	`&lo("$s1")`,$acc2
406	movzb	`&hi("$s3")`,$acc0
407	shl	\$16,$acc1
408	movzb	($sbox,$acc2,1),$acc2	#$t3
409	movzb	($sbox,$acc0,1),$acc0	#$t0
410	xor	$acc1,$t0
411
412	movzb	`&hi("$s0")`,$acc1
413	shr	\$8,$s2
414	shr	\$8,$s1
415	movzb	($sbox,$acc1,1),$acc1	#$t1
416	movzb	($sbox,$s2,1),$s3	#$t3
417	movzb	($sbox,$s1,1),$s2	#$t2
418	shl	\$16,$t4
419	shl	\$16,$t5
420	shl	\$16,$acc2
421	xor	$t4,$t1
422	xor	$t5,$t2
423	xor	$acc2,$t3
424
425	shl	\$24,$acc0
426	shl	\$24,$acc1
427	shl	\$24,$s3
428	xor	$acc0,$t0
429	shl	\$24,$s2
430	xor	$acc1,$t1
431	mov	$t0,$s0
432	mov	$t1,$s1
433	xor	$t2,$s2
434	xor	$t3,$s3
435___
436}
437
438sub enctransform_ref()
439{ my $sn = shift;
440  my ($acc,$r2,$tmp)=("%r8d","%r9d","%r13d");
441
442$code.=<<___;
443	mov	$sn,$acc
444	and	\$0x80808080,$acc
445	mov	$acc,$tmp
446	shr	\$7,$tmp
447	lea	($sn,$sn),$r2
448	sub	$tmp,$acc
449	and	\$0xfefefefe,$r2
450	and	\$0x1b1b1b1b,$acc
451	mov	$sn,$tmp
452	xor	$acc,$r2
453
454	xor	$r2,$sn
455	rol	\$24,$sn
456	xor	$r2,$sn
457	ror	\$16,$tmp
458	xor	$tmp,$sn
459	ror	\$8,$tmp
460	xor	$tmp,$sn
461___
462}
463
464# unlike decrypt case it does not pay off to parallelize enctransform
465sub enctransform()
466{ my ($t3,$r20,$r21)=($acc2,"%r8d","%r9d");
467
468$code.=<<___;
469	mov	$s0,$acc0
470	mov	$s1,$acc1
471	and	\$0x80808080,$acc0
472	and	\$0x80808080,$acc1
473	mov	$acc0,$t0
474	mov	$acc1,$t1
475	shr	\$7,$t0
476	lea	($s0,$s0),$r20
477	shr	\$7,$t1
478	lea	($s1,$s1),$r21
479	sub	$t0,$acc0
480	sub	$t1,$acc1
481	and	\$0xfefefefe,$r20
482	and	\$0xfefefefe,$r21
483	and	\$0x1b1b1b1b,$acc0
484	and	\$0x1b1b1b1b,$acc1
485	mov	$s0,$t0
486	mov	$s1,$t1
487	xor	$acc0,$r20
488	xor	$acc1,$r21
489
490	xor	$r20,$s0
491	xor	$r21,$s1
492	 mov	$s2,$acc0
493	 mov	$s3,$acc1
494	rol	\$24,$s0
495	rol	\$24,$s1
496	 and	\$0x80808080,$acc0
497	 and	\$0x80808080,$acc1
498	xor	$r20,$s0
499	xor	$r21,$s1
500	 mov	$acc0,$t2
501	 mov	$acc1,$t3
502	ror	\$16,$t0
503	ror	\$16,$t1
504	 shr	\$7,$t2
505	 lea	($s2,$s2),$r20
506	xor	$t0,$s0
507	xor	$t1,$s1
508	 shr	\$7,$t3
509	 lea	($s3,$s3),$r21
510	ror	\$8,$t0
511	ror	\$8,$t1
512	 sub	$t2,$acc0
513	 sub	$t3,$acc1
514	xor	$t0,$s0
515	xor	$t1,$s1
516
517	and	\$0xfefefefe,$r20
518	and	\$0xfefefefe,$r21
519	and	\$0x1b1b1b1b,$acc0
520	and	\$0x1b1b1b1b,$acc1
521	mov	$s2,$t2
522	mov	$s3,$t3
523	xor	$acc0,$r20
524	xor	$acc1,$r21
525
526	xor	$r20,$s2
527	xor	$r21,$s3
528	rol	\$24,$s2
529	rol	\$24,$s3
530	xor	$r20,$s2
531	xor	$r21,$s3
532	mov	0($sbox),$acc0			# prefetch Te4
533	ror	\$16,$t2
534	ror	\$16,$t3
535	mov	64($sbox),$acc1
536	xor	$t2,$s2
537	xor	$t3,$s3
538	mov	128($sbox),$r20
539	ror	\$8,$t2
540	ror	\$8,$t3
541	mov	192($sbox),$r21
542	xor	$t2,$s2
543	xor	$t3,$s3
544___
545}
546
547$code.=<<___;
548.type	_x86_64_AES_encrypt_compact,\@abi-omnipotent
549.align	16
550_x86_64_AES_encrypt_compact:
551	lea	128($sbox),$inp			# size optimization
552	mov	0-128($inp),$acc1		# prefetch Te4
553	mov	32-128($inp),$acc2
554	mov	64-128($inp),$t0
555	mov	96-128($inp),$t1
556	mov	128-128($inp),$acc1
557	mov	160-128($inp),$acc2
558	mov	192-128($inp),$t0
559	mov	224-128($inp),$t1
560	jmp	.Lenc_loop_compact
561.align	16
562.Lenc_loop_compact:
563		xor	0($key),$s0		# xor with key
564		xor	4($key),$s1
565		xor	8($key),$s2
566		xor	12($key),$s3
567		lea	16($key),$key
568___
569		&enccompactvert();
570$code.=<<___;
571		cmp	16(%rsp),$key
572		je	.Lenc_compact_done
573___
574		&enctransform();
575$code.=<<___;
576	jmp	.Lenc_loop_compact
577.align	16
578.Lenc_compact_done:
579	xor	0($key),$s0
580	xor	4($key),$s1
581	xor	8($key),$s2
582	xor	12($key),$s3
583	.byte	0xf3,0xc3			# rep ret
584.size	_x86_64_AES_encrypt_compact,.-_x86_64_AES_encrypt_compact
585___
586
587# void AES_encrypt (const void *inp,void *out,const AES_KEY *key);
588$code.=<<___;
589.globl	AES_encrypt
590.type	AES_encrypt,\@function,3
591.align	16
592.globl	asm_AES_encrypt
593.hidden	asm_AES_encrypt
594asm_AES_encrypt:
595AES_encrypt:
596	push	%rbx
597	push	%rbp
598	push	%r12
599	push	%r13
600	push	%r14
601	push	%r15
602
603	# allocate frame "above" key schedule
604	mov	%rsp,%r10
605	lea	-63(%rdx),%rcx	# %rdx is key argument
606	and	\$-64,%rsp
607	sub	%rsp,%rcx
608	neg	%rcx
609	and	\$0x3c0,%rcx
610	sub	%rcx,%rsp
611	sub	\$32,%rsp
612
613	mov	%rsi,16(%rsp)	# save out
614	mov	%r10,24(%rsp)	# save real stack pointer
615.Lenc_prologue:
616
617	mov	%rdx,$key
618	mov	240($key),$rnds	# load rounds
619
620	mov	0(%rdi),$s0	# load input vector
621	mov	4(%rdi),$s1
622	mov	8(%rdi),$s2
623	mov	12(%rdi),$s3
624
625	shl	\$4,$rnds
626	lea	($key,$rnds),%rbp
627	mov	$key,(%rsp)	# key schedule
628	mov	%rbp,8(%rsp)	# end of key schedule
629
630	# pick Te4 copy which can't "overlap" with stack frame or key schedule
631	lea	.LAES_Te+2048(%rip),$sbox
632	lea	768(%rsp),%rbp
633	sub	$sbox,%rbp
634	and	\$0x300,%rbp
635	lea	($sbox,%rbp),$sbox
636
637	call	_x86_64_AES_encrypt_compact
638
639	mov	16(%rsp),$out	# restore out
640	mov	24(%rsp),%rsi	# restore saved stack pointer
641	mov	$s0,0($out)	# write output vector
642	mov	$s1,4($out)
643	mov	$s2,8($out)
644	mov	$s3,12($out)
645
646	mov	(%rsi),%r15
647	mov	8(%rsi),%r14
648	mov	16(%rsi),%r13
649	mov	24(%rsi),%r12
650	mov	32(%rsi),%rbp
651	mov	40(%rsi),%rbx
652	lea	48(%rsi),%rsp
653.Lenc_epilogue:
654	ret
655.size	AES_encrypt,.-AES_encrypt
656___
657
658#------------------------------------------------------------------#
659
660sub decvert()
661{ my $t3="%r8d";	# zaps $inp!
662
663$code.=<<___;
664	# favor 3-way issue Opteron pipeline...
665	movzb	`&lo("$s0")`,$acc0
666	movzb	`&lo("$s1")`,$acc1
667	movzb	`&lo("$s2")`,$acc2
668	mov	0($sbox,$acc0,8),$t0
669	mov	0($sbox,$acc1,8),$t1
670	mov	0($sbox,$acc2,8),$t2
671
672	movzb	`&hi("$s3")`,$acc0
673	movzb	`&hi("$s0")`,$acc1
674	movzb	`&lo("$s3")`,$acc2
675	xor	3($sbox,$acc0,8),$t0
676	xor	3($sbox,$acc1,8),$t1
677	mov	0($sbox,$acc2,8),$t3
678
679	movzb	`&hi("$s1")`,$acc0
680	shr	\$16,$s0
681	movzb	`&hi("$s2")`,$acc2
682	xor	3($sbox,$acc0,8),$t2
683	shr	\$16,$s3
684	xor	3($sbox,$acc2,8),$t3
685
686	shr	\$16,$s1
687	lea	16($key),$key
688	shr	\$16,$s2
689
690	movzb	`&lo("$s2")`,$acc0
691	movzb	`&lo("$s3")`,$acc1
692	movzb	`&lo("$s0")`,$acc2
693	xor	2($sbox,$acc0,8),$t0
694	xor	2($sbox,$acc1,8),$t1
695	xor	2($sbox,$acc2,8),$t2
696
697	movzb	`&hi("$s1")`,$acc0
698	movzb	`&hi("$s2")`,$acc1
699	movzb	`&lo("$s1")`,$acc2
700	xor	1($sbox,$acc0,8),$t0
701	xor	1($sbox,$acc1,8),$t1
702	xor	2($sbox,$acc2,8),$t3
703
704	movzb	`&hi("$s3")`,$acc0
705	mov	12($key),$s3
706	movzb	`&hi("$s0")`,$acc2
707	xor	1($sbox,$acc0,8),$t2
708	mov	0($key),$s0
709	xor	1($sbox,$acc2,8),$t3
710
711	xor	$t0,$s0
712	mov	4($key),$s1
713	mov	8($key),$s2
714	xor	$t2,$s2
715	xor	$t1,$s1
716	xor	$t3,$s3
717___
718}
719
720sub declastvert()
721{ my $t3="%r8d";	# zaps $inp!
722
723$code.=<<___;
724	lea	2048($sbox),$sbox	# size optimization
725	movzb	`&lo("$s0")`,$acc0
726	movzb	`&lo("$s1")`,$acc1
727	movzb	`&lo("$s2")`,$acc2
728	movzb	($sbox,$acc0,1),$t0
729	movzb	($sbox,$acc1,1),$t1
730	movzb	($sbox,$acc2,1),$t2
731
732	movzb	`&lo("$s3")`,$acc0
733	movzb	`&hi("$s3")`,$acc1
734	movzb	`&hi("$s0")`,$acc2
735	movzb	($sbox,$acc0,1),$t3
736	movzb	($sbox,$acc1,1),$acc1	#$t0
737	movzb	($sbox,$acc2,1),$acc2	#$t1
738
739	shl	\$8,$acc1
740	shl	\$8,$acc2
741
742	xor	$acc1,$t0
743	xor	$acc2,$t1
744	shr	\$16,$s3
745
746	movzb	`&hi("$s1")`,$acc0
747	movzb	`&hi("$s2")`,$acc1
748	shr	\$16,$s0
749	movzb	($sbox,$acc0,1),$acc0	#$t2
750	movzb	($sbox,$acc1,1),$acc1	#$t3
751
752	shl	\$8,$acc0
753	shl	\$8,$acc1
754	shr	\$16,$s1
755	xor	$acc0,$t2
756	xor	$acc1,$t3
757	shr	\$16,$s2
758
759	movzb	`&lo("$s2")`,$acc0
760	movzb	`&lo("$s3")`,$acc1
761	movzb	`&lo("$s0")`,$acc2
762	movzb	($sbox,$acc0,1),$acc0	#$t0
763	movzb	($sbox,$acc1,1),$acc1	#$t1
764	movzb	($sbox,$acc2,1),$acc2	#$t2
765
766	shl	\$16,$acc0
767	shl	\$16,$acc1
768	shl	\$16,$acc2
769
770	xor	$acc0,$t0
771	xor	$acc1,$t1
772	xor	$acc2,$t2
773
774	movzb	`&lo("$s1")`,$acc0
775	movzb	`&hi("$s1")`,$acc1
776	movzb	`&hi("$s2")`,$acc2
777	movzb	($sbox,$acc0,1),$acc0	#$t3
778	movzb	($sbox,$acc1,1),$acc1	#$t0
779	movzb	($sbox,$acc2,1),$acc2	#$t1
780
781	shl	\$16,$acc0
782	shl	\$24,$acc1
783	shl	\$24,$acc2
784
785	xor	$acc0,$t3
786	xor	$acc1,$t0
787	xor	$acc2,$t1
788
789	movzb	`&hi("$s3")`,$acc0
790	movzb	`&hi("$s0")`,$acc1
791	mov	16+12($key),$s3
792	movzb	($sbox,$acc0,1),$acc0	#$t2
793	movzb	($sbox,$acc1,1),$acc1	#$t3
794	mov	16+0($key),$s0
795
796	shl	\$24,$acc0
797	shl	\$24,$acc1
798
799	xor	$acc0,$t2
800	xor	$acc1,$t3
801
802	mov	16+4($key),$s1
803	mov	16+8($key),$s2
804	lea	-2048($sbox),$sbox
805	xor	$t0,$s0
806	xor	$t1,$s1
807	xor	$t2,$s2
808	xor	$t3,$s3
809___
810}
811
812sub decstep()
813{ my ($i,@s) = @_;
814  my $tmp0=$acc0;
815  my $tmp1=$acc1;
816  my $tmp2=$acc2;
817  my $out=($t0,$t1,$t2,$s[0])[$i];
818
819	$code.="	mov	$s[0],$out\n"		if ($i!=3);
820			$tmp1=$s[2]			if ($i==3);
821	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
822	$code.="	and	\$0xFF,$out\n";
823
824	$code.="	mov	0($sbox,$out,8),$out\n";
825	$code.="	shr	\$16,$tmp1\n";
826			$tmp2=$s[3]			if ($i==3);
827	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
828
829			$tmp0=$s[1]			if ($i==3);
830	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
831	$code.="	and	\$0xFF,$tmp1\n";
832	$code.="	shr	\$24,$tmp2\n";
833
834	$code.="	xor	3($sbox,$tmp0,8),$out\n";
835	$code.="	xor	2($sbox,$tmp1,8),$out\n";
836	$code.="	xor	1($sbox,$tmp2,8),$out\n";
837
838	$code.="	mov	$t2,$s[1]\n"		if ($i==3);
839	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
840	$code.="	mov	$t0,$s[3]\n"		if ($i==3);
841	$code.="\n";
842}
843
844sub declast()
845{ my ($i,@s)=@_;
846  my $tmp0=$acc0;
847  my $tmp1=$acc1;
848  my $tmp2=$acc2;
849  my $out=($t0,$t1,$t2,$s[0])[$i];
850
851	$code.="	mov	$s[0],$out\n"		if ($i!=3);
852			$tmp1=$s[2]			if ($i==3);
853	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
854	$code.="	and	\$0xFF,$out\n";
855
856	$code.="	movzb	2048($sbox,$out,1),$out\n";
857	$code.="	shr	\$16,$tmp1\n";
858			$tmp2=$s[3]			if ($i==3);
859	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
860
861			$tmp0=$s[1]			if ($i==3);
862	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
863	$code.="	and	\$0xFF,$tmp1\n";
864	$code.="	shr	\$24,$tmp2\n";
865
866	$code.="	movzb	2048($sbox,$tmp0,1),$tmp0\n";
867	$code.="	movzb	2048($sbox,$tmp1,1),$tmp1\n";
868	$code.="	movzb	2048($sbox,$tmp2,1),$tmp2\n";
869
870	$code.="	shl	\$8,$tmp0\n";
871	$code.="	shl	\$16,$tmp1\n";
872	$code.="	shl	\$24,$tmp2\n";
873
874	$code.="	xor	$tmp0,$out\n";
875	$code.="	mov	$t2,$s[1]\n"		if ($i==3);
876	$code.="	xor	$tmp1,$out\n";
877	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
878	$code.="	xor	$tmp2,$out\n";
879	$code.="	mov	$t0,$s[3]\n"		if ($i==3);
880	$code.="\n";
881}
882
883$code.=<<___;
884.type	_x86_64_AES_decrypt,\@abi-omnipotent
885.align	16
886_x86_64_AES_decrypt:
887	xor	0($key),$s0			# xor with key
888	xor	4($key),$s1
889	xor	8($key),$s2
890	xor	12($key),$s3
891
892	mov	240($key),$rnds			# load key->rounds
893	sub	\$1,$rnds
894	jmp	.Ldec_loop
895.align	16
896.Ldec_loop:
897___
898	if ($verticalspin) { &decvert(); }
899	else {	&decstep(0,$s0,$s3,$s2,$s1);
900		&decstep(1,$s1,$s0,$s3,$s2);
901		&decstep(2,$s2,$s1,$s0,$s3);
902		&decstep(3,$s3,$s2,$s1,$s0);
903		$code.=<<___;
904		lea	16($key),$key
905		xor	0($key),$s0			# xor with key
906		xor	4($key),$s1
907		xor	8($key),$s2
908		xor	12($key),$s3
909___
910	}
911$code.=<<___;
912	sub	\$1,$rnds
913	jnz	.Ldec_loop
914___
915	if ($verticalspin) { &declastvert(); }
916	else {	&declast(0,$s0,$s3,$s2,$s1);
917		&declast(1,$s1,$s0,$s3,$s2);
918		&declast(2,$s2,$s1,$s0,$s3);
919		&declast(3,$s3,$s2,$s1,$s0);
920		$code.=<<___;
921		xor	16+0($key),$s0			# xor with key
922		xor	16+4($key),$s1
923		xor	16+8($key),$s2
924		xor	16+12($key),$s3
925___
926	}
927$code.=<<___;
928	.byte	0xf3,0xc3			# rep ret
929.size	_x86_64_AES_decrypt,.-_x86_64_AES_decrypt
930___
931
932sub deccompactvert()
933{ my ($t3,$t4,$t5)=("%r8d","%r9d","%r13d");
934
935$code.=<<___;
936	movzb	`&lo("$s0")`,$t0
937	movzb	`&lo("$s1")`,$t1
938	movzb	`&lo("$s2")`,$t2
939	movzb	($sbox,$t0,1),$t0
940	movzb	($sbox,$t1,1),$t1
941	movzb	($sbox,$t2,1),$t2
942
943	movzb	`&lo("$s3")`,$t3
944	movzb	`&hi("$s3")`,$acc0
945	movzb	`&hi("$s0")`,$acc1
946	movzb	($sbox,$t3,1),$t3
947	movzb	($sbox,$acc0,1),$t4	#$t0
948	movzb	($sbox,$acc1,1),$t5	#$t1
949
950	movzb	`&hi("$s1")`,$acc2
951	movzb	`&hi("$s2")`,$acc0
952	shr	\$16,$s2
953	movzb	($sbox,$acc2,1),$acc2	#$t2
954	movzb	($sbox,$acc0,1),$acc0	#$t3
955	shr	\$16,$s3
956
957	movzb	`&lo("$s2")`,$acc1
958	shl	\$8,$t4
959	shl	\$8,$t5
960	movzb	($sbox,$acc1,1),$acc1	#$t0
961	xor	$t4,$t0
962	xor	$t5,$t1
963
964	movzb	`&lo("$s3")`,$t4
965	shr	\$16,$s0
966	shr	\$16,$s1
967	movzb	`&lo("$s0")`,$t5
968	shl	\$8,$acc2
969	shl	\$8,$acc0
970	movzb	($sbox,$t4,1),$t4	#$t1
971	movzb	($sbox,$t5,1),$t5	#$t2
972	xor	$acc2,$t2
973	xor	$acc0,$t3
974
975	movzb	`&lo("$s1")`,$acc2
976	movzb	`&hi("$s1")`,$acc0
977	shl	\$16,$acc1
978	movzb	($sbox,$acc2,1),$acc2	#$t3
979	movzb	($sbox,$acc0,1),$acc0	#$t0
980	xor	$acc1,$t0
981
982	movzb	`&hi("$s2")`,$acc1
983	shl	\$16,$t4
984	shl	\$16,$t5
985	movzb	($sbox,$acc1,1),$s1	#$t1
986	xor	$t4,$t1
987	xor	$t5,$t2
988
989	movzb	`&hi("$s3")`,$acc1
990	shr	\$8,$s0
991	shl	\$16,$acc2
992	movzb	($sbox,$acc1,1),$s2	#$t2
993	movzb	($sbox,$s0,1),$s3	#$t3
994	xor	$acc2,$t3
995
996	shl	\$24,$acc0
997	shl	\$24,$s1
998	shl	\$24,$s2
999	xor	$acc0,$t0
1000	shl	\$24,$s3
1001	xor	$t1,$s1
1002	mov	$t0,$s0
1003	xor	$t2,$s2
1004	xor	$t3,$s3
1005___
1006}
1007
1008# parallelized version! input is pair of 64-bit values: %rax=s1.s0
1009# and %rcx=s3.s2, output is four 32-bit values in %eax=s0, %ebx=s1,
1010# %ecx=s2 and %edx=s3.
1011sub dectransform()
1012{ my ($tp10,$tp20,$tp40,$tp80,$acc0)=("%rax","%r8", "%r9", "%r10","%rbx");
1013  my ($tp18,$tp28,$tp48,$tp88,$acc8)=("%rcx","%r11","%r12","%r13","%rdx");
1014  my $prefetch = shift;
1015
1016$code.=<<___;
1017	mov	$tp10,$acc0
1018	mov	$tp18,$acc8
1019	and	$mask80,$acc0
1020	and	$mask80,$acc8
1021	mov	$acc0,$tp40
1022	mov	$acc8,$tp48
1023	shr	\$7,$tp40
1024	lea	($tp10,$tp10),$tp20
1025	shr	\$7,$tp48
1026	lea	($tp18,$tp18),$tp28
1027	sub	$tp40,$acc0
1028	sub	$tp48,$acc8
1029	and	$maskfe,$tp20
1030	and	$maskfe,$tp28
1031	and	$mask1b,$acc0
1032	and	$mask1b,$acc8
1033	xor	$tp20,$acc0
1034	xor	$tp28,$acc8
1035	mov	$acc0,$tp20
1036	mov	$acc8,$tp28
1037
1038	and	$mask80,$acc0
1039	and	$mask80,$acc8
1040	mov	$acc0,$tp80
1041	mov	$acc8,$tp88
1042	shr	\$7,$tp80
1043	lea	($tp20,$tp20),$tp40
1044	shr	\$7,$tp88
1045	lea	($tp28,$tp28),$tp48
1046	sub	$tp80,$acc0
1047	sub	$tp88,$acc8
1048	and	$maskfe,$tp40
1049	and	$maskfe,$tp48
1050	and	$mask1b,$acc0
1051	and	$mask1b,$acc8
1052	xor	$tp40,$acc0
1053	xor	$tp48,$acc8
1054	mov	$acc0,$tp40
1055	mov	$acc8,$tp48
1056
1057	and	$mask80,$acc0
1058	and	$mask80,$acc8
1059	mov	$acc0,$tp80
1060	mov	$acc8,$tp88
1061	shr	\$7,$tp80
1062	 xor	$tp10,$tp20		# tp2^=tp1
1063	shr	\$7,$tp88
1064	 xor	$tp18,$tp28		# tp2^=tp1
1065	sub	$tp80,$acc0
1066	sub	$tp88,$acc8
1067	lea	($tp40,$tp40),$tp80
1068	lea	($tp48,$tp48),$tp88
1069	 xor	$tp10,$tp40		# tp4^=tp1
1070	 xor	$tp18,$tp48		# tp4^=tp1
1071	and	$maskfe,$tp80
1072	and	$maskfe,$tp88
1073	and	$mask1b,$acc0
1074	and	$mask1b,$acc8
1075	xor	$acc0,$tp80
1076	xor	$acc8,$tp88
1077
1078	xor	$tp80,$tp10		# tp1^=tp8
1079	xor	$tp88,$tp18		# tp1^=tp8
1080	xor	$tp80,$tp20		# tp2^tp1^=tp8
1081	xor	$tp88,$tp28		# tp2^tp1^=tp8
1082	mov	$tp10,$acc0
1083	mov	$tp18,$acc8
1084	xor	$tp80,$tp40		# tp4^tp1^=tp8
1085	xor	$tp88,$tp48		# tp4^tp1^=tp8
1086	shr	\$32,$acc0
1087	shr	\$32,$acc8
1088	xor	$tp20,$tp80		# tp8^=tp8^tp2^tp1=tp2^tp1
1089	xor	$tp28,$tp88		# tp8^=tp8^tp2^tp1=tp2^tp1
1090	rol	\$8,`&LO("$tp10")`	# ROTATE(tp1^tp8,8)
1091	rol	\$8,`&LO("$tp18")`	# ROTATE(tp1^tp8,8)
1092	xor	$tp40,$tp80		# tp2^tp1^=tp8^tp4^tp1=tp8^tp4^tp2
1093	xor	$tp48,$tp88		# tp2^tp1^=tp8^tp4^tp1=tp8^tp4^tp2
1094
1095	rol	\$8,`&LO("$acc0")`	# ROTATE(tp1^tp8,8)
1096	rol	\$8,`&LO("$acc8")`	# ROTATE(tp1^tp8,8)
1097	xor	`&LO("$tp80")`,`&LO("$tp10")`
1098	xor	`&LO("$tp88")`,`&LO("$tp18")`
1099	shr	\$32,$tp80
1100	shr	\$32,$tp88
1101	xor	`&LO("$tp80")`,`&LO("$acc0")`
1102	xor	`&LO("$tp88")`,`&LO("$acc8")`
1103
1104	mov	$tp20,$tp80
1105	mov	$tp28,$tp88
1106	shr	\$32,$tp80
1107	shr	\$32,$tp88
1108	rol	\$24,`&LO("$tp20")`	# ROTATE(tp2^tp1^tp8,24)
1109	rol	\$24,`&LO("$tp28")`	# ROTATE(tp2^tp1^tp8,24)
1110	rol	\$24,`&LO("$tp80")`	# ROTATE(tp2^tp1^tp8,24)
1111	rol	\$24,`&LO("$tp88")`	# ROTATE(tp2^tp1^tp8,24)
1112	xor	`&LO("$tp20")`,`&LO("$tp10")`
1113	xor	`&LO("$tp28")`,`&LO("$tp18")`
1114	mov	$tp40,$tp20
1115	mov	$tp48,$tp28
1116	xor	`&LO("$tp80")`,`&LO("$acc0")`
1117	xor	`&LO("$tp88")`,`&LO("$acc8")`
1118
1119	`"mov	0($sbox),$mask80"	if ($prefetch)`
1120	shr	\$32,$tp20
1121	shr	\$32,$tp28
1122	`"mov	64($sbox),$maskfe"	if ($prefetch)`
1123	rol	\$16,`&LO("$tp40")`	# ROTATE(tp4^tp1^tp8,16)
1124	rol	\$16,`&LO("$tp48")`	# ROTATE(tp4^tp1^tp8,16)
1125	`"mov	128($sbox),$mask1b"	if ($prefetch)`
1126	rol	\$16,`&LO("$tp20")`	# ROTATE(tp4^tp1^tp8,16)
1127	rol	\$16,`&LO("$tp28")`	# ROTATE(tp4^tp1^tp8,16)
1128	`"mov	192($sbox),$tp80"	if ($prefetch)`
1129	xor	`&LO("$tp40")`,`&LO("$tp10")`
1130	xor	`&LO("$tp48")`,`&LO("$tp18")`
1131	`"mov	256($sbox),$tp88"	if ($prefetch)`
1132	xor	`&LO("$tp20")`,`&LO("$acc0")`
1133	xor	`&LO("$tp28")`,`&LO("$acc8")`
1134___
1135}
1136
1137$code.=<<___;
1138.type	_x86_64_AES_decrypt_compact,\@abi-omnipotent
1139.align	16
1140_x86_64_AES_decrypt_compact:
1141	lea	128($sbox),$inp			# size optimization
1142	mov	0-128($inp),$acc1		# prefetch Td4
1143	mov	32-128($inp),$acc2
1144	mov	64-128($inp),$t0
1145	mov	96-128($inp),$t1
1146	mov	128-128($inp),$acc1
1147	mov	160-128($inp),$acc2
1148	mov	192-128($inp),$t0
1149	mov	224-128($inp),$t1
1150	jmp	.Ldec_loop_compact
1151
1152.align	16
1153.Ldec_loop_compact:
1154		xor	0($key),$s0		# xor with key
1155		xor	4($key),$s1
1156		xor	8($key),$s2
1157		xor	12($key),$s3
1158		lea	16($key),$key
1159___
1160		&deccompactvert();
1161$code.=<<___;
1162		cmp	16(%rsp),$key
1163		je	.Ldec_compact_done
1164
1165		mov	256+0($sbox),$mask80
1166		shl	\$32,%rbx
1167		shl	\$32,%rdx
1168		mov	256+8($sbox),$maskfe
1169		or	%rbx,%rax
1170		or	%rdx,%rcx
1171		mov	256+16($sbox),$mask1b
1172___
1173		&dectransform(1);
1174$code.=<<___;
1175	jmp	.Ldec_loop_compact
1176.align	16
1177.Ldec_compact_done:
1178	xor	0($key),$s0
1179	xor	4($key),$s1
1180	xor	8($key),$s2
1181	xor	12($key),$s3
1182	.byte	0xf3,0xc3			# rep ret
1183.size	_x86_64_AES_decrypt_compact,.-_x86_64_AES_decrypt_compact
1184___
1185
1186# void AES_decrypt (const void *inp,void *out,const AES_KEY *key);
1187$code.=<<___;
1188.globl	AES_decrypt
1189.type	AES_decrypt,\@function,3
1190.align	16
1191.globl	asm_AES_decrypt
1192.hidden	asm_AES_decrypt
1193asm_AES_decrypt:
1194AES_decrypt:
1195	push	%rbx
1196	push	%rbp
1197	push	%r12
1198	push	%r13
1199	push	%r14
1200	push	%r15
1201
1202	# allocate frame "above" key schedule
1203	mov	%rsp,%r10
1204	lea	-63(%rdx),%rcx	# %rdx is key argument
1205	and	\$-64,%rsp
1206	sub	%rsp,%rcx
1207	neg	%rcx
1208	and	\$0x3c0,%rcx
1209	sub	%rcx,%rsp
1210	sub	\$32,%rsp
1211
1212	mov	%rsi,16(%rsp)	# save out
1213	mov	%r10,24(%rsp)	# save real stack pointer
1214.Ldec_prologue:
1215
1216	mov	%rdx,$key
1217	mov	240($key),$rnds	# load rounds
1218
1219	mov	0(%rdi),$s0	# load input vector
1220	mov	4(%rdi),$s1
1221	mov	8(%rdi),$s2
1222	mov	12(%rdi),$s3
1223
1224	shl	\$4,$rnds
1225	lea	($key,$rnds),%rbp
1226	mov	$key,(%rsp)	# key schedule
1227	mov	%rbp,8(%rsp)	# end of key schedule
1228
1229	# pick Td4 copy which can't "overlap" with stack frame or key schedule
1230	lea	.LAES_Td+2048(%rip),$sbox
1231	lea	768(%rsp),%rbp
1232	sub	$sbox,%rbp
1233	and	\$0x300,%rbp
1234	lea	($sbox,%rbp),$sbox
1235	shr	\$3,%rbp	# recall "magic" constants!
1236	add	%rbp,$sbox
1237
1238	call	_x86_64_AES_decrypt_compact
1239
1240	mov	16(%rsp),$out	# restore out
1241	mov	24(%rsp),%rsi	# restore saved stack pointer
1242	mov	$s0,0($out)	# write output vector
1243	mov	$s1,4($out)
1244	mov	$s2,8($out)
1245	mov	$s3,12($out)
1246
1247	mov	(%rsi),%r15
1248	mov	8(%rsi),%r14
1249	mov	16(%rsi),%r13
1250	mov	24(%rsi),%r12
1251	mov	32(%rsi),%rbp
1252	mov	40(%rsi),%rbx
1253	lea	48(%rsi),%rsp
1254.Ldec_epilogue:
1255	ret
1256.size	AES_decrypt,.-AES_decrypt
1257___
1258#------------------------------------------------------------------#
1259
1260sub enckey()
1261{
1262$code.=<<___;
1263	movz	%dl,%esi		# rk[i]>>0
1264	movzb	-128(%rbp,%rsi),%ebx
1265	movz	%dh,%esi		# rk[i]>>8
1266	shl	\$24,%ebx
1267	xor	%ebx,%eax
1268
1269	movzb	-128(%rbp,%rsi),%ebx
1270	shr	\$16,%edx
1271	movz	%dl,%esi		# rk[i]>>16
1272	xor	%ebx,%eax
1273
1274	movzb	-128(%rbp,%rsi),%ebx
1275	movz	%dh,%esi		# rk[i]>>24
1276	shl	\$8,%ebx
1277	xor	%ebx,%eax
1278
1279	movzb	-128(%rbp,%rsi),%ebx
1280	shl	\$16,%ebx
1281	xor	%ebx,%eax
1282
1283	xor	1024-128(%rbp,%rcx,4),%eax		# rcon
1284___
1285}
1286
1287# int private_AES_set_encrypt_key(const unsigned char *userKey, const int bits,
1288#                        AES_KEY *key)
1289$code.=<<___;
1290.globl	private_AES_set_encrypt_key
1291.type	private_AES_set_encrypt_key,\@function,3
1292.align	16
1293private_AES_set_encrypt_key:
1294	push	%rbx
1295	push	%rbp
1296	push	%r12			# redundant, but allows to share
1297	push	%r13			# exception handler...
1298	push	%r14
1299	push	%r15
1300	sub	\$8,%rsp
1301.Lenc_key_prologue:
1302
1303	call	_x86_64_AES_set_encrypt_key
1304
1305	mov	8(%rsp),%r15
1306	mov	16(%rsp),%r14
1307	mov	24(%rsp),%r13
1308	mov	32(%rsp),%r12
1309	mov	40(%rsp),%rbp
1310	mov	48(%rsp),%rbx
1311	add	\$56,%rsp
1312.Lenc_key_epilogue:
1313	ret
1314.size	private_AES_set_encrypt_key,.-private_AES_set_encrypt_key
1315
1316.type	_x86_64_AES_set_encrypt_key,\@abi-omnipotent
1317.align	16
1318_x86_64_AES_set_encrypt_key:
1319	mov	%esi,%ecx			# %ecx=bits
1320	mov	%rdi,%rsi			# %rsi=userKey
1321	mov	%rdx,%rdi			# %rdi=key
1322
1323	test	\$-1,%rsi
1324	jz	.Lbadpointer
1325	test	\$-1,%rdi
1326	jz	.Lbadpointer
1327
1328	lea	.LAES_Te(%rip),%rbp
1329	lea	2048+128(%rbp),%rbp
1330
1331	# prefetch Te4
1332	mov	0-128(%rbp),%eax
1333	mov	32-128(%rbp),%ebx
1334	mov	64-128(%rbp),%r8d
1335	mov	96-128(%rbp),%edx
1336	mov	128-128(%rbp),%eax
1337	mov	160-128(%rbp),%ebx
1338	mov	192-128(%rbp),%r8d
1339	mov	224-128(%rbp),%edx
1340
1341	cmp	\$128,%ecx
1342	je	.L10rounds
1343	cmp	\$192,%ecx
1344	je	.L12rounds
1345	cmp	\$256,%ecx
1346	je	.L14rounds
1347	mov	\$-2,%rax			# invalid number of bits
1348	jmp	.Lexit
1349
1350.L10rounds:
1351	mov	0(%rsi),%rax			# copy first 4 dwords
1352	mov	8(%rsi),%rdx
1353	mov	%rax,0(%rdi)
1354	mov	%rdx,8(%rdi)
1355
1356	shr	\$32,%rdx
1357	xor	%ecx,%ecx
1358	jmp	.L10shortcut
1359.align	4
1360.L10loop:
1361		mov	0(%rdi),%eax			# rk[0]
1362		mov	12(%rdi),%edx			# rk[3]
1363.L10shortcut:
1364___
1365		&enckey	();
1366$code.=<<___;
1367		mov	%eax,16(%rdi)			# rk[4]
1368		xor	4(%rdi),%eax
1369		mov	%eax,20(%rdi)			# rk[5]
1370		xor	8(%rdi),%eax
1371		mov	%eax,24(%rdi)			# rk[6]
1372		xor	12(%rdi),%eax
1373		mov	%eax,28(%rdi)			# rk[7]
1374		add	\$1,%ecx
1375		lea	16(%rdi),%rdi
1376		cmp	\$10,%ecx
1377	jl	.L10loop
1378
1379	movl	\$10,80(%rdi)			# setup number of rounds
1380	xor	%rax,%rax
1381	jmp	.Lexit
1382
1383.L12rounds:
1384	mov	0(%rsi),%rax			# copy first 6 dwords
1385	mov	8(%rsi),%rbx
1386	mov	16(%rsi),%rdx
1387	mov	%rax,0(%rdi)
1388	mov	%rbx,8(%rdi)
1389	mov	%rdx,16(%rdi)
1390
1391	shr	\$32,%rdx
1392	xor	%ecx,%ecx
1393	jmp	.L12shortcut
1394.align	4
1395.L12loop:
1396		mov	0(%rdi),%eax			# rk[0]
1397		mov	20(%rdi),%edx			# rk[5]
1398.L12shortcut:
1399___
1400		&enckey	();
1401$code.=<<___;
1402		mov	%eax,24(%rdi)			# rk[6]
1403		xor	4(%rdi),%eax
1404		mov	%eax,28(%rdi)			# rk[7]
1405		xor	8(%rdi),%eax
1406		mov	%eax,32(%rdi)			# rk[8]
1407		xor	12(%rdi),%eax
1408		mov	%eax,36(%rdi)			# rk[9]
1409
1410		cmp	\$7,%ecx
1411		je	.L12break
1412		add	\$1,%ecx
1413
1414		xor	16(%rdi),%eax
1415		mov	%eax,40(%rdi)			# rk[10]
1416		xor	20(%rdi),%eax
1417		mov	%eax,44(%rdi)			# rk[11]
1418
1419		lea	24(%rdi),%rdi
1420	jmp	.L12loop
1421.L12break:
1422	movl	\$12,72(%rdi)		# setup number of rounds
1423	xor	%rax,%rax
1424	jmp	.Lexit
1425
1426.L14rounds:
1427	mov	0(%rsi),%rax			# copy first 8 dwords
1428	mov	8(%rsi),%rbx
1429	mov	16(%rsi),%rcx
1430	mov	24(%rsi),%rdx
1431	mov	%rax,0(%rdi)
1432	mov	%rbx,8(%rdi)
1433	mov	%rcx,16(%rdi)
1434	mov	%rdx,24(%rdi)
1435
1436	shr	\$32,%rdx
1437	xor	%ecx,%ecx
1438	jmp	.L14shortcut
1439.align	4
1440.L14loop:
1441		mov	0(%rdi),%eax			# rk[0]
1442		mov	28(%rdi),%edx			# rk[4]
1443.L14shortcut:
1444___
1445		&enckey	();
1446$code.=<<___;
1447		mov	%eax,32(%rdi)			# rk[8]
1448		xor	4(%rdi),%eax
1449		mov	%eax,36(%rdi)			# rk[9]
1450		xor	8(%rdi),%eax
1451		mov	%eax,40(%rdi)			# rk[10]
1452		xor	12(%rdi),%eax
1453		mov	%eax,44(%rdi)			# rk[11]
1454
1455		cmp	\$6,%ecx
1456		je	.L14break
1457		add	\$1,%ecx
1458
1459		mov	%eax,%edx
1460		mov	16(%rdi),%eax			# rk[4]
1461		movz	%dl,%esi			# rk[11]>>0
1462		movzb	-128(%rbp,%rsi),%ebx
1463		movz	%dh,%esi			# rk[11]>>8
1464		xor	%ebx,%eax
1465
1466		movzb	-128(%rbp,%rsi),%ebx
1467		shr	\$16,%edx
1468		shl	\$8,%ebx
1469		movz	%dl,%esi			# rk[11]>>16
1470		xor	%ebx,%eax
1471
1472		movzb	-128(%rbp,%rsi),%ebx
1473		movz	%dh,%esi			# rk[11]>>24
1474		shl	\$16,%ebx
1475		xor	%ebx,%eax
1476
1477		movzb	-128(%rbp,%rsi),%ebx
1478		shl	\$24,%ebx
1479		xor	%ebx,%eax
1480
1481		mov	%eax,48(%rdi)			# rk[12]
1482		xor	20(%rdi),%eax
1483		mov	%eax,52(%rdi)			# rk[13]
1484		xor	24(%rdi),%eax
1485		mov	%eax,56(%rdi)			# rk[14]
1486		xor	28(%rdi),%eax
1487		mov	%eax,60(%rdi)			# rk[15]
1488
1489		lea	32(%rdi),%rdi
1490	jmp	.L14loop
1491.L14break:
1492	movl	\$14,48(%rdi)		# setup number of rounds
1493	xor	%rax,%rax
1494	jmp	.Lexit
1495
1496.Lbadpointer:
1497	mov	\$-1,%rax
1498.Lexit:
1499	.byte	0xf3,0xc3			# rep ret
1500.size	_x86_64_AES_set_encrypt_key,.-_x86_64_AES_set_encrypt_key
1501___
1502
1503sub deckey_ref()
1504{ my ($i,$ptr,$te,$td) = @_;
1505  my ($tp1,$tp2,$tp4,$tp8,$acc)=("%eax","%ebx","%edi","%edx","%r8d");
1506$code.=<<___;
1507	mov	$i($ptr),$tp1
1508	mov	$tp1,$acc
1509	and	\$0x80808080,$acc
1510	mov	$acc,$tp4
1511	shr	\$7,$tp4
1512	lea	0($tp1,$tp1),$tp2
1513	sub	$tp4,$acc
1514	and	\$0xfefefefe,$tp2
1515	and	\$0x1b1b1b1b,$acc
1516	xor	$tp2,$acc
1517	mov	$acc,$tp2
1518
1519	and	\$0x80808080,$acc
1520	mov	$acc,$tp8
1521	shr	\$7,$tp8
1522	lea	0($tp2,$tp2),$tp4
1523	sub	$tp8,$acc
1524	and	\$0xfefefefe,$tp4
1525	and	\$0x1b1b1b1b,$acc
1526	 xor	$tp1,$tp2		# tp2^tp1
1527	xor	$tp4,$acc
1528	mov	$acc,$tp4
1529
1530	and	\$0x80808080,$acc
1531	mov	$acc,$tp8
1532	shr	\$7,$tp8
1533	sub	$tp8,$acc
1534	lea	0($tp4,$tp4),$tp8
1535	 xor	$tp1,$tp4		# tp4^tp1
1536	and	\$0xfefefefe,$tp8
1537	and	\$0x1b1b1b1b,$acc
1538	xor	$acc,$tp8
1539
1540	xor	$tp8,$tp1		# tp1^tp8
1541	rol	\$8,$tp1		# ROTATE(tp1^tp8,8)
1542	xor	$tp8,$tp2		# tp2^tp1^tp8
1543	xor	$tp8,$tp4		# tp4^tp1^tp8
1544	xor	$tp2,$tp8
1545	xor	$tp4,$tp8		# tp8^(tp8^tp4^tp1)^(tp8^tp2^tp1)=tp8^tp4^tp2
1546
1547	xor	$tp8,$tp1
1548	rol	\$24,$tp2		# ROTATE(tp2^tp1^tp8,24)
1549	xor	$tp2,$tp1
1550	rol	\$16,$tp4		# ROTATE(tp4^tp1^tp8,16)
1551	xor	$tp4,$tp1
1552
1553	mov	$tp1,$i($ptr)
1554___
1555}
1556
1557# int private_AES_set_decrypt_key(const unsigned char *userKey, const int bits,
1558#                        AES_KEY *key)
1559$code.=<<___;
1560.globl	private_AES_set_decrypt_key
1561.type	private_AES_set_decrypt_key,\@function,3
1562.align	16
1563private_AES_set_decrypt_key:
1564	push	%rbx
1565	push	%rbp
1566	push	%r12
1567	push	%r13
1568	push	%r14
1569	push	%r15
1570	push	%rdx			# save key schedule
1571.Ldec_key_prologue:
1572
1573	call	_x86_64_AES_set_encrypt_key
1574	mov	(%rsp),%r8		# restore key schedule
1575	cmp	\$0,%eax
1576	jne	.Labort
1577
1578	mov	240(%r8),%r14d		# pull number of rounds
1579	xor	%rdi,%rdi
1580	lea	(%rdi,%r14d,4),%rcx
1581	mov	%r8,%rsi
1582	lea	(%r8,%rcx,4),%rdi	# pointer to last chunk
1583.align	4
1584.Linvert:
1585		mov	0(%rsi),%rax
1586		mov	8(%rsi),%rbx
1587		mov	0(%rdi),%rcx
1588		mov	8(%rdi),%rdx
1589		mov	%rax,0(%rdi)
1590		mov	%rbx,8(%rdi)
1591		mov	%rcx,0(%rsi)
1592		mov	%rdx,8(%rsi)
1593		lea	16(%rsi),%rsi
1594		lea	-16(%rdi),%rdi
1595		cmp	%rsi,%rdi
1596	jne	.Linvert
1597
1598	lea	.LAES_Te+2048+1024(%rip),%rax	# rcon
1599
1600	mov	40(%rax),$mask80
1601	mov	48(%rax),$maskfe
1602	mov	56(%rax),$mask1b
1603
1604	mov	%r8,$key
1605	sub	\$1,%r14d
1606.align	4
1607.Lpermute:
1608		lea	16($key),$key
1609		mov	0($key),%rax
1610		mov	8($key),%rcx
1611___
1612		&dectransform ();
1613$code.=<<___;
1614		mov	%eax,0($key)
1615		mov	%ebx,4($key)
1616		mov	%ecx,8($key)
1617		mov	%edx,12($key)
1618		sub	\$1,%r14d
1619	jnz	.Lpermute
1620
1621	xor	%rax,%rax
1622.Labort:
1623	mov	8(%rsp),%r15
1624	mov	16(%rsp),%r14
1625	mov	24(%rsp),%r13
1626	mov	32(%rsp),%r12
1627	mov	40(%rsp),%rbp
1628	mov	48(%rsp),%rbx
1629	add	\$56,%rsp
1630.Ldec_key_epilogue:
1631	ret
1632.size	private_AES_set_decrypt_key,.-private_AES_set_decrypt_key
1633___
1634
1635# void AES_cbc_encrypt (const void char *inp, unsigned char *out,
1636#			size_t length, const AES_KEY *key,
1637#			unsigned char *ivp,const int enc);
1638{
1639# stack frame layout
1640# -8(%rsp)		return address
1641my $keyp="0(%rsp)";		# one to pass as $key
1642my $keyend="8(%rsp)";		# &(keyp->rd_key[4*keyp->rounds])
1643my $_rsp="16(%rsp)";		# saved %rsp
1644my $_inp="24(%rsp)";		# copy of 1st parameter, inp
1645my $_out="32(%rsp)";		# copy of 2nd parameter, out
1646my $_len="40(%rsp)";		# copy of 3rd parameter, length
1647my $_key="48(%rsp)";		# copy of 4th parameter, key
1648my $_ivp="56(%rsp)";		# copy of 5th parameter, ivp
1649my $ivec="64(%rsp)";		# ivec[16]
1650my $aes_key="80(%rsp)";		# copy of aes_key
1651my $mark="80+240(%rsp)";	# copy of aes_key->rounds
1652
1653$code.=<<___;
1654.globl	AES_cbc_encrypt
1655.type	AES_cbc_encrypt,\@function,6
1656.align	16
1657.extern	OPENSSL_ia32cap_P
1658.globl	asm_AES_cbc_encrypt
1659.hidden	asm_AES_cbc_encrypt
1660asm_AES_cbc_encrypt:
1661AES_cbc_encrypt:
1662	cmp	\$0,%rdx	# check length
1663	je	.Lcbc_epilogue
1664	pushfq
1665	push	%rbx
1666	push	%rbp
1667	push	%r12
1668	push	%r13
1669	push	%r14
1670	push	%r15
1671.Lcbc_prologue:
1672
1673	cld
1674	mov	%r9d,%r9d	# clear upper half of enc
1675
1676	lea	.LAES_Te(%rip),$sbox
1677	cmp	\$0,%r9
1678	jne	.Lcbc_picked_te
1679	lea	.LAES_Td(%rip),$sbox
1680.Lcbc_picked_te:
1681
1682	mov	OPENSSL_ia32cap_P(%rip),%r10d
1683	cmp	\$$speed_limit,%rdx
1684	jb	.Lcbc_slow_prologue
1685	test	\$15,%rdx
1686	jnz	.Lcbc_slow_prologue
1687	bt	\$28,%r10d
1688	jc	.Lcbc_slow_prologue
1689
1690	# allocate aligned stack frame...
1691	lea	-88-248(%rsp),$key
1692	and	\$-64,$key
1693
1694	# ... and make sure it doesn't alias with AES_T[ed] modulo 4096
1695	mov	$sbox,%r10
1696	lea	2304($sbox),%r11
1697	mov	$key,%r12
1698	and	\$0xFFF,%r10	# s = $sbox&0xfff
1699	and	\$0xFFF,%r11	# e = ($sbox+2048)&0xfff
1700	and	\$0xFFF,%r12	# p = %rsp&0xfff
1701
1702	cmp	%r11,%r12	# if (p=>e) %rsp =- (p-e);
1703	jb	.Lcbc_te_break_out
1704	sub	%r11,%r12
1705	sub	%r12,$key
1706	jmp	.Lcbc_te_ok
1707.Lcbc_te_break_out:		# else %rsp -= (p-s)&0xfff + framesz
1708	sub	%r10,%r12
1709	and	\$0xFFF,%r12
1710	add	\$320,%r12
1711	sub	%r12,$key
1712.align	4
1713.Lcbc_te_ok:
1714
1715	xchg	%rsp,$key
1716	#add	\$8,%rsp	# reserve for return address!
1717	mov	$key,$_rsp	# save %rsp
1718.Lcbc_fast_body:
1719	mov	%rdi,$_inp	# save copy of inp
1720	mov	%rsi,$_out	# save copy of out
1721	mov	%rdx,$_len	# save copy of len
1722	mov	%rcx,$_key	# save copy of key
1723	mov	%r8,$_ivp	# save copy of ivp
1724	movl	\$0,$mark	# copy of aes_key->rounds = 0;
1725	mov	%r8,%rbp	# rearrange input arguments
1726	mov	%r9,%rbx
1727	mov	%rsi,$out
1728	mov	%rdi,$inp
1729	mov	%rcx,$key
1730
1731	mov	240($key),%eax		# key->rounds
1732	# do we copy key schedule to stack?
1733	mov	$key,%r10
1734	sub	$sbox,%r10
1735	and	\$0xfff,%r10
1736	cmp	\$2304,%r10
1737	jb	.Lcbc_do_ecopy
1738	cmp	\$4096-248,%r10
1739	jb	.Lcbc_skip_ecopy
1740.align	4
1741.Lcbc_do_ecopy:
1742		mov	$key,%rsi
1743		lea	$aes_key,%rdi
1744		lea	$aes_key,$key
1745		mov	\$240/8,%ecx
1746		.long	0x90A548F3	# rep movsq
1747		mov	%eax,(%rdi)	# copy aes_key->rounds
1748.Lcbc_skip_ecopy:
1749	mov	$key,$keyp	# save key pointer
1750
1751	mov	\$18,%ecx
1752.align	4
1753.Lcbc_prefetch_te:
1754		mov	0($sbox),%r10
1755		mov	32($sbox),%r11
1756		mov	64($sbox),%r12
1757		mov	96($sbox),%r13
1758		lea	128($sbox),$sbox
1759		sub	\$1,%ecx
1760	jnz	.Lcbc_prefetch_te
1761	lea	-2304($sbox),$sbox
1762
1763	cmp	\$0,%rbx
1764	je	.LFAST_DECRYPT
1765
1766#----------------------------- ENCRYPT -----------------------------#
1767	mov	0(%rbp),$s0		# load iv
1768	mov	4(%rbp),$s1
1769	mov	8(%rbp),$s2
1770	mov	12(%rbp),$s3
1771
1772.align	4
1773.Lcbc_fast_enc_loop:
1774		xor	0($inp),$s0
1775		xor	4($inp),$s1
1776		xor	8($inp),$s2
1777		xor	12($inp),$s3
1778		mov	$keyp,$key	# restore key
1779		mov	$inp,$_inp	# if ($verticalspin) save inp
1780
1781		call	_x86_64_AES_encrypt
1782
1783		mov	$_inp,$inp	# if ($verticalspin) restore inp
1784		mov	$_len,%r10
1785		mov	$s0,0($out)
1786		mov	$s1,4($out)
1787		mov	$s2,8($out)
1788		mov	$s3,12($out)
1789
1790		lea	16($inp),$inp
1791		lea	16($out),$out
1792		sub	\$16,%r10
1793		test	\$-16,%r10
1794		mov	%r10,$_len
1795	jnz	.Lcbc_fast_enc_loop
1796	mov	$_ivp,%rbp	# restore ivp
1797	mov	$s0,0(%rbp)	# save ivec
1798	mov	$s1,4(%rbp)
1799	mov	$s2,8(%rbp)
1800	mov	$s3,12(%rbp)
1801
1802	jmp	.Lcbc_fast_cleanup
1803
1804#----------------------------- DECRYPT -----------------------------#
1805.align	16
1806.LFAST_DECRYPT:
1807	cmp	$inp,$out
1808	je	.Lcbc_fast_dec_in_place
1809
1810	mov	%rbp,$ivec
1811.align	4
1812.Lcbc_fast_dec_loop:
1813		mov	0($inp),$s0	# read input
1814		mov	4($inp),$s1
1815		mov	8($inp),$s2
1816		mov	12($inp),$s3
1817		mov	$keyp,$key	# restore key
1818		mov	$inp,$_inp	# if ($verticalspin) save inp
1819
1820		call	_x86_64_AES_decrypt
1821
1822		mov	$ivec,%rbp	# load ivp
1823		mov	$_inp,$inp	# if ($verticalspin) restore inp
1824		mov	$_len,%r10	# load len
1825		xor	0(%rbp),$s0	# xor iv
1826		xor	4(%rbp),$s1
1827		xor	8(%rbp),$s2
1828		xor	12(%rbp),$s3
1829		mov	$inp,%rbp	# current input, next iv
1830
1831		sub	\$16,%r10
1832		mov	%r10,$_len	# update len
1833		mov	%rbp,$ivec	# update ivp
1834
1835		mov	$s0,0($out)	# write output
1836		mov	$s1,4($out)
1837		mov	$s2,8($out)
1838		mov	$s3,12($out)
1839
1840		lea	16($inp),$inp
1841		lea	16($out),$out
1842	jnz	.Lcbc_fast_dec_loop
1843	mov	$_ivp,%r12		# load user ivp
1844	mov	0(%rbp),%r10		# load iv
1845	mov	8(%rbp),%r11
1846	mov	%r10,0(%r12)		# copy back to user
1847	mov	%r11,8(%r12)
1848	jmp	.Lcbc_fast_cleanup
1849
1850.align	16
1851.Lcbc_fast_dec_in_place:
1852	mov	0(%rbp),%r10		# copy iv to stack
1853	mov	8(%rbp),%r11
1854	mov	%r10,0+$ivec
1855	mov	%r11,8+$ivec
1856.align	4
1857.Lcbc_fast_dec_in_place_loop:
1858		mov	0($inp),$s0	# load input
1859		mov	4($inp),$s1
1860		mov	8($inp),$s2
1861		mov	12($inp),$s3
1862		mov	$keyp,$key	# restore key
1863		mov	$inp,$_inp	# if ($verticalspin) save inp
1864
1865		call	_x86_64_AES_decrypt
1866
1867		mov	$_inp,$inp	# if ($verticalspin) restore inp
1868		mov	$_len,%r10
1869		xor	0+$ivec,$s0
1870		xor	4+$ivec,$s1
1871		xor	8+$ivec,$s2
1872		xor	12+$ivec,$s3
1873
1874		mov	0($inp),%r11	# load input
1875		mov	8($inp),%r12
1876		sub	\$16,%r10
1877		jz	.Lcbc_fast_dec_in_place_done
1878
1879		mov	%r11,0+$ivec	# copy input to iv
1880		mov	%r12,8+$ivec
1881
1882		mov	$s0,0($out)	# save output [zaps input]
1883		mov	$s1,4($out)
1884		mov	$s2,8($out)
1885		mov	$s3,12($out)
1886
1887		lea	16($inp),$inp
1888		lea	16($out),$out
1889		mov	%r10,$_len
1890	jmp	.Lcbc_fast_dec_in_place_loop
1891.Lcbc_fast_dec_in_place_done:
1892	mov	$_ivp,%rdi
1893	mov	%r11,0(%rdi)	# copy iv back to user
1894	mov	%r12,8(%rdi)
1895
1896	mov	$s0,0($out)	# save output [zaps input]
1897	mov	$s1,4($out)
1898	mov	$s2,8($out)
1899	mov	$s3,12($out)
1900
1901.align	4
1902.Lcbc_fast_cleanup:
1903	cmpl	\$0,$mark	# was the key schedule copied?
1904	lea	$aes_key,%rdi
1905	je	.Lcbc_exit
1906		mov	\$240/8,%ecx
1907		xor	%rax,%rax
1908		.long	0x90AB48F3	# rep stosq
1909
1910	jmp	.Lcbc_exit
1911
1912#--------------------------- SLOW ROUTINE ---------------------------#
1913.align	16
1914.Lcbc_slow_prologue:
1915	# allocate aligned stack frame...
1916	lea	-88(%rsp),%rbp
1917	and	\$-64,%rbp
1918	# ... just "above" key schedule
1919	lea	-88-63(%rcx),%r10
1920	sub	%rbp,%r10
1921	neg	%r10
1922	and	\$0x3c0,%r10
1923	sub	%r10,%rbp
1924
1925	xchg	%rsp,%rbp
1926	#add	\$8,%rsp	# reserve for return address!
1927	mov	%rbp,$_rsp	# save %rsp
1928.Lcbc_slow_body:
1929	#mov	%rdi,$_inp	# save copy of inp
1930	#mov	%rsi,$_out	# save copy of out
1931	#mov	%rdx,$_len	# save copy of len
1932	#mov	%rcx,$_key	# save copy of key
1933	mov	%r8,$_ivp	# save copy of ivp
1934	mov	%r8,%rbp	# rearrange input arguments
1935	mov	%r9,%rbx
1936	mov	%rsi,$out
1937	mov	%rdi,$inp
1938	mov	%rcx,$key
1939	mov	%rdx,%r10
1940
1941	mov	240($key),%eax
1942	mov	$key,$keyp	# save key pointer
1943	shl	\$4,%eax
1944	lea	($key,%rax),%rax
1945	mov	%rax,$keyend
1946
1947	# pick Te4 copy which can't "overlap" with stack frame or key scdedule
1948	lea	2048($sbox),$sbox
1949	lea	768-8(%rsp),%rax
1950	sub	$sbox,%rax
1951	and	\$0x300,%rax
1952	lea	($sbox,%rax),$sbox
1953
1954	cmp	\$0,%rbx
1955	je	.LSLOW_DECRYPT
1956
1957#--------------------------- SLOW ENCRYPT ---------------------------#
1958	test	\$-16,%r10		# check upon length
1959	mov	0(%rbp),$s0		# load iv
1960	mov	4(%rbp),$s1
1961	mov	8(%rbp),$s2
1962	mov	12(%rbp),$s3
1963	jz	.Lcbc_slow_enc_tail	# short input...
1964
1965.align	4
1966.Lcbc_slow_enc_loop:
1967		xor	0($inp),$s0
1968		xor	4($inp),$s1
1969		xor	8($inp),$s2
1970		xor	12($inp),$s3
1971		mov	$keyp,$key	# restore key
1972		mov	$inp,$_inp	# save inp
1973		mov	$out,$_out	# save out
1974		mov	%r10,$_len	# save len
1975
1976		call	_x86_64_AES_encrypt_compact
1977
1978		mov	$_inp,$inp	# restore inp
1979		mov	$_out,$out	# restore out
1980		mov	$_len,%r10	# restore len
1981		mov	$s0,0($out)
1982		mov	$s1,4($out)
1983		mov	$s2,8($out)
1984		mov	$s3,12($out)
1985
1986		lea	16($inp),$inp
1987		lea	16($out),$out
1988		sub	\$16,%r10
1989		test	\$-16,%r10
1990	jnz	.Lcbc_slow_enc_loop
1991	test	\$15,%r10
1992	jnz	.Lcbc_slow_enc_tail
1993	mov	$_ivp,%rbp	# restore ivp
1994	mov	$s0,0(%rbp)	# save ivec
1995	mov	$s1,4(%rbp)
1996	mov	$s2,8(%rbp)
1997	mov	$s3,12(%rbp)
1998
1999	jmp	.Lcbc_exit
2000
2001.align	4
2002.Lcbc_slow_enc_tail:
2003	mov	%rax,%r11
2004	mov	%rcx,%r12
2005	mov	%r10,%rcx
2006	mov	$inp,%rsi
2007	mov	$out,%rdi
2008	.long	0x9066A4F3		# rep movsb
2009	mov	\$16,%rcx		# zero tail
2010	sub	%r10,%rcx
2011	xor	%rax,%rax
2012	.long	0x9066AAF3		# rep stosb
2013	mov	$out,$inp		# this is not a mistake!
2014	mov	\$16,%r10		# len=16
2015	mov	%r11,%rax
2016	mov	%r12,%rcx
2017	jmp	.Lcbc_slow_enc_loop	# one more spin...
2018#--------------------------- SLOW DECRYPT ---------------------------#
2019.align	16
2020.LSLOW_DECRYPT:
2021	shr	\$3,%rax
2022	add	%rax,$sbox		# recall "magic" constants!
2023
2024	mov	0(%rbp),%r11		# copy iv to stack
2025	mov	8(%rbp),%r12
2026	mov	%r11,0+$ivec
2027	mov	%r12,8+$ivec
2028
2029.align	4
2030.Lcbc_slow_dec_loop:
2031		mov	0($inp),$s0	# load input
2032		mov	4($inp),$s1
2033		mov	8($inp),$s2
2034		mov	12($inp),$s3
2035		mov	$keyp,$key	# restore key
2036		mov	$inp,$_inp	# save inp
2037		mov	$out,$_out	# save out
2038		mov	%r10,$_len	# save len
2039
2040		call	_x86_64_AES_decrypt_compact
2041
2042		mov	$_inp,$inp	# restore inp
2043		mov	$_out,$out	# restore out
2044		mov	$_len,%r10
2045		xor	0+$ivec,$s0
2046		xor	4+$ivec,$s1
2047		xor	8+$ivec,$s2
2048		xor	12+$ivec,$s3
2049
2050		mov	0($inp),%r11	# load input
2051		mov	8($inp),%r12
2052		sub	\$16,%r10
2053		jc	.Lcbc_slow_dec_partial
2054		jz	.Lcbc_slow_dec_done
2055
2056		mov	%r11,0+$ivec	# copy input to iv
2057		mov	%r12,8+$ivec
2058
2059		mov	$s0,0($out)	# save output [can zap input]
2060		mov	$s1,4($out)
2061		mov	$s2,8($out)
2062		mov	$s3,12($out)
2063
2064		lea	16($inp),$inp
2065		lea	16($out),$out
2066	jmp	.Lcbc_slow_dec_loop
2067.Lcbc_slow_dec_done:
2068	mov	$_ivp,%rdi
2069	mov	%r11,0(%rdi)		# copy iv back to user
2070	mov	%r12,8(%rdi)
2071
2072	mov	$s0,0($out)		# save output [can zap input]
2073	mov	$s1,4($out)
2074	mov	$s2,8($out)
2075	mov	$s3,12($out)
2076
2077	jmp	.Lcbc_exit
2078
2079.align	4
2080.Lcbc_slow_dec_partial:
2081	mov	$_ivp,%rdi
2082	mov	%r11,0(%rdi)		# copy iv back to user
2083	mov	%r12,8(%rdi)
2084
2085	mov	$s0,0+$ivec		# save output to stack
2086	mov	$s1,4+$ivec
2087	mov	$s2,8+$ivec
2088	mov	$s3,12+$ivec
2089
2090	mov	$out,%rdi
2091	lea	$ivec,%rsi
2092	lea	16(%r10),%rcx
2093	.long	0x9066A4F3	# rep movsb
2094	jmp	.Lcbc_exit
2095
2096.align	16
2097.Lcbc_exit:
2098	mov	$_rsp,%rsi
2099	mov	(%rsi),%r15
2100	mov	8(%rsi),%r14
2101	mov	16(%rsi),%r13
2102	mov	24(%rsi),%r12
2103	mov	32(%rsi),%rbp
2104	mov	40(%rsi),%rbx
2105	lea	48(%rsi),%rsp
2106.Lcbc_popfq:
2107	popfq
2108.Lcbc_epilogue:
2109	ret
2110.size	AES_cbc_encrypt,.-AES_cbc_encrypt
2111___
2112}
2113
2114$code.=<<___;
2115.align	64
2116.LAES_Te:
2117___
2118	&_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6);
2119	&_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591);
2120	&_data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56);
2121	&_data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec);
2122	&_data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa);
2123	&_data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb);
2124	&_data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45);
2125	&_data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b);
2126	&_data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c);
2127	&_data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83);
2128	&_data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9);
2129	&_data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a);
2130	&_data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d);
2131	&_data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f);
2132	&_data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df);
2133	&_data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea);
2134	&_data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34);
2135	&_data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b);
2136	&_data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d);
2137	&_data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413);
2138	&_data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1);
2139	&_data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6);
2140	&_data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972);
2141	&_data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85);
2142	&_data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed);
2143	&_data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511);
2144	&_data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe);
2145	&_data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b);
2146	&_data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05);
2147	&_data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1);
2148	&_data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142);
2149	&_data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf);
2150	&_data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3);
2151	&_data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e);
2152	&_data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a);
2153	&_data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6);
2154	&_data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3);
2155	&_data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b);
2156	&_data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428);
2157	&_data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad);
2158	&_data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14);
2159	&_data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8);
2160	&_data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4);
2161	&_data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2);
2162	&_data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda);
2163	&_data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949);
2164	&_data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf);
2165	&_data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810);
2166	&_data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c);
2167	&_data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697);
2168	&_data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e);
2169	&_data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f);
2170	&_data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc);
2171	&_data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c);
2172	&_data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969);
2173	&_data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27);
2174	&_data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122);
2175	&_data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433);
2176	&_data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9);
2177	&_data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5);
2178	&_data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a);
2179	&_data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0);
2180	&_data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e);
2181	&_data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c);
2182
2183#Te4	# four copies of Te4 to choose from to avoid L1 aliasing
2184	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2185	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2186	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2187	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2188	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2189	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2190	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2191	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2192	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2193	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2194	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2195	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2196	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2197	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2198	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2199	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2200	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2201	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2202	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2203	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2204	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2205	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2206	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2207	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2208	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2209	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2210	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2211	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2212	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2213	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2214	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2215	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2216
2217	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2218	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2219	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2220	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2221	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2222	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2223	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2224	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2225	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2226	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2227	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2228	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2229	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2230	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2231	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2232	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2233	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2234	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2235	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2236	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2237	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2238	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2239	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2240	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2241	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2242	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2243	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2244	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2245	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2246	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2247	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2248	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2249
2250	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2251	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2252	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2253	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2254	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2255	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2256	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2257	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2258	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2259	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2260	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2261	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2262	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2263	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2264	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2265	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2266	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2267	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2268	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2269	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2270	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2271	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2272	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2273	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2274	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2275	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2276	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2277	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2278	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2279	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2280	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2281	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2282
2283	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2284	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2285	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2286	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2287	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2288	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2289	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2290	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2291	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2292	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2293	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2294	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2295	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2296	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2297	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2298	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2299	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2300	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2301	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2302	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2303	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2304	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2305	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2306	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2307	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2308	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2309	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2310	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2311	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2312	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2313	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2314	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2315#rcon:
2316$code.=<<___;
2317	.long	0x00000001, 0x00000002, 0x00000004, 0x00000008
2318	.long	0x00000010, 0x00000020, 0x00000040, 0x00000080
2319	.long	0x0000001b, 0x00000036, 0x80808080, 0x80808080
2320	.long	0xfefefefe, 0xfefefefe, 0x1b1b1b1b, 0x1b1b1b1b
2321___
2322$code.=<<___;
2323.align	64
2324.LAES_Td:
2325___
2326	&_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a);
2327	&_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b);
2328	&_data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5);
2329	&_data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5);
2330	&_data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d);
2331	&_data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b);
2332	&_data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295);
2333	&_data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e);
2334	&_data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927);
2335	&_data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d);
2336	&_data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362);
2337	&_data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9);
2338	&_data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52);
2339	&_data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566);
2340	&_data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3);
2341	&_data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed);
2342	&_data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e);
2343	&_data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4);
2344	&_data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4);
2345	&_data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd);
2346	&_data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d);
2347	&_data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060);
2348	&_data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967);
2349	&_data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879);
2350	&_data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000);
2351	&_data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c);
2352	&_data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36);
2353	&_data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624);
2354	&_data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b);
2355	&_data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c);
2356	&_data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12);
2357	&_data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14);
2358	&_data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3);
2359	&_data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b);
2360	&_data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8);
2361	&_data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684);
2362	&_data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7);
2363	&_data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177);
2364	&_data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947);
2365	&_data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322);
2366	&_data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498);
2367	&_data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f);
2368	&_data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54);
2369	&_data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382);
2370	&_data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf);
2371	&_data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb);
2372	&_data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83);
2373	&_data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef);
2374	&_data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029);
2375	&_data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235);
2376	&_data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733);
2377	&_data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117);
2378	&_data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4);
2379	&_data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546);
2380	&_data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb);
2381	&_data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d);
2382	&_data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb);
2383	&_data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a);
2384	&_data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773);
2385	&_data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478);
2386	&_data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2);
2387	&_data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff);
2388	&_data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664);
2389	&_data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0);
2390
2391#Td4:	# four copies of Td4 to choose from to avoid L1 aliasing
2392	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2393	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2394	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2395	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2396	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2397	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2398	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2399	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2400	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2401	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2402	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2403	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2404	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2405	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2406	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2407	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2408	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2409	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2410	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2411	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2412	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2413	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2414	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2415	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2416	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2417	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2418	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2419	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2420	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2421	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2422	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2423	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2424$code.=<<___;
2425	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2426	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2427___
2428	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2429	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2430	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2431	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2432	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2433	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2434	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2435	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2436	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2437	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2438	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2439	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2440	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2441	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2442	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2443	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2444	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2445	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2446	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2447	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2448	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2449	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2450	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2451	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2452	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2453	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2454	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2455	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2456	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2457	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2458	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2459	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2460$code.=<<___;
2461	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2462	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2463___
2464	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2465	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2466	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2467	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2468	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2469	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2470	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2471	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2472	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2473	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2474	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2475	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2476	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2477	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2478	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2479	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2480	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2481	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2482	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2483	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2484	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2485	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2486	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2487	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2488	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2489	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2490	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2491	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2492	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2493	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2494	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2495	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2496$code.=<<___;
2497	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2498	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2499___
2500	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2501	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2502	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2503	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2504	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2505	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2506	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2507	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2508	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2509	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2510	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2511	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2512	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2513	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2514	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2515	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2516	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2517	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2518	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2519	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2520	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2521	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2522	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2523	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2524	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2525	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2526	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2527	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2528	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2529	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2530	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2531	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2532$code.=<<___;
2533	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2534	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2535.asciz  "AES for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
2536.align	64
2537___
2538
2539# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
2540#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
2541if ($win64) {
2542$rec="%rcx";
2543$frame="%rdx";
2544$context="%r8";
2545$disp="%r9";
2546
2547$code.=<<___;
2548.extern	__imp_RtlVirtualUnwind
2549.type	block_se_handler,\@abi-omnipotent
2550.align	16
2551block_se_handler:
2552	push	%rsi
2553	push	%rdi
2554	push	%rbx
2555	push	%rbp
2556	push	%r12
2557	push	%r13
2558	push	%r14
2559	push	%r15
2560	pushfq
2561	sub	\$64,%rsp
2562
2563	mov	120($context),%rax	# pull context->Rax
2564	mov	248($context),%rbx	# pull context->Rip
2565
2566	mov	8($disp),%rsi		# disp->ImageBase
2567	mov	56($disp),%r11		# disp->HandlerData
2568
2569	mov	0(%r11),%r10d		# HandlerData[0]
2570	lea	(%rsi,%r10),%r10	# prologue label
2571	cmp	%r10,%rbx		# context->Rip<prologue label
2572	jb	.Lin_block_prologue
2573
2574	mov	152($context),%rax	# pull context->Rsp
2575
2576	mov	4(%r11),%r10d		# HandlerData[1]
2577	lea	(%rsi,%r10),%r10	# epilogue label
2578	cmp	%r10,%rbx		# context->Rip>=epilogue label
2579	jae	.Lin_block_prologue
2580
2581	mov	24(%rax),%rax		# pull saved real stack pointer
2582	lea	48(%rax),%rax		# adjust...
2583
2584	mov	-8(%rax),%rbx
2585	mov	-16(%rax),%rbp
2586	mov	-24(%rax),%r12
2587	mov	-32(%rax),%r13
2588	mov	-40(%rax),%r14
2589	mov	-48(%rax),%r15
2590	mov	%rbx,144($context)	# restore context->Rbx
2591	mov	%rbp,160($context)	# restore context->Rbp
2592	mov	%r12,216($context)	# restore context->R12
2593	mov	%r13,224($context)	# restore context->R13
2594	mov	%r14,232($context)	# restore context->R14
2595	mov	%r15,240($context)	# restore context->R15
2596
2597.Lin_block_prologue:
2598	mov	8(%rax),%rdi
2599	mov	16(%rax),%rsi
2600	mov	%rax,152($context)	# restore context->Rsp
2601	mov	%rsi,168($context)	# restore context->Rsi
2602	mov	%rdi,176($context)	# restore context->Rdi
2603
2604	jmp	.Lcommon_seh_exit
2605.size	block_se_handler,.-block_se_handler
2606
2607.type	key_se_handler,\@abi-omnipotent
2608.align	16
2609key_se_handler:
2610	push	%rsi
2611	push	%rdi
2612	push	%rbx
2613	push	%rbp
2614	push	%r12
2615	push	%r13
2616	push	%r14
2617	push	%r15
2618	pushfq
2619	sub	\$64,%rsp
2620
2621	mov	120($context),%rax	# pull context->Rax
2622	mov	248($context),%rbx	# pull context->Rip
2623
2624	mov	8($disp),%rsi		# disp->ImageBase
2625	mov	56($disp),%r11		# disp->HandlerData
2626
2627	mov	0(%r11),%r10d		# HandlerData[0]
2628	lea	(%rsi,%r10),%r10	# prologue label
2629	cmp	%r10,%rbx		# context->Rip<prologue label
2630	jb	.Lin_key_prologue
2631
2632	mov	152($context),%rax	# pull context->Rsp
2633
2634	mov	4(%r11),%r10d		# HandlerData[1]
2635	lea	(%rsi,%r10),%r10	# epilogue label
2636	cmp	%r10,%rbx		# context->Rip>=epilogue label
2637	jae	.Lin_key_prologue
2638
2639	lea	56(%rax),%rax
2640
2641	mov	-8(%rax),%rbx
2642	mov	-16(%rax),%rbp
2643	mov	-24(%rax),%r12
2644	mov	-32(%rax),%r13
2645	mov	-40(%rax),%r14
2646	mov	-48(%rax),%r15
2647	mov	%rbx,144($context)	# restore context->Rbx
2648	mov	%rbp,160($context)	# restore context->Rbp
2649	mov	%r12,216($context)	# restore context->R12
2650	mov	%r13,224($context)	# restore context->R13
2651	mov	%r14,232($context)	# restore context->R14
2652	mov	%r15,240($context)	# restore context->R15
2653
2654.Lin_key_prologue:
2655	mov	8(%rax),%rdi
2656	mov	16(%rax),%rsi
2657	mov	%rax,152($context)	# restore context->Rsp
2658	mov	%rsi,168($context)	# restore context->Rsi
2659	mov	%rdi,176($context)	# restore context->Rdi
2660
2661	jmp	.Lcommon_seh_exit
2662.size	key_se_handler,.-key_se_handler
2663
2664.type	cbc_se_handler,\@abi-omnipotent
2665.align	16
2666cbc_se_handler:
2667	push	%rsi
2668	push	%rdi
2669	push	%rbx
2670	push	%rbp
2671	push	%r12
2672	push	%r13
2673	push	%r14
2674	push	%r15
2675	pushfq
2676	sub	\$64,%rsp
2677
2678	mov	120($context),%rax	# pull context->Rax
2679	mov	248($context),%rbx	# pull context->Rip
2680
2681	lea	.Lcbc_prologue(%rip),%r10
2682	cmp	%r10,%rbx		# context->Rip<.Lcbc_prologue
2683	jb	.Lin_cbc_prologue
2684
2685	lea	.Lcbc_fast_body(%rip),%r10
2686	cmp	%r10,%rbx		# context->Rip<.Lcbc_fast_body
2687	jb	.Lin_cbc_frame_setup
2688
2689	lea	.Lcbc_slow_prologue(%rip),%r10
2690	cmp	%r10,%rbx		# context->Rip<.Lcbc_slow_prologue
2691	jb	.Lin_cbc_body
2692
2693	lea	.Lcbc_slow_body(%rip),%r10
2694	cmp	%r10,%rbx		# context->Rip<.Lcbc_slow_body
2695	jb	.Lin_cbc_frame_setup
2696
2697.Lin_cbc_body:
2698	mov	152($context),%rax	# pull context->Rsp
2699
2700	lea	.Lcbc_epilogue(%rip),%r10
2701	cmp	%r10,%rbx		# context->Rip>=.Lcbc_epilogue
2702	jae	.Lin_cbc_prologue
2703
2704	lea	8(%rax),%rax
2705
2706	lea	.Lcbc_popfq(%rip),%r10
2707	cmp	%r10,%rbx		# context->Rip>=.Lcbc_popfq
2708	jae	.Lin_cbc_prologue
2709
2710	mov	`16-8`(%rax),%rax	# biased $_rsp
2711	lea	56(%rax),%rax
2712
2713.Lin_cbc_frame_setup:
2714	mov	-16(%rax),%rbx
2715	mov	-24(%rax),%rbp
2716	mov	-32(%rax),%r12
2717	mov	-40(%rax),%r13
2718	mov	-48(%rax),%r14
2719	mov	-56(%rax),%r15
2720	mov	%rbx,144($context)	# restore context->Rbx
2721	mov	%rbp,160($context)	# restore context->Rbp
2722	mov	%r12,216($context)	# restore context->R12
2723	mov	%r13,224($context)	# restore context->R13
2724	mov	%r14,232($context)	# restore context->R14
2725	mov	%r15,240($context)	# restore context->R15
2726
2727.Lin_cbc_prologue:
2728	mov	8(%rax),%rdi
2729	mov	16(%rax),%rsi
2730	mov	%rax,152($context)	# restore context->Rsp
2731	mov	%rsi,168($context)	# restore context->Rsi
2732	mov	%rdi,176($context)	# restore context->Rdi
2733
2734.Lcommon_seh_exit:
2735
2736	mov	40($disp),%rdi		# disp->ContextRecord
2737	mov	$context,%rsi		# context
2738	mov	\$`1232/8`,%ecx		# sizeof(CONTEXT)
2739	.long	0xa548f3fc		# cld; rep movsq
2740
2741	mov	$disp,%rsi
2742	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
2743	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
2744	mov	0(%rsi),%r8		# arg3, disp->ControlPc
2745	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
2746	mov	40(%rsi),%r10		# disp->ContextRecord
2747	lea	56(%rsi),%r11		# &disp->HandlerData
2748	lea	24(%rsi),%r12		# &disp->EstablisherFrame
2749	mov	%r10,32(%rsp)		# arg5
2750	mov	%r11,40(%rsp)		# arg6
2751	mov	%r12,48(%rsp)		# arg7
2752	mov	%rcx,56(%rsp)		# arg8, (NULL)
2753	call	*__imp_RtlVirtualUnwind(%rip)
2754
2755	mov	\$1,%eax		# ExceptionContinueSearch
2756	add	\$64,%rsp
2757	popfq
2758	pop	%r15
2759	pop	%r14
2760	pop	%r13
2761	pop	%r12
2762	pop	%rbp
2763	pop	%rbx
2764	pop	%rdi
2765	pop	%rsi
2766	ret
2767.size	cbc_se_handler,.-cbc_se_handler
2768
2769.section	.pdata
2770.align	4
2771	.rva	.LSEH_begin_AES_encrypt
2772	.rva	.LSEH_end_AES_encrypt
2773	.rva	.LSEH_info_AES_encrypt
2774
2775	.rva	.LSEH_begin_AES_decrypt
2776	.rva	.LSEH_end_AES_decrypt
2777	.rva	.LSEH_info_AES_decrypt
2778
2779	.rva	.LSEH_begin_private_AES_set_encrypt_key
2780	.rva	.LSEH_end_private_AES_set_encrypt_key
2781	.rva	.LSEH_info_private_AES_set_encrypt_key
2782
2783	.rva	.LSEH_begin_private_AES_set_decrypt_key
2784	.rva	.LSEH_end_private_AES_set_decrypt_key
2785	.rva	.LSEH_info_private_AES_set_decrypt_key
2786
2787	.rva	.LSEH_begin_AES_cbc_encrypt
2788	.rva	.LSEH_end_AES_cbc_encrypt
2789	.rva	.LSEH_info_AES_cbc_encrypt
2790
2791.section	.xdata
2792.align	8
2793.LSEH_info_AES_encrypt:
2794	.byte	9,0,0,0
2795	.rva	block_se_handler
2796	.rva	.Lenc_prologue,.Lenc_epilogue	# HandlerData[]
2797.LSEH_info_AES_decrypt:
2798	.byte	9,0,0,0
2799	.rva	block_se_handler
2800	.rva	.Ldec_prologue,.Ldec_epilogue	# HandlerData[]
2801.LSEH_info_private_AES_set_encrypt_key:
2802	.byte	9,0,0,0
2803	.rva	key_se_handler
2804	.rva	.Lenc_key_prologue,.Lenc_key_epilogue	# HandlerData[]
2805.LSEH_info_private_AES_set_decrypt_key:
2806	.byte	9,0,0,0
2807	.rva	key_se_handler
2808	.rva	.Ldec_key_prologue,.Ldec_key_epilogue	# HandlerData[]
2809.LSEH_info_AES_cbc_encrypt:
2810	.byte	9,0,0,0
2811	.rva	cbc_se_handler
2812___
2813}
2814
2815$code =~ s/\`([^\`]*)\`/eval($1)/gem;
2816
2817print $code;
2818
2819close STDOUT;
2820