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