xref: /openbsd-src/lib/libcrypto/aes/asm/aes-x86_64.pl (revision 4a615b8755ec141a62a4a26da5bf86944339541e) 
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	_CET_ENDBR
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	retq
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	($sbox,$t0,1),$t0
371	movzb	($sbox,$t1,1),$t1
372	movzb	($sbox,$t2,1),$t2
373
374	movzb	`&lo("$s3")`,$t3
375	movzb	`&hi("$s1")`,$acc0
376	movzb	`&hi("$s2")`,$acc1
377	movzb	($sbox,$t3,1),$t3
378	movzb	($sbox,$acc0,1),$t4	#$t0
379	movzb	($sbox,$acc1,1),$t5	#$t1
380
381	movzb	`&hi("$s3")`,$acc2
382	movzb	`&hi("$s0")`,$acc0
383	shr	\$16,$s2
384	movzb	($sbox,$acc2,1),$acc2	#$t2
385	movzb	($sbox,$acc0,1),$acc0	#$t3
386	shr	\$16,$s3
387
388	movzb	`&lo("$s2")`,$acc1
389	shl	\$8,$t4
390	shl	\$8,$t5
391	movzb	($sbox,$acc1,1),$acc1	#$t0
392	xor	$t4,$t0
393	xor	$t5,$t1
394
395	movzb	`&lo("$s3")`,$t4
396	shr	\$16,$s0
397	shr	\$16,$s1
398	movzb	`&lo("$s0")`,$t5
399	shl	\$8,$acc2
400	shl	\$8,$acc0
401	movzb	($sbox,$t4,1),$t4	#$t1
402	movzb	($sbox,$t5,1),$t5	#$t2
403	xor	$acc2,$t2
404	xor	$acc0,$t3
405
406	movzb	`&lo("$s1")`,$acc2
407	movzb	`&hi("$s3")`,$acc0
408	shl	\$16,$acc1
409	movzb	($sbox,$acc2,1),$acc2	#$t3
410	movzb	($sbox,$acc0,1),$acc0	#$t0
411	xor	$acc1,$t0
412
413	movzb	`&hi("$s0")`,$acc1
414	shr	\$8,$s2
415	shr	\$8,$s1
416	movzb	($sbox,$acc1,1),$acc1	#$t1
417	movzb	($sbox,$s2,1),$s3	#$t3
418	movzb	($sbox,$s1,1),$s2	#$t2
419	shl	\$16,$t4
420	shl	\$16,$t5
421	shl	\$16,$acc2
422	xor	$t4,$t1
423	xor	$t5,$t2
424	xor	$acc2,$t3
425
426	shl	\$24,$acc0
427	shl	\$24,$acc1
428	shl	\$24,$s3
429	xor	$acc0,$t0
430	shl	\$24,$s2
431	xor	$acc1,$t1
432	mov	$t0,$s0
433	mov	$t1,$s1
434	xor	$t2,$s2
435	xor	$t3,$s3
436___
437}
438
439sub enctransform_ref()
440{ my $sn = shift;
441  my ($acc,$r2,$tmp)=("%r8d","%r9d","%r13d");
442
443$code.=<<___;
444	mov	$sn,$acc
445	and	\$0x80808080,$acc
446	mov	$acc,$tmp
447	shr	\$7,$tmp
448	lea	($sn,$sn),$r2
449	sub	$tmp,$acc
450	and	\$0xfefefefe,$r2
451	and	\$0x1b1b1b1b,$acc
452	mov	$sn,$tmp
453	xor	$acc,$r2
454
455	xor	$r2,$sn
456	rol	\$24,$sn
457	xor	$r2,$sn
458	ror	\$16,$tmp
459	xor	$tmp,$sn
460	ror	\$8,$tmp
461	xor	$tmp,$sn
462___
463}
464
465# unlike decrypt case it does not pay off to parallelize enctransform
466sub enctransform()
467{ my ($t3,$r20,$r21)=($acc2,"%r8d","%r9d");
468
469$code.=<<___;
470	mov	$s0,$acc0
471	mov	$s1,$acc1
472	and	\$0x80808080,$acc0
473	and	\$0x80808080,$acc1
474	mov	$acc0,$t0
475	mov	$acc1,$t1
476	shr	\$7,$t0
477	lea	($s0,$s0),$r20
478	shr	\$7,$t1
479	lea	($s1,$s1),$r21
480	sub	$t0,$acc0
481	sub	$t1,$acc1
482	and	\$0xfefefefe,$r20
483	and	\$0xfefefefe,$r21
484	and	\$0x1b1b1b1b,$acc0
485	and	\$0x1b1b1b1b,$acc1
486	mov	$s0,$t0
487	mov	$s1,$t1
488	xor	$acc0,$r20
489	xor	$acc1,$r21
490
491	xor	$r20,$s0
492	xor	$r21,$s1
493	 mov	$s2,$acc0
494	 mov	$s3,$acc1
495	rol	\$24,$s0
496	rol	\$24,$s1
497	 and	\$0x80808080,$acc0
498	 and	\$0x80808080,$acc1
499	xor	$r20,$s0
500	xor	$r21,$s1
501	 mov	$acc0,$t2
502	 mov	$acc1,$t3
503	ror	\$16,$t0
504	ror	\$16,$t1
505	 shr	\$7,$t2
506	 lea	($s2,$s2),$r20
507	xor	$t0,$s0
508	xor	$t1,$s1
509	 shr	\$7,$t3
510	 lea	($s3,$s3),$r21
511	ror	\$8,$t0
512	ror	\$8,$t1
513	 sub	$t2,$acc0
514	 sub	$t3,$acc1
515	xor	$t0,$s0
516	xor	$t1,$s1
517
518	and	\$0xfefefefe,$r20
519	and	\$0xfefefefe,$r21
520	and	\$0x1b1b1b1b,$acc0
521	and	\$0x1b1b1b1b,$acc1
522	mov	$s2,$t2
523	mov	$s3,$t3
524	xor	$acc0,$r20
525	xor	$acc1,$r21
526
527	xor	$r20,$s2
528	xor	$r21,$s3
529	rol	\$24,$s2
530	rol	\$24,$s3
531	xor	$r20,$s2
532	xor	$r21,$s3
533	mov	0($sbox),$acc0			# prefetch Te4
534	ror	\$16,$t2
535	ror	\$16,$t3
536	mov	64($sbox),$acc1
537	xor	$t2,$s2
538	xor	$t3,$s3
539	mov	128($sbox),$r20
540	ror	\$8,$t2
541	ror	\$8,$t3
542	mov	192($sbox),$r21
543	xor	$t2,$s2
544	xor	$t3,$s3
545___
546}
547
548$code.=<<___;
549.type	_x86_64_AES_encrypt_compact,\@abi-omnipotent
550.align	16
551_x86_64_AES_encrypt_compact:
552	_CET_ENDBR
553	lea	128($sbox),$inp			# size optimization
554	mov	0-128($inp),$acc1		# prefetch Te4
555	mov	32-128($inp),$acc2
556	mov	64-128($inp),$t0
557	mov	96-128($inp),$t1
558	mov	128-128($inp),$acc1
559	mov	160-128($inp),$acc2
560	mov	192-128($inp),$t0
561	mov	224-128($inp),$t1
562	jmp	.Lenc_loop_compact
563.align	16
564.Lenc_loop_compact:
565		xor	0($key),$s0		# xor with key
566		xor	4($key),$s1
567		xor	8($key),$s2
568		xor	12($key),$s3
569		lea	16($key),$key
570___
571		&enccompactvert();
572$code.=<<___;
573		cmp	16(%rsp),$key
574		je	.Lenc_compact_done
575___
576		&enctransform();
577$code.=<<___;
578	jmp	.Lenc_loop_compact
579.align	16
580.Lenc_compact_done:
581	xor	0($key),$s0
582	xor	4($key),$s1
583	xor	8($key),$s2
584	xor	12($key),$s3
585	retq
586.size	_x86_64_AES_encrypt_compact,.-_x86_64_AES_encrypt_compact
587___
588
589# void aes_encrypt_internal(const void *inp, void *out, const AES_KEY *key);
590$code.=<<___;
591.globl	aes_encrypt_internal
592.type	aes_encrypt_internal,\@function,3
593.align	16
594.globl	asm_AES_encrypt
595.hidden	asm_AES_encrypt
596asm_AES_encrypt:
597aes_encrypt_internal:
598	_CET_ENDBR
599	push	%rbx
600	push	%rbp
601	push	%r12
602	push	%r13
603	push	%r14
604	push	%r15
605
606	# allocate frame "above" key schedule
607	mov	%rsp,%r10
608	lea	-63(%rdx),%rcx	# %rdx is key argument
609	and	\$-64,%rsp
610	sub	%rsp,%rcx
611	neg	%rcx
612	and	\$0x3c0,%rcx
613	sub	%rcx,%rsp
614	sub	\$32,%rsp
615
616	mov	%rsi,16(%rsp)	# save out
617	mov	%r10,24(%rsp)	# save real stack pointer
618.Lenc_prologue:
619
620	mov	%rdx,$key
621	mov	240($key),$rnds	# load rounds
622
623	mov	0(%rdi),$s0	# load input vector
624	mov	4(%rdi),$s1
625	mov	8(%rdi),$s2
626	mov	12(%rdi),$s3
627
628	shl	\$4,$rnds
629	lea	($key,$rnds),%rbp
630	mov	$key,(%rsp)	# key schedule
631	mov	%rbp,8(%rsp)	# end of key schedule
632
633	# pick Te4 copy which can't "overlap" with stack frame or key schedule
634	lea	.LAES_Te+2048(%rip),$sbox
635	lea	768(%rsp),%rbp
636	sub	$sbox,%rbp
637	and	\$0x300,%rbp
638	lea	($sbox,%rbp),$sbox
639
640	call	_x86_64_AES_encrypt_compact
641
642	mov	16(%rsp),$out	# restore out
643	mov	24(%rsp),%rsi	# restore saved stack pointer
644	mov	$s0,0($out)	# write output vector
645	mov	$s1,4($out)
646	mov	$s2,8($out)
647	mov	$s3,12($out)
648
649	mov	(%rsi),%r15
650	mov	8(%rsi),%r14
651	mov	16(%rsi),%r13
652	mov	24(%rsi),%r12
653	mov	32(%rsi),%rbp
654	mov	40(%rsi),%rbx
655	lea	48(%rsi),%rsp
656.Lenc_epilogue:
657	ret
658.size	aes_encrypt_internal,.-aes_encrypt_internal
659___
660
661#------------------------------------------------------------------#
662
663sub decvert()
664{ my $t3="%r8d";	# zaps $inp!
665
666$code.=<<___;
667	# favor 3-way issue Opteron pipeline...
668	movzb	`&lo("$s0")`,$acc0
669	movzb	`&lo("$s1")`,$acc1
670	movzb	`&lo("$s2")`,$acc2
671	mov	0($sbox,$acc0,8),$t0
672	mov	0($sbox,$acc1,8),$t1
673	mov	0($sbox,$acc2,8),$t2
674
675	movzb	`&hi("$s3")`,$acc0
676	movzb	`&hi("$s0")`,$acc1
677	movzb	`&lo("$s3")`,$acc2
678	xor	3($sbox,$acc0,8),$t0
679	xor	3($sbox,$acc1,8),$t1
680	mov	0($sbox,$acc2,8),$t3
681
682	movzb	`&hi("$s1")`,$acc0
683	shr	\$16,$s0
684	movzb	`&hi("$s2")`,$acc2
685	xor	3($sbox,$acc0,8),$t2
686	shr	\$16,$s3
687	xor	3($sbox,$acc2,8),$t3
688
689	shr	\$16,$s1
690	lea	16($key),$key
691	shr	\$16,$s2
692
693	movzb	`&lo("$s2")`,$acc0
694	movzb	`&lo("$s3")`,$acc1
695	movzb	`&lo("$s0")`,$acc2
696	xor	2($sbox,$acc0,8),$t0
697	xor	2($sbox,$acc1,8),$t1
698	xor	2($sbox,$acc2,8),$t2
699
700	movzb	`&hi("$s1")`,$acc0
701	movzb	`&hi("$s2")`,$acc1
702	movzb	`&lo("$s1")`,$acc2
703	xor	1($sbox,$acc0,8),$t0
704	xor	1($sbox,$acc1,8),$t1
705	xor	2($sbox,$acc2,8),$t3
706
707	movzb	`&hi("$s3")`,$acc0
708	mov	12($key),$s3
709	movzb	`&hi("$s0")`,$acc2
710	xor	1($sbox,$acc0,8),$t2
711	mov	0($key),$s0
712	xor	1($sbox,$acc2,8),$t3
713
714	xor	$t0,$s0
715	mov	4($key),$s1
716	mov	8($key),$s2
717	xor	$t2,$s2
718	xor	$t1,$s1
719	xor	$t3,$s3
720___
721}
722
723sub declastvert()
724{ my $t3="%r8d";	# zaps $inp!
725
726$code.=<<___;
727	lea	2048($sbox),$sbox	# size optimization
728	movzb	`&lo("$s0")`,$acc0
729	movzb	`&lo("$s1")`,$acc1
730	movzb	`&lo("$s2")`,$acc2
731	movzb	($sbox,$acc0,1),$t0
732	movzb	($sbox,$acc1,1),$t1
733	movzb	($sbox,$acc2,1),$t2
734
735	movzb	`&lo("$s3")`,$acc0
736	movzb	`&hi("$s3")`,$acc1
737	movzb	`&hi("$s0")`,$acc2
738	movzb	($sbox,$acc0,1),$t3
739	movzb	($sbox,$acc1,1),$acc1	#$t0
740	movzb	($sbox,$acc2,1),$acc2	#$t1
741
742	shl	\$8,$acc1
743	shl	\$8,$acc2
744
745	xor	$acc1,$t0
746	xor	$acc2,$t1
747	shr	\$16,$s3
748
749	movzb	`&hi("$s1")`,$acc0
750	movzb	`&hi("$s2")`,$acc1
751	shr	\$16,$s0
752	movzb	($sbox,$acc0,1),$acc0	#$t2
753	movzb	($sbox,$acc1,1),$acc1	#$t3
754
755	shl	\$8,$acc0
756	shl	\$8,$acc1
757	shr	\$16,$s1
758	xor	$acc0,$t2
759	xor	$acc1,$t3
760	shr	\$16,$s2
761
762	movzb	`&lo("$s2")`,$acc0
763	movzb	`&lo("$s3")`,$acc1
764	movzb	`&lo("$s0")`,$acc2
765	movzb	($sbox,$acc0,1),$acc0	#$t0
766	movzb	($sbox,$acc1,1),$acc1	#$t1
767	movzb	($sbox,$acc2,1),$acc2	#$t2
768
769	shl	\$16,$acc0
770	shl	\$16,$acc1
771	shl	\$16,$acc2
772
773	xor	$acc0,$t0
774	xor	$acc1,$t1
775	xor	$acc2,$t2
776
777	movzb	`&lo("$s1")`,$acc0
778	movzb	`&hi("$s1")`,$acc1
779	movzb	`&hi("$s2")`,$acc2
780	movzb	($sbox,$acc0,1),$acc0	#$t3
781	movzb	($sbox,$acc1,1),$acc1	#$t0
782	movzb	($sbox,$acc2,1),$acc2	#$t1
783
784	shl	\$16,$acc0
785	shl	\$24,$acc1
786	shl	\$24,$acc2
787
788	xor	$acc0,$t3
789	xor	$acc1,$t0
790	xor	$acc2,$t1
791
792	movzb	`&hi("$s3")`,$acc0
793	movzb	`&hi("$s0")`,$acc1
794	mov	16+12($key),$s3
795	movzb	($sbox,$acc0,1),$acc0	#$t2
796	movzb	($sbox,$acc1,1),$acc1	#$t3
797	mov	16+0($key),$s0
798
799	shl	\$24,$acc0
800	shl	\$24,$acc1
801
802	xor	$acc0,$t2
803	xor	$acc1,$t3
804
805	mov	16+4($key),$s1
806	mov	16+8($key),$s2
807	lea	-2048($sbox),$sbox
808	xor	$t0,$s0
809	xor	$t1,$s1
810	xor	$t2,$s2
811	xor	$t3,$s3
812___
813}
814
815sub decstep()
816{ my ($i,@s) = @_;
817  my $tmp0=$acc0;
818  my $tmp1=$acc1;
819  my $tmp2=$acc2;
820  my $out=($t0,$t1,$t2,$s[0])[$i];
821
822	$code.="	mov	$s[0],$out\n"		if ($i!=3);
823			$tmp1=$s[2]			if ($i==3);
824	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
825	$code.="	and	\$0xFF,$out\n";
826
827	$code.="	mov	0($sbox,$out,8),$out\n";
828	$code.="	shr	\$16,$tmp1\n";
829			$tmp2=$s[3]			if ($i==3);
830	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
831
832			$tmp0=$s[1]			if ($i==3);
833	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
834	$code.="	and	\$0xFF,$tmp1\n";
835	$code.="	shr	\$24,$tmp2\n";
836
837	$code.="	xor	3($sbox,$tmp0,8),$out\n";
838	$code.="	xor	2($sbox,$tmp1,8),$out\n";
839	$code.="	xor	1($sbox,$tmp2,8),$out\n";
840
841	$code.="	mov	$t2,$s[1]\n"		if ($i==3);
842	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
843	$code.="	mov	$t0,$s[3]\n"		if ($i==3);
844	$code.="\n";
845}
846
847sub declast()
848{ my ($i,@s)=@_;
849  my $tmp0=$acc0;
850  my $tmp1=$acc1;
851  my $tmp2=$acc2;
852  my $out=($t0,$t1,$t2,$s[0])[$i];
853
854	$code.="	mov	$s[0],$out\n"		if ($i!=3);
855			$tmp1=$s[2]			if ($i==3);
856	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
857	$code.="	and	\$0xFF,$out\n";
858
859	$code.="	movzb	2048($sbox,$out,1),$out\n";
860	$code.="	shr	\$16,$tmp1\n";
861			$tmp2=$s[3]			if ($i==3);
862	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
863
864			$tmp0=$s[1]			if ($i==3);
865	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
866	$code.="	and	\$0xFF,$tmp1\n";
867	$code.="	shr	\$24,$tmp2\n";
868
869	$code.="	movzb	2048($sbox,$tmp0,1),$tmp0\n";
870	$code.="	movzb	2048($sbox,$tmp1,1),$tmp1\n";
871	$code.="	movzb	2048($sbox,$tmp2,1),$tmp2\n";
872
873	$code.="	shl	\$8,$tmp0\n";
874	$code.="	shl	\$16,$tmp1\n";
875	$code.="	shl	\$24,$tmp2\n";
876
877	$code.="	xor	$tmp0,$out\n";
878	$code.="	mov	$t2,$s[1]\n"		if ($i==3);
879	$code.="	xor	$tmp1,$out\n";
880	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
881	$code.="	xor	$tmp2,$out\n";
882	$code.="	mov	$t0,$s[3]\n"		if ($i==3);
883	$code.="\n";
884}
885
886$code.=<<___;
887.type	_x86_64_AES_decrypt,\@abi-omnipotent
888.align	16
889_x86_64_AES_decrypt:
890	_CET_ENDBR
891	xor	0($key),$s0			# xor with key
892	xor	4($key),$s1
893	xor	8($key),$s2
894	xor	12($key),$s3
895
896	mov	240($key),$rnds			# load key->rounds
897	sub	\$1,$rnds
898	jmp	.Ldec_loop
899.align	16
900.Ldec_loop:
901___
902	if ($verticalspin) { &decvert(); }
903	else {	&decstep(0,$s0,$s3,$s2,$s1);
904		&decstep(1,$s1,$s0,$s3,$s2);
905		&decstep(2,$s2,$s1,$s0,$s3);
906		&decstep(3,$s3,$s2,$s1,$s0);
907		$code.=<<___;
908		lea	16($key),$key
909		xor	0($key),$s0			# xor with key
910		xor	4($key),$s1
911		xor	8($key),$s2
912		xor	12($key),$s3
913___
914	}
915$code.=<<___;
916	sub	\$1,$rnds
917	jnz	.Ldec_loop
918___
919	if ($verticalspin) { &declastvert(); }
920	else {	&declast(0,$s0,$s3,$s2,$s1);
921		&declast(1,$s1,$s0,$s3,$s2);
922		&declast(2,$s2,$s1,$s0,$s3);
923		&declast(3,$s3,$s2,$s1,$s0);
924		$code.=<<___;
925		xor	16+0($key),$s0			# xor with key
926		xor	16+4($key),$s1
927		xor	16+8($key),$s2
928		xor	16+12($key),$s3
929___
930	}
931$code.=<<___;
932	retq
933.size	_x86_64_AES_decrypt,.-_x86_64_AES_decrypt
934___
935
936sub deccompactvert()
937{ my ($t3,$t4,$t5)=("%r8d","%r9d","%r13d");
938
939$code.=<<___;
940	movzb	`&lo("$s0")`,$t0
941	movzb	`&lo("$s1")`,$t1
942	movzb	`&lo("$s2")`,$t2
943	movzb	($sbox,$t0,1),$t0
944	movzb	($sbox,$t1,1),$t1
945	movzb	($sbox,$t2,1),$t2
946
947	movzb	`&lo("$s3")`,$t3
948	movzb	`&hi("$s3")`,$acc0
949	movzb	`&hi("$s0")`,$acc1
950	movzb	($sbox,$t3,1),$t3
951	movzb	($sbox,$acc0,1),$t4	#$t0
952	movzb	($sbox,$acc1,1),$t5	#$t1
953
954	movzb	`&hi("$s1")`,$acc2
955	movzb	`&hi("$s2")`,$acc0
956	shr	\$16,$s2
957	movzb	($sbox,$acc2,1),$acc2	#$t2
958	movzb	($sbox,$acc0,1),$acc0	#$t3
959	shr	\$16,$s3
960
961	movzb	`&lo("$s2")`,$acc1
962	shl	\$8,$t4
963	shl	\$8,$t5
964	movzb	($sbox,$acc1,1),$acc1	#$t0
965	xor	$t4,$t0
966	xor	$t5,$t1
967
968	movzb	`&lo("$s3")`,$t4
969	shr	\$16,$s0
970	shr	\$16,$s1
971	movzb	`&lo("$s0")`,$t5
972	shl	\$8,$acc2
973	shl	\$8,$acc0
974	movzb	($sbox,$t4,1),$t4	#$t1
975	movzb	($sbox,$t5,1),$t5	#$t2
976	xor	$acc2,$t2
977	xor	$acc0,$t3
978
979	movzb	`&lo("$s1")`,$acc2
980	movzb	`&hi("$s1")`,$acc0
981	shl	\$16,$acc1
982	movzb	($sbox,$acc2,1),$acc2	#$t3
983	movzb	($sbox,$acc0,1),$acc0	#$t0
984	xor	$acc1,$t0
985
986	movzb	`&hi("$s2")`,$acc1
987	shl	\$16,$t4
988	shl	\$16,$t5
989	movzb	($sbox,$acc1,1),$s1	#$t1
990	xor	$t4,$t1
991	xor	$t5,$t2
992
993	movzb	`&hi("$s3")`,$acc1
994	shr	\$8,$s0
995	shl	\$16,$acc2
996	movzb	($sbox,$acc1,1),$s2	#$t2
997	movzb	($sbox,$s0,1),$s3	#$t3
998	xor	$acc2,$t3
999
1000	shl	\$24,$acc0
1001	shl	\$24,$s1
1002	shl	\$24,$s2
1003	xor	$acc0,$t0
1004	shl	\$24,$s3
1005	xor	$t1,$s1
1006	mov	$t0,$s0
1007	xor	$t2,$s2
1008	xor	$t3,$s3
1009___
1010}
1011
1012# parallelized version! input is pair of 64-bit values: %rax=s1.s0
1013# and %rcx=s3.s2, output is four 32-bit values in %eax=s0, %ebx=s1,
1014# %ecx=s2 and %edx=s3.
1015sub dectransform()
1016{ my ($tp10,$tp20,$tp40,$tp80,$acc0)=("%rax","%r8", "%r9", "%r10","%rbx");
1017  my ($tp18,$tp28,$tp48,$tp88,$acc8)=("%rcx","%r11","%r12","%r13","%rdx");
1018  my $prefetch = shift;
1019
1020$code.=<<___;
1021	mov	$tp10,$acc0
1022	mov	$tp18,$acc8
1023	and	$mask80,$acc0
1024	and	$mask80,$acc8
1025	mov	$acc0,$tp40
1026	mov	$acc8,$tp48
1027	shr	\$7,$tp40
1028	lea	($tp10,$tp10),$tp20
1029	shr	\$7,$tp48
1030	lea	($tp18,$tp18),$tp28
1031	sub	$tp40,$acc0
1032	sub	$tp48,$acc8
1033	and	$maskfe,$tp20
1034	and	$maskfe,$tp28
1035	and	$mask1b,$acc0
1036	and	$mask1b,$acc8
1037	xor	$tp20,$acc0
1038	xor	$tp28,$acc8
1039	mov	$acc0,$tp20
1040	mov	$acc8,$tp28
1041
1042	and	$mask80,$acc0
1043	and	$mask80,$acc8
1044	mov	$acc0,$tp80
1045	mov	$acc8,$tp88
1046	shr	\$7,$tp80
1047	lea	($tp20,$tp20),$tp40
1048	shr	\$7,$tp88
1049	lea	($tp28,$tp28),$tp48
1050	sub	$tp80,$acc0
1051	sub	$tp88,$acc8
1052	and	$maskfe,$tp40
1053	and	$maskfe,$tp48
1054	and	$mask1b,$acc0
1055	and	$mask1b,$acc8
1056	xor	$tp40,$acc0
1057	xor	$tp48,$acc8
1058	mov	$acc0,$tp40
1059	mov	$acc8,$tp48
1060
1061	and	$mask80,$acc0
1062	and	$mask80,$acc8
1063	mov	$acc0,$tp80
1064	mov	$acc8,$tp88
1065	shr	\$7,$tp80
1066	 xor	$tp10,$tp20		# tp2^=tp1
1067	shr	\$7,$tp88
1068	 xor	$tp18,$tp28		# tp2^=tp1
1069	sub	$tp80,$acc0
1070	sub	$tp88,$acc8
1071	lea	($tp40,$tp40),$tp80
1072	lea	($tp48,$tp48),$tp88
1073	 xor	$tp10,$tp40		# tp4^=tp1
1074	 xor	$tp18,$tp48		# tp4^=tp1
1075	and	$maskfe,$tp80
1076	and	$maskfe,$tp88
1077	and	$mask1b,$acc0
1078	and	$mask1b,$acc8
1079	xor	$acc0,$tp80
1080	xor	$acc8,$tp88
1081
1082	xor	$tp80,$tp10		# tp1^=tp8
1083	xor	$tp88,$tp18		# tp1^=tp8
1084	xor	$tp80,$tp20		# tp2^tp1^=tp8
1085	xor	$tp88,$tp28		# tp2^tp1^=tp8
1086	mov	$tp10,$acc0
1087	mov	$tp18,$acc8
1088	xor	$tp80,$tp40		# tp4^tp1^=tp8
1089	xor	$tp88,$tp48		# tp4^tp1^=tp8
1090	shr	\$32,$acc0
1091	shr	\$32,$acc8
1092	xor	$tp20,$tp80		# tp8^=tp8^tp2^tp1=tp2^tp1
1093	xor	$tp28,$tp88		# tp8^=tp8^tp2^tp1=tp2^tp1
1094	rol	\$8,`&LO("$tp10")`	# ROTATE(tp1^tp8,8)
1095	rol	\$8,`&LO("$tp18")`	# ROTATE(tp1^tp8,8)
1096	xor	$tp40,$tp80		# tp2^tp1^=tp8^tp4^tp1=tp8^tp4^tp2
1097	xor	$tp48,$tp88		# tp2^tp1^=tp8^tp4^tp1=tp8^tp4^tp2
1098
1099	rol	\$8,`&LO("$acc0")`	# ROTATE(tp1^tp8,8)
1100	rol	\$8,`&LO("$acc8")`	# ROTATE(tp1^tp8,8)
1101	xor	`&LO("$tp80")`,`&LO("$tp10")`
1102	xor	`&LO("$tp88")`,`&LO("$tp18")`
1103	shr	\$32,$tp80
1104	shr	\$32,$tp88
1105	xor	`&LO("$tp80")`,`&LO("$acc0")`
1106	xor	`&LO("$tp88")`,`&LO("$acc8")`
1107
1108	mov	$tp20,$tp80
1109	mov	$tp28,$tp88
1110	shr	\$32,$tp80
1111	shr	\$32,$tp88
1112	rol	\$24,`&LO("$tp20")`	# ROTATE(tp2^tp1^tp8,24)
1113	rol	\$24,`&LO("$tp28")`	# ROTATE(tp2^tp1^tp8,24)
1114	rol	\$24,`&LO("$tp80")`	# ROTATE(tp2^tp1^tp8,24)
1115	rol	\$24,`&LO("$tp88")`	# ROTATE(tp2^tp1^tp8,24)
1116	xor	`&LO("$tp20")`,`&LO("$tp10")`
1117	xor	`&LO("$tp28")`,`&LO("$tp18")`
1118	mov	$tp40,$tp20
1119	mov	$tp48,$tp28
1120	xor	`&LO("$tp80")`,`&LO("$acc0")`
1121	xor	`&LO("$tp88")`,`&LO("$acc8")`
1122
1123	`"mov	0($sbox),$mask80"	if ($prefetch)`
1124	shr	\$32,$tp20
1125	shr	\$32,$tp28
1126	`"mov	64($sbox),$maskfe"	if ($prefetch)`
1127	rol	\$16,`&LO("$tp40")`	# ROTATE(tp4^tp1^tp8,16)
1128	rol	\$16,`&LO("$tp48")`	# ROTATE(tp4^tp1^tp8,16)
1129	`"mov	128($sbox),$mask1b"	if ($prefetch)`
1130	rol	\$16,`&LO("$tp20")`	# ROTATE(tp4^tp1^tp8,16)
1131	rol	\$16,`&LO("$tp28")`	# ROTATE(tp4^tp1^tp8,16)
1132	`"mov	192($sbox),$tp80"	if ($prefetch)`
1133	xor	`&LO("$tp40")`,`&LO("$tp10")`
1134	xor	`&LO("$tp48")`,`&LO("$tp18")`
1135	`"mov	256($sbox),$tp88"	if ($prefetch)`
1136	xor	`&LO("$tp20")`,`&LO("$acc0")`
1137	xor	`&LO("$tp28")`,`&LO("$acc8")`
1138___
1139}
1140
1141$code.=<<___;
1142.type	_x86_64_AES_decrypt_compact,\@abi-omnipotent
1143.align	16
1144_x86_64_AES_decrypt_compact:
1145	_CET_ENDBR
1146	lea	128($sbox),$inp			# size optimization
1147	mov	0-128($inp),$acc1		# prefetch Td4
1148	mov	32-128($inp),$acc2
1149	mov	64-128($inp),$t0
1150	mov	96-128($inp),$t1
1151	mov	128-128($inp),$acc1
1152	mov	160-128($inp),$acc2
1153	mov	192-128($inp),$t0
1154	mov	224-128($inp),$t1
1155	jmp	.Ldec_loop_compact
1156
1157.align	16
1158.Ldec_loop_compact:
1159		xor	0($key),$s0		# xor with key
1160		xor	4($key),$s1
1161		xor	8($key),$s2
1162		xor	12($key),$s3
1163		lea	16($key),$key
1164___
1165		&deccompactvert();
1166$code.=<<___;
1167		cmp	16(%rsp),$key
1168		je	.Ldec_compact_done
1169
1170		mov	256+0($sbox),$mask80
1171		shl	\$32,%rbx
1172		shl	\$32,%rdx
1173		mov	256+8($sbox),$maskfe
1174		or	%rbx,%rax
1175		or	%rdx,%rcx
1176		mov	256+16($sbox),$mask1b
1177___
1178		&dectransform(1);
1179$code.=<<___;
1180	jmp	.Ldec_loop_compact
1181.align	16
1182.Ldec_compact_done:
1183	xor	0($key),$s0
1184	xor	4($key),$s1
1185	xor	8($key),$s2
1186	xor	12($key),$s3
1187	retq
1188.size	_x86_64_AES_decrypt_compact,.-_x86_64_AES_decrypt_compact
1189___
1190
1191# void aes_decrypt_internal(const void *inp, void *out, const AES_KEY *key);
1192$code.=<<___;
1193.globl	aes_decrypt_internal
1194.type	aes_decrypt_internal,\@function,3
1195.align	16
1196.globl	asm_AES_decrypt
1197.hidden	asm_AES_decrypt
1198asm_AES_decrypt:
1199aes_decrypt_internal:
1200	_CET_ENDBR
1201	push	%rbx
1202	push	%rbp
1203	push	%r12
1204	push	%r13
1205	push	%r14
1206	push	%r15
1207
1208	# allocate frame "above" key schedule
1209	mov	%rsp,%r10
1210	lea	-63(%rdx),%rcx	# %rdx is key argument
1211	and	\$-64,%rsp
1212	sub	%rsp,%rcx
1213	neg	%rcx
1214	and	\$0x3c0,%rcx
1215	sub	%rcx,%rsp
1216	sub	\$32,%rsp
1217
1218	mov	%rsi,16(%rsp)	# save out
1219	mov	%r10,24(%rsp)	# save real stack pointer
1220.Ldec_prologue:
1221
1222	mov	%rdx,$key
1223	mov	240($key),$rnds	# load rounds
1224
1225	mov	0(%rdi),$s0	# load input vector
1226	mov	4(%rdi),$s1
1227	mov	8(%rdi),$s2
1228	mov	12(%rdi),$s3
1229
1230	shl	\$4,$rnds
1231	lea	($key,$rnds),%rbp
1232	mov	$key,(%rsp)	# key schedule
1233	mov	%rbp,8(%rsp)	# end of key schedule
1234
1235	# pick Td4 copy which can't "overlap" with stack frame or key schedule
1236	lea	.LAES_Td+2048(%rip),$sbox
1237	lea	768(%rsp),%rbp
1238	sub	$sbox,%rbp
1239	and	\$0x300,%rbp
1240	lea	($sbox,%rbp),$sbox
1241	shr	\$3,%rbp	# recall "magic" constants!
1242	add	%rbp,$sbox
1243
1244	call	_x86_64_AES_decrypt_compact
1245
1246	mov	16(%rsp),$out	# restore out
1247	mov	24(%rsp),%rsi	# restore saved stack pointer
1248	mov	$s0,0($out)	# write output vector
1249	mov	$s1,4($out)
1250	mov	$s2,8($out)
1251	mov	$s3,12($out)
1252
1253	mov	(%rsi),%r15
1254	mov	8(%rsi),%r14
1255	mov	16(%rsi),%r13
1256	mov	24(%rsi),%r12
1257	mov	32(%rsi),%rbp
1258	mov	40(%rsi),%rbx
1259	lea	48(%rsi),%rsp
1260.Ldec_epilogue:
1261	ret
1262.size	aes_decrypt_internal,.-aes_decrypt_internal
1263___
1264#------------------------------------------------------------------#
1265
1266sub enckey()
1267{
1268$code.=<<___;
1269	movz	%dl,%esi		# rk[i]>>0
1270	movzb	-128(%rbp,%rsi),%ebx
1271	movz	%dh,%esi		# rk[i]>>8
1272	shl	\$24,%ebx
1273	xor	%ebx,%eax
1274
1275	movzb	-128(%rbp,%rsi),%ebx
1276	shr	\$16,%edx
1277	movz	%dl,%esi		# rk[i]>>16
1278	xor	%ebx,%eax
1279
1280	movzb	-128(%rbp,%rsi),%ebx
1281	movz	%dh,%esi		# rk[i]>>24
1282	shl	\$8,%ebx
1283	xor	%ebx,%eax
1284
1285	movzb	-128(%rbp,%rsi),%ebx
1286	shl	\$16,%ebx
1287	xor	%ebx,%eax
1288
1289	xor	1024-128(%rbp,%rcx,4),%eax		# rcon
1290___
1291}
1292
1293# int aes_set_encrypt_key_internal(const unsigned char *userKey, const int bits,
1294#     AES_KEY *key)
1295$code.=<<___;
1296.globl	aes_set_encrypt_key_internal
1297.type	aes_set_encrypt_key_internal,\@function,3
1298.align	16
1299aes_set_encrypt_key_internal:
1300	_CET_ENDBR
1301	push	%rbx
1302	push	%rbp
1303	push	%r12			# redundant, but allows to share
1304	push	%r13			# exception handler...
1305	push	%r14
1306	push	%r15
1307	sub	\$8,%rsp
1308.Lenc_key_prologue:
1309
1310	call	_x86_64_AES_set_encrypt_key
1311
1312	mov	8(%rsp),%r15
1313	mov	16(%rsp),%r14
1314	mov	24(%rsp),%r13
1315	mov	32(%rsp),%r12
1316	mov	40(%rsp),%rbp
1317	mov	48(%rsp),%rbx
1318	add	\$56,%rsp
1319.Lenc_key_epilogue:
1320	ret
1321.size	aes_set_encrypt_key_internal,.-aes_set_encrypt_key_internal
1322
1323.type	_x86_64_AES_set_encrypt_key,\@abi-omnipotent
1324.align	16
1325_x86_64_AES_set_encrypt_key:
1326	_CET_ENDBR
1327	mov	%esi,%ecx			# %ecx=bits
1328	mov	%rdi,%rsi			# %rsi=userKey
1329	mov	%rdx,%rdi			# %rdi=key
1330
1331	test	\$-1,%rsi
1332	jz	.Lbadpointer
1333	test	\$-1,%rdi
1334	jz	.Lbadpointer
1335
1336	lea	.LAES_Te(%rip),%rbp
1337	lea	2048+128(%rbp),%rbp
1338
1339	# prefetch Te4
1340	mov	0-128(%rbp),%eax
1341	mov	32-128(%rbp),%ebx
1342	mov	64-128(%rbp),%r8d
1343	mov	96-128(%rbp),%edx
1344	mov	128-128(%rbp),%eax
1345	mov	160-128(%rbp),%ebx
1346	mov	192-128(%rbp),%r8d
1347	mov	224-128(%rbp),%edx
1348
1349	cmp	\$128,%ecx
1350	je	.L10rounds
1351	cmp	\$192,%ecx
1352	je	.L12rounds
1353	cmp	\$256,%ecx
1354	je	.L14rounds
1355	mov	\$-2,%rax			# invalid number of bits
1356	jmp	.Lexit
1357
1358.L10rounds:
1359	mov	0(%rsi),%rax			# copy first 4 dwords
1360	mov	8(%rsi),%rdx
1361	mov	%rax,0(%rdi)
1362	mov	%rdx,8(%rdi)
1363
1364	shr	\$32,%rdx
1365	xor	%ecx,%ecx
1366	jmp	.L10shortcut
1367.align	4
1368.L10loop:
1369		mov	0(%rdi),%eax			# rk[0]
1370		mov	12(%rdi),%edx			# rk[3]
1371.L10shortcut:
1372___
1373		&enckey	();
1374$code.=<<___;
1375		mov	%eax,16(%rdi)			# rk[4]
1376		xor	4(%rdi),%eax
1377		mov	%eax,20(%rdi)			# rk[5]
1378		xor	8(%rdi),%eax
1379		mov	%eax,24(%rdi)			# rk[6]
1380		xor	12(%rdi),%eax
1381		mov	%eax,28(%rdi)			# rk[7]
1382		add	\$1,%ecx
1383		lea	16(%rdi),%rdi
1384		cmp	\$10,%ecx
1385	jl	.L10loop
1386
1387	movl	\$10,80(%rdi)			# setup number of rounds
1388	xor	%rax,%rax
1389	jmp	.Lexit
1390
1391.L12rounds:
1392	mov	0(%rsi),%rax			# copy first 6 dwords
1393	mov	8(%rsi),%rbx
1394	mov	16(%rsi),%rdx
1395	mov	%rax,0(%rdi)
1396	mov	%rbx,8(%rdi)
1397	mov	%rdx,16(%rdi)
1398
1399	shr	\$32,%rdx
1400	xor	%ecx,%ecx
1401	jmp	.L12shortcut
1402.align	4
1403.L12loop:
1404		mov	0(%rdi),%eax			# rk[0]
1405		mov	20(%rdi),%edx			# rk[5]
1406.L12shortcut:
1407___
1408		&enckey	();
1409$code.=<<___;
1410		mov	%eax,24(%rdi)			# rk[6]
1411		xor	4(%rdi),%eax
1412		mov	%eax,28(%rdi)			# rk[7]
1413		xor	8(%rdi),%eax
1414		mov	%eax,32(%rdi)			# rk[8]
1415		xor	12(%rdi),%eax
1416		mov	%eax,36(%rdi)			# rk[9]
1417
1418		cmp	\$7,%ecx
1419		je	.L12break
1420		add	\$1,%ecx
1421
1422		xor	16(%rdi),%eax
1423		mov	%eax,40(%rdi)			# rk[10]
1424		xor	20(%rdi),%eax
1425		mov	%eax,44(%rdi)			# rk[11]
1426
1427		lea	24(%rdi),%rdi
1428	jmp	.L12loop
1429.L12break:
1430	movl	\$12,72(%rdi)		# setup number of rounds
1431	xor	%rax,%rax
1432	jmp	.Lexit
1433
1434.L14rounds:
1435	mov	0(%rsi),%rax			# copy first 8 dwords
1436	mov	8(%rsi),%rbx
1437	mov	16(%rsi),%rcx
1438	mov	24(%rsi),%rdx
1439	mov	%rax,0(%rdi)
1440	mov	%rbx,8(%rdi)
1441	mov	%rcx,16(%rdi)
1442	mov	%rdx,24(%rdi)
1443
1444	shr	\$32,%rdx
1445	xor	%ecx,%ecx
1446	jmp	.L14shortcut
1447.align	4
1448.L14loop:
1449		mov	0(%rdi),%eax			# rk[0]
1450		mov	28(%rdi),%edx			# rk[4]
1451.L14shortcut:
1452___
1453		&enckey	();
1454$code.=<<___;
1455		mov	%eax,32(%rdi)			# rk[8]
1456		xor	4(%rdi),%eax
1457		mov	%eax,36(%rdi)			# rk[9]
1458		xor	8(%rdi),%eax
1459		mov	%eax,40(%rdi)			# rk[10]
1460		xor	12(%rdi),%eax
1461		mov	%eax,44(%rdi)			# rk[11]
1462
1463		cmp	\$6,%ecx
1464		je	.L14break
1465		add	\$1,%ecx
1466
1467		mov	%eax,%edx
1468		mov	16(%rdi),%eax			# rk[4]
1469		movz	%dl,%esi			# rk[11]>>0
1470		movzb	-128(%rbp,%rsi),%ebx
1471		movz	%dh,%esi			# rk[11]>>8
1472		xor	%ebx,%eax
1473
1474		movzb	-128(%rbp,%rsi),%ebx
1475		shr	\$16,%edx
1476		shl	\$8,%ebx
1477		movz	%dl,%esi			# rk[11]>>16
1478		xor	%ebx,%eax
1479
1480		movzb	-128(%rbp,%rsi),%ebx
1481		movz	%dh,%esi			# rk[11]>>24
1482		shl	\$16,%ebx
1483		xor	%ebx,%eax
1484
1485		movzb	-128(%rbp,%rsi),%ebx
1486		shl	\$24,%ebx
1487		xor	%ebx,%eax
1488
1489		mov	%eax,48(%rdi)			# rk[12]
1490		xor	20(%rdi),%eax
1491		mov	%eax,52(%rdi)			# rk[13]
1492		xor	24(%rdi),%eax
1493		mov	%eax,56(%rdi)			# rk[14]
1494		xor	28(%rdi),%eax
1495		mov	%eax,60(%rdi)			# rk[15]
1496
1497		lea	32(%rdi),%rdi
1498	jmp	.L14loop
1499.L14break:
1500	movl	\$14,48(%rdi)		# setup number of rounds
1501	xor	%rax,%rax
1502	jmp	.Lexit
1503
1504.Lbadpointer:
1505	mov	\$-1,%rax
1506.Lexit:
1507	retq
1508.size	_x86_64_AES_set_encrypt_key,.-_x86_64_AES_set_encrypt_key
1509___
1510
1511sub deckey_ref()
1512{ my ($i,$ptr,$te,$td) = @_;
1513  my ($tp1,$tp2,$tp4,$tp8,$acc)=("%eax","%ebx","%edi","%edx","%r8d");
1514$code.=<<___;
1515	mov	$i($ptr),$tp1
1516	mov	$tp1,$acc
1517	and	\$0x80808080,$acc
1518	mov	$acc,$tp4
1519	shr	\$7,$tp4
1520	lea	0($tp1,$tp1),$tp2
1521	sub	$tp4,$acc
1522	and	\$0xfefefefe,$tp2
1523	and	\$0x1b1b1b1b,$acc
1524	xor	$tp2,$acc
1525	mov	$acc,$tp2
1526
1527	and	\$0x80808080,$acc
1528	mov	$acc,$tp8
1529	shr	\$7,$tp8
1530	lea	0($tp2,$tp2),$tp4
1531	sub	$tp8,$acc
1532	and	\$0xfefefefe,$tp4
1533	and	\$0x1b1b1b1b,$acc
1534	 xor	$tp1,$tp2		# tp2^tp1
1535	xor	$tp4,$acc
1536	mov	$acc,$tp4
1537
1538	and	\$0x80808080,$acc
1539	mov	$acc,$tp8
1540	shr	\$7,$tp8
1541	sub	$tp8,$acc
1542	lea	0($tp4,$tp4),$tp8
1543	 xor	$tp1,$tp4		# tp4^tp1
1544	and	\$0xfefefefe,$tp8
1545	and	\$0x1b1b1b1b,$acc
1546	xor	$acc,$tp8
1547
1548	xor	$tp8,$tp1		# tp1^tp8
1549	rol	\$8,$tp1		# ROTATE(tp1^tp8,8)
1550	xor	$tp8,$tp2		# tp2^tp1^tp8
1551	xor	$tp8,$tp4		# tp4^tp1^tp8
1552	xor	$tp2,$tp8
1553	xor	$tp4,$tp8		# tp8^(tp8^tp4^tp1)^(tp8^tp2^tp1)=tp8^tp4^tp2
1554
1555	xor	$tp8,$tp1
1556	rol	\$24,$tp2		# ROTATE(tp2^tp1^tp8,24)
1557	xor	$tp2,$tp1
1558	rol	\$16,$tp4		# ROTATE(tp4^tp1^tp8,16)
1559	xor	$tp4,$tp1
1560
1561	mov	$tp1,$i($ptr)
1562___
1563}
1564
1565# int aes_set_decrypt_key_internal(const unsigned char *userKey, const int bits,
1566#     AES_KEY *key)
1567$code.=<<___;
1568.globl	aes_set_decrypt_key_internal
1569.type	aes_set_decrypt_key_internal,\@function,3
1570.align	16
1571aes_set_decrypt_key_internal:
1572	_CET_ENDBR
1573	push	%rbx
1574	push	%rbp
1575	push	%r12
1576	push	%r13
1577	push	%r14
1578	push	%r15
1579	push	%rdx			# save key schedule
1580.Ldec_key_prologue:
1581
1582	call	_x86_64_AES_set_encrypt_key
1583	mov	(%rsp),%r8		# restore key schedule
1584	cmp	\$0,%eax
1585	jne	.Labort
1586
1587	mov	240(%r8),%r14d		# pull number of rounds
1588	xor	%rdi,%rdi
1589	lea	(%rdi,%r14d,4),%rcx
1590	mov	%r8,%rsi
1591	lea	(%r8,%rcx,4),%rdi	# pointer to last chunk
1592.align	4
1593.Linvert:
1594		mov	0(%rsi),%rax
1595		mov	8(%rsi),%rbx
1596		mov	0(%rdi),%rcx
1597		mov	8(%rdi),%rdx
1598		mov	%rax,0(%rdi)
1599		mov	%rbx,8(%rdi)
1600		mov	%rcx,0(%rsi)
1601		mov	%rdx,8(%rsi)
1602		lea	16(%rsi),%rsi
1603		lea	-16(%rdi),%rdi
1604		cmp	%rsi,%rdi
1605	jne	.Linvert
1606
1607	lea	.LAES_Te+2048+1024(%rip),%rax	# rcon
1608
1609	mov	40(%rax),$mask80
1610	mov	48(%rax),$maskfe
1611	mov	56(%rax),$mask1b
1612
1613	mov	%r8,$key
1614	sub	\$1,%r14d
1615.align	4
1616.Lpermute:
1617		lea	16($key),$key
1618		mov	0($key),%rax
1619		mov	8($key),%rcx
1620___
1621		&dectransform ();
1622$code.=<<___;
1623		mov	%eax,0($key)
1624		mov	%ebx,4($key)
1625		mov	%ecx,8($key)
1626		mov	%edx,12($key)
1627		sub	\$1,%r14d
1628	jnz	.Lpermute
1629
1630	xor	%rax,%rax
1631.Labort:
1632	mov	8(%rsp),%r15
1633	mov	16(%rsp),%r14
1634	mov	24(%rsp),%r13
1635	mov	32(%rsp),%r12
1636	mov	40(%rsp),%rbp
1637	mov	48(%rsp),%rbx
1638	add	\$56,%rsp
1639.Ldec_key_epilogue:
1640	ret
1641.size	aes_set_decrypt_key_internal,.-aes_set_decrypt_key_internal
1642___
1643
1644# void aes_cbc_encrypt_internal(const void char *inp, unsigned char *out,
1645#     size_t length, const AES_KEY *key, unsigned char *ivp,const int enc);
1646{
1647# stack frame layout
1648# -8(%rsp)		return address
1649my $keyp="0(%rsp)";		# one to pass as $key
1650my $keyend="8(%rsp)";		# &(keyp->rd_key[4*keyp->rounds])
1651my $_rsp="16(%rsp)";		# saved %rsp
1652my $_inp="24(%rsp)";		# copy of 1st parameter, inp
1653my $_out="32(%rsp)";		# copy of 2nd parameter, out
1654my $_len="40(%rsp)";		# copy of 3rd parameter, length
1655my $_key="48(%rsp)";		# copy of 4th parameter, key
1656my $_ivp="56(%rsp)";		# copy of 5th parameter, ivp
1657my $ivec="64(%rsp)";		# ivec[16]
1658my $aes_key="80(%rsp)";		# copy of aes_key
1659my $mark="80+240(%rsp)";	# copy of aes_key->rounds
1660
1661$code.=<<___;
1662.globl	aes_cbc_encrypt_internal
1663.type	aes_cbc_encrypt_internal,\@function,6
1664.align	16
1665.extern	OPENSSL_ia32cap_P
1666.hidden	OPENSSL_ia32cap_P
1667.globl	asm_AES_cbc_encrypt
1668.hidden	asm_AES_cbc_encrypt
1669asm_AES_cbc_encrypt:
1670aes_cbc_encrypt_internal:
1671	_CET_ENDBR
1672	cmp	\$0,%rdx	# check length
1673	je	.Lcbc_epilogue
1674	pushfq
1675	push	%rbx
1676	push	%rbp
1677	push	%r12
1678	push	%r13
1679	push	%r14
1680	push	%r15
1681.Lcbc_prologue:
1682
1683	cld
1684	mov	%r9d,%r9d	# clear upper half of enc
1685
1686	lea	.LAES_Te(%rip),$sbox
1687	cmp	\$0,%r9
1688	jne	.Lcbc_picked_te
1689	lea	.LAES_Td(%rip),$sbox
1690.Lcbc_picked_te:
1691
1692	mov	OPENSSL_ia32cap_P(%rip),%r10d
1693	cmp	\$$speed_limit,%rdx
1694	jb	.Lcbc_slow_prologue
1695	test	\$15,%rdx
1696	jnz	.Lcbc_slow_prologue
1697	bt	\$IA32CAP_BIT0_HT,%r10d
1698	jc	.Lcbc_slow_prologue
1699
1700	# allocate aligned stack frame...
1701	lea	-88-248(%rsp),$key
1702	and	\$-64,$key
1703
1704	# ... and make sure it doesn't alias with AES_T[ed] modulo 4096
1705	mov	$sbox,%r10
1706	lea	2304($sbox),%r11
1707	mov	$key,%r12
1708	and	\$0xFFF,%r10	# s = $sbox&0xfff
1709	and	\$0xFFF,%r11	# e = ($sbox+2048)&0xfff
1710	and	\$0xFFF,%r12	# p = %rsp&0xfff
1711
1712	cmp	%r11,%r12	# if (p=>e) %rsp =- (p-e);
1713	jb	.Lcbc_te_break_out
1714	sub	%r11,%r12
1715	sub	%r12,$key
1716	jmp	.Lcbc_te_ok
1717.Lcbc_te_break_out:		# else %rsp -= (p-s)&0xfff + framesz
1718	sub	%r10,%r12
1719	and	\$0xFFF,%r12
1720	add	\$320,%r12
1721	sub	%r12,$key
1722.align	4
1723.Lcbc_te_ok:
1724
1725	xchg	%rsp,$key
1726	#add	\$8,%rsp	# reserve for return address!
1727	mov	$key,$_rsp	# save %rsp
1728.Lcbc_fast_body:
1729	mov	%rdi,$_inp	# save copy of inp
1730	mov	%rsi,$_out	# save copy of out
1731	mov	%rdx,$_len	# save copy of len
1732	mov	%rcx,$_key	# save copy of key
1733	mov	%r8,$_ivp	# save copy of ivp
1734	movl	\$0,$mark	# copy of aes_key->rounds = 0;
1735	mov	%r8,%rbp	# rearrange input arguments
1736	mov	%r9,%rbx
1737	mov	%rsi,$out
1738	mov	%rdi,$inp
1739	mov	%rcx,$key
1740
1741	mov	240($key),%eax		# key->rounds
1742	# do we copy key schedule to stack?
1743	mov	$key,%r10
1744	sub	$sbox,%r10
1745	and	\$0xfff,%r10
1746	cmp	\$2304,%r10
1747	jb	.Lcbc_do_ecopy
1748	cmp	\$4096-248,%r10
1749	jb	.Lcbc_skip_ecopy
1750.align	4
1751.Lcbc_do_ecopy:
1752		mov	$key,%rsi
1753		lea	$aes_key,%rdi
1754		lea	$aes_key,$key
1755		mov	\$240/8,%ecx
1756		.long	0x90A548F3	# rep movsq
1757		mov	%eax,(%rdi)	# copy aes_key->rounds
1758.Lcbc_skip_ecopy:
1759	mov	$key,$keyp	# save key pointer
1760
1761	mov	\$18,%ecx
1762.align	4
1763.Lcbc_prefetch_te:
1764		mov	0($sbox),%r10
1765		mov	32($sbox),%r11
1766		mov	64($sbox),%r12
1767		mov	96($sbox),%r13
1768		lea	128($sbox),$sbox
1769		sub	\$1,%ecx
1770	jnz	.Lcbc_prefetch_te
1771	lea	-2304($sbox),$sbox
1772
1773	cmp	\$0,%rbx
1774	je	.LFAST_DECRYPT
1775
1776#----------------------------- ENCRYPT -----------------------------#
1777	mov	0(%rbp),$s0		# load iv
1778	mov	4(%rbp),$s1
1779	mov	8(%rbp),$s2
1780	mov	12(%rbp),$s3
1781
1782.align	4
1783.Lcbc_fast_enc_loop:
1784		xor	0($inp),$s0
1785		xor	4($inp),$s1
1786		xor	8($inp),$s2
1787		xor	12($inp),$s3
1788		mov	$keyp,$key	# restore key
1789		mov	$inp,$_inp	# if ($verticalspin) save inp
1790
1791		call	_x86_64_AES_encrypt
1792
1793		mov	$_inp,$inp	# if ($verticalspin) restore inp
1794		mov	$_len,%r10
1795		mov	$s0,0($out)
1796		mov	$s1,4($out)
1797		mov	$s2,8($out)
1798		mov	$s3,12($out)
1799
1800		lea	16($inp),$inp
1801		lea	16($out),$out
1802		sub	\$16,%r10
1803		test	\$-16,%r10
1804		mov	%r10,$_len
1805	jnz	.Lcbc_fast_enc_loop
1806	mov	$_ivp,%rbp	# restore ivp
1807	mov	$s0,0(%rbp)	# save ivec
1808	mov	$s1,4(%rbp)
1809	mov	$s2,8(%rbp)
1810	mov	$s3,12(%rbp)
1811
1812	jmp	.Lcbc_fast_cleanup
1813
1814#----------------------------- DECRYPT -----------------------------#
1815.align	16
1816.LFAST_DECRYPT:
1817	cmp	$inp,$out
1818	je	.Lcbc_fast_dec_in_place
1819
1820	mov	%rbp,$ivec
1821.align	4
1822.Lcbc_fast_dec_loop:
1823		mov	0($inp),$s0	# read input
1824		mov	4($inp),$s1
1825		mov	8($inp),$s2
1826		mov	12($inp),$s3
1827		mov	$keyp,$key	# restore key
1828		mov	$inp,$_inp	# if ($verticalspin) save inp
1829
1830		call	_x86_64_AES_decrypt
1831
1832		mov	$ivec,%rbp	# load ivp
1833		mov	$_inp,$inp	# if ($verticalspin) restore inp
1834		mov	$_len,%r10	# load len
1835		xor	0(%rbp),$s0	# xor iv
1836		xor	4(%rbp),$s1
1837		xor	8(%rbp),$s2
1838		xor	12(%rbp),$s3
1839		mov	$inp,%rbp	# current input, next iv
1840
1841		sub	\$16,%r10
1842		mov	%r10,$_len	# update len
1843		mov	%rbp,$ivec	# update ivp
1844
1845		mov	$s0,0($out)	# write output
1846		mov	$s1,4($out)
1847		mov	$s2,8($out)
1848		mov	$s3,12($out)
1849
1850		lea	16($inp),$inp
1851		lea	16($out),$out
1852	jnz	.Lcbc_fast_dec_loop
1853	mov	$_ivp,%r12		# load user ivp
1854	mov	0(%rbp),%r10		# load iv
1855	mov	8(%rbp),%r11
1856	mov	%r10,0(%r12)		# copy back to user
1857	mov	%r11,8(%r12)
1858	jmp	.Lcbc_fast_cleanup
1859
1860.align	16
1861.Lcbc_fast_dec_in_place:
1862	mov	0(%rbp),%r10		# copy iv to stack
1863	mov	8(%rbp),%r11
1864	mov	%r10,0+$ivec
1865	mov	%r11,8+$ivec
1866.align	4
1867.Lcbc_fast_dec_in_place_loop:
1868		mov	0($inp),$s0	# load input
1869		mov	4($inp),$s1
1870		mov	8($inp),$s2
1871		mov	12($inp),$s3
1872		mov	$keyp,$key	# restore key
1873		mov	$inp,$_inp	# if ($verticalspin) save inp
1874
1875		call	_x86_64_AES_decrypt
1876
1877		mov	$_inp,$inp	# if ($verticalspin) restore inp
1878		mov	$_len,%r10
1879		xor	0+$ivec,$s0
1880		xor	4+$ivec,$s1
1881		xor	8+$ivec,$s2
1882		xor	12+$ivec,$s3
1883
1884		mov	0($inp),%r11	# load input
1885		mov	8($inp),%r12
1886		sub	\$16,%r10
1887		jz	.Lcbc_fast_dec_in_place_done
1888
1889		mov	%r11,0+$ivec	# copy input to iv
1890		mov	%r12,8+$ivec
1891
1892		mov	$s0,0($out)	# save output [zaps input]
1893		mov	$s1,4($out)
1894		mov	$s2,8($out)
1895		mov	$s3,12($out)
1896
1897		lea	16($inp),$inp
1898		lea	16($out),$out
1899		mov	%r10,$_len
1900	jmp	.Lcbc_fast_dec_in_place_loop
1901.Lcbc_fast_dec_in_place_done:
1902	mov	$_ivp,%rdi
1903	mov	%r11,0(%rdi)	# copy iv back to user
1904	mov	%r12,8(%rdi)
1905
1906	mov	$s0,0($out)	# save output [zaps input]
1907	mov	$s1,4($out)
1908	mov	$s2,8($out)
1909	mov	$s3,12($out)
1910
1911.align	4
1912.Lcbc_fast_cleanup:
1913	cmpl	\$0,$mark	# was the key schedule copied?
1914	lea	$aes_key,%rdi
1915	je	.Lcbc_exit
1916		mov	\$240/8,%ecx
1917		xor	%rax,%rax
1918		.long	0x90AB48F3	# rep stosq
1919
1920	jmp	.Lcbc_exit
1921
1922#--------------------------- SLOW ROUTINE ---------------------------#
1923.align	16
1924.Lcbc_slow_prologue:
1925	# allocate aligned stack frame...
1926	lea	-88(%rsp),%rbp
1927	and	\$-64,%rbp
1928	# ... just "above" key schedule
1929	lea	-88-63(%rcx),%r10
1930	sub	%rbp,%r10
1931	neg	%r10
1932	and	\$0x3c0,%r10
1933	sub	%r10,%rbp
1934
1935	xchg	%rsp,%rbp
1936	#add	\$8,%rsp	# reserve for return address!
1937	mov	%rbp,$_rsp	# save %rsp
1938.Lcbc_slow_body:
1939	#mov	%rdi,$_inp	# save copy of inp
1940	#mov	%rsi,$_out	# save copy of out
1941	#mov	%rdx,$_len	# save copy of len
1942	#mov	%rcx,$_key	# save copy of key
1943	mov	%r8,$_ivp	# save copy of ivp
1944	mov	%r8,%rbp	# rearrange input arguments
1945	mov	%r9,%rbx
1946	mov	%rsi,$out
1947	mov	%rdi,$inp
1948	mov	%rcx,$key
1949	mov	%rdx,%r10
1950
1951	mov	240($key),%eax
1952	mov	$key,$keyp	# save key pointer
1953	shl	\$4,%eax
1954	lea	($key,%rax),%rax
1955	mov	%rax,$keyend
1956
1957	# pick Te4 copy which can't "overlap" with stack frame or key schedule
1958	lea	2048($sbox),$sbox
1959	lea	768-8(%rsp),%rax
1960	sub	$sbox,%rax
1961	and	\$0x300,%rax
1962	lea	($sbox,%rax),$sbox
1963
1964	cmp	\$0,%rbx
1965	je	.LSLOW_DECRYPT
1966
1967#--------------------------- SLOW ENCRYPT ---------------------------#
1968	test	\$-16,%r10		# check upon length
1969	mov	0(%rbp),$s0		# load iv
1970	mov	4(%rbp),$s1
1971	mov	8(%rbp),$s2
1972	mov	12(%rbp),$s3
1973	jz	.Lcbc_slow_enc_tail	# short input...
1974
1975.align	4
1976.Lcbc_slow_enc_loop:
1977		xor	0($inp),$s0
1978		xor	4($inp),$s1
1979		xor	8($inp),$s2
1980		xor	12($inp),$s3
1981		mov	$keyp,$key	# restore key
1982		mov	$inp,$_inp	# save inp
1983		mov	$out,$_out	# save out
1984		mov	%r10,$_len	# save len
1985
1986		call	_x86_64_AES_encrypt_compact
1987
1988		mov	$_inp,$inp	# restore inp
1989		mov	$_out,$out	# restore out
1990		mov	$_len,%r10	# restore len
1991		mov	$s0,0($out)
1992		mov	$s1,4($out)
1993		mov	$s2,8($out)
1994		mov	$s3,12($out)
1995
1996		lea	16($inp),$inp
1997		lea	16($out),$out
1998		sub	\$16,%r10
1999		test	\$-16,%r10
2000	jnz	.Lcbc_slow_enc_loop
2001	test	\$15,%r10
2002	jnz	.Lcbc_slow_enc_tail
2003	mov	$_ivp,%rbp	# restore ivp
2004	mov	$s0,0(%rbp)	# save ivec
2005	mov	$s1,4(%rbp)
2006	mov	$s2,8(%rbp)
2007	mov	$s3,12(%rbp)
2008
2009	jmp	.Lcbc_exit
2010
2011.align	4
2012.Lcbc_slow_enc_tail:
2013	mov	%rax,%r11
2014	mov	%rcx,%r12
2015	mov	%r10,%rcx
2016	mov	$inp,%rsi
2017	mov	$out,%rdi
2018	.long	0x9066A4F3		# rep movsb
2019	mov	\$16,%rcx		# zero tail
2020	sub	%r10,%rcx
2021	xor	%rax,%rax
2022	.long	0x9066AAF3		# rep stosb
2023	mov	$out,$inp		# this is not a mistake!
2024	mov	\$16,%r10		# len=16
2025	mov	%r11,%rax
2026	mov	%r12,%rcx
2027	jmp	.Lcbc_slow_enc_loop	# one more spin...
2028#--------------------------- SLOW DECRYPT ---------------------------#
2029.align	16
2030.LSLOW_DECRYPT:
2031	shr	\$3,%rax
2032	add	%rax,$sbox		# recall "magic" constants!
2033
2034	mov	0(%rbp),%r11		# copy iv to stack
2035	mov	8(%rbp),%r12
2036	mov	%r11,0+$ivec
2037	mov	%r12,8+$ivec
2038
2039.align	4
2040.Lcbc_slow_dec_loop:
2041		mov	0($inp),$s0	# load input
2042		mov	4($inp),$s1
2043		mov	8($inp),$s2
2044		mov	12($inp),$s3
2045		mov	$keyp,$key	# restore key
2046		mov	$inp,$_inp	# save inp
2047		mov	$out,$_out	# save out
2048		mov	%r10,$_len	# save len
2049
2050		call	_x86_64_AES_decrypt_compact
2051
2052		mov	$_inp,$inp	# restore inp
2053		mov	$_out,$out	# restore out
2054		mov	$_len,%r10
2055		xor	0+$ivec,$s0
2056		xor	4+$ivec,$s1
2057		xor	8+$ivec,$s2
2058		xor	12+$ivec,$s3
2059
2060		mov	0($inp),%r11	# load input
2061		mov	8($inp),%r12
2062		sub	\$16,%r10
2063		jc	.Lcbc_slow_dec_partial
2064		jz	.Lcbc_slow_dec_done
2065
2066		mov	%r11,0+$ivec	# copy input to iv
2067		mov	%r12,8+$ivec
2068
2069		mov	$s0,0($out)	# save output [can zap input]
2070		mov	$s1,4($out)
2071		mov	$s2,8($out)
2072		mov	$s3,12($out)
2073
2074		lea	16($inp),$inp
2075		lea	16($out),$out
2076	jmp	.Lcbc_slow_dec_loop
2077.Lcbc_slow_dec_done:
2078	mov	$_ivp,%rdi
2079	mov	%r11,0(%rdi)		# copy iv back to user
2080	mov	%r12,8(%rdi)
2081
2082	mov	$s0,0($out)		# save output [can zap input]
2083	mov	$s1,4($out)
2084	mov	$s2,8($out)
2085	mov	$s3,12($out)
2086
2087	jmp	.Lcbc_exit
2088
2089.align	4
2090.Lcbc_slow_dec_partial:
2091	mov	$_ivp,%rdi
2092	mov	%r11,0(%rdi)		# copy iv back to user
2093	mov	%r12,8(%rdi)
2094
2095	mov	$s0,0+$ivec		# save output to stack
2096	mov	$s1,4+$ivec
2097	mov	$s2,8+$ivec
2098	mov	$s3,12+$ivec
2099
2100	mov	$out,%rdi
2101	lea	$ivec,%rsi
2102	lea	16(%r10),%rcx
2103	.long	0x9066A4F3	# rep movsb
2104	jmp	.Lcbc_exit
2105
2106.align	16
2107.Lcbc_exit:
2108	mov	$_rsp,%rsi
2109	mov	(%rsi),%r15
2110	mov	8(%rsi),%r14
2111	mov	16(%rsi),%r13
2112	mov	24(%rsi),%r12
2113	mov	32(%rsi),%rbp
2114	mov	40(%rsi),%rbx
2115	lea	48(%rsi),%rsp
2116.Lcbc_popfq:
2117	popfq
2118.Lcbc_epilogue:
2119	ret
2120.size	aes_cbc_encrypt_internal,.-aes_cbc_encrypt_internal
2121___
2122}
2123
2124$code.=<<___;
2125.section .rodata
2126.align	64
2127.LAES_Te:
2128___
2129	&_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6);
2130	&_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591);
2131	&_data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56);
2132	&_data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec);
2133	&_data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa);
2134	&_data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb);
2135	&_data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45);
2136	&_data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b);
2137	&_data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c);
2138	&_data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83);
2139	&_data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9);
2140	&_data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a);
2141	&_data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d);
2142	&_data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f);
2143	&_data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df);
2144	&_data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea);
2145	&_data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34);
2146	&_data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b);
2147	&_data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d);
2148	&_data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413);
2149	&_data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1);
2150	&_data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6);
2151	&_data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972);
2152	&_data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85);
2153	&_data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed);
2154	&_data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511);
2155	&_data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe);
2156	&_data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b);
2157	&_data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05);
2158	&_data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1);
2159	&_data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142);
2160	&_data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf);
2161	&_data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3);
2162	&_data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e);
2163	&_data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a);
2164	&_data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6);
2165	&_data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3);
2166	&_data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b);
2167	&_data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428);
2168	&_data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad);
2169	&_data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14);
2170	&_data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8);
2171	&_data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4);
2172	&_data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2);
2173	&_data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda);
2174	&_data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949);
2175	&_data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf);
2176	&_data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810);
2177	&_data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c);
2178	&_data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697);
2179	&_data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e);
2180	&_data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f);
2181	&_data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc);
2182	&_data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c);
2183	&_data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969);
2184	&_data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27);
2185	&_data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122);
2186	&_data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433);
2187	&_data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9);
2188	&_data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5);
2189	&_data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a);
2190	&_data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0);
2191	&_data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e);
2192	&_data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c);
2193
2194#Te4	# four copies of Te4 to choose from to avoid L1 aliasing
2195	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2196	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2197	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2198	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2199	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2200	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2201	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2202	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2203	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2204	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2205	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2206	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2207	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2208	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2209	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2210	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2211	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2212	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2213	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2214	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2215	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2216	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2217	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2218	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2219	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2220	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2221	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2222	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2223	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2224	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2225	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2226	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2227
2228	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2229	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2230	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2231	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2232	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2233	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2234	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2235	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2236	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2237	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2238	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2239	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2240	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2241	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2242	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2243	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2244	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2245	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2246	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2247	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2248	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2249	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2250	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2251	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2252	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2253	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2254	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2255	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2256	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2257	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2258	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2259	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2260
2261	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2262	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2263	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2264	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2265	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2266	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2267	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2268	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2269	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2270	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2271	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2272	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2273	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2274	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2275	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2276	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2277	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2278	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2279	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2280	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2281	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2282	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2283	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2284	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2285	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2286	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2287	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2288	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2289	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2290	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2291	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2292	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2293
2294	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
2295	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
2296	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
2297	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
2298	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
2299	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
2300	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
2301	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
2302	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
2303	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
2304	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
2305	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
2306	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
2307	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
2308	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
2309	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
2310	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
2311	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
2312	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
2313	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
2314	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
2315	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
2316	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
2317	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
2318	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
2319	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
2320	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
2321	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
2322	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
2323	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
2324	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
2325	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
2326#rcon:
2327$code.=<<___;
2328	.long	0x00000001, 0x00000002, 0x00000004, 0x00000008
2329	.long	0x00000010, 0x00000020, 0x00000040, 0x00000080
2330	.long	0x0000001b, 0x00000036, 0x80808080, 0x80808080
2331	.long	0xfefefefe, 0xfefefefe, 0x1b1b1b1b, 0x1b1b1b1b
2332___
2333$code.=<<___;
2334.align	64
2335.LAES_Td:
2336___
2337	&_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a);
2338	&_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b);
2339	&_data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5);
2340	&_data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5);
2341	&_data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d);
2342	&_data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b);
2343	&_data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295);
2344	&_data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e);
2345	&_data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927);
2346	&_data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d);
2347	&_data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362);
2348	&_data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9);
2349	&_data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52);
2350	&_data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566);
2351	&_data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3);
2352	&_data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed);
2353	&_data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e);
2354	&_data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4);
2355	&_data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4);
2356	&_data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd);
2357	&_data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d);
2358	&_data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060);
2359	&_data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967);
2360	&_data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879);
2361	&_data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000);
2362	&_data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c);
2363	&_data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36);
2364	&_data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624);
2365	&_data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b);
2366	&_data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c);
2367	&_data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12);
2368	&_data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14);
2369	&_data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3);
2370	&_data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b);
2371	&_data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8);
2372	&_data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684);
2373	&_data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7);
2374	&_data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177);
2375	&_data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947);
2376	&_data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322);
2377	&_data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498);
2378	&_data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f);
2379	&_data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54);
2380	&_data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382);
2381	&_data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf);
2382	&_data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb);
2383	&_data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83);
2384	&_data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef);
2385	&_data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029);
2386	&_data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235);
2387	&_data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733);
2388	&_data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117);
2389	&_data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4);
2390	&_data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546);
2391	&_data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb);
2392	&_data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d);
2393	&_data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb);
2394	&_data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a);
2395	&_data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773);
2396	&_data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478);
2397	&_data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2);
2398	&_data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff);
2399	&_data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664);
2400	&_data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0);
2401
2402#Td4:	# four copies of Td4 to choose from to avoid L1 aliasing
2403	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2404	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2405	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2406	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2407	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2408	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2409	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2410	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2411	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2412	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2413	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2414	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2415	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2416	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2417	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2418	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2419	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2420	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2421	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2422	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2423	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2424	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2425	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2426	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2427	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2428	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2429	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2430	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2431	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2432	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2433	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2434	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2435$code.=<<___;
2436	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2437	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2438___
2439	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2440	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2441	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2442	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2443	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2444	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2445	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2446	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2447	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2448	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2449	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2450	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2451	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2452	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2453	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2454	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2455	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2456	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2457	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2458	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2459	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2460	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2461	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2462	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2463	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2464	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2465	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2466	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2467	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2468	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2469	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2470	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2471$code.=<<___;
2472	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2473	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2474___
2475	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2476	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2477	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2478	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2479	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2480	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2481	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2482	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2483	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2484	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2485	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2486	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2487	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2488	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2489	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2490	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2491	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2492	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2493	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2494	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2495	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2496	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2497	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2498	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2499	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2500	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2501	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2502	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2503	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2504	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2505	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2506	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2507$code.=<<___;
2508	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2509	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2510___
2511	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
2512	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
2513	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
2514	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
2515	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
2516	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
2517	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
2518	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
2519	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
2520	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
2521	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
2522	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
2523	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
2524	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
2525	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
2526	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
2527	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
2528	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
2529	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
2530	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
2531	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
2532	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
2533	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
2534	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
2535	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
2536	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
2537	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
2538	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
2539	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
2540	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
2541	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
2542	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
2543$code.=<<___;
2544	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
2545	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
2546.align	64
2547.text
2548___
2549
2550# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
2551#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
2552if ($win64) {
2553$rec="%rcx";
2554$frame="%rdx";
2555$context="%r8";
2556$disp="%r9";
2557
2558$code.=<<___;
2559.extern	__imp_RtlVirtualUnwind
2560.type	block_se_handler,\@abi-omnipotent
2561.align	16
2562block_se_handler:
2563	_CET_ENDBR
2564	push	%rsi
2565	push	%rdi
2566	push	%rbx
2567	push	%rbp
2568	push	%r12
2569	push	%r13
2570	push	%r14
2571	push	%r15
2572	pushfq
2573	sub	\$64,%rsp
2574
2575	mov	120($context),%rax	# pull context->Rax
2576	mov	248($context),%rbx	# pull context->Rip
2577
2578	mov	8($disp),%rsi		# disp->ImageBase
2579	mov	56($disp),%r11		# disp->HandlerData
2580
2581	mov	0(%r11),%r10d		# HandlerData[0]
2582	lea	(%rsi,%r10),%r10	# prologue label
2583	cmp	%r10,%rbx		# context->Rip<prologue label
2584	jb	.Lin_block_prologue
2585
2586	mov	152($context),%rax	# pull context->Rsp
2587
2588	mov	4(%r11),%r10d		# HandlerData[1]
2589	lea	(%rsi,%r10),%r10	# epilogue label
2590	cmp	%r10,%rbx		# context->Rip>=epilogue label
2591	jae	.Lin_block_prologue
2592
2593	mov	24(%rax),%rax		# pull saved real stack pointer
2594	lea	48(%rax),%rax		# adjust...
2595
2596	mov	-8(%rax),%rbx
2597	mov	-16(%rax),%rbp
2598	mov	-24(%rax),%r12
2599	mov	-32(%rax),%r13
2600	mov	-40(%rax),%r14
2601	mov	-48(%rax),%r15
2602	mov	%rbx,144($context)	# restore context->Rbx
2603	mov	%rbp,160($context)	# restore context->Rbp
2604	mov	%r12,216($context)	# restore context->R12
2605	mov	%r13,224($context)	# restore context->R13
2606	mov	%r14,232($context)	# restore context->R14
2607	mov	%r15,240($context)	# restore context->R15
2608
2609.Lin_block_prologue:
2610	mov	8(%rax),%rdi
2611	mov	16(%rax),%rsi
2612	mov	%rax,152($context)	# restore context->Rsp
2613	mov	%rsi,168($context)	# restore context->Rsi
2614	mov	%rdi,176($context)	# restore context->Rdi
2615
2616	jmp	.Lcommon_seh_exit
2617.size	block_se_handler,.-block_se_handler
2618
2619.type	key_se_handler,\@abi-omnipotent
2620.align	16
2621key_se_handler:
2622	_CET_ENDBR
2623	push	%rsi
2624	push	%rdi
2625	push	%rbx
2626	push	%rbp
2627	push	%r12
2628	push	%r13
2629	push	%r14
2630	push	%r15
2631	pushfq
2632	sub	\$64,%rsp
2633
2634	mov	120($context),%rax	# pull context->Rax
2635	mov	248($context),%rbx	# pull context->Rip
2636
2637	mov	8($disp),%rsi		# disp->ImageBase
2638	mov	56($disp),%r11		# disp->HandlerData
2639
2640	mov	0(%r11),%r10d		# HandlerData[0]
2641	lea	(%rsi,%r10),%r10	# prologue label
2642	cmp	%r10,%rbx		# context->Rip<prologue label
2643	jb	.Lin_key_prologue
2644
2645	mov	152($context),%rax	# pull context->Rsp
2646
2647	mov	4(%r11),%r10d		# HandlerData[1]
2648	lea	(%rsi,%r10),%r10	# epilogue label
2649	cmp	%r10,%rbx		# context->Rip>=epilogue label
2650	jae	.Lin_key_prologue
2651
2652	lea	56(%rax),%rax
2653
2654	mov	-8(%rax),%rbx
2655	mov	-16(%rax),%rbp
2656	mov	-24(%rax),%r12
2657	mov	-32(%rax),%r13
2658	mov	-40(%rax),%r14
2659	mov	-48(%rax),%r15
2660	mov	%rbx,144($context)	# restore context->Rbx
2661	mov	%rbp,160($context)	# restore context->Rbp
2662	mov	%r12,216($context)	# restore context->R12
2663	mov	%r13,224($context)	# restore context->R13
2664	mov	%r14,232($context)	# restore context->R14
2665	mov	%r15,240($context)	# restore context->R15
2666
2667.Lin_key_prologue:
2668	mov	8(%rax),%rdi
2669	mov	16(%rax),%rsi
2670	mov	%rax,152($context)	# restore context->Rsp
2671	mov	%rsi,168($context)	# restore context->Rsi
2672	mov	%rdi,176($context)	# restore context->Rdi
2673
2674	jmp	.Lcommon_seh_exit
2675.size	key_se_handler,.-key_se_handler
2676
2677.type	cbc_se_handler,\@abi-omnipotent
2678.align	16
2679cbc_se_handler:
2680	_CET_ENDBR
2681	push	%rsi
2682	push	%rdi
2683	push	%rbx
2684	push	%rbp
2685	push	%r12
2686	push	%r13
2687	push	%r14
2688	push	%r15
2689	pushfq
2690	sub	\$64,%rsp
2691
2692	mov	120($context),%rax	# pull context->Rax
2693	mov	248($context),%rbx	# pull context->Rip
2694
2695	lea	.Lcbc_prologue(%rip),%r10
2696	cmp	%r10,%rbx		# context->Rip<.Lcbc_prologue
2697	jb	.Lin_cbc_prologue
2698
2699	lea	.Lcbc_fast_body(%rip),%r10
2700	cmp	%r10,%rbx		# context->Rip<.Lcbc_fast_body
2701	jb	.Lin_cbc_frame_setup
2702
2703	lea	.Lcbc_slow_prologue(%rip),%r10
2704	cmp	%r10,%rbx		# context->Rip<.Lcbc_slow_prologue
2705	jb	.Lin_cbc_body
2706
2707	lea	.Lcbc_slow_body(%rip),%r10
2708	cmp	%r10,%rbx		# context->Rip<.Lcbc_slow_body
2709	jb	.Lin_cbc_frame_setup
2710
2711.Lin_cbc_body:
2712	mov	152($context),%rax	# pull context->Rsp
2713
2714	lea	.Lcbc_epilogue(%rip),%r10
2715	cmp	%r10,%rbx		# context->Rip>=.Lcbc_epilogue
2716	jae	.Lin_cbc_prologue
2717
2718	lea	8(%rax),%rax
2719
2720	lea	.Lcbc_popfq(%rip),%r10
2721	cmp	%r10,%rbx		# context->Rip>=.Lcbc_popfq
2722	jae	.Lin_cbc_prologue
2723
2724	mov	`16-8`(%rax),%rax	# biased $_rsp
2725	lea	56(%rax),%rax
2726
2727.Lin_cbc_frame_setup:
2728	mov	-16(%rax),%rbx
2729	mov	-24(%rax),%rbp
2730	mov	-32(%rax),%r12
2731	mov	-40(%rax),%r13
2732	mov	-48(%rax),%r14
2733	mov	-56(%rax),%r15
2734	mov	%rbx,144($context)	# restore context->Rbx
2735	mov	%rbp,160($context)	# restore context->Rbp
2736	mov	%r12,216($context)	# restore context->R12
2737	mov	%r13,224($context)	# restore context->R13
2738	mov	%r14,232($context)	# restore context->R14
2739	mov	%r15,240($context)	# restore context->R15
2740
2741.Lin_cbc_prologue:
2742	mov	8(%rax),%rdi
2743	mov	16(%rax),%rsi
2744	mov	%rax,152($context)	# restore context->Rsp
2745	mov	%rsi,168($context)	# restore context->Rsi
2746	mov	%rdi,176($context)	# restore context->Rdi
2747
2748.Lcommon_seh_exit:
2749
2750	mov	40($disp),%rdi		# disp->ContextRecord
2751	mov	$context,%rsi		# context
2752	mov	\$`1232/8`,%ecx		# sizeof(CONTEXT)
2753	.long	0xa548f3fc		# cld; rep movsq
2754
2755	mov	$disp,%rsi
2756	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
2757	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
2758	mov	0(%rsi),%r8		# arg3, disp->ControlPc
2759	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
2760	mov	40(%rsi),%r10		# disp->ContextRecord
2761	lea	56(%rsi),%r11		# &disp->HandlerData
2762	lea	24(%rsi),%r12		# &disp->EstablisherFrame
2763	mov	%r10,32(%rsp)		# arg5
2764	mov	%r11,40(%rsp)		# arg6
2765	mov	%r12,48(%rsp)		# arg7
2766	mov	%rcx,56(%rsp)		# arg8, (NULL)
2767	call	*__imp_RtlVirtualUnwind(%rip)
2768
2769	mov	\$1,%eax		# ExceptionContinueSearch
2770	add	\$64,%rsp
2771	popfq
2772	pop	%r15
2773	pop	%r14
2774	pop	%r13
2775	pop	%r12
2776	pop	%rbp
2777	pop	%rbx
2778	pop	%rdi
2779	pop	%rsi
2780	ret
2781.size	cbc_se_handler,.-cbc_se_handler
2782
2783.section	.pdata
2784.align	4
2785	.rva	.LSEH_begin_aes_encrypt_internal
2786	.rva	.LSEH_end_aes_encrypt_internal
2787	.rva	.LSEH_info_aes_encrypt_internal
2788
2789	.rva	.LSEH_begin_aes_decrypt_internal
2790	.rva	.LSEH_end_aes_decrypt_internal
2791	.rva	.LSEH_info_aes_decrypt_internal
2792
2793	.rva	.LSEH_begin_aes_set_encrypt_key_internal
2794	.rva	.LSEH_end_aes_set_encrypt_key_internal
2795	.rva	.LSEH_info_aes_set_encrypt_key_internal
2796
2797	.rva	.LSEH_begin_aes_set_decrypt_key_internal
2798	.rva	.LSEH_end_aes_set_decrypt_key_internal
2799	.rva	.LSEH_info_aes_set_decrypt_key_internal
2800
2801	.rva	.LSEH_begin_aes_cbc_encrypt_internal
2802	.rva	.LSEH_end_aes_cbc_encrypt_internal
2803	.rva	.LSEH_info_aes_cbc_encrypt_internal
2804
2805.section	.xdata
2806.align	8
2807.LSEH_info_aes_encrypt_internal:
2808	.byte	9,0,0,0
2809	.rva	block_se_handler
2810	.rva	.Lenc_prologue,.Lenc_epilogue	# HandlerData[]
2811.LSEH_info_aes_decrypt_internal:
2812	.byte	9,0,0,0
2813	.rva	block_se_handler
2814	.rva	.Ldec_prologue,.Ldec_epilogue	# HandlerData[]
2815.LSEH_info_aes_set_encrypt_key_internal:
2816	.byte	9,0,0,0
2817	.rva	key_se_handler
2818	.rva	.Lenc_key_prologue,.Lenc_key_epilogue	# HandlerData[]
2819.LSEH_info_aes_set_decrypt_key_internal:
2820	.byte	9,0,0,0
2821	.rva	key_se_handler
2822	.rva	.Ldec_key_prologue,.Ldec_key_epilogue	# HandlerData[]
2823.LSEH_info_aes_cbc_encrypt_internal:
2824	.byte	9,0,0,0
2825	.rva	cbc_se_handler
2826___
2827}
2828
2829$code =~ s/\`([^\`]*)\`/eval($1)/gem;
2830
2831print "#include \"x86_arch.h\"\n";
2832print $code;
2833
2834close STDOUT;
2835