xref: /netbsd-src/crypto/external/bsd/openssl/dist/crypto/aes/asm/aesni-sha1-x86_64.pl (revision 16dce51364ebe8aeafbae46bc5aa167b8115bc45)
1#! /usr/bin/env perl
2# Copyright 2011-2016 The OpenSSL Project Authors. All Rights Reserved.
3#
4# Licensed under the OpenSSL license (the "License").  You may not use
5# this file except in compliance with the License.  You can obtain a copy
6# in the file LICENSE in the source distribution or at
7# https://www.openssl.org/source/license.html
8
9#
10# ====================================================================
11# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
12# project. The module is, however, dual licensed under OpenSSL and
13# CRYPTOGAMS licenses depending on where you obtain it. For further
14# details see http://www.openssl.org/~appro/cryptogams/.
15# ====================================================================
16#
17# June 2011
18#
19# This is AESNI-CBC+SHA1 "stitch" implementation. The idea, as spelled
20# in http://download.intel.com/design/intarch/papers/323686.pdf, is
21# that since AESNI-CBC encrypt exhibit *very* low instruction-level
22# parallelism, interleaving it with another algorithm would allow to
23# utilize processor resources better and achieve better performance.
24# SHA1 instruction sequences(*) are taken from sha1-x86_64.pl and
25# AESNI code is weaved into it. Below are performance numbers in
26# cycles per processed byte, less is better, for standalone AESNI-CBC
27# encrypt, sum of the latter and standalone SHA1, and "stitched"
28# subroutine:
29#
30#		AES-128-CBC	+SHA1		stitch      gain
31# Westmere	3.77[+5.3]	9.07		6.55	    +38%
32# Sandy Bridge	5.05[+5.0(6.1)]	10.06(11.15)	5.98(7.05)  +68%(+58%)
33# Ivy Bridge	5.05[+4.6]	9.65		5.54        +74%
34# Haswell	4.43[+3.6(4.2)]	8.00(8.58)	4.55(5.21)  +75%(+65%)
35# Skylake	2.63[+3.5(4.1)]	6.17(6.69)	4.23(4.44)  +46%(+51%)
36# Bulldozer	5.77[+6.0]	11.72		6.37        +84%
37# Ryzen(**)	2.71[+1.93]	4.64		2.74        +69%
38# Goldmont(**)	3.82[+1.70]	5.52		4.20        +31%
39#
40#		AES-192-CBC
41# Westmere	4.51		9.81		6.80	    +44%
42# Sandy Bridge	6.05		11.06(12.15)	6.11(7.19)  +81%(+69%)
43# Ivy Bridge	6.05		10.65		6.07        +75%
44# Haswell	5.29		8.86(9.44)	5.32(5.32)  +67%(+77%)
45# Bulldozer	6.89		12.84		6.96        +84%
46#
47#		AES-256-CBC
48# Westmere	5.25		10.55		7.21	    +46%
49# Sandy Bridge	7.05		12.06(13.15)	7.12(7.72)  +69%(+70%)
50# Ivy Bridge	7.05		11.65		7.12        +64%
51# Haswell	6.19		9.76(10.34)	6.21(6.25)  +57%(+65%)
52# Skylake	3.62		7.16(7.68)	4.56(4.76)  +57%(+61%)
53# Bulldozer	8.00		13.95		8.25        +69%
54# Ryzen(**)	3.71		5.64		3.72        +52%
55# Goldmont(**)	5.35		7.05		5.76        +22%
56#
57# (*)	There are two code paths: SSSE3 and AVX. See sha1-568.pl for
58#	background information. Above numbers in parentheses are SSSE3
59#	results collected on AVX-capable CPU, i.e. apply on OSes that
60#	don't support AVX.
61# (**)	SHAEXT results.
62#
63# Needless to mention that it makes no sense to implement "stitched"
64# *decrypt* subroutine. Because *both* AESNI-CBC decrypt and SHA1
65# fully utilize parallelism, so stitching would not give any gain
66# anyway. Well, there might be some, e.g. because of better cache
67# locality... For reference, here are performance results for
68# standalone AESNI-CBC decrypt:
69#
70#		AES-128-CBC	AES-192-CBC	AES-256-CBC
71# Westmere	1.25		1.50		1.75
72# Sandy Bridge	0.74		0.91		1.09
73# Ivy Bridge	0.74		0.90		1.11
74# Haswell	0.63		0.76		0.88
75# Bulldozer	0.70		0.85		0.99
76
77# And indeed:
78#
79#		AES-256-CBC	+SHA1		stitch      gain
80# Westmere	1.75		7.20		6.68        +7.8%
81# Sandy Bridge	1.09		6.09(7.22)	5.82(6.95)  +4.6%(+3.9%)
82# Ivy Bridge	1.11		5.70		5.45        +4.6%
83# Haswell	0.88		4.45(5.00)	4.39(4.69)  +1.4%(*)(+6.6%)
84# Bulldozer	0.99		6.95		5.95        +17%(**)
85#
86# (*)	Tiny improvement coefficient on Haswell is because we compare
87#	AVX1 stitch to sum with AVX2 SHA1.
88# (**)	Execution is fully dominated by integer code sequence and
89#	SIMD still hardly shows [in single-process benchmark;-]
90
91$flavour = shift;
92$output  = shift;
93if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
94
95$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
96
97$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
98( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
99( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
100die "can't locate x86_64-xlate.pl";
101
102$avx=1 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
103		=~ /GNU assembler version ([2-9]\.[0-9]+)/ &&
104	   $1>=2.19);
105$avx=1 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
106	   `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/ &&
107	   $1>=2.09);
108$avx=1 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
109	   `ml64 2>&1` =~ /Version ([0-9]+)\./ &&
110	   $1>=10);
111$avx=1 if (!$avx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9]\.[0-9]+)/ && $2>=3.0);
112
113$shaext=1;	### set to zero if compiling for 1.0.1
114
115$stitched_decrypt=0;
116
117open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
118*STDOUT=*OUT;
119
120# void aesni_cbc_sha1_enc(const void *inp,
121#			void *out,
122#			size_t length,
123#			const AES_KEY *key,
124#			unsigned char *iv,
125#			SHA_CTX *ctx,
126#			const void *in0);
127
128$code.=<<___;
129.text
130.extern	OPENSSL_ia32cap_P
131
132.globl	aesni_cbc_sha1_enc
133.type	aesni_cbc_sha1_enc,\@abi-omnipotent
134.align	32
135aesni_cbc_sha1_enc:
136	# caller should check for SSSE3 and AES-NI bits
137	mov	OPENSSL_ia32cap_P+0(%rip),%r10d
138	mov	OPENSSL_ia32cap_P+4(%rip),%r11
139___
140$code.=<<___ if ($shaext);
141	bt	\$61,%r11		# check SHA bit
142	jc	aesni_cbc_sha1_enc_shaext
143___
144$code.=<<___ if ($avx);
145	and	\$`1<<28`,%r11d		# mask AVX bit
146	and	\$`1<<30`,%r10d		# mask "Intel CPU" bit
147	or	%r11d,%r10d
148	cmp	\$`1<<28|1<<30`,%r10d
149	je	aesni_cbc_sha1_enc_avx
150___
151$code.=<<___;
152	jmp	aesni_cbc_sha1_enc_ssse3
153	ret
154.size	aesni_cbc_sha1_enc,.-aesni_cbc_sha1_enc
155___
156
157my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
158
159my $Xi=4;
160my @X=map("%xmm$_",(4..7,0..3));
161my @Tx=map("%xmm$_",(8..10));
162my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp");	# size optimization
163my @T=("%esi","%edi");
164my $j=0; my $jj=0; my $r=0; my $sn=0; my $rx=0;
165my $K_XX_XX="%r11";
166my ($rndkey0,$iv,$in)=map("%xmm$_",(11..13));			# for enc
167my @rndkey=("%xmm14","%xmm15");					# for enc
168my ($inout0,$inout1,$inout2,$inout3)=map("%xmm$_",(12..15));	# for dec
169
170if (1) {	# reassign for Atom Silvermont
171    # The goal is to minimize amount of instructions with more than
172    # 3 prefix bytes. Or in more practical terms to keep AES-NI *and*
173    # SSSE3 instructions to upper half of the register bank.
174    @X=map("%xmm$_",(8..11,4..7));
175    @Tx=map("%xmm$_",(12,13,3));
176    ($iv,$in,$rndkey0)=map("%xmm$_",(2,14,15));
177    @rndkey=("%xmm0","%xmm1");
178}
179
180sub AUTOLOAD()		# thunk [simplified] 32-bit style perlasm
181{ my $opcode = $AUTOLOAD; $opcode =~ s/.*:://;
182  my $arg = pop;
183    $arg = "\$$arg" if ($arg*1 eq $arg);
184    $code .= "\t$opcode\t".join(',',$arg,reverse @_)."\n";
185}
186
187my $_rol=sub { &rol(@_) };
188my $_ror=sub { &ror(@_) };
189
190$code.=<<___;
191.type	aesni_cbc_sha1_enc_ssse3,\@function,6
192.align	32
193aesni_cbc_sha1_enc_ssse3:
194	mov	`($win64?56:8)`(%rsp),$inp	# load 7th argument
195	#shr	\$6,$len			# debugging artefact
196	#jz	.Lepilogue_ssse3		# debugging artefact
197	push	%rbx
198	push	%rbp
199	push	%r12
200	push	%r13
201	push	%r14
202	push	%r15
203	lea	`-104-($win64?10*16:0)`(%rsp),%rsp
204	#mov	$in0,$inp			# debugging artefact
205	#lea	64(%rsp),$ctx			# debugging artefact
206___
207$code.=<<___ if ($win64);
208	movaps	%xmm6,96+0(%rsp)
209	movaps	%xmm7,96+16(%rsp)
210	movaps	%xmm8,96+32(%rsp)
211	movaps	%xmm9,96+48(%rsp)
212	movaps	%xmm10,96+64(%rsp)
213	movaps	%xmm11,96+80(%rsp)
214	movaps	%xmm12,96+96(%rsp)
215	movaps	%xmm13,96+112(%rsp)
216	movaps	%xmm14,96+128(%rsp)
217	movaps	%xmm15,96+144(%rsp)
218.Lprologue_ssse3:
219___
220$code.=<<___;
221	mov	$in0,%r12			# reassign arguments
222	mov	$out,%r13
223	mov	$len,%r14
224	lea	112($key),%r15			# size optimization
225	movdqu	($ivp),$iv			# load IV
226	mov	$ivp,88(%rsp)			# save $ivp
227___
228($in0,$out,$len,$key)=map("%r$_",(12..15));	# reassign arguments
229my $rounds="${ivp}d";
230$code.=<<___;
231	shl	\$6,$len
232	sub	$in0,$out
233	mov	240-112($key),$rounds
234	add	$inp,$len		# end of input
235
236	lea	K_XX_XX(%rip),$K_XX_XX
237	mov	0($ctx),$A		# load context
238	mov	4($ctx),$B
239	mov	8($ctx),$C
240	mov	12($ctx),$D
241	mov	$B,@T[0]		# magic seed
242	mov	16($ctx),$E
243	mov	$C,@T[1]
244	xor	$D,@T[1]
245	and	@T[1],@T[0]
246
247	movdqa	64($K_XX_XX),@Tx[2]	# pbswap mask
248	movdqa	0($K_XX_XX),@Tx[1]	# K_00_19
249	movdqu	0($inp),@X[-4&7]	# load input to %xmm[0-3]
250	movdqu	16($inp),@X[-3&7]
251	movdqu	32($inp),@X[-2&7]
252	movdqu	48($inp),@X[-1&7]
253	pshufb	@Tx[2],@X[-4&7]		# byte swap
254	pshufb	@Tx[2],@X[-3&7]
255	pshufb	@Tx[2],@X[-2&7]
256	add	\$64,$inp
257	paddd	@Tx[1],@X[-4&7]		# add K_00_19
258	pshufb	@Tx[2],@X[-1&7]
259	paddd	@Tx[1],@X[-3&7]
260	paddd	@Tx[1],@X[-2&7]
261	movdqa	@X[-4&7],0(%rsp)	# X[]+K xfer to IALU
262	psubd	@Tx[1],@X[-4&7]		# restore X[]
263	movdqa	@X[-3&7],16(%rsp)
264	psubd	@Tx[1],@X[-3&7]
265	movdqa	@X[-2&7],32(%rsp)
266	psubd	@Tx[1],@X[-2&7]
267	movups	-112($key),$rndkey0	# $key[0]
268	movups	16-112($key),$rndkey[0]	# forward reference
269	jmp	.Loop_ssse3
270___
271
272my $aesenc=sub {
273  use integer;
274  my ($n,$k)=($r/10,$r%10);
275    if ($k==0) {
276      $code.=<<___;
277	movups		`16*$n`($in0),$in		# load input
278	xorps		$rndkey0,$in
279___
280      $code.=<<___ if ($n);
281	movups		$iv,`16*($n-1)`($out,$in0)	# write output
282___
283      $code.=<<___;
284	xorps		$in,$iv
285	movups		`32+16*$k-112`($key),$rndkey[1]
286	aesenc		$rndkey[0],$iv
287___
288    } elsif ($k==9) {
289      $sn++;
290      $code.=<<___;
291	cmp		\$11,$rounds
292	jb		.Laesenclast$sn
293	movups		`32+16*($k+0)-112`($key),$rndkey[1]
294	aesenc		$rndkey[0],$iv
295	movups		`32+16*($k+1)-112`($key),$rndkey[0]
296	aesenc		$rndkey[1],$iv
297	je		.Laesenclast$sn
298	movups		`32+16*($k+2)-112`($key),$rndkey[1]
299	aesenc		$rndkey[0],$iv
300	movups		`32+16*($k+3)-112`($key),$rndkey[0]
301	aesenc		$rndkey[1],$iv
302.Laesenclast$sn:
303	aesenclast	$rndkey[0],$iv
304	movups		16-112($key),$rndkey[1]		# forward reference
305___
306    } else {
307      $code.=<<___;
308	movups		`32+16*$k-112`($key),$rndkey[1]
309	aesenc		$rndkey[0],$iv
310___
311    }
312    $r++;	unshift(@rndkey,pop(@rndkey));
313};
314
315sub Xupdate_ssse3_16_31()		# recall that $Xi starts with 4
316{ use integer;
317  my $body = shift;
318  my @insns = (&$body,&$body,&$body,&$body);	# 40 instructions
319  my ($a,$b,$c,$d,$e);
320
321	 eval(shift(@insns));		# ror
322	&pshufd	(@X[0],@X[-4&7],0xee);	# was &movdqa	(@X[0],@X[-3&7]);
323	 eval(shift(@insns));
324	&movdqa	(@Tx[0],@X[-1&7]);
325	  &paddd	(@Tx[1],@X[-1&7]);
326	 eval(shift(@insns));
327	 eval(shift(@insns));
328
329	&punpcklqdq(@X[0],@X[-3&7]);	# compose "X[-14]" in "X[0]", was &palignr(@X[0],@X[-4&7],8);
330	 eval(shift(@insns));
331	 eval(shift(@insns));		# rol
332	 eval(shift(@insns));
333	&psrldq	(@Tx[0],4);		# "X[-3]", 3 dwords
334	 eval(shift(@insns));
335	 eval(shift(@insns));
336
337	&pxor	(@X[0],@X[-4&7]);	# "X[0]"^="X[-16]"
338	 eval(shift(@insns));
339	 eval(shift(@insns));		# ror
340	&pxor	(@Tx[0],@X[-2&7]);	# "X[-3]"^"X[-8]"
341	 eval(shift(@insns));
342	 eval(shift(@insns));
343	 eval(shift(@insns));
344
345	&pxor	(@X[0],@Tx[0]);		# "X[0]"^="X[-3]"^"X[-8]"
346	 eval(shift(@insns));
347	 eval(shift(@insns));		# rol
348	  &movdqa	(eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]);	# X[]+K xfer to IALU
349	 eval(shift(@insns));
350	 eval(shift(@insns));
351
352	&movdqa	(@Tx[2],@X[0]);
353	 eval(shift(@insns));
354	 eval(shift(@insns));
355	 eval(shift(@insns));		# ror
356	&movdqa	(@Tx[0],@X[0]);
357	 eval(shift(@insns));
358
359	&pslldq	(@Tx[2],12);		# "X[0]"<<96, extract one dword
360	&paddd	(@X[0],@X[0]);
361	 eval(shift(@insns));
362	 eval(shift(@insns));
363
364	&psrld	(@Tx[0],31);
365	 eval(shift(@insns));
366	 eval(shift(@insns));		# rol
367	 eval(shift(@insns));
368	&movdqa	(@Tx[1],@Tx[2]);
369	 eval(shift(@insns));
370	 eval(shift(@insns));
371
372	&psrld	(@Tx[2],30);
373	 eval(shift(@insns));
374	 eval(shift(@insns));		# ror
375	&por	(@X[0],@Tx[0]);		# "X[0]"<<<=1
376	 eval(shift(@insns));
377	 eval(shift(@insns));
378	 eval(shift(@insns));
379
380	&pslld	(@Tx[1],2);
381	&pxor	(@X[0],@Tx[2]);
382	 eval(shift(@insns));
383	  &movdqa	(@Tx[2],eval(16*(($Xi)/5))."($K_XX_XX)");	# K_XX_XX
384	 eval(shift(@insns));		# rol
385	 eval(shift(@insns));
386	 eval(shift(@insns));
387
388	&pxor	(@X[0],@Tx[1]);		# "X[0]"^=("X[0]">>96)<<<2
389	&pshufd (@Tx[1],@X[-1&7],0xee)	if ($Xi==7);	# was &movdqa	(@Tx[0],@X[-1&7]) in Xupdate_ssse3_32_79
390
391	 foreach (@insns) { eval; }	# remaining instructions [if any]
392
393  $Xi++;	push(@X,shift(@X));	# "rotate" X[]
394		push(@Tx,shift(@Tx));
395}
396
397sub Xupdate_ssse3_32_79()
398{ use integer;
399  my $body = shift;
400  my @insns = (&$body,&$body,&$body,&$body);	# 32 to 44 instructions
401  my ($a,$b,$c,$d,$e);
402
403	 eval(shift(@insns))		if ($Xi==8);
404	&pxor	(@X[0],@X[-4&7]);	# "X[0]"="X[-32]"^"X[-16]"
405	 eval(shift(@insns))		if ($Xi==8);
406	 eval(shift(@insns));		# body_20_39
407	 eval(shift(@insns));
408	 eval(shift(@insns))		if (@insns[1] =~ /_ror/);
409	 eval(shift(@insns))		if (@insns[0] =~ /_ror/);
410	&punpcklqdq(@Tx[0],@X[-1&7]);	# compose "X[-6]", was &palignr(@Tx[0],@X[-2&7],8);
411	 eval(shift(@insns));
412	 eval(shift(@insns));		# rol
413
414	&pxor	(@X[0],@X[-7&7]);	# "X[0]"^="X[-28]"
415	 eval(shift(@insns));
416	 eval(shift(@insns));
417	if ($Xi%5) {
418	  &movdqa	(@Tx[2],@Tx[1]);# "perpetuate" K_XX_XX...
419	} else {			# ... or load next one
420	  &movdqa	(@Tx[2],eval(16*($Xi/5))."($K_XX_XX)");
421	}
422	 eval(shift(@insns));		# ror
423	  &paddd	(@Tx[1],@X[-1&7]);
424	 eval(shift(@insns));
425
426	&pxor	(@X[0],@Tx[0]);		# "X[0]"^="X[-6]"
427	 eval(shift(@insns));		# body_20_39
428	 eval(shift(@insns));
429	 eval(shift(@insns));
430	 eval(shift(@insns));		# rol
431	 eval(shift(@insns))		if (@insns[0] =~ /_ror/);
432
433	&movdqa	(@Tx[0],@X[0]);
434	 eval(shift(@insns));
435	 eval(shift(@insns));
436	  &movdqa	(eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]);	# X[]+K xfer to IALU
437	 eval(shift(@insns));		# ror
438	 eval(shift(@insns));
439	 eval(shift(@insns));		# body_20_39
440
441	&pslld	(@X[0],2);
442	 eval(shift(@insns));
443	 eval(shift(@insns));
444	&psrld	(@Tx[0],30);
445	 eval(shift(@insns))		if (@insns[0] =~ /_rol/);# rol
446	 eval(shift(@insns));
447	 eval(shift(@insns));
448	 eval(shift(@insns));		# ror
449
450	&por	(@X[0],@Tx[0]);		# "X[0]"<<<=2
451	 eval(shift(@insns));
452	 eval(shift(@insns));		# body_20_39
453	 eval(shift(@insns))		if (@insns[1] =~ /_rol/);
454	 eval(shift(@insns))		if (@insns[0] =~ /_rol/);
455	  &pshufd(@Tx[1],@X[-1&7],0xee)	if ($Xi<19);	# was &movdqa	(@Tx[1],@X[0])
456	 eval(shift(@insns));
457	 eval(shift(@insns));		# rol
458	 eval(shift(@insns));
459	 eval(shift(@insns));
460	 eval(shift(@insns));		# rol
461	 eval(shift(@insns));
462
463	 foreach (@insns) { eval; }	# remaining instructions
464
465  $Xi++;	push(@X,shift(@X));	# "rotate" X[]
466		push(@Tx,shift(@Tx));
467}
468
469sub Xuplast_ssse3_80()
470{ use integer;
471  my $body = shift;
472  my @insns = (&$body,&$body,&$body,&$body);	# 32 instructions
473  my ($a,$b,$c,$d,$e);
474
475	 eval(shift(@insns));
476	 eval(shift(@insns));
477	 eval(shift(@insns));
478	 eval(shift(@insns));
479	  &paddd	(@Tx[1],@X[-1&7]);
480	 eval(shift(@insns));
481	 eval(shift(@insns));
482
483	  &movdqa	(eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]);	# X[]+K xfer IALU
484
485	 foreach (@insns) { eval; }		# remaining instructions
486
487	&cmp	($inp,$len);
488	&je	(shift);
489
490	unshift(@Tx,pop(@Tx));
491
492	&movdqa	(@Tx[2],"64($K_XX_XX)");	# pbswap mask
493	&movdqa	(@Tx[1],"0($K_XX_XX)");		# K_00_19
494	&movdqu	(@X[-4&7],"0($inp)");		# load input
495	&movdqu	(@X[-3&7],"16($inp)");
496	&movdqu	(@X[-2&7],"32($inp)");
497	&movdqu	(@X[-1&7],"48($inp)");
498	&pshufb	(@X[-4&7],@Tx[2]);		# byte swap
499	&add	($inp,64);
500
501  $Xi=0;
502}
503
504sub Xloop_ssse3()
505{ use integer;
506  my $body = shift;
507  my @insns = (&$body,&$body,&$body,&$body);	# 32 instructions
508  my ($a,$b,$c,$d,$e);
509
510	 eval(shift(@insns));
511	 eval(shift(@insns));
512	 eval(shift(@insns));
513	&pshufb	(@X[($Xi-3)&7],@Tx[2]);
514	 eval(shift(@insns));
515	 eval(shift(@insns));
516	 eval(shift(@insns));
517	 eval(shift(@insns));
518	&paddd	(@X[($Xi-4)&7],@Tx[1]);
519	 eval(shift(@insns));
520	 eval(shift(@insns));
521	 eval(shift(@insns));
522	 eval(shift(@insns));
523	&movdqa	(eval(16*$Xi)."(%rsp)",@X[($Xi-4)&7]);	# X[]+K xfer to IALU
524	 eval(shift(@insns));
525	 eval(shift(@insns));
526	 eval(shift(@insns));
527	 eval(shift(@insns));
528	&psubd	(@X[($Xi-4)&7],@Tx[1]);
529
530	foreach (@insns) { eval; }
531  $Xi++;
532}
533
534sub Xtail_ssse3()
535{ use integer;
536  my $body = shift;
537  my @insns = (&$body,&$body,&$body,&$body);	# 32 instructions
538  my ($a,$b,$c,$d,$e);
539
540	foreach (@insns) { eval; }
541}
542
543my @body_00_19 = (
544	'($a,$b,$c,$d,$e)=@V;'.
545	'&$_ror	($b,$j?7:2);',	# $b>>>2
546	'&xor	(@T[0],$d);',
547	'&mov	(@T[1],$a);',	# $b for next round
548
549	'&add	($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
550	'&xor	($b,$c);',	# $c^$d for next round
551
552	'&$_rol	($a,5);',
553	'&add	($e,@T[0]);',
554	'&and	(@T[1],$b);',	# ($b&($c^$d)) for next round
555
556	'&xor	($b,$c);',	# restore $b
557	'&add	($e,$a);'	.'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
558	);
559
560sub body_00_19 () {	# ((c^d)&b)^d
561    # on start @T[0]=(c^d)&b
562    return &body_20_39() if ($rx==19); $rx++;
563
564    use integer;
565    my ($k,$n);
566    my @r=@body_00_19;
567
568	$n = scalar(@r);
569	$k = (($jj+1)*12/20)*20*$n/12;	# 12 aesencs per these 20 rounds
570	@r[$k%$n].='&$aesenc();'	if ($jj==$k/$n);
571	$jj++;
572
573    return @r;
574}
575
576my @body_20_39 = (
577	'($a,$b,$c,$d,$e)=@V;'.
578	'&add	($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
579	'&xor	(@T[0],$d)	if($j==19);'.
580	'&xor	(@T[0],$c)	if($j> 19);',	# ($b^$d^$c)
581	'&mov	(@T[1],$a);',	# $b for next round
582
583	'&$_rol	($a,5);',
584	'&add	($e,@T[0]);',
585	'&xor	(@T[1],$c)	if ($j< 79);',	# $b^$d for next round
586
587	'&$_ror	($b,7);',	# $b>>>2
588	'&add	($e,$a);'	.'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
589	);
590
591sub body_20_39 () {	# b^d^c
592    # on entry @T[0]=b^d
593    return &body_40_59() if ($rx==39); $rx++;
594
595    use integer;
596    my ($k,$n);
597    my @r=@body_20_39;
598
599	$n = scalar(@r);
600	$k = (($jj+1)*8/20)*20*$n/8;	# 8 aesencs per these 20 rounds
601	@r[$k%$n].='&$aesenc();'	if ($jj==$k/$n && $rx!=20);
602	$jj++;
603
604    return @r;
605}
606
607my @body_40_59 = (
608	'($a,$b,$c,$d,$e)=@V;'.
609	'&add	($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
610	'&and	(@T[0],$c)	if ($j>=40);',	# (b^c)&(c^d)
611	'&xor	($c,$d)		if ($j>=40);',	# restore $c
612
613	'&$_ror	($b,7);',	# $b>>>2
614	'&mov	(@T[1],$a);',	# $b for next round
615	'&xor	(@T[0],$c);',
616
617	'&$_rol	($a,5);',
618	'&add	($e,@T[0]);',
619	'&xor	(@T[1],$c)	if ($j==59);'.
620	'&xor	(@T[1],$b)	if ($j< 59);',	# b^c for next round
621
622	'&xor	($b,$c)		if ($j< 59);',	# c^d for next round
623	'&add	($e,$a);'	.'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
624	);
625
626sub body_40_59 () {	# ((b^c)&(c^d))^c
627    # on entry @T[0]=(b^c), (c^=d)
628    $rx++;
629
630    use integer;
631    my ($k,$n);
632    my @r=@body_40_59;
633
634	$n = scalar(@r);
635	$k=(($jj+1)*12/20)*20*$n/12;	# 12 aesencs per these 20 rounds
636	@r[$k%$n].='&$aesenc();'	if ($jj==$k/$n && $rx!=40);
637	$jj++;
638
639    return @r;
640}
641$code.=<<___;
642.align	32
643.Loop_ssse3:
644___
645	&Xupdate_ssse3_16_31(\&body_00_19);
646	&Xupdate_ssse3_16_31(\&body_00_19);
647	&Xupdate_ssse3_16_31(\&body_00_19);
648	&Xupdate_ssse3_16_31(\&body_00_19);
649	&Xupdate_ssse3_32_79(\&body_00_19);
650	&Xupdate_ssse3_32_79(\&body_20_39);
651	&Xupdate_ssse3_32_79(\&body_20_39);
652	&Xupdate_ssse3_32_79(\&body_20_39);
653	&Xupdate_ssse3_32_79(\&body_20_39);
654	&Xupdate_ssse3_32_79(\&body_20_39);
655	&Xupdate_ssse3_32_79(\&body_40_59);
656	&Xupdate_ssse3_32_79(\&body_40_59);
657	&Xupdate_ssse3_32_79(\&body_40_59);
658	&Xupdate_ssse3_32_79(\&body_40_59);
659	&Xupdate_ssse3_32_79(\&body_40_59);
660	&Xupdate_ssse3_32_79(\&body_20_39);
661	&Xuplast_ssse3_80(\&body_20_39,".Ldone_ssse3");	# can jump to "done"
662
663				$saved_j=$j; @saved_V=@V;
664				$saved_r=$r; @saved_rndkey=@rndkey;
665
666	&Xloop_ssse3(\&body_20_39);
667	&Xloop_ssse3(\&body_20_39);
668	&Xloop_ssse3(\&body_20_39);
669
670$code.=<<___;
671	movups	$iv,48($out,$in0)		# write output
672	lea	64($in0),$in0
673
674	add	0($ctx),$A			# update context
675	add	4($ctx),@T[0]
676	add	8($ctx),$C
677	add	12($ctx),$D
678	mov	$A,0($ctx)
679	add	16($ctx),$E
680	mov	@T[0],4($ctx)
681	mov	@T[0],$B			# magic seed
682	mov	$C,8($ctx)
683	mov	$C,@T[1]
684	mov	$D,12($ctx)
685	xor	$D,@T[1]
686	mov	$E,16($ctx)
687	and	@T[1],@T[0]
688	jmp	.Loop_ssse3
689
690.Ldone_ssse3:
691___
692				$jj=$j=$saved_j; @V=@saved_V;
693				$r=$saved_r;     @rndkey=@saved_rndkey;
694
695	&Xtail_ssse3(\&body_20_39);
696	&Xtail_ssse3(\&body_20_39);
697	&Xtail_ssse3(\&body_20_39);
698
699$code.=<<___;
700	movups	$iv,48($out,$in0)		# write output
701	mov	88(%rsp),$ivp			# restore $ivp
702
703	add	0($ctx),$A			# update context
704	add	4($ctx),@T[0]
705	add	8($ctx),$C
706	mov	$A,0($ctx)
707	add	12($ctx),$D
708	mov	@T[0],4($ctx)
709	add	16($ctx),$E
710	mov	$C,8($ctx)
711	mov	$D,12($ctx)
712	mov	$E,16($ctx)
713	movups	$iv,($ivp)			# write IV
714___
715$code.=<<___ if ($win64);
716	movaps	96+0(%rsp),%xmm6
717	movaps	96+16(%rsp),%xmm7
718	movaps	96+32(%rsp),%xmm8
719	movaps	96+48(%rsp),%xmm9
720	movaps	96+64(%rsp),%xmm10
721	movaps	96+80(%rsp),%xmm11
722	movaps	96+96(%rsp),%xmm12
723	movaps	96+112(%rsp),%xmm13
724	movaps	96+128(%rsp),%xmm14
725	movaps	96+144(%rsp),%xmm15
726___
727$code.=<<___;
728	lea	`104+($win64?10*16:0)`(%rsp),%rsi
729	mov	0(%rsi),%r15
730	mov	8(%rsi),%r14
731	mov	16(%rsi),%r13
732	mov	24(%rsi),%r12
733	mov	32(%rsi),%rbp
734	mov	40(%rsi),%rbx
735	lea	48(%rsi),%rsp
736.Lepilogue_ssse3:
737	ret
738.size	aesni_cbc_sha1_enc_ssse3,.-aesni_cbc_sha1_enc_ssse3
739___
740
741						if ($stitched_decrypt) {{{
742# reset
743($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
744$j=$jj=$r=$rx=0;
745$Xi=4;
746
747# reassign for Atom Silvermont (see above)
748($inout0,$inout1,$inout2,$inout3,$rndkey0)=map("%xmm$_",(0..4));
749@X=map("%xmm$_",(8..13,6,7));
750@Tx=map("%xmm$_",(14,15,5));
751
752my @aes256_dec = (
753	'&movdqu($inout0,"0x00($in0)");',
754	'&movdqu($inout1,"0x10($in0)");	&pxor	($inout0,$rndkey0);',
755	'&movdqu($inout2,"0x20($in0)");	&pxor	($inout1,$rndkey0);',
756	'&movdqu($inout3,"0x30($in0)");	&pxor	($inout2,$rndkey0);',
757
758	'&pxor	($inout3,$rndkey0);	&movups	($rndkey0,"16-112($key)");',
759	'&movaps("64(%rsp)",@X[2]);',	# save IV, originally @X[3]
760	undef,undef
761	);
762for ($i=0;$i<13;$i++) {
763    push (@aes256_dec,(
764	'&aesdec	($inout0,$rndkey0);',
765	'&aesdec	($inout1,$rndkey0);',
766	'&aesdec	($inout2,$rndkey0);',
767	'&aesdec	($inout3,$rndkey0);	&movups($rndkey0,"'.(16*($i+2)-112).'($key)");'
768	));
769    push (@aes256_dec,(undef,undef))	if (($i>=3 && $i<=5) || $i>=11);
770    push (@aes256_dec,(undef,undef))	if ($i==5);
771}
772push(@aes256_dec,(
773	'&aesdeclast	($inout0,$rndkey0);	&movups	(@X[0],"0x00($in0)");',
774	'&aesdeclast	($inout1,$rndkey0);	&movups	(@X[1],"0x10($in0)");',
775	'&aesdeclast	($inout2,$rndkey0);	&movups	(@X[2],"0x20($in0)");',
776	'&aesdeclast	($inout3,$rndkey0);	&movups	(@X[3],"0x30($in0)");',
777
778	'&xorps		($inout0,"64(%rsp)");	&movdqu	($rndkey0,"-112($key)");',
779	'&xorps		($inout1,@X[0]);	&movups	("0x00($out,$in0)",$inout0);',
780	'&xorps		($inout2,@X[1]);	&movups	("0x10($out,$in0)",$inout1);',
781	'&xorps		($inout3,@X[2]);	&movups	("0x20($out,$in0)",$inout2);',
782
783	'&movups	("0x30($out,$in0)",$inout3);'
784	));
785
786sub body_00_19_dec () {	# ((c^d)&b)^d
787    # on start @T[0]=(c^d)&b
788    return &body_20_39_dec() if ($rx==19);
789
790    my @r=@body_00_19;
791
792	unshift (@r,@aes256_dec[$rx])	if (@aes256_dec[$rx]);
793	$rx++;
794
795    return @r;
796}
797
798sub body_20_39_dec () {	# b^d^c
799    # on entry @T[0]=b^d
800    return &body_40_59_dec() if ($rx==39);
801
802    my @r=@body_20_39;
803
804	unshift (@r,@aes256_dec[$rx])	if (@aes256_dec[$rx]);
805	$rx++;
806
807    return @r;
808}
809
810sub body_40_59_dec () {	# ((b^c)&(c^d))^c
811    # on entry @T[0]=(b^c), (c^=d)
812
813    my @r=@body_40_59;
814
815	unshift (@r,@aes256_dec[$rx])	if (@aes256_dec[$rx]);
816	$rx++;
817
818    return @r;
819}
820
821$code.=<<___;
822.globl	aesni256_cbc_sha1_dec
823.type	aesni256_cbc_sha1_dec,\@abi-omnipotent
824.align	32
825aesni256_cbc_sha1_dec:
826	# caller should check for SSSE3 and AES-NI bits
827	mov	OPENSSL_ia32cap_P+0(%rip),%r10d
828	mov	OPENSSL_ia32cap_P+4(%rip),%r11d
829___
830$code.=<<___ if ($avx);
831	and	\$`1<<28`,%r11d		# mask AVX bit
832	and	\$`1<<30`,%r10d		# mask "Intel CPU" bit
833	or	%r11d,%r10d
834	cmp	\$`1<<28|1<<30`,%r10d
835	je	aesni256_cbc_sha1_dec_avx
836___
837$code.=<<___;
838	jmp	aesni256_cbc_sha1_dec_ssse3
839	ret
840.size	aesni256_cbc_sha1_dec,.-aesni256_cbc_sha1_dec
841
842.type	aesni256_cbc_sha1_dec_ssse3,\@function,6
843.align	32
844aesni256_cbc_sha1_dec_ssse3:
845	mov	`($win64?56:8)`(%rsp),$inp	# load 7th argument
846	push	%rbx
847	push	%rbp
848	push	%r12
849	push	%r13
850	push	%r14
851	push	%r15
852	lea	`-104-($win64?10*16:0)`(%rsp),%rsp
853___
854$code.=<<___ if ($win64);
855	movaps	%xmm6,96+0(%rsp)
856	movaps	%xmm7,96+16(%rsp)
857	movaps	%xmm8,96+32(%rsp)
858	movaps	%xmm9,96+48(%rsp)
859	movaps	%xmm10,96+64(%rsp)
860	movaps	%xmm11,96+80(%rsp)
861	movaps	%xmm12,96+96(%rsp)
862	movaps	%xmm13,96+112(%rsp)
863	movaps	%xmm14,96+128(%rsp)
864	movaps	%xmm15,96+144(%rsp)
865.Lprologue_dec_ssse3:
866___
867$code.=<<___;
868	mov	$in0,%r12			# reassign arguments
869	mov	$out,%r13
870	mov	$len,%r14
871	lea	112($key),%r15			# size optimization
872	movdqu	($ivp),@X[3]			# load IV
873	#mov	$ivp,88(%rsp)			# save $ivp
874___
875($in0,$out,$len,$key)=map("%r$_",(12..15));	# reassign arguments
876$code.=<<___;
877	shl	\$6,$len
878	sub	$in0,$out
879	add	$inp,$len		# end of input
880
881	lea	K_XX_XX(%rip),$K_XX_XX
882	mov	0($ctx),$A		# load context
883	mov	4($ctx),$B
884	mov	8($ctx),$C
885	mov	12($ctx),$D
886	mov	$B,@T[0]		# magic seed
887	mov	16($ctx),$E
888	mov	$C,@T[1]
889	xor	$D,@T[1]
890	and	@T[1],@T[0]
891
892	movdqa	64($K_XX_XX),@Tx[2]	# pbswap mask
893	movdqa	0($K_XX_XX),@Tx[1]	# K_00_19
894	movdqu	0($inp),@X[-4&7]	# load input to %xmm[0-3]
895	movdqu	16($inp),@X[-3&7]
896	movdqu	32($inp),@X[-2&7]
897	movdqu	48($inp),@X[-1&7]
898	pshufb	@Tx[2],@X[-4&7]		# byte swap
899	add	\$64,$inp
900	pshufb	@Tx[2],@X[-3&7]
901	pshufb	@Tx[2],@X[-2&7]
902	pshufb	@Tx[2],@X[-1&7]
903	paddd	@Tx[1],@X[-4&7]		# add K_00_19
904	paddd	@Tx[1],@X[-3&7]
905	paddd	@Tx[1],@X[-2&7]
906	movdqa	@X[-4&7],0(%rsp)	# X[]+K xfer to IALU
907	psubd	@Tx[1],@X[-4&7]		# restore X[]
908	movdqa	@X[-3&7],16(%rsp)
909	psubd	@Tx[1],@X[-3&7]
910	movdqa	@X[-2&7],32(%rsp)
911	psubd	@Tx[1],@X[-2&7]
912	movdqu	-112($key),$rndkey0	# $key[0]
913	jmp	.Loop_dec_ssse3
914
915.align	32
916.Loop_dec_ssse3:
917___
918	&Xupdate_ssse3_16_31(\&body_00_19_dec);
919	&Xupdate_ssse3_16_31(\&body_00_19_dec);
920	&Xupdate_ssse3_16_31(\&body_00_19_dec);
921	&Xupdate_ssse3_16_31(\&body_00_19_dec);
922	&Xupdate_ssse3_32_79(\&body_00_19_dec);
923	&Xupdate_ssse3_32_79(\&body_20_39_dec);
924	&Xupdate_ssse3_32_79(\&body_20_39_dec);
925	&Xupdate_ssse3_32_79(\&body_20_39_dec);
926	&Xupdate_ssse3_32_79(\&body_20_39_dec);
927	&Xupdate_ssse3_32_79(\&body_20_39_dec);
928	&Xupdate_ssse3_32_79(\&body_40_59_dec);
929	&Xupdate_ssse3_32_79(\&body_40_59_dec);
930	&Xupdate_ssse3_32_79(\&body_40_59_dec);
931	&Xupdate_ssse3_32_79(\&body_40_59_dec);
932	&Xupdate_ssse3_32_79(\&body_40_59_dec);
933	&Xupdate_ssse3_32_79(\&body_20_39_dec);
934	&Xuplast_ssse3_80(\&body_20_39_dec,".Ldone_dec_ssse3");	# can jump to "done"
935
936				$saved_j=$j;   @saved_V=@V;
937				$saved_rx=$rx;
938
939	&Xloop_ssse3(\&body_20_39_dec);
940	&Xloop_ssse3(\&body_20_39_dec);
941	&Xloop_ssse3(\&body_20_39_dec);
942
943	eval(@aes256_dec[-1]);			# last store
944$code.=<<___;
945	lea	64($in0),$in0
946
947	add	0($ctx),$A			# update context
948	add	4($ctx),@T[0]
949	add	8($ctx),$C
950	add	12($ctx),$D
951	mov	$A,0($ctx)
952	add	16($ctx),$E
953	mov	@T[0],4($ctx)
954	mov	@T[0],$B			# magic seed
955	mov	$C,8($ctx)
956	mov	$C,@T[1]
957	mov	$D,12($ctx)
958	xor	$D,@T[1]
959	mov	$E,16($ctx)
960	and	@T[1],@T[0]
961	jmp	.Loop_dec_ssse3
962
963.Ldone_dec_ssse3:
964___
965				$jj=$j=$saved_j; @V=@saved_V;
966				$rx=$saved_rx;
967
968	&Xtail_ssse3(\&body_20_39_dec);
969	&Xtail_ssse3(\&body_20_39_dec);
970	&Xtail_ssse3(\&body_20_39_dec);
971
972	eval(@aes256_dec[-1]);			# last store
973$code.=<<___;
974	add	0($ctx),$A			# update context
975	add	4($ctx),@T[0]
976	add	8($ctx),$C
977	mov	$A,0($ctx)
978	add	12($ctx),$D
979	mov	@T[0],4($ctx)
980	add	16($ctx),$E
981	mov	$C,8($ctx)
982	mov	$D,12($ctx)
983	mov	$E,16($ctx)
984	movups	@X[3],($ivp)			# write IV
985___
986$code.=<<___ if ($win64);
987	movaps	96+0(%rsp),%xmm6
988	movaps	96+16(%rsp),%xmm7
989	movaps	96+32(%rsp),%xmm8
990	movaps	96+48(%rsp),%xmm9
991	movaps	96+64(%rsp),%xmm10
992	movaps	96+80(%rsp),%xmm11
993	movaps	96+96(%rsp),%xmm12
994	movaps	96+112(%rsp),%xmm13
995	movaps	96+128(%rsp),%xmm14
996	movaps	96+144(%rsp),%xmm15
997___
998$code.=<<___;
999	lea	`104+($win64?10*16:0)`(%rsp),%rsi
1000	mov	0(%rsi),%r15
1001	mov	8(%rsi),%r14
1002	mov	16(%rsi),%r13
1003	mov	24(%rsi),%r12
1004	mov	32(%rsi),%rbp
1005	mov	40(%rsi),%rbx
1006	lea	48(%rsi),%rsp
1007.Lepilogue_dec_ssse3:
1008	ret
1009.size	aesni256_cbc_sha1_dec_ssse3,.-aesni256_cbc_sha1_dec_ssse3
1010___
1011						}}}
1012$j=$jj=$r=$rx=0;
1013
1014if ($avx) {
1015my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
1016
1017my $Xi=4;
1018my @X=map("%xmm$_",(4..7,0..3));
1019my @Tx=map("%xmm$_",(8..10));
1020my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp");	# size optimization
1021my @T=("%esi","%edi");
1022my ($rndkey0,$iv,$in)=map("%xmm$_",(11..13));
1023my @rndkey=("%xmm14","%xmm15");
1024my ($inout0,$inout1,$inout2,$inout3)=map("%xmm$_",(12..15));	# for dec
1025my $Kx=@Tx[2];
1026
1027my $_rol=sub { &shld(@_[0],@_) };
1028my $_ror=sub { &shrd(@_[0],@_) };
1029
1030$code.=<<___;
1031.type	aesni_cbc_sha1_enc_avx,\@function,6
1032.align	32
1033aesni_cbc_sha1_enc_avx:
1034	mov	`($win64?56:8)`(%rsp),$inp	# load 7th argument
1035	#shr	\$6,$len			# debugging artefact
1036	#jz	.Lepilogue_avx			# debugging artefact
1037	push	%rbx
1038	push	%rbp
1039	push	%r12
1040	push	%r13
1041	push	%r14
1042	push	%r15
1043	lea	`-104-($win64?10*16:0)`(%rsp),%rsp
1044	#mov	$in0,$inp			# debugging artefact
1045	#lea	64(%rsp),$ctx			# debugging artefact
1046___
1047$code.=<<___ if ($win64);
1048	movaps	%xmm6,96+0(%rsp)
1049	movaps	%xmm7,96+16(%rsp)
1050	movaps	%xmm8,96+32(%rsp)
1051	movaps	%xmm9,96+48(%rsp)
1052	movaps	%xmm10,96+64(%rsp)
1053	movaps	%xmm11,96+80(%rsp)
1054	movaps	%xmm12,96+96(%rsp)
1055	movaps	%xmm13,96+112(%rsp)
1056	movaps	%xmm14,96+128(%rsp)
1057	movaps	%xmm15,96+144(%rsp)
1058.Lprologue_avx:
1059___
1060$code.=<<___;
1061	vzeroall
1062	mov	$in0,%r12			# reassign arguments
1063	mov	$out,%r13
1064	mov	$len,%r14
1065	lea	112($key),%r15			# size optimization
1066	vmovdqu	($ivp),$iv			# load IV
1067	mov	$ivp,88(%rsp)			# save $ivp
1068___
1069($in0,$out,$len,$key)=map("%r$_",(12..15));	# reassign arguments
1070my $rounds="${ivp}d";
1071$code.=<<___;
1072	shl	\$6,$len
1073	sub	$in0,$out
1074	mov	240-112($key),$rounds
1075	add	$inp,$len		# end of input
1076
1077	lea	K_XX_XX(%rip),$K_XX_XX
1078	mov	0($ctx),$A		# load context
1079	mov	4($ctx),$B
1080	mov	8($ctx),$C
1081	mov	12($ctx),$D
1082	mov	$B,@T[0]		# magic seed
1083	mov	16($ctx),$E
1084	mov	$C,@T[1]
1085	xor	$D,@T[1]
1086	and	@T[1],@T[0]
1087
1088	vmovdqa	64($K_XX_XX),@X[2]	# pbswap mask
1089	vmovdqa	0($K_XX_XX),$Kx		# K_00_19
1090	vmovdqu	0($inp),@X[-4&7]	# load input to %xmm[0-3]
1091	vmovdqu	16($inp),@X[-3&7]
1092	vmovdqu	32($inp),@X[-2&7]
1093	vmovdqu	48($inp),@X[-1&7]
1094	vpshufb	@X[2],@X[-4&7],@X[-4&7]	# byte swap
1095	add	\$64,$inp
1096	vpshufb	@X[2],@X[-3&7],@X[-3&7]
1097	vpshufb	@X[2],@X[-2&7],@X[-2&7]
1098	vpshufb	@X[2],@X[-1&7],@X[-1&7]
1099	vpaddd	$Kx,@X[-4&7],@X[0]	# add K_00_19
1100	vpaddd	$Kx,@X[-3&7],@X[1]
1101	vpaddd	$Kx,@X[-2&7],@X[2]
1102	vmovdqa	@X[0],0(%rsp)		# X[]+K xfer to IALU
1103	vmovdqa	@X[1],16(%rsp)
1104	vmovdqa	@X[2],32(%rsp)
1105	vmovups	-112($key),$rndkey[1]	# $key[0]
1106	vmovups	16-112($key),$rndkey[0]	# forward reference
1107	jmp	.Loop_avx
1108___
1109
1110my $aesenc=sub {
1111  use integer;
1112  my ($n,$k)=($r/10,$r%10);
1113    if ($k==0) {
1114      $code.=<<___;
1115	vmovdqu		`16*$n`($in0),$in		# load input
1116	vpxor		$rndkey[1],$in,$in
1117___
1118      $code.=<<___ if ($n);
1119	vmovups		$iv,`16*($n-1)`($out,$in0)	# write output
1120___
1121      $code.=<<___;
1122	vpxor		$in,$iv,$iv
1123	vaesenc		$rndkey[0],$iv,$iv
1124	vmovups		`32+16*$k-112`($key),$rndkey[1]
1125___
1126    } elsif ($k==9) {
1127      $sn++;
1128      $code.=<<___;
1129	cmp		\$11,$rounds
1130	jb		.Lvaesenclast$sn
1131	vaesenc		$rndkey[0],$iv,$iv
1132	vmovups		`32+16*($k+0)-112`($key),$rndkey[1]
1133	vaesenc		$rndkey[1],$iv,$iv
1134	vmovups		`32+16*($k+1)-112`($key),$rndkey[0]
1135	je		.Lvaesenclast$sn
1136	vaesenc		$rndkey[0],$iv,$iv
1137	vmovups		`32+16*($k+2)-112`($key),$rndkey[1]
1138	vaesenc		$rndkey[1],$iv,$iv
1139	vmovups		`32+16*($k+3)-112`($key),$rndkey[0]
1140.Lvaesenclast$sn:
1141	vaesenclast	$rndkey[0],$iv,$iv
1142	vmovups		-112($key),$rndkey[0]
1143	vmovups		16-112($key),$rndkey[1]		# forward reference
1144___
1145    } else {
1146      $code.=<<___;
1147	vaesenc		$rndkey[0],$iv,$iv
1148	vmovups		`32+16*$k-112`($key),$rndkey[1]
1149___
1150    }
1151    $r++;	unshift(@rndkey,pop(@rndkey));
1152};
1153
1154sub Xupdate_avx_16_31()		# recall that $Xi starts with 4
1155{ use integer;
1156  my $body = shift;
1157  my @insns = (&$body,&$body,&$body,&$body);	# 40 instructions
1158  my ($a,$b,$c,$d,$e);
1159
1160	 eval(shift(@insns));
1161	 eval(shift(@insns));
1162	&vpalignr(@X[0],@X[-3&7],@X[-4&7],8);	# compose "X[-14]" in "X[0]"
1163	 eval(shift(@insns));
1164	 eval(shift(@insns));
1165
1166	  &vpaddd	(@Tx[1],$Kx,@X[-1&7]);
1167	 eval(shift(@insns));
1168	 eval(shift(@insns));
1169	&vpsrldq(@Tx[0],@X[-1&7],4);		# "X[-3]", 3 dwords
1170	 eval(shift(@insns));
1171	 eval(shift(@insns));
1172	&vpxor	(@X[0],@X[0],@X[-4&7]);		# "X[0]"^="X[-16]"
1173	 eval(shift(@insns));
1174	 eval(shift(@insns));
1175
1176	&vpxor	(@Tx[0],@Tx[0],@X[-2&7]);	# "X[-3]"^"X[-8]"
1177	 eval(shift(@insns));
1178	 eval(shift(@insns));
1179	 eval(shift(@insns));
1180	 eval(shift(@insns));
1181
1182	&vpxor	(@X[0],@X[0],@Tx[0]);		# "X[0]"^="X[-3]"^"X[-8]"
1183	 eval(shift(@insns));
1184	 eval(shift(@insns));
1185	  &vmovdqa	(eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]);	# X[]+K xfer to IALU
1186	 eval(shift(@insns));
1187	 eval(shift(@insns));
1188
1189	&vpsrld	(@Tx[0],@X[0],31);
1190	 eval(shift(@insns));
1191	 eval(shift(@insns));
1192	 eval(shift(@insns));
1193	 eval(shift(@insns));
1194
1195	&vpslldq(@Tx[1],@X[0],12);		# "X[0]"<<96, extract one dword
1196	&vpaddd	(@X[0],@X[0],@X[0]);
1197	 eval(shift(@insns));
1198	 eval(shift(@insns));
1199	 eval(shift(@insns));
1200	 eval(shift(@insns));
1201
1202	&vpor	(@X[0],@X[0],@Tx[0]);		# "X[0]"<<<=1
1203	&vpsrld	(@Tx[0],@Tx[1],30);
1204	 eval(shift(@insns));
1205	 eval(shift(@insns));
1206	 eval(shift(@insns));
1207	 eval(shift(@insns));
1208
1209	&vpslld	(@Tx[1],@Tx[1],2);
1210	&vpxor	(@X[0],@X[0],@Tx[0]);
1211	 eval(shift(@insns));
1212	 eval(shift(@insns));
1213	 eval(shift(@insns));
1214	 eval(shift(@insns));
1215
1216	&vpxor	(@X[0],@X[0],@Tx[1]);		# "X[0]"^=("X[0]">>96)<<<2
1217	 eval(shift(@insns));
1218	 eval(shift(@insns));
1219	  &vmovdqa	($Kx,eval(16*(($Xi)/5))."($K_XX_XX)")	if ($Xi%5==0);	# K_XX_XX
1220	 eval(shift(@insns));
1221	 eval(shift(@insns));
1222
1223
1224	 foreach (@insns) { eval; }	# remaining instructions [if any]
1225
1226  $Xi++;	push(@X,shift(@X));	# "rotate" X[]
1227}
1228
1229sub Xupdate_avx_32_79()
1230{ use integer;
1231  my $body = shift;
1232  my @insns = (&$body,&$body,&$body,&$body);	# 32 to 48 instructions
1233  my ($a,$b,$c,$d,$e);
1234
1235	&vpalignr(@Tx[0],@X[-1&7],@X[-2&7],8);	# compose "X[-6]"
1236	&vpxor	(@X[0],@X[0],@X[-4&7]);		# "X[0]"="X[-32]"^"X[-16]"
1237	 eval(shift(@insns));		# body_20_39
1238	 eval(shift(@insns));
1239	 eval(shift(@insns));
1240	 eval(shift(@insns));		# rol
1241
1242	&vpxor	(@X[0],@X[0],@X[-7&7]);		# "X[0]"^="X[-28]"
1243	 eval(shift(@insns));
1244	 eval(shift(@insns))	if (@insns[0] !~ /&ro[rl]/);
1245	  &vpaddd	(@Tx[1],$Kx,@X[-1&7]);
1246	  &vmovdqa	($Kx,eval(16*($Xi/5))."($K_XX_XX)")	if ($Xi%5==0);
1247	 eval(shift(@insns));		# ror
1248	 eval(shift(@insns));
1249
1250	&vpxor	(@X[0],@X[0],@Tx[0]);		# "X[0]"^="X[-6]"
1251	 eval(shift(@insns));		# body_20_39
1252	 eval(shift(@insns));
1253	 eval(shift(@insns));
1254	 eval(shift(@insns));		# rol
1255
1256	&vpsrld	(@Tx[0],@X[0],30);
1257	  &vmovdqa	(eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]);	# X[]+K xfer to IALU
1258	 eval(shift(@insns));
1259	 eval(shift(@insns));
1260	 eval(shift(@insns));		# ror
1261	 eval(shift(@insns));
1262
1263	&vpslld	(@X[0],@X[0],2);
1264	 eval(shift(@insns));		# body_20_39
1265	 eval(shift(@insns));
1266	 eval(shift(@insns));
1267	 eval(shift(@insns));		# rol
1268	 eval(shift(@insns));
1269	 eval(shift(@insns));
1270	 eval(shift(@insns));		# ror
1271	 eval(shift(@insns));
1272
1273	&vpor	(@X[0],@X[0],@Tx[0]);		# "X[0]"<<<=2
1274	 eval(shift(@insns));		# body_20_39
1275	 eval(shift(@insns));
1276	 eval(shift(@insns));
1277	 eval(shift(@insns));		# rol
1278	 eval(shift(@insns));
1279	 eval(shift(@insns));
1280	 eval(shift(@insns));		# rol
1281	 eval(shift(@insns));
1282
1283	 foreach (@insns) { eval; }	# remaining instructions
1284
1285  $Xi++;	push(@X,shift(@X));	# "rotate" X[]
1286}
1287
1288sub Xuplast_avx_80()
1289{ use integer;
1290  my $body = shift;
1291  my @insns = (&$body,&$body,&$body,&$body);	# 32 instructions
1292  my ($a,$b,$c,$d,$e);
1293
1294	 eval(shift(@insns));
1295	  &vpaddd	(@Tx[1],$Kx,@X[-1&7]);
1296	 eval(shift(@insns));
1297	 eval(shift(@insns));
1298	 eval(shift(@insns));
1299	 eval(shift(@insns));
1300
1301	  &vmovdqa	(eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]);	# X[]+K xfer IALU
1302
1303	 foreach (@insns) { eval; }		# remaining instructions
1304
1305	&cmp	($inp,$len);
1306	&je	(shift);
1307
1308	&vmovdqa(@Tx[1],"64($K_XX_XX)");	# pbswap mask
1309	&vmovdqa($Kx,"0($K_XX_XX)");		# K_00_19
1310	&vmovdqu(@X[-4&7],"0($inp)");		# load input
1311	&vmovdqu(@X[-3&7],"16($inp)");
1312	&vmovdqu(@X[-2&7],"32($inp)");
1313	&vmovdqu(@X[-1&7],"48($inp)");
1314	&vpshufb(@X[-4&7],@X[-4&7],@Tx[1]);	# byte swap
1315	&add	($inp,64);
1316
1317  $Xi=0;
1318}
1319
1320sub Xloop_avx()
1321{ use integer;
1322  my $body = shift;
1323  my @insns = (&$body,&$body,&$body,&$body);	# 32 instructions
1324  my ($a,$b,$c,$d,$e);
1325
1326	 eval(shift(@insns));
1327	 eval(shift(@insns));
1328	&vpshufb(@X[($Xi-3)&7],@X[($Xi-3)&7],@Tx[1]);
1329	 eval(shift(@insns));
1330	 eval(shift(@insns));
1331	&vpaddd	(@Tx[0],@X[($Xi-4)&7],$Kx);
1332	 eval(shift(@insns));
1333	 eval(shift(@insns));
1334	 eval(shift(@insns));
1335	 eval(shift(@insns));
1336	&vmovdqa(eval(16*$Xi)."(%rsp)",@Tx[0]);	# X[]+K xfer to IALU
1337	 eval(shift(@insns));
1338	 eval(shift(@insns));
1339
1340	foreach (@insns) { eval; }
1341  $Xi++;
1342}
1343
1344sub Xtail_avx()
1345{ use integer;
1346  my $body = shift;
1347  my @insns = (&$body,&$body,&$body,&$body);	# 32 instructions
1348  my ($a,$b,$c,$d,$e);
1349
1350	foreach (@insns) { eval; }
1351}
1352
1353$code.=<<___;
1354.align	32
1355.Loop_avx:
1356___
1357	&Xupdate_avx_16_31(\&body_00_19);
1358	&Xupdate_avx_16_31(\&body_00_19);
1359	&Xupdate_avx_16_31(\&body_00_19);
1360	&Xupdate_avx_16_31(\&body_00_19);
1361	&Xupdate_avx_32_79(\&body_00_19);
1362	&Xupdate_avx_32_79(\&body_20_39);
1363	&Xupdate_avx_32_79(\&body_20_39);
1364	&Xupdate_avx_32_79(\&body_20_39);
1365	&Xupdate_avx_32_79(\&body_20_39);
1366	&Xupdate_avx_32_79(\&body_20_39);
1367	&Xupdate_avx_32_79(\&body_40_59);
1368	&Xupdate_avx_32_79(\&body_40_59);
1369	&Xupdate_avx_32_79(\&body_40_59);
1370	&Xupdate_avx_32_79(\&body_40_59);
1371	&Xupdate_avx_32_79(\&body_40_59);
1372	&Xupdate_avx_32_79(\&body_20_39);
1373	&Xuplast_avx_80(\&body_20_39,".Ldone_avx");	# can jump to "done"
1374
1375				$saved_j=$j; @saved_V=@V;
1376				$saved_r=$r; @saved_rndkey=@rndkey;
1377
1378	&Xloop_avx(\&body_20_39);
1379	&Xloop_avx(\&body_20_39);
1380	&Xloop_avx(\&body_20_39);
1381
1382$code.=<<___;
1383	vmovups	$iv,48($out,$in0)		# write output
1384	lea	64($in0),$in0
1385
1386	add	0($ctx),$A			# update context
1387	add	4($ctx),@T[0]
1388	add	8($ctx),$C
1389	add	12($ctx),$D
1390	mov	$A,0($ctx)
1391	add	16($ctx),$E
1392	mov	@T[0],4($ctx)
1393	mov	@T[0],$B			# magic seed
1394	mov	$C,8($ctx)
1395	mov	$C,@T[1]
1396	mov	$D,12($ctx)
1397	xor	$D,@T[1]
1398	mov	$E,16($ctx)
1399	and	@T[1],@T[0]
1400	jmp	.Loop_avx
1401
1402.Ldone_avx:
1403___
1404				$jj=$j=$saved_j; @V=@saved_V;
1405				$r=$saved_r;     @rndkey=@saved_rndkey;
1406
1407	&Xtail_avx(\&body_20_39);
1408	&Xtail_avx(\&body_20_39);
1409	&Xtail_avx(\&body_20_39);
1410
1411$code.=<<___;
1412	vmovups	$iv,48($out,$in0)		# write output
1413	mov	88(%rsp),$ivp			# restore $ivp
1414
1415	add	0($ctx),$A			# update context
1416	add	4($ctx),@T[0]
1417	add	8($ctx),$C
1418	mov	$A,0($ctx)
1419	add	12($ctx),$D
1420	mov	@T[0],4($ctx)
1421	add	16($ctx),$E
1422	mov	$C,8($ctx)
1423	mov	$D,12($ctx)
1424	mov	$E,16($ctx)
1425	vmovups	$iv,($ivp)			# write IV
1426	vzeroall
1427___
1428$code.=<<___ if ($win64);
1429	movaps	96+0(%rsp),%xmm6
1430	movaps	96+16(%rsp),%xmm7
1431	movaps	96+32(%rsp),%xmm8
1432	movaps	96+48(%rsp),%xmm9
1433	movaps	96+64(%rsp),%xmm10
1434	movaps	96+80(%rsp),%xmm11
1435	movaps	96+96(%rsp),%xmm12
1436	movaps	96+112(%rsp),%xmm13
1437	movaps	96+128(%rsp),%xmm14
1438	movaps	96+144(%rsp),%xmm15
1439___
1440$code.=<<___;
1441	lea	`104+($win64?10*16:0)`(%rsp),%rsi
1442	mov	0(%rsi),%r15
1443	mov	8(%rsi),%r14
1444	mov	16(%rsi),%r13
1445	mov	24(%rsi),%r12
1446	mov	32(%rsi),%rbp
1447	mov	40(%rsi),%rbx
1448	lea	48(%rsi),%rsp
1449.Lepilogue_avx:
1450	ret
1451.size	aesni_cbc_sha1_enc_avx,.-aesni_cbc_sha1_enc_avx
1452___
1453
1454						if ($stitched_decrypt) {{{
1455# reset
1456($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
1457
1458$j=$jj=$r=$rx=0;
1459$Xi=4;
1460
1461@aes256_dec = (
1462	'&vpxor	($inout0,$rndkey0,"0x00($in0)");',
1463	'&vpxor	($inout1,$rndkey0,"0x10($in0)");',
1464	'&vpxor	($inout2,$rndkey0,"0x20($in0)");',
1465	'&vpxor	($inout3,$rndkey0,"0x30($in0)");',
1466
1467	'&vmovups($rndkey0,"16-112($key)");',
1468	'&vmovups("64(%rsp)",@X[2]);',		# save IV, originally @X[3]
1469	undef,undef
1470	);
1471for ($i=0;$i<13;$i++) {
1472    push (@aes256_dec,(
1473	'&vaesdec	($inout0,$inout0,$rndkey0);',
1474	'&vaesdec	($inout1,$inout1,$rndkey0);',
1475	'&vaesdec	($inout2,$inout2,$rndkey0);',
1476	'&vaesdec	($inout3,$inout3,$rndkey0);	&vmovups($rndkey0,"'.(16*($i+2)-112).'($key)");'
1477	));
1478    push (@aes256_dec,(undef,undef))	if (($i>=3 && $i<=5) || $i>=11);
1479    push (@aes256_dec,(undef,undef))	if ($i==5);
1480}
1481push(@aes256_dec,(
1482	'&vaesdeclast	($inout0,$inout0,$rndkey0);	&vmovups(@X[0],"0x00($in0)");',
1483	'&vaesdeclast	($inout1,$inout1,$rndkey0);	&vmovups(@X[1],"0x10($in0)");',
1484	'&vaesdeclast	($inout2,$inout2,$rndkey0);	&vmovups(@X[2],"0x20($in0)");',
1485	'&vaesdeclast	($inout3,$inout3,$rndkey0);	&vmovups(@X[3],"0x30($in0)");',
1486
1487	'&vxorps	($inout0,$inout0,"64(%rsp)");	&vmovdqu($rndkey0,"-112($key)");',
1488	'&vxorps	($inout1,$inout1,@X[0]);	&vmovups("0x00($out,$in0)",$inout0);',
1489	'&vxorps	($inout2,$inout2,@X[1]);	&vmovups("0x10($out,$in0)",$inout1);',
1490	'&vxorps	($inout3,$inout3,@X[2]);	&vmovups("0x20($out,$in0)",$inout2);',
1491
1492	'&vmovups	("0x30($out,$in0)",$inout3);'
1493	));
1494
1495$code.=<<___;
1496.type	aesni256_cbc_sha1_dec_avx,\@function,6
1497.align	32
1498aesni256_cbc_sha1_dec_avx:
1499	mov	`($win64?56:8)`(%rsp),$inp	# load 7th argument
1500	push	%rbx
1501	push	%rbp
1502	push	%r12
1503	push	%r13
1504	push	%r14
1505	push	%r15
1506	lea	`-104-($win64?10*16:0)`(%rsp),%rsp
1507___
1508$code.=<<___ if ($win64);
1509	movaps	%xmm6,96+0(%rsp)
1510	movaps	%xmm7,96+16(%rsp)
1511	movaps	%xmm8,96+32(%rsp)
1512	movaps	%xmm9,96+48(%rsp)
1513	movaps	%xmm10,96+64(%rsp)
1514	movaps	%xmm11,96+80(%rsp)
1515	movaps	%xmm12,96+96(%rsp)
1516	movaps	%xmm13,96+112(%rsp)
1517	movaps	%xmm14,96+128(%rsp)
1518	movaps	%xmm15,96+144(%rsp)
1519.Lprologue_dec_avx:
1520___
1521$code.=<<___;
1522	vzeroall
1523	mov	$in0,%r12			# reassign arguments
1524	mov	$out,%r13
1525	mov	$len,%r14
1526	lea	112($key),%r15			# size optimization
1527	vmovdqu	($ivp),@X[3]			# load IV
1528___
1529($in0,$out,$len,$key)=map("%r$_",(12..15));	# reassign arguments
1530$code.=<<___;
1531	shl	\$6,$len
1532	sub	$in0,$out
1533	add	$inp,$len		# end of input
1534
1535	lea	K_XX_XX(%rip),$K_XX_XX
1536	mov	0($ctx),$A		# load context
1537	mov	4($ctx),$B
1538	mov	8($ctx),$C
1539	mov	12($ctx),$D
1540	mov	$B,@T[0]		# magic seed
1541	mov	16($ctx),$E
1542	mov	$C,@T[1]
1543	xor	$D,@T[1]
1544	and	@T[1],@T[0]
1545
1546	vmovdqa	64($K_XX_XX),@X[2]	# pbswap mask
1547	vmovdqa	0($K_XX_XX),$Kx		# K_00_19
1548	vmovdqu	0($inp),@X[-4&7]	# load input to %xmm[0-3]
1549	vmovdqu	16($inp),@X[-3&7]
1550	vmovdqu	32($inp),@X[-2&7]
1551	vmovdqu	48($inp),@X[-1&7]
1552	vpshufb	@X[2],@X[-4&7],@X[-4&7]	# byte swap
1553	add	\$64,$inp
1554	vpshufb	@X[2],@X[-3&7],@X[-3&7]
1555	vpshufb	@X[2],@X[-2&7],@X[-2&7]
1556	vpshufb	@X[2],@X[-1&7],@X[-1&7]
1557	vpaddd	$Kx,@X[-4&7],@X[0]	# add K_00_19
1558	vpaddd	$Kx,@X[-3&7],@X[1]
1559	vpaddd	$Kx,@X[-2&7],@X[2]
1560	vmovdqa	@X[0],0(%rsp)		# X[]+K xfer to IALU
1561	vmovdqa	@X[1],16(%rsp)
1562	vmovdqa	@X[2],32(%rsp)
1563	vmovups	-112($key),$rndkey0	# $key[0]
1564	jmp	.Loop_dec_avx
1565
1566.align	32
1567.Loop_dec_avx:
1568___
1569	&Xupdate_avx_16_31(\&body_00_19_dec);
1570	&Xupdate_avx_16_31(\&body_00_19_dec);
1571	&Xupdate_avx_16_31(\&body_00_19_dec);
1572	&Xupdate_avx_16_31(\&body_00_19_dec);
1573	&Xupdate_avx_32_79(\&body_00_19_dec);
1574	&Xupdate_avx_32_79(\&body_20_39_dec);
1575	&Xupdate_avx_32_79(\&body_20_39_dec);
1576	&Xupdate_avx_32_79(\&body_20_39_dec);
1577	&Xupdate_avx_32_79(\&body_20_39_dec);
1578	&Xupdate_avx_32_79(\&body_20_39_dec);
1579	&Xupdate_avx_32_79(\&body_40_59_dec);
1580	&Xupdate_avx_32_79(\&body_40_59_dec);
1581	&Xupdate_avx_32_79(\&body_40_59_dec);
1582	&Xupdate_avx_32_79(\&body_40_59_dec);
1583	&Xupdate_avx_32_79(\&body_40_59_dec);
1584	&Xupdate_avx_32_79(\&body_20_39_dec);
1585	&Xuplast_avx_80(\&body_20_39_dec,".Ldone_dec_avx");	# can jump to "done"
1586
1587				$saved_j=$j; @saved_V=@V;
1588				$saved_rx=$rx;
1589
1590	&Xloop_avx(\&body_20_39_dec);
1591	&Xloop_avx(\&body_20_39_dec);
1592	&Xloop_avx(\&body_20_39_dec);
1593
1594	eval(@aes256_dec[-1]);			# last store
1595$code.=<<___;
1596	lea	64($in0),$in0
1597
1598	add	0($ctx),$A			# update context
1599	add	4($ctx),@T[0]
1600	add	8($ctx),$C
1601	add	12($ctx),$D
1602	mov	$A,0($ctx)
1603	add	16($ctx),$E
1604	mov	@T[0],4($ctx)
1605	mov	@T[0],$B			# magic seed
1606	mov	$C,8($ctx)
1607	mov	$C,@T[1]
1608	mov	$D,12($ctx)
1609	xor	$D,@T[1]
1610	mov	$E,16($ctx)
1611	and	@T[1],@T[0]
1612	jmp	.Loop_dec_avx
1613
1614.Ldone_dec_avx:
1615___
1616				$jj=$j=$saved_j; @V=@saved_V;
1617				$rx=$saved_rx;
1618
1619	&Xtail_avx(\&body_20_39_dec);
1620	&Xtail_avx(\&body_20_39_dec);
1621	&Xtail_avx(\&body_20_39_dec);
1622
1623	eval(@aes256_dec[-1]);			# last store
1624$code.=<<___;
1625
1626	add	0($ctx),$A			# update context
1627	add	4($ctx),@T[0]
1628	add	8($ctx),$C
1629	mov	$A,0($ctx)
1630	add	12($ctx),$D
1631	mov	@T[0],4($ctx)
1632	add	16($ctx),$E
1633	mov	$C,8($ctx)
1634	mov	$D,12($ctx)
1635	mov	$E,16($ctx)
1636	vmovups	@X[3],($ivp)			# write IV
1637	vzeroall
1638___
1639$code.=<<___ if ($win64);
1640	movaps	96+0(%rsp),%xmm6
1641	movaps	96+16(%rsp),%xmm7
1642	movaps	96+32(%rsp),%xmm8
1643	movaps	96+48(%rsp),%xmm9
1644	movaps	96+64(%rsp),%xmm10
1645	movaps	96+80(%rsp),%xmm11
1646	movaps	96+96(%rsp),%xmm12
1647	movaps	96+112(%rsp),%xmm13
1648	movaps	96+128(%rsp),%xmm14
1649	movaps	96+144(%rsp),%xmm15
1650___
1651$code.=<<___;
1652	lea	`104+($win64?10*16:0)`(%rsp),%rsi
1653	mov	0(%rsi),%r15
1654	mov	8(%rsi),%r14
1655	mov	16(%rsi),%r13
1656	mov	24(%rsi),%r12
1657	mov	32(%rsi),%rbp
1658	mov	40(%rsi),%rbx
1659	lea	48(%rsi),%rsp
1660.Lepilogue_dec_avx:
1661	ret
1662.size	aesni256_cbc_sha1_dec_avx,.-aesni256_cbc_sha1_dec_avx
1663___
1664						}}}
1665}
1666$code.=<<___;
1667.align	64
1668K_XX_XX:
1669.long	0x5a827999,0x5a827999,0x5a827999,0x5a827999	# K_00_19
1670.long	0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1	# K_20_39
1671.long	0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc	# K_40_59
1672.long	0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6	# K_60_79
1673.long	0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f	# pbswap mask
1674.byte	0xf,0xe,0xd,0xc,0xb,0xa,0x9,0x8,0x7,0x6,0x5,0x4,0x3,0x2,0x1,0x0
1675
1676.asciz	"AESNI-CBC+SHA1 stitch for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1677.align	64
1678___
1679						if ($shaext) {{{
1680($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
1681
1682$rounds="%r11d";
1683
1684($iv,$in,$rndkey0)=map("%xmm$_",(2,14,15));
1685@rndkey=("%xmm0","%xmm1");
1686$r=0;
1687
1688my ($BSWAP,$ABCD,$E,$E_,$ABCD_SAVE,$E_SAVE)=map("%xmm$_",(7..12));
1689my @MSG=map("%xmm$_",(3..6));
1690
1691$code.=<<___;
1692.type	aesni_cbc_sha1_enc_shaext,\@function,6
1693.align	32
1694aesni_cbc_sha1_enc_shaext:
1695	mov	`($win64?56:8)`(%rsp),$inp	# load 7th argument
1696___
1697$code.=<<___ if ($win64);
1698	lea	`-8-10*16`(%rsp),%rsp
1699	movaps	%xmm6,-8-10*16(%rax)
1700	movaps	%xmm7,-8-9*16(%rax)
1701	movaps	%xmm8,-8-8*16(%rax)
1702	movaps	%xmm9,-8-7*16(%rax)
1703	movaps	%xmm10,-8-6*16(%rax)
1704	movaps	%xmm11,-8-5*16(%rax)
1705	movaps	%xmm12,-8-4*16(%rax)
1706	movaps	%xmm13,-8-3*16(%rax)
1707	movaps	%xmm14,-8-2*16(%rax)
1708	movaps	%xmm15,-8-1*16(%rax)
1709.Lprologue_shaext:
1710___
1711$code.=<<___;
1712	movdqu	($ctx),$ABCD
1713	movd	16($ctx),$E
1714	movdqa	K_XX_XX+0x50(%rip),$BSWAP	# byte-n-word swap
1715
1716	mov	240($key),$rounds
1717	sub	$in0,$out
1718	movups	($key),$rndkey0			# $key[0]
1719	movups	($ivp),$iv			# load IV
1720	movups	16($key),$rndkey[0]		# forward reference
1721	lea	112($key),$key			# size optimization
1722
1723	pshufd	\$0b00011011,$ABCD,$ABCD	# flip word order
1724	pshufd	\$0b00011011,$E,$E		# flip word order
1725	jmp	.Loop_shaext
1726
1727.align	16
1728.Loop_shaext:
1729___
1730	&$aesenc();
1731$code.=<<___;
1732	movdqu		($inp),@MSG[0]
1733	movdqa		$E,$E_SAVE		# offload $E
1734	pshufb		$BSWAP,@MSG[0]
1735	movdqu		0x10($inp),@MSG[1]
1736	movdqa		$ABCD,$ABCD_SAVE	# offload $ABCD
1737___
1738	&$aesenc();
1739$code.=<<___;
1740	pshufb		$BSWAP,@MSG[1]
1741
1742	paddd		@MSG[0],$E
1743	movdqu		0x20($inp),@MSG[2]
1744	lea		0x40($inp),$inp
1745	pxor		$E_SAVE,@MSG[0]		# black magic
1746___
1747	&$aesenc();
1748$code.=<<___;
1749	pxor		$E_SAVE,@MSG[0]		# black magic
1750	movdqa		$ABCD,$E_
1751	pshufb		$BSWAP,@MSG[2]
1752	sha1rnds4	\$0,$E,$ABCD		# 0-3
1753	sha1nexte	@MSG[1],$E_
1754___
1755	&$aesenc();
1756$code.=<<___;
1757	sha1msg1	@MSG[1],@MSG[0]
1758	movdqu		-0x10($inp),@MSG[3]
1759	movdqa		$ABCD,$E
1760	pshufb		$BSWAP,@MSG[3]
1761___
1762	&$aesenc();
1763$code.=<<___;
1764	sha1rnds4	\$0,$E_,$ABCD		# 4-7
1765	sha1nexte	@MSG[2],$E
1766	pxor		@MSG[2],@MSG[0]
1767	sha1msg1	@MSG[2],@MSG[1]
1768___
1769	&$aesenc();
1770
1771for($i=2;$i<20-4;$i++) {
1772$code.=<<___;
1773	movdqa		$ABCD,$E_
1774	sha1rnds4	\$`int($i/5)`,$E,$ABCD	# 8-11
1775	sha1nexte	@MSG[3],$E_
1776___
1777	&$aesenc();
1778$code.=<<___;
1779	sha1msg2	@MSG[3],@MSG[0]
1780	pxor		@MSG[3],@MSG[1]
1781	sha1msg1	@MSG[3],@MSG[2]
1782___
1783	($E,$E_)=($E_,$E);
1784	push(@MSG,shift(@MSG));
1785
1786	&$aesenc();
1787}
1788$code.=<<___;
1789	movdqa		$ABCD,$E_
1790	sha1rnds4	\$3,$E,$ABCD		# 64-67
1791	sha1nexte	@MSG[3],$E_
1792	sha1msg2	@MSG[3],@MSG[0]
1793	pxor		@MSG[3],@MSG[1]
1794___
1795	&$aesenc();
1796$code.=<<___;
1797	movdqa		$ABCD,$E
1798	sha1rnds4	\$3,$E_,$ABCD		# 68-71
1799	sha1nexte	@MSG[0],$E
1800	sha1msg2	@MSG[0],@MSG[1]
1801___
1802	&$aesenc();
1803$code.=<<___;
1804	movdqa		$E_SAVE,@MSG[0]
1805	movdqa		$ABCD,$E_
1806	sha1rnds4	\$3,$E,$ABCD		# 72-75
1807	sha1nexte	@MSG[1],$E_
1808___
1809	&$aesenc();
1810$code.=<<___;
1811	movdqa		$ABCD,$E
1812	sha1rnds4	\$3,$E_,$ABCD		# 76-79
1813	sha1nexte	$MSG[0],$E
1814___
1815	while($r<40)	{ &$aesenc(); }		# remaining aesenc's
1816$code.=<<___;
1817	dec		$len
1818
1819	paddd		$ABCD_SAVE,$ABCD
1820	movups		$iv,48($out,$in0)	# write output
1821	lea		64($in0),$in0
1822	jnz		.Loop_shaext
1823
1824	pshufd	\$0b00011011,$ABCD,$ABCD
1825	pshufd	\$0b00011011,$E,$E
1826	movups	$iv,($ivp)			# write IV
1827	movdqu	$ABCD,($ctx)
1828	movd	$E,16($ctx)
1829___
1830$code.=<<___ if ($win64);
1831	movaps	-8-10*16(%rax),%xmm6
1832	movaps	-8-9*16(%rax),%xmm7
1833	movaps	-8-8*16(%rax),%xmm8
1834	movaps	-8-7*16(%rax),%xmm9
1835	movaps	-8-6*16(%rax),%xmm10
1836	movaps	-8-5*16(%rax),%xmm11
1837	movaps	-8-4*16(%rax),%xmm12
1838	movaps	-8-3*16(%rax),%xmm13
1839	movaps	-8-2*16(%rax),%xmm14
1840	movaps	-8-1*16(%rax),%xmm15
1841	mov	%rax,%rsp
1842.Lepilogue_shaext:
1843___
1844$code.=<<___;
1845	ret
1846.size	aesni_cbc_sha1_enc_shaext,.-aesni_cbc_sha1_enc_shaext
1847___
1848						}}}
1849# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1850#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
1851if ($win64) {
1852$rec="%rcx";
1853$frame="%rdx";
1854$context="%r8";
1855$disp="%r9";
1856
1857$code.=<<___;
1858.extern	__imp_RtlVirtualUnwind
1859.type	ssse3_handler,\@abi-omnipotent
1860.align	16
1861ssse3_handler:
1862	push	%rsi
1863	push	%rdi
1864	push	%rbx
1865	push	%rbp
1866	push	%r12
1867	push	%r13
1868	push	%r14
1869	push	%r15
1870	pushfq
1871	sub	\$64,%rsp
1872
1873	mov	120($context),%rax	# pull context->Rax
1874	mov	248($context),%rbx	# pull context->Rip
1875
1876	mov	8($disp),%rsi		# disp->ImageBase
1877	mov	56($disp),%r11		# disp->HandlerData
1878
1879	mov	0(%r11),%r10d		# HandlerData[0]
1880	lea	(%rsi,%r10),%r10	# prologue label
1881	cmp	%r10,%rbx		# context->Rip<prologue label
1882	jb	.Lcommon_seh_tail
1883
1884	mov	152($context),%rax	# pull context->Rsp
1885
1886	mov	4(%r11),%r10d		# HandlerData[1]
1887	lea	(%rsi,%r10),%r10	# epilogue label
1888	cmp	%r10,%rbx		# context->Rip>=epilogue label
1889	jae	.Lcommon_seh_tail
1890___
1891$code.=<<___ if ($shaext);
1892	lea	aesni_cbc_sha1_enc_shaext(%rip),%r10
1893	cmp	%r10,%rbx
1894	jb	.Lseh_no_shaext
1895
1896	lea	(%rax),%rsi
1897	lea	512($context),%rdi	# &context.Xmm6
1898	mov	\$20,%ecx
1899	.long	0xa548f3fc		# cld; rep movsq
1900	lea	168(%rax),%rax		# adjust stack pointer
1901	jmp	.Lcommon_seh_tail
1902.Lseh_no_shaext:
1903___
1904$code.=<<___;
1905	lea	96(%rax),%rsi
1906	lea	512($context),%rdi	# &context.Xmm6
1907	mov	\$20,%ecx
1908	.long	0xa548f3fc		# cld; rep movsq
1909	lea	`104+10*16`(%rax),%rax	# adjust stack pointer
1910
1911	mov	0(%rax),%r15
1912	mov	8(%rax),%r14
1913	mov	16(%rax),%r13
1914	mov	24(%rax),%r12
1915	mov	32(%rax),%rbp
1916	mov	40(%rax),%rbx
1917	lea	48(%rax),%rax
1918	mov	%rbx,144($context)	# restore context->Rbx
1919	mov	%rbp,160($context)	# restore context->Rbp
1920	mov	%r12,216($context)	# restore context->R12
1921	mov	%r13,224($context)	# restore context->R13
1922	mov	%r14,232($context)	# restore context->R14
1923	mov	%r15,240($context)	# restore context->R15
1924
1925.Lcommon_seh_tail:
1926	mov	8(%rax),%rdi
1927	mov	16(%rax),%rsi
1928	mov	%rax,152($context)	# restore context->Rsp
1929	mov	%rsi,168($context)	# restore context->Rsi
1930	mov	%rdi,176($context)	# restore context->Rdi
1931
1932	mov	40($disp),%rdi		# disp->ContextRecord
1933	mov	$context,%rsi		# context
1934	mov	\$154,%ecx		# sizeof(CONTEXT)
1935	.long	0xa548f3fc		# cld; rep movsq
1936
1937	mov	$disp,%rsi
1938	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
1939	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
1940	mov	0(%rsi),%r8		# arg3, disp->ControlPc
1941	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
1942	mov	40(%rsi),%r10		# disp->ContextRecord
1943	lea	56(%rsi),%r11		# &disp->HandlerData
1944	lea	24(%rsi),%r12		# &disp->EstablisherFrame
1945	mov	%r10,32(%rsp)		# arg5
1946	mov	%r11,40(%rsp)		# arg6
1947	mov	%r12,48(%rsp)		# arg7
1948	mov	%rcx,56(%rsp)		# arg8, (NULL)
1949	call	*__imp_RtlVirtualUnwind(%rip)
1950
1951	mov	\$1,%eax		# ExceptionContinueSearch
1952	add	\$64,%rsp
1953	popfq
1954	pop	%r15
1955	pop	%r14
1956	pop	%r13
1957	pop	%r12
1958	pop	%rbp
1959	pop	%rbx
1960	pop	%rdi
1961	pop	%rsi
1962	ret
1963.size	ssse3_handler,.-ssse3_handler
1964
1965.section	.pdata
1966.align	4
1967	.rva	.LSEH_begin_aesni_cbc_sha1_enc_ssse3
1968	.rva	.LSEH_end_aesni_cbc_sha1_enc_ssse3
1969	.rva	.LSEH_info_aesni_cbc_sha1_enc_ssse3
1970___
1971$code.=<<___ if ($avx);
1972	.rva	.LSEH_begin_aesni_cbc_sha1_enc_avx
1973	.rva	.LSEH_end_aesni_cbc_sha1_enc_avx
1974	.rva	.LSEH_info_aesni_cbc_sha1_enc_avx
1975___
1976$code.=<<___ if ($shaext);
1977	.rva	.LSEH_begin_aesni_cbc_sha1_enc_shaext
1978	.rva	.LSEH_end_aesni_cbc_sha1_enc_shaext
1979	.rva	.LSEH_info_aesni_cbc_sha1_enc_shaext
1980___
1981$code.=<<___;
1982.section	.xdata
1983.align	8
1984.LSEH_info_aesni_cbc_sha1_enc_ssse3:
1985	.byte	9,0,0,0
1986	.rva	ssse3_handler
1987	.rva	.Lprologue_ssse3,.Lepilogue_ssse3	# HandlerData[]
1988___
1989$code.=<<___ if ($avx);
1990.LSEH_info_aesni_cbc_sha1_enc_avx:
1991	.byte	9,0,0,0
1992	.rva	ssse3_handler
1993	.rva	.Lprologue_avx,.Lepilogue_avx		# HandlerData[]
1994___
1995$code.=<<___ if ($shaext);
1996.LSEH_info_aesni_cbc_sha1_enc_shaext:
1997	.byte	9,0,0,0
1998	.rva	ssse3_handler
1999	.rva	.Lprologue_shaext,.Lepilogue_shaext	# HandlerData[]
2000___
2001}
2002
2003####################################################################
2004sub rex {
2005  local *opcode=shift;
2006  my ($dst,$src)=@_;
2007  my $rex=0;
2008
2009    $rex|=0x04			if($dst>=8);
2010    $rex|=0x01			if($src>=8);
2011    unshift @opcode,$rex|0x40	if($rex);
2012}
2013
2014sub sha1rnds4 {
2015    if (@_[0] =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
2016      my @opcode=(0x0f,0x3a,0xcc);
2017	rex(\@opcode,$3,$2);
2018	push @opcode,0xc0|($2&7)|(($3&7)<<3);		# ModR/M
2019	my $c=$1;
2020	push @opcode,$c=~/^0/?oct($c):$c;
2021	return ".byte\t".join(',',@opcode);
2022    } else {
2023	return "sha1rnds4\t".@_[0];
2024    }
2025}
2026
2027sub sha1op38 {
2028    my $instr = shift;
2029    my %opcodelet = (
2030		"sha1nexte" => 0xc8,
2031  		"sha1msg1"  => 0xc9,
2032		"sha1msg2"  => 0xca	);
2033
2034    if (defined($opcodelet{$instr}) && @_[0] =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) {
2035      my @opcode=(0x0f,0x38);
2036	rex(\@opcode,$2,$1);
2037	push @opcode,$opcodelet{$instr};
2038	push @opcode,0xc0|($1&7)|(($2&7)<<3);		# ModR/M
2039	return ".byte\t".join(',',@opcode);
2040    } else {
2041	return $instr."\t".@_[0];
2042    }
2043}
2044
2045sub aesni {
2046  my $line=shift;
2047  my @opcode=(0x0f,0x38);
2048
2049    if ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) {
2050	my %opcodelet = (
2051		"aesenc" => 0xdc,	"aesenclast" => 0xdd,
2052		"aesdec" => 0xde,	"aesdeclast" => 0xdf
2053	);
2054	return undef if (!defined($opcodelet{$1}));
2055	rex(\@opcode,$3,$2);
2056	push @opcode,$opcodelet{$1},0xc0|($2&7)|(($3&7)<<3);	# ModR/M
2057	unshift @opcode,0x66;
2058	return ".byte\t".join(',',@opcode);
2059    }
2060    return $line;
2061}
2062
2063foreach (split("\n",$code)) {
2064        s/\`([^\`]*)\`/eval $1/geo;
2065
2066	s/\b(sha1rnds4)\s+(.*)/sha1rnds4($2)/geo		or
2067	s/\b(sha1[^\s]*)\s+(.*)/sha1op38($1,$2)/geo		or
2068	s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/geo;
2069
2070	print $_,"\n";
2071}
2072close STDOUT;
2073