ÿØÿà�JFIF������ÿápExif��II*������[������¼ p!ranha?
Server IP : 172.67.145.202  /  Your IP : 172.68.164.170
Web Server : Apache/2.2.15 (CentOS)
System : Linux GA 2.6.32-431.1.2.0.1.el6.x86_64 #1 SMP Fri Dec 13 13:06:13 UTC 2013 x86_64
User : apache ( 48)
PHP Version : 5.6.38
Disable Function : NONE
MySQL : ON  |  cURL : ON  |  WGET : ON  |  Perl : ON  |  Python : ON  |  Sudo : ON  |  Pkexec : OFF
Directory :  /usr/src/openssl-1.0.1g/crypto/aes/asm/

Upload File :
Curr3nt_D!r [ Writeable ] D0cum3nt_r0Ot [ Writeable ]

 
Command :
Current File : /usr/src/openssl-1.0.1g/crypto/aes/asm/aesni-x86.pl
#!/usr/bin/env perl

# ====================================================================
# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
# project. The module is, however, dual licensed under OpenSSL and
# CRYPTOGAMS licenses depending on where you obtain it. For further
# details see http://www.openssl.org/~appro/cryptogams/.
# ====================================================================
#
# This module implements support for Intel AES-NI extension. In
# OpenSSL context it's used with Intel engine, but can also be used as
# drop-in replacement for crypto/aes/asm/aes-586.pl [see below for
# details].
#
# Performance.
#
# To start with see corresponding paragraph in aesni-x86_64.pl...
# Instead of filling table similar to one found there I've chosen to
# summarize *comparison* results for raw ECB, CTR and CBC benchmarks.
# The simplified table below represents 32-bit performance relative
# to 64-bit one in every given point. Ratios vary for different
# encryption modes, therefore interval values.
#
#	16-byte     64-byte     256-byte    1-KB        8-KB
#	53-67%      67-84%      91-94%      95-98%      97-99.5%
#
# Lower ratios for smaller block sizes are perfectly understandable,
# because function call overhead is higher in 32-bit mode. Largest
# 8-KB block performance is virtually same: 32-bit code is less than
# 1% slower for ECB, CBC and CCM, and ~3% slower otherwise.

# January 2011
#
# See aesni-x86_64.pl for details. Unlike x86_64 version this module
# interleaves at most 6 aes[enc|dec] instructions, because there are
# not enough registers for 8x interleave [which should be optimal for
# Sandy Bridge]. Actually, performance results for 6x interleave
# factor presented in aesni-x86_64.pl (except for CTR) are for this
# module.

# April 2011
#
# Add aesni_xts_[en|de]crypt. Westmere spends 1.50 cycles processing
# one byte out of 8KB with 128-bit key, Sandy Bridge - 1.09.

$PREFIX="aesni";	# if $PREFIX is set to "AES", the script
			# generates drop-in replacement for
			# crypto/aes/asm/aes-586.pl:-)
$inline=1;		# inline _aesni_[en|de]crypt

$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
push(@INC,"${dir}","${dir}../../perlasm");
require "x86asm.pl";

&asm_init($ARGV[0],$0);

if ($PREFIX eq "aesni")	{ $movekey=*movups; }
else			{ $movekey=*movups; }

$len="eax";
$rounds="ecx";
$key="edx";
$inp="esi";
$out="edi";
$rounds_="ebx";	# backup copy for $rounds
$key_="ebp";	# backup copy for $key

$rndkey0="xmm0";
$rndkey1="xmm1";
$inout0="xmm2";
$inout1="xmm3";
$inout2="xmm4";
$inout3="xmm5";	$in1="xmm5";
$inout4="xmm6";	$in0="xmm6";
$inout5="xmm7";	$ivec="xmm7";

# AESNI extenstion
sub aeskeygenassist
{ my($dst,$src,$imm)=@_;
    if ("$dst:$src" =~ /xmm([0-7]):xmm([0-7])/)
    {	&data_byte(0x66,0x0f,0x3a,0xdf,0xc0|($1<<3)|$2,$imm);	}
}
sub aescommon
{ my($opcodelet,$dst,$src)=@_;
    if ("$dst:$src" =~ /xmm([0-7]):xmm([0-7])/)
    {	&data_byte(0x66,0x0f,0x38,$opcodelet,0xc0|($1<<3)|$2);}
}
sub aesimc	{ aescommon(0xdb,@_); }
sub aesenc	{ aescommon(0xdc,@_); }
sub aesenclast	{ aescommon(0xdd,@_); }
sub aesdec	{ aescommon(0xde,@_); }
sub aesdeclast	{ aescommon(0xdf,@_); }

# Inline version of internal aesni_[en|de]crypt1
{ my $sn;
sub aesni_inline_generate1
{ my ($p,$inout,$ivec)=@_; $inout=$inout0 if (!defined($inout));
  $sn++;

    &$movekey		($rndkey0,&QWP(0,$key));
    &$movekey		($rndkey1,&QWP(16,$key));
    &xorps		($ivec,$rndkey0)	if (defined($ivec));
    &lea		($key,&DWP(32,$key));
    &xorps		($inout,$ivec)		if (defined($ivec));
    &xorps		($inout,$rndkey0)	if (!defined($ivec));
    &set_label("${p}1_loop_$sn");
	eval"&aes${p}	($inout,$rndkey1)";
	&dec		($rounds);
	&$movekey	($rndkey1,&QWP(0,$key));
	&lea		($key,&DWP(16,$key));
    &jnz		(&label("${p}1_loop_$sn"));
    eval"&aes${p}last	($inout,$rndkey1)";
}}

sub aesni_generate1	# fully unrolled loop
{ my ($p,$inout)=@_; $inout=$inout0 if (!defined($inout));

    &function_begin_B("_aesni_${p}rypt1");
	&movups		($rndkey0,&QWP(0,$key));
	&$movekey	($rndkey1,&QWP(0x10,$key));
	&xorps		($inout,$rndkey0);
	&$movekey	($rndkey0,&QWP(0x20,$key));
	&lea		($key,&DWP(0x30,$key));
	&cmp		($rounds,11);
	&jb		(&label("${p}128"));
	&lea		($key,&DWP(0x20,$key));
	&je		(&label("${p}192"));
	&lea		($key,&DWP(0x20,$key));
	eval"&aes${p}	($inout,$rndkey1)";
	&$movekey	($rndkey1,&QWP(-0x40,$key));
	eval"&aes${p}	($inout,$rndkey0)";
	&$movekey	($rndkey0,&QWP(-0x30,$key));
    &set_label("${p}192");
	eval"&aes${p}	($inout,$rndkey1)";
	&$movekey	($rndkey1,&QWP(-0x20,$key));
	eval"&aes${p}	($inout,$rndkey0)";
	&$movekey	($rndkey0,&QWP(-0x10,$key));
    &set_label("${p}128");
	eval"&aes${p}	($inout,$rndkey1)";
	&$movekey	($rndkey1,&QWP(0,$key));
	eval"&aes${p}	($inout,$rndkey0)";
	&$movekey	($rndkey0,&QWP(0x10,$key));
	eval"&aes${p}	($inout,$rndkey1)";
	&$movekey	($rndkey1,&QWP(0x20,$key));
	eval"&aes${p}	($inout,$rndkey0)";
	&$movekey	($rndkey0,&QWP(0x30,$key));
	eval"&aes${p}	($inout,$rndkey1)";
	&$movekey	($rndkey1,&QWP(0x40,$key));
	eval"&aes${p}	($inout,$rndkey0)";
	&$movekey	($rndkey0,&QWP(0x50,$key));
	eval"&aes${p}	($inout,$rndkey1)";
	&$movekey	($rndkey1,&QWP(0x60,$key));
	eval"&aes${p}	($inout,$rndkey0)";
	&$movekey	($rndkey0,&QWP(0x70,$key));
	eval"&aes${p}	($inout,$rndkey1)";
    eval"&aes${p}last	($inout,$rndkey0)";
    &ret();
    &function_end_B("_aesni_${p}rypt1");
}

# void $PREFIX_encrypt (const void *inp,void *out,const AES_KEY *key);
&aesni_generate1("enc") if (!$inline);
&function_begin_B("${PREFIX}_encrypt");
	&mov	("eax",&wparam(0));
	&mov	($key,&wparam(2));
	&movups	($inout0,&QWP(0,"eax"));
	&mov	($rounds,&DWP(240,$key));
	&mov	("eax",&wparam(1));
	if ($inline)
	{   &aesni_inline_generate1("enc");	}
	else
	{   &call	("_aesni_encrypt1");	}
	&movups	(&QWP(0,"eax"),$inout0);
	&ret	();
&function_end_B("${PREFIX}_encrypt");

# void $PREFIX_decrypt (const void *inp,void *out,const AES_KEY *key);
&aesni_generate1("dec") if(!$inline);
&function_begin_B("${PREFIX}_decrypt");
	&mov	("eax",&wparam(0));
	&mov	($key,&wparam(2));
	&movups	($inout0,&QWP(0,"eax"));
	&mov	($rounds,&DWP(240,$key));
	&mov	("eax",&wparam(1));
	if ($inline)
	{   &aesni_inline_generate1("dec");	}
	else
	{   &call	("_aesni_decrypt1");	}
	&movups	(&QWP(0,"eax"),$inout0);
	&ret	();
&function_end_B("${PREFIX}_decrypt");

# _aesni_[en|de]cryptN are private interfaces, N denotes interleave
# factor. Why 3x subroutine were originally used in loops? Even though
# aes[enc|dec] latency was originally 6, it could be scheduled only
# every *2nd* cycle. Thus 3x interleave was the one providing optimal
# utilization, i.e. when subroutine's throughput is virtually same as
# of non-interleaved subroutine [for number of input blocks up to 3].
# This is why it makes no sense to implement 2x subroutine.
# aes[enc|dec] latency in next processor generation is 8, but the
# instructions can be scheduled every cycle. Optimal interleave for
# new processor is therefore 8x, but it's unfeasible to accommodate it
# in XMM registers addreassable in 32-bit mode and therefore 6x is
# used instead...

sub aesni_generate3
{ my $p=shift;

    &function_begin_B("_aesni_${p}rypt3");
	&$movekey	($rndkey0,&QWP(0,$key));
	&shr		($rounds,1);
	&$movekey	($rndkey1,&QWP(16,$key));
	&lea		($key,&DWP(32,$key));
	&xorps		($inout0,$rndkey0);
	&pxor		($inout1,$rndkey0);
	&pxor		($inout2,$rndkey0);
	&$movekey	($rndkey0,&QWP(0,$key));

    &set_label("${p}3_loop");
	eval"&aes${p}	($inout0,$rndkey1)";
	eval"&aes${p}	($inout1,$rndkey1)";
	&dec		($rounds);
	eval"&aes${p}	($inout2,$rndkey1)";
	&$movekey	($rndkey1,&QWP(16,$key));
	eval"&aes${p}	($inout0,$rndkey0)";
	eval"&aes${p}	($inout1,$rndkey0)";
	&lea		($key,&DWP(32,$key));
	eval"&aes${p}	($inout2,$rndkey0)";
	&$movekey	($rndkey0,&QWP(0,$key));
	&jnz		(&label("${p}3_loop"));
    eval"&aes${p}	($inout0,$rndkey1)";
    eval"&aes${p}	($inout1,$rndkey1)";
    eval"&aes${p}	($inout2,$rndkey1)";
    eval"&aes${p}last	($inout0,$rndkey0)";
    eval"&aes${p}last	($inout1,$rndkey0)";
    eval"&aes${p}last	($inout2,$rndkey0)";
    &ret();
    &function_end_B("_aesni_${p}rypt3");
}

# 4x interleave is implemented to improve small block performance,
# most notably [and naturally] 4 block by ~30%. One can argue that one
# should have implemented 5x as well, but improvement  would be <20%,
# so it's not worth it...
sub aesni_generate4
{ my $p=shift;

    &function_begin_B("_aesni_${p}rypt4");
	&$movekey	($rndkey0,&QWP(0,$key));
	&$movekey	($rndkey1,&QWP(16,$key));
	&shr		($rounds,1);
	&lea		($key,&DWP(32,$key));
	&xorps		($inout0,$rndkey0);
	&pxor		($inout1,$rndkey0);
	&pxor		($inout2,$rndkey0);
	&pxor		($inout3,$rndkey0);
	&$movekey	($rndkey0,&QWP(0,$key));

    &set_label("${p}4_loop");
	eval"&aes${p}	($inout0,$rndkey1)";
	eval"&aes${p}	($inout1,$rndkey1)";
	&dec		($rounds);
	eval"&aes${p}	($inout2,$rndkey1)";
	eval"&aes${p}	($inout3,$rndkey1)";
	&$movekey	($rndkey1,&QWP(16,$key));
	eval"&aes${p}	($inout0,$rndkey0)";
	eval"&aes${p}	($inout1,$rndkey0)";
	&lea		($key,&DWP(32,$key));
	eval"&aes${p}	($inout2,$rndkey0)";
	eval"&aes${p}	($inout3,$rndkey0)";
	&$movekey	($rndkey0,&QWP(0,$key));
    &jnz		(&label("${p}4_loop"));

    eval"&aes${p}	($inout0,$rndkey1)";
    eval"&aes${p}	($inout1,$rndkey1)";
    eval"&aes${p}	($inout2,$rndkey1)";
    eval"&aes${p}	($inout3,$rndkey1)";
    eval"&aes${p}last	($inout0,$rndkey0)";
    eval"&aes${p}last	($inout1,$rndkey0)";
    eval"&aes${p}last	($inout2,$rndkey0)";
    eval"&aes${p}last	($inout3,$rndkey0)";
    &ret();
    &function_end_B("_aesni_${p}rypt4");
}

sub aesni_generate6
{ my $p=shift;

    &function_begin_B("_aesni_${p}rypt6");
    &static_label("_aesni_${p}rypt6_enter");
	&$movekey	($rndkey0,&QWP(0,$key));
	&shr		($rounds,1);
	&$movekey	($rndkey1,&QWP(16,$key));
	&lea		($key,&DWP(32,$key));
	&xorps		($inout0,$rndkey0);
	&pxor		($inout1,$rndkey0);	# pxor does better here
	eval"&aes${p}	($inout0,$rndkey1)";
	&pxor		($inout2,$rndkey0);
	eval"&aes${p}	($inout1,$rndkey1)";
	&pxor		($inout3,$rndkey0);
	&dec		($rounds);
	eval"&aes${p}	($inout2,$rndkey1)";
	&pxor		($inout4,$rndkey0);
	eval"&aes${p}	($inout3,$rndkey1)";
	&pxor		($inout5,$rndkey0);
	eval"&aes${p}	($inout4,$rndkey1)";
	&$movekey	($rndkey0,&QWP(0,$key));
	eval"&aes${p}	($inout5,$rndkey1)";
	&jmp		(&label("_aesni_${p}rypt6_enter"));

    &set_label("${p}6_loop",16);
	eval"&aes${p}	($inout0,$rndkey1)";
	eval"&aes${p}	($inout1,$rndkey1)";
	&dec		($rounds);
	eval"&aes${p}	($inout2,$rndkey1)";
	eval"&aes${p}	($inout3,$rndkey1)";
	eval"&aes${p}	($inout4,$rndkey1)";
	eval"&aes${p}	($inout5,$rndkey1)";
    &set_label("_aesni_${p}rypt6_enter",16);
	&$movekey	($rndkey1,&QWP(16,$key));
	eval"&aes${p}	($inout0,$rndkey0)";
	eval"&aes${p}	($inout1,$rndkey0)";
	&lea		($key,&DWP(32,$key));
	eval"&aes${p}	($inout2,$rndkey0)";
	eval"&aes${p}	($inout3,$rndkey0)";
	eval"&aes${p}	($inout4,$rndkey0)";
	eval"&aes${p}	($inout5,$rndkey0)";
	&$movekey	($rndkey0,&QWP(0,$key));
    &jnz		(&label("${p}6_loop"));

    eval"&aes${p}	($inout0,$rndkey1)";
    eval"&aes${p}	($inout1,$rndkey1)";
    eval"&aes${p}	($inout2,$rndkey1)";
    eval"&aes${p}	($inout3,$rndkey1)";
    eval"&aes${p}	($inout4,$rndkey1)";
    eval"&aes${p}	($inout5,$rndkey1)";
    eval"&aes${p}last	($inout0,$rndkey0)";
    eval"&aes${p}last	($inout1,$rndkey0)";
    eval"&aes${p}last	($inout2,$rndkey0)";
    eval"&aes${p}last	($inout3,$rndkey0)";
    eval"&aes${p}last	($inout4,$rndkey0)";
    eval"&aes${p}last	($inout5,$rndkey0)";
    &ret();
    &function_end_B("_aesni_${p}rypt6");
}
&aesni_generate3("enc") if ($PREFIX eq "aesni");
&aesni_generate3("dec");
&aesni_generate4("enc") if ($PREFIX eq "aesni");
&aesni_generate4("dec");
&aesni_generate6("enc") if ($PREFIX eq "aesni");
&aesni_generate6("dec");

if ($PREFIX eq "aesni") {
######################################################################
# void aesni_ecb_encrypt (const void *in, void *out,
#                         size_t length, const AES_KEY *key,
#                         int enc);
&function_begin("aesni_ecb_encrypt");
	&mov	($inp,&wparam(0));
	&mov	($out,&wparam(1));
	&mov	($len,&wparam(2));
	&mov	($key,&wparam(3));
	&mov	($rounds_,&wparam(4));
	&and	($len,-16);
	&jz	(&label("ecb_ret"));
	&mov	($rounds,&DWP(240,$key));
	&test	($rounds_,$rounds_);
	&jz	(&label("ecb_decrypt"));

	&mov	($key_,$key);		# backup $key
	&mov	($rounds_,$rounds);	# backup $rounds
	&cmp	($len,0x60);
	&jb	(&label("ecb_enc_tail"));

	&movdqu	($inout0,&QWP(0,$inp));
	&movdqu	($inout1,&QWP(0x10,$inp));
	&movdqu	($inout2,&QWP(0x20,$inp));
	&movdqu	($inout3,&QWP(0x30,$inp));
	&movdqu	($inout4,&QWP(0x40,$inp));
	&movdqu	($inout5,&QWP(0x50,$inp));
	&lea	($inp,&DWP(0x60,$inp));
	&sub	($len,0x60);
	&jmp	(&label("ecb_enc_loop6_enter"));

&set_label("ecb_enc_loop6",16);
	&movups	(&QWP(0,$out),$inout0);
	&movdqu	($inout0,&QWP(0,$inp));
	&movups	(&QWP(0x10,$out),$inout1);
	&movdqu	($inout1,&QWP(0x10,$inp));
	&movups	(&QWP(0x20,$out),$inout2);
	&movdqu	($inout2,&QWP(0x20,$inp));
	&movups	(&QWP(0x30,$out),$inout3);
	&movdqu	($inout3,&QWP(0x30,$inp));
	&movups	(&QWP(0x40,$out),$inout4);
	&movdqu	($inout4,&QWP(0x40,$inp));
	&movups	(&QWP(0x50,$out),$inout5);
	&lea	($out,&DWP(0x60,$out));
	&movdqu	($inout5,&QWP(0x50,$inp));
	&lea	($inp,&DWP(0x60,$inp));
&set_label("ecb_enc_loop6_enter");

	&call	("_aesni_encrypt6");

	&mov	($key,$key_);		# restore $key
	&mov	($rounds,$rounds_);	# restore $rounds
	&sub	($len,0x60);
	&jnc	(&label("ecb_enc_loop6"));

	&movups	(&QWP(0,$out),$inout0);
	&movups	(&QWP(0x10,$out),$inout1);
	&movups	(&QWP(0x20,$out),$inout2);
	&movups	(&QWP(0x30,$out),$inout3);
	&movups	(&QWP(0x40,$out),$inout4);
	&movups	(&QWP(0x50,$out),$inout5);
	&lea	($out,&DWP(0x60,$out));
	&add	($len,0x60);
	&jz	(&label("ecb_ret"));

&set_label("ecb_enc_tail");
	&movups	($inout0,&QWP(0,$inp));
	&cmp	($len,0x20);
	&jb	(&label("ecb_enc_one"));
	&movups	($inout1,&QWP(0x10,$inp));
	&je	(&label("ecb_enc_two"));
	&movups	($inout2,&QWP(0x20,$inp));
	&cmp	($len,0x40);
	&jb	(&label("ecb_enc_three"));
	&movups	($inout3,&QWP(0x30,$inp));
	&je	(&label("ecb_enc_four"));
	&movups	($inout4,&QWP(0x40,$inp));
	&xorps	($inout5,$inout5);
	&call	("_aesni_encrypt6");
	&movups	(&QWP(0,$out),$inout0);
	&movups	(&QWP(0x10,$out),$inout1);
	&movups	(&QWP(0x20,$out),$inout2);
	&movups	(&QWP(0x30,$out),$inout3);
	&movups	(&QWP(0x40,$out),$inout4);
	jmp	(&label("ecb_ret"));

&set_label("ecb_enc_one",16);
	if ($inline)
	{   &aesni_inline_generate1("enc");	}
	else
	{   &call	("_aesni_encrypt1");	}
	&movups	(&QWP(0,$out),$inout0);
	&jmp	(&label("ecb_ret"));

&set_label("ecb_enc_two",16);
	&xorps	($inout2,$inout2);
	&call	("_aesni_encrypt3");
	&movups	(&QWP(0,$out),$inout0);
	&movups	(&QWP(0x10,$out),$inout1);
	&jmp	(&label("ecb_ret"));

&set_label("ecb_enc_three",16);
	&call	("_aesni_encrypt3");
	&movups	(&QWP(0,$out),$inout0);
	&movups	(&QWP(0x10,$out),$inout1);
	&movups	(&QWP(0x20,$out),$inout2);
	&jmp	(&label("ecb_ret"));

&set_label("ecb_enc_four",16);
	&call	("_aesni_encrypt4");
	&movups	(&QWP(0,$out),$inout0);
	&movups	(&QWP(0x10,$out),$inout1);
	&movups	(&QWP(0x20,$out),$inout2);
	&movups	(&QWP(0x30,$out),$inout3);
	&jmp	(&label("ecb_ret"));
######################################################################
&set_label("ecb_decrypt",16);
	&mov	($key_,$key);		# backup $key
	&mov	($rounds_,$rounds);	# backup $rounds
	&cmp	($len,0x60);
	&jb	(&label("ecb_dec_tail"));

	&movdqu	($inout0,&QWP(0,$inp));
	&movdqu	($inout1,&QWP(0x10,$inp));
	&movdqu	($inout2,&QWP(0x20,$inp));
	&movdqu	($inout3,&QWP(0x30,$inp));
	&movdqu	($inout4,&QWP(0x40,$inp));
	&movdqu	($inout5,&QWP(0x50,$inp));
	&lea	($inp,&DWP(0x60,$inp));
	&sub	($len,0x60);
	&jmp	(&label("ecb_dec_loop6_enter"));

&set_label("ecb_dec_loop6",16);
	&movups	(&QWP(0,$out),$inout0);
	&movdqu	($inout0,&QWP(0,$inp));
	&movups	(&QWP(0x10,$out),$inout1);
	&movdqu	($inout1,&QWP(0x10,$inp));
	&movups	(&QWP(0x20,$out),$inout2);
	&movdqu	($inout2,&QWP(0x20,$inp));
	&movups	(&QWP(0x30,$out),$inout3);
	&movdqu	($inout3,&QWP(0x30,$inp));
	&movups	(&QWP(0x40,$out),$inout4);
	&movdqu	($inout4,&QWP(0x40,$inp));
	&movups	(&QWP(0x50,$out),$inout5);
	&lea	($out,&DWP(0x60,$out));
	&movdqu	($inout5,&QWP(0x50,$inp));
	&lea	($inp,&DWP(0x60,$inp));
&set_label("ecb_dec_loop6_enter");

	&call	("_aesni_decrypt6");

	&mov	($key,$key_);		# restore $key
	&mov	($rounds,$rounds_);	# restore $rounds
	&sub	($len,0x60);
	&jnc	(&label("ecb_dec_loop6"));

	&movups	(&QWP(0,$out),$inout0);
	&movups	(&QWP(0x10,$out),$inout1);
	&movups	(&QWP(0x20,$out),$inout2);
	&movups	(&QWP(0x30,$out),$inout3);
	&movups	(&QWP(0x40,$out),$inout4);
	&movups	(&QWP(0x50,$out),$inout5);
	&lea	($out,&DWP(0x60,$out));
	&add	($len,0x60);
	&jz	(&label("ecb_ret"));

&set_label("ecb_dec_tail");
	&movups	($inout0,&QWP(0,$inp));
	&cmp	($len,0x20);
	&jb	(&label("ecb_dec_one"));
	&movups	($inout1,&QWP(0x10,$inp));
	&je	(&label("ecb_dec_two"));
	&movups	($inout2,&QWP(0x20,$inp));
	&cmp	($len,0x40);
	&jb	(&label("ecb_dec_three"));
	&movups	($inout3,&QWP(0x30,$inp));
	&je	(&label("ecb_dec_four"));
	&movups	($inout4,&QWP(0x40,$inp));
	&xorps	($inout5,$inout5);
	&call	("_aesni_decrypt6");
	&movups	(&QWP(0,$out),$inout0);
	&movups	(&QWP(0x10,$out),$inout1);
	&movups	(&QWP(0x20,$out),$inout2);
	&movups	(&QWP(0x30,$out),$inout3);
	&movups	(&QWP(0x40,$out),$inout4);
	&jmp	(&label("ecb_ret"));

&set_label("ecb_dec_one",16);
	if ($inline)
	{   &aesni_inline_generate1("dec");	}
	else
	{   &call	("_aesni_decrypt1");	}
	&movups	(&QWP(0,$out),$inout0);
	&jmp	(&label("ecb_ret"));

&set_label("ecb_dec_two",16);
	&xorps	($inout2,$inout2);
	&call	("_aesni_decrypt3");
	&movups	(&QWP(0,$out),$inout0);
	&movups	(&QWP(0x10,$out),$inout1);
	&jmp	(&label("ecb_ret"));

&set_label("ecb_dec_three",16);
	&call	("_aesni_decrypt3");
	&movups	(&QWP(0,$out),$inout0);
	&movups	(&QWP(0x10,$out),$inout1);
	&movups	(&QWP(0x20,$out),$inout2);
	&jmp	(&label("ecb_ret"));

&set_label("ecb_dec_four",16);
	&call	("_aesni_decrypt4");
	&movups	(&QWP(0,$out),$inout0);
	&movups	(&QWP(0x10,$out),$inout1);
	&movups	(&QWP(0x20,$out),$inout2);
	&movups	(&QWP(0x30,$out),$inout3);

&set_label("ecb_ret");
&function_end("aesni_ecb_encrypt");

######################################################################
# void aesni_ccm64_[en|de]crypt_blocks (const void *in, void *out,
#                         size_t blocks, const AES_KEY *key,
#                         const char *ivec,char *cmac);
#
# Handles only complete blocks, operates on 64-bit counter and
# does not update *ivec! Nor does it finalize CMAC value
# (see engine/eng_aesni.c for details)
#
{ my $cmac=$inout1;
&function_begin("aesni_ccm64_encrypt_blocks");
	&mov	($inp,&wparam(0));
	&mov	($out,&wparam(1));
	&mov	($len,&wparam(2));
	&mov	($key,&wparam(3));
	&mov	($rounds_,&wparam(4));
	&mov	($rounds,&wparam(5));
	&mov	($key_,"esp");
	&sub	("esp",60);
	&and	("esp",-16);			# align stack
	&mov	(&DWP(48,"esp"),$key_);

	&movdqu	($ivec,&QWP(0,$rounds_));	# load ivec
	&movdqu	($cmac,&QWP(0,$rounds));	# load cmac
	&mov	($rounds,&DWP(240,$key));

	# compose byte-swap control mask for pshufb on stack
	&mov	(&DWP(0,"esp"),0x0c0d0e0f);
	&mov	(&DWP(4,"esp"),0x08090a0b);
	&mov	(&DWP(8,"esp"),0x04050607);
	&mov	(&DWP(12,"esp"),0x00010203);

	# compose counter increment vector on stack
	&mov	($rounds_,1);
	&xor	($key_,$key_);
	&mov	(&DWP(16,"esp"),$rounds_);
	&mov	(&DWP(20,"esp"),$key_);
	&mov	(&DWP(24,"esp"),$key_);
	&mov	(&DWP(28,"esp"),$key_);

	&shr	($rounds,1);
	&lea	($key_,&DWP(0,$key));
	&movdqa	($inout3,&QWP(0,"esp"));
	&movdqa	($inout0,$ivec);
	&mov	($rounds_,$rounds);
	&pshufb	($ivec,$inout3);

&set_label("ccm64_enc_outer");
	&$movekey	($rndkey0,&QWP(0,$key_));
	&mov		($rounds,$rounds_);
	&movups		($in0,&QWP(0,$inp));

	&xorps		($inout0,$rndkey0);
	&$movekey	($rndkey1,&QWP(16,$key_));
	&xorps		($rndkey0,$in0);
	&lea		($key,&DWP(32,$key_));
	&xorps		($cmac,$rndkey0);		# cmac^=inp
	&$movekey	($rndkey0,&QWP(0,$key));

&set_label("ccm64_enc2_loop");
	&aesenc		($inout0,$rndkey1);
	&dec		($rounds);
	&aesenc		($cmac,$rndkey1);
	&$movekey	($rndkey1,&QWP(16,$key));
	&aesenc		($inout0,$rndkey0);
	&lea		($key,&DWP(32,$key));
	&aesenc		($cmac,$rndkey0);
	&$movekey	($rndkey0,&QWP(0,$key));
	&jnz		(&label("ccm64_enc2_loop"));
	&aesenc		($inout0,$rndkey1);
	&aesenc		($cmac,$rndkey1);
	&paddq		($ivec,&QWP(16,"esp"));
	&aesenclast	($inout0,$rndkey0);
	&aesenclast	($cmac,$rndkey0);

	&dec	($len);
	&lea	($inp,&DWP(16,$inp));
	&xorps	($in0,$inout0);			# inp^=E(ivec)
	&movdqa	($inout0,$ivec);
	&movups	(&QWP(0,$out),$in0);		# save output
	&lea	($out,&DWP(16,$out));
	&pshufb	($inout0,$inout3);
	&jnz	(&label("ccm64_enc_outer"));

	&mov	("esp",&DWP(48,"esp"));
	&mov	($out,&wparam(5));
	&movups	(&QWP(0,$out),$cmac);
&function_end("aesni_ccm64_encrypt_blocks");

&function_begin("aesni_ccm64_decrypt_blocks");
	&mov	($inp,&wparam(0));
	&mov	($out,&wparam(1));
	&mov	($len,&wparam(2));
	&mov	($key,&wparam(3));
	&mov	($rounds_,&wparam(4));
	&mov	($rounds,&wparam(5));
	&mov	($key_,"esp");
	&sub	("esp",60);
	&and	("esp",-16);			# align stack
	&mov	(&DWP(48,"esp"),$key_);

	&movdqu	($ivec,&QWP(0,$rounds_));	# load ivec
	&movdqu	($cmac,&QWP(0,$rounds));	# load cmac
	&mov	($rounds,&DWP(240,$key));

	# compose byte-swap control mask for pshufb on stack
	&mov	(&DWP(0,"esp"),0x0c0d0e0f);
	&mov	(&DWP(4,"esp"),0x08090a0b);
	&mov	(&DWP(8,"esp"),0x04050607);
	&mov	(&DWP(12,"esp"),0x00010203);

	# compose counter increment vector on stack
	&mov	($rounds_,1);
	&xor	($key_,$key_);
	&mov	(&DWP(16,"esp"),$rounds_);
	&mov	(&DWP(20,"esp"),$key_);
	&mov	(&DWP(24,"esp"),$key_);
	&mov	(&DWP(28,"esp"),$key_);

	&movdqa	($inout3,&QWP(0,"esp"));	# bswap mask
	&movdqa	($inout0,$ivec);

	&mov	($key_,$key);
	&mov	($rounds_,$rounds);

	&pshufb	($ivec,$inout3);
	if ($inline)
	{   &aesni_inline_generate1("enc");	}
	else
	{   &call	("_aesni_encrypt1");	}
	&movups	($in0,&QWP(0,$inp));		# load inp
	&paddq	($ivec,&QWP(16,"esp"));
	&lea	($inp,&QWP(16,$inp));
	&jmp	(&label("ccm64_dec_outer"));

&set_label("ccm64_dec_outer",16);
	&xorps	($in0,$inout0);			# inp ^= E(ivec)
	&movdqa	($inout0,$ivec);
	&mov	($rounds,$rounds_);
	&movups	(&QWP(0,$out),$in0);		# save output
	&lea	($out,&DWP(16,$out));
	&pshufb	($inout0,$inout3);

	&sub	($len,1);
	&jz	(&label("ccm64_dec_break"));

	&$movekey	($rndkey0,&QWP(0,$key_));
	&shr		($rounds,1);
	&$movekey	($rndkey1,&QWP(16,$key_));
	&xorps		($in0,$rndkey0);
	&lea		($key,&DWP(32,$key_));
	&xorps		($inout0,$rndkey0);
	&xorps		($cmac,$in0);		# cmac^=out
	&$movekey	($rndkey0,&QWP(0,$key));

&set_label("ccm64_dec2_loop");
	&aesenc		($inout0,$rndkey1);
	&dec		($rounds);
	&aesenc		($cmac,$rndkey1);
	&$movekey	($rndkey1,&QWP(16,$key));
	&aesenc		($inout0,$rndkey0);
	&lea		($key,&DWP(32,$key));
	&aesenc		($cmac,$rndkey0);
	&$movekey	($rndkey0,&QWP(0,$key));
	&jnz		(&label("ccm64_dec2_loop"));
	&movups		($in0,&QWP(0,$inp));	# load inp
	&paddq		($ivec,&QWP(16,"esp"));
	&aesenc		($inout0,$rndkey1);
	&aesenc		($cmac,$rndkey1);
	&lea		($inp,&QWP(16,$inp));
	&aesenclast	($inout0,$rndkey0);
	&aesenclast	($cmac,$rndkey0);
	&jmp	(&label("ccm64_dec_outer"));

&set_label("ccm64_dec_break",16);
	&mov	($key,$key_);
	if ($inline)
	{   &aesni_inline_generate1("enc",$cmac,$in0);	}
	else
	{   &call	("_aesni_encrypt1",$cmac);	}

	&mov	("esp",&DWP(48,"esp"));
	&mov	($out,&wparam(5));
	&movups	(&QWP(0,$out),$cmac);
&function_end("aesni_ccm64_decrypt_blocks");
}

######################################################################
# void aesni_ctr32_encrypt_blocks (const void *in, void *out,
#                         size_t blocks, const AES_KEY *key,
#                         const char *ivec);
#
# Handles only complete blocks, operates on 32-bit counter and
# does not update *ivec! (see engine/eng_aesni.c for details)
#
# stack layout:
#	0	pshufb mask
#	16	vector addend: 0,6,6,6
# 	32	counter-less ivec
#	48	1st triplet of counter vector
#	64	2nd triplet of counter vector
#	80	saved %esp

&function_begin("aesni_ctr32_encrypt_blocks");
	&mov	($inp,&wparam(0));
	&mov	($out,&wparam(1));
	&mov	($len,&wparam(2));
	&mov	($key,&wparam(3));
	&mov	($rounds_,&wparam(4));
	&mov	($key_,"esp");
	&sub	("esp",88);
	&and	("esp",-16);			# align stack
	&mov	(&DWP(80,"esp"),$key_);

	&cmp	($len,1);
	&je	(&label("ctr32_one_shortcut"));

	&movdqu	($inout5,&QWP(0,$rounds_));	# load ivec

	# compose byte-swap control mask for pshufb on stack
	&mov	(&DWP(0,"esp"),0x0c0d0e0f);
	&mov	(&DWP(4,"esp"),0x08090a0b);
	&mov	(&DWP(8,"esp"),0x04050607);
	&mov	(&DWP(12,"esp"),0x00010203);

	# compose counter increment vector on stack
	&mov	($rounds,6);
	&xor	($key_,$key_);
	&mov	(&DWP(16,"esp"),$rounds);
	&mov	(&DWP(20,"esp"),$rounds);
	&mov	(&DWP(24,"esp"),$rounds);
	&mov	(&DWP(28,"esp"),$key_);

	&pextrd	($rounds_,$inout5,3);		# pull 32-bit counter
	&pinsrd	($inout5,$key_,3);		# wipe 32-bit counter

	&mov	($rounds,&DWP(240,$key));	# key->rounds

	# compose 2 vectors of 3x32-bit counters
	&bswap	($rounds_);
	&pxor	($rndkey1,$rndkey1);
	&pxor	($rndkey0,$rndkey0);
	&movdqa	($inout0,&QWP(0,"esp"));	# load byte-swap mask
	&pinsrd	($rndkey1,$rounds_,0);
	&lea	($key_,&DWP(3,$rounds_));
	&pinsrd	($rndkey0,$key_,0);
	&inc	($rounds_);
	&pinsrd	($rndkey1,$rounds_,1);
	&inc	($key_);
	&pinsrd	($rndkey0,$key_,1);
	&inc	($rounds_);
	&pinsrd	($rndkey1,$rounds_,2);
	&inc	($key_);
	&pinsrd	($rndkey0,$key_,2);
	&movdqa	(&QWP(48,"esp"),$rndkey1);	# save 1st triplet
	&pshufb	($rndkey1,$inout0);		# byte swap
	&movdqa	(&QWP(64,"esp"),$rndkey0);	# save 2nd triplet
	&pshufb	($rndkey0,$inout0);		# byte swap

	&pshufd	($inout0,$rndkey1,3<<6);	# place counter to upper dword
	&pshufd	($inout1,$rndkey1,2<<6);
	&cmp	($len,6);
	&jb	(&label("ctr32_tail"));
	&movdqa	(&QWP(32,"esp"),$inout5);	# save counter-less ivec
	&shr	($rounds,1);
	&mov	($key_,$key);			# backup $key
	&mov	($rounds_,$rounds);		# backup $rounds
	&sub	($len,6);
	&jmp	(&label("ctr32_loop6"));

&set_label("ctr32_loop6",16);
	&pshufd	($inout2,$rndkey1,1<<6);
	&movdqa	($rndkey1,&QWP(32,"esp"));	# pull counter-less ivec
	&pshufd	($inout3,$rndkey0,3<<6);
	&por	($inout0,$rndkey1);		# merge counter-less ivec
	&pshufd	($inout4,$rndkey0,2<<6);
	&por	($inout1,$rndkey1);
	&pshufd	($inout5,$rndkey0,1<<6);
	&por	($inout2,$rndkey1);
	&por	($inout3,$rndkey1);
	&por	($inout4,$rndkey1);
	&por	($inout5,$rndkey1);

	# inlining _aesni_encrypt6's prologue gives ~4% improvement...
	&$movekey	($rndkey0,&QWP(0,$key_));
	&$movekey	($rndkey1,&QWP(16,$key_));
	&lea		($key,&DWP(32,$key_));
	&dec		($rounds);
	&pxor		($inout0,$rndkey0);
	&pxor		($inout1,$rndkey0);
	&aesenc		($inout0,$rndkey1);
	&pxor		($inout2,$rndkey0);
	&aesenc		($inout1,$rndkey1);
	&pxor		($inout3,$rndkey0);
	&aesenc		($inout2,$rndkey1);
	&pxor		($inout4,$rndkey0);
	&aesenc		($inout3,$rndkey1);
	&pxor		($inout5,$rndkey0);
	&aesenc		($inout4,$rndkey1);
	&$movekey	($rndkey0,&QWP(0,$key));
	&aesenc		($inout5,$rndkey1);

	&call		(&label("_aesni_encrypt6_enter"));

	&movups	($rndkey1,&QWP(0,$inp));
	&movups	($rndkey0,&QWP(0x10,$inp));
	&xorps	($inout0,$rndkey1);
	&movups	($rndkey1,&QWP(0x20,$inp));
	&xorps	($inout1,$rndkey0);
	&movups	(&QWP(0,$out),$inout0);
	&movdqa	($rndkey0,&QWP(16,"esp"));	# load increment
	&xorps	($inout2,$rndkey1);
	&movdqa	($rndkey1,&QWP(48,"esp"));	# load 1st triplet
	&movups	(&QWP(0x10,$out),$inout1);
	&movups	(&QWP(0x20,$out),$inout2);

	&paddd	($rndkey1,$rndkey0);		# 1st triplet increment
	&paddd	($rndkey0,&QWP(64,"esp"));	# 2nd triplet increment
	&movdqa	($inout0,&QWP(0,"esp"));	# load byte swap mask

	&movups	($inout1,&QWP(0x30,$inp));
	&movups	($inout2,&QWP(0x40,$inp));
	&xorps	($inout3,$inout1);
	&movups	($inout1,&QWP(0x50,$inp));
	&lea	($inp,&DWP(0x60,$inp));
	&movdqa	(&QWP(48,"esp"),$rndkey1);	# save 1st triplet
	&pshufb	($rndkey1,$inout0);		# byte swap
	&xorps	($inout4,$inout2);
	&movups	(&QWP(0x30,$out),$inout3);
	&xorps	($inout5,$inout1);
	&movdqa	(&QWP(64,"esp"),$rndkey0);	# save 2nd triplet
	&pshufb	($rndkey0,$inout0);		# byte swap
	&movups	(&QWP(0x40,$out),$inout4);
	&pshufd	($inout0,$rndkey1,3<<6);
	&movups	(&QWP(0x50,$out),$inout5);
	&lea	($out,&DWP(0x60,$out));

	&mov	($rounds,$rounds_);
	&pshufd	($inout1,$rndkey1,2<<6);
	&sub	($len,6);
	&jnc	(&label("ctr32_loop6"));

	&add	($len,6);
	&jz	(&label("ctr32_ret"));
	&mov	($key,$key_);
	&lea	($rounds,&DWP(1,"",$rounds,2));	# restore $rounds
	&movdqa	($inout5,&QWP(32,"esp"));	# pull count-less ivec

&set_label("ctr32_tail");
	&por	($inout0,$inout5);
	&cmp	($len,2);
	&jb	(&label("ctr32_one"));

	&pshufd	($inout2,$rndkey1,1<<6);
	&por	($inout1,$inout5);
	&je	(&label("ctr32_two"));

	&pshufd	($inout3,$rndkey0,3<<6);
	&por	($inout2,$inout5);
	&cmp	($len,4);
	&jb	(&label("ctr32_three"));

	&pshufd	($inout4,$rndkey0,2<<6);
	&por	($inout3,$inout5);
	&je	(&label("ctr32_four"));

	&por	($inout4,$inout5);
	&call	("_aesni_encrypt6");
	&movups	($rndkey1,&QWP(0,$inp));
	&movups	($rndkey0,&QWP(0x10,$inp));
	&xorps	($inout0,$rndkey1);
	&movups	($rndkey1,&QWP(0x20,$inp));
	&xorps	($inout1,$rndkey0);
	&movups	($rndkey0,&QWP(0x30,$inp));
	&xorps	($inout2,$rndkey1);
	&movups	($rndkey1,&QWP(0x40,$inp));
	&xorps	($inout3,$rndkey0);
	&movups	(&QWP(0,$out),$inout0);
	&xorps	($inout4,$rndkey1);
	&movups	(&QWP(0x10,$out),$inout1);
	&movups	(&QWP(0x20,$out),$inout2);
	&movups	(&QWP(0x30,$out),$inout3);
	&movups	(&QWP(0x40,$out),$inout4);
	&jmp	(&label("ctr32_ret"));

&set_label("ctr32_one_shortcut",16);
	&movups	($inout0,&QWP(0,$rounds_));	# load ivec
	&mov	($rounds,&DWP(240,$key));
	
&set_label("ctr32_one");
	if ($inline)
	{   &aesni_inline_generate1("enc");	}
	else
	{   &call	("_aesni_encrypt1");	}
	&movups	($in0,&QWP(0,$inp));
	&xorps	($in0,$inout0);
	&movups	(&QWP(0,$out),$in0);
	&jmp	(&label("ctr32_ret"));

&set_label("ctr32_two",16);
	&call	("_aesni_encrypt3");
	&movups	($inout3,&QWP(0,$inp));
	&movups	($inout4,&QWP(0x10,$inp));
	&xorps	($inout0,$inout3);
	&xorps	($inout1,$inout4);
	&movups	(&QWP(0,$out),$inout0);
	&movups	(&QWP(0x10,$out),$inout1);
	&jmp	(&label("ctr32_ret"));

&set_label("ctr32_three",16);
	&call	("_aesni_encrypt3");
	&movups	($inout3,&QWP(0,$inp));
	&movups	($inout4,&QWP(0x10,$inp));
	&xorps	($inout0,$inout3);
	&movups	($inout5,&QWP(0x20,$inp));
	&xorps	($inout1,$inout4);
	&movups	(&QWP(0,$out),$inout0);
	&xorps	($inout2,$inout5);
	&movups	(&QWP(0x10,$out),$inout1);
	&movups	(&QWP(0x20,$out),$inout2);
	&jmp	(&label("ctr32_ret"));

&set_label("ctr32_four",16);
	&call	("_aesni_encrypt4");
	&movups	($inout4,&QWP(0,$inp));
	&movups	($inout5,&QWP(0x10,$inp));
	&movups	($rndkey1,&QWP(0x20,$inp));
	&xorps	($inout0,$inout4);
	&movups	($rndkey0,&QWP(0x30,$inp));
	&xorps	($inout1,$inout5);
	&movups	(&QWP(0,$out),$inout0);
	&xorps	($inout2,$rndkey1);
	&movups	(&QWP(0x10,$out),$inout1);
	&xorps	($inout3,$rndkey0);
	&movups	(&QWP(0x20,$out),$inout2);
	&movups	(&QWP(0x30,$out),$inout3);

&set_label("ctr32_ret");
	&mov	("esp",&DWP(80,"esp"));
&function_end("aesni_ctr32_encrypt_blocks");

######################################################################
# void aesni_xts_[en|de]crypt(const char *inp,char *out,size_t len,
#	const AES_KEY *key1, const AES_KEY *key2
#	const unsigned char iv[16]);
#
{ my ($tweak,$twtmp,$twres,$twmask)=($rndkey1,$rndkey0,$inout0,$inout1);

&function_begin("aesni_xts_encrypt");
	&mov	($key,&wparam(4));		# key2
	&mov	($inp,&wparam(5));		# clear-text tweak

	&mov	($rounds,&DWP(240,$key));	# key2->rounds
	&movups	($inout0,&QWP(0,$inp));
	if ($inline)
	{   &aesni_inline_generate1("enc");	}
	else
	{   &call	("_aesni_encrypt1");	}

	&mov	($inp,&wparam(0));
	&mov	($out,&wparam(1));
	&mov	($len,&wparam(2));
	&mov	($key,&wparam(3));		# key1

	&mov	($key_,"esp");
	&sub	("esp",16*7+8);
	&mov	($rounds,&DWP(240,$key));	# key1->rounds
	&and	("esp",-16);			# align stack

	&mov	(&DWP(16*6+0,"esp"),0x87);	# compose the magic constant
	&mov	(&DWP(16*6+4,"esp"),0);
	&mov	(&DWP(16*6+8,"esp"),1);
	&mov	(&DWP(16*6+12,"esp"),0);
	&mov	(&DWP(16*7+0,"esp"),$len);	# save original $len
	&mov	(&DWP(16*7+4,"esp"),$key_);	# save original %esp

	&movdqa	($tweak,$inout0);
	&pxor	($twtmp,$twtmp);
	&movdqa	($twmask,&QWP(6*16,"esp"));	# 0x0...010...87
	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits

	&and	($len,-16);
	&mov	($key_,$key);			# backup $key
	&mov	($rounds_,$rounds);		# backup $rounds
	&sub	($len,16*6);
	&jc	(&label("xts_enc_short"));

	&shr	($rounds,1);
	&mov	($rounds_,$rounds);
	&jmp	(&label("xts_enc_loop6"));

&set_label("xts_enc_loop6",16);
	for ($i=0;$i<4;$i++) {
	    &pshufd	($twres,$twtmp,0x13);
	    &pxor	($twtmp,$twtmp);
	    &movdqa	(&QWP(16*$i,"esp"),$tweak);
	    &paddq	($tweak,$tweak);	# &psllq($tweak,1);
	    &pand	($twres,$twmask);	# isolate carry and residue
	    &pcmpgtd	($twtmp,$tweak);	# broadcast upper bits
	    &pxor	($tweak,$twres);
	}
	&pshufd	($inout5,$twtmp,0x13);
	&movdqa	(&QWP(16*$i++,"esp"),$tweak);
	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
	 &$movekey	($rndkey0,&QWP(0,$key_));
	&pand	($inout5,$twmask);		# isolate carry and residue
	 &movups	($inout0,&QWP(0,$inp));	# load input
	&pxor	($inout5,$tweak);

	# inline _aesni_encrypt6 prologue and flip xor with tweak and key[0]
	&movdqu	($inout1,&QWP(16*1,$inp));
	 &xorps		($inout0,$rndkey0);	# input^=rndkey[0]
	&movdqu	($inout2,&QWP(16*2,$inp));
	 &pxor		($inout1,$rndkey0);
	&movdqu	($inout3,&QWP(16*3,$inp));
	 &pxor		($inout2,$rndkey0);
	&movdqu	($inout4,&QWP(16*4,$inp));
	 &pxor		($inout3,$rndkey0);
	&movdqu	($rndkey1,&QWP(16*5,$inp));
	 &pxor		($inout4,$rndkey0);
	&lea	($inp,&DWP(16*6,$inp));
	&pxor	($inout0,&QWP(16*0,"esp"));	# input^=tweak
	&movdqa	(&QWP(16*$i,"esp"),$inout5);	# save last tweak
	&pxor	($inout5,$rndkey1);

	 &$movekey	($rndkey1,&QWP(16,$key_));
	 &lea		($key,&DWP(32,$key_));
	&pxor	($inout1,&QWP(16*1,"esp"));
	 &aesenc	($inout0,$rndkey1);
	&pxor	($inout2,&QWP(16*2,"esp"));
	 &aesenc	($inout1,$rndkey1);
	&pxor	($inout3,&QWP(16*3,"esp"));
	 &dec		($rounds);
	 &aesenc	($inout2,$rndkey1);
	&pxor	($inout4,&QWP(16*4,"esp"));
	 &aesenc	($inout3,$rndkey1);
	&pxor		($inout5,$rndkey0);
	 &aesenc	($inout4,$rndkey1);
	 &$movekey	($rndkey0,&QWP(0,$key));
	 &aesenc	($inout5,$rndkey1);
	&call		(&label("_aesni_encrypt6_enter"));

	&movdqa	($tweak,&QWP(16*5,"esp"));	# last tweak
       &pxor	($twtmp,$twtmp);
	&xorps	($inout0,&QWP(16*0,"esp"));	# output^=tweak
       &pcmpgtd	($twtmp,$tweak);		# broadcast upper bits
	&xorps	($inout1,&QWP(16*1,"esp"));
	&movups	(&QWP(16*0,$out),$inout0);	# write output
	&xorps	($inout2,&QWP(16*2,"esp"));
	&movups	(&QWP(16*1,$out),$inout1);
	&xorps	($inout3,&QWP(16*3,"esp"));
	&movups	(&QWP(16*2,$out),$inout2);
	&xorps	($inout4,&QWP(16*4,"esp"));
	&movups	(&QWP(16*3,$out),$inout3);
	&xorps	($inout5,$tweak);
	&movups	(&QWP(16*4,$out),$inout4);
       &pshufd	($twres,$twtmp,0x13);
	&movups	(&QWP(16*5,$out),$inout5);
	&lea	($out,&DWP(16*6,$out));
       &movdqa	($twmask,&QWP(16*6,"esp"));	# 0x0...010...87

	&pxor	($twtmp,$twtmp);
	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
	&pand	($twres,$twmask);		# isolate carry and residue
	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits
	&mov	($rounds,$rounds_);		# restore $rounds
	&pxor	($tweak,$twres);

	&sub	($len,16*6);
	&jnc	(&label("xts_enc_loop6"));

	&lea	($rounds,&DWP(1,"",$rounds,2));	# restore $rounds
	&mov	($key,$key_);			# restore $key
	&mov	($rounds_,$rounds);

&set_label("xts_enc_short");
	&add	($len,16*6);
	&jz	(&label("xts_enc_done6x"));

	&movdqa	($inout3,$tweak);		# put aside previous tweak
	&cmp	($len,0x20);
	&jb	(&label("xts_enc_one"));

	&pshufd	($twres,$twtmp,0x13);
	&pxor	($twtmp,$twtmp);
	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
	&pand	($twres,$twmask);		# isolate carry and residue
	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits
	&pxor	($tweak,$twres);
	&je	(&label("xts_enc_two"));

	&pshufd	($twres,$twtmp,0x13);
	&pxor	($twtmp,$twtmp);
	&movdqa	($inout4,$tweak);		# put aside previous tweak
	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
	&pand	($twres,$twmask);		# isolate carry and residue
	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits
	&pxor	($tweak,$twres);
	&cmp	($len,0x40);
	&jb	(&label("xts_enc_three"));

	&pshufd	($twres,$twtmp,0x13);
	&pxor	($twtmp,$twtmp);
	&movdqa	($inout5,$tweak);		# put aside previous tweak
	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
	&pand	($twres,$twmask);		# isolate carry and residue
	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits
	&pxor	($tweak,$twres);
	&movdqa	(&QWP(16*0,"esp"),$inout3);
	&movdqa	(&QWP(16*1,"esp"),$inout4);
	&je	(&label("xts_enc_four"));

	&movdqa	(&QWP(16*2,"esp"),$inout5);
	&pshufd	($inout5,$twtmp,0x13);
	&movdqa	(&QWP(16*3,"esp"),$tweak);
	&paddq	($tweak,$tweak);		# &psllq($inout0,1);
	&pand	($inout5,$twmask);		# isolate carry and residue
	&pxor	($inout5,$tweak);

	&movdqu	($inout0,&QWP(16*0,$inp));	# load input
	&movdqu	($inout1,&QWP(16*1,$inp));
	&movdqu	($inout2,&QWP(16*2,$inp));
	&pxor	($inout0,&QWP(16*0,"esp"));	# input^=tweak
	&movdqu	($inout3,&QWP(16*3,$inp));
	&pxor	($inout1,&QWP(16*1,"esp"));
	&movdqu	($inout4,&QWP(16*4,$inp));
	&pxor	($inout2,&QWP(16*2,"esp"));
	&lea	($inp,&DWP(16*5,$inp));
	&pxor	($inout3,&QWP(16*3,"esp"));
	&movdqa	(&QWP(16*4,"esp"),$inout5);	# save last tweak
	&pxor	($inout4,$inout5);

	&call	("_aesni_encrypt6");

	&movaps	($tweak,&QWP(16*4,"esp"));	# last tweak
	&xorps	($inout0,&QWP(16*0,"esp"));	# output^=tweak
	&xorps	($inout1,&QWP(16*1,"esp"));
	&xorps	($inout2,&QWP(16*2,"esp"));
	&movups	(&QWP(16*0,$out),$inout0);	# write output
	&xorps	($inout3,&QWP(16*3,"esp"));
	&movups	(&QWP(16*1,$out),$inout1);
	&xorps	($inout4,$tweak);
	&movups	(&QWP(16*2,$out),$inout2);
	&movups	(&QWP(16*3,$out),$inout3);
	&movups	(&QWP(16*4,$out),$inout4);
	&lea	($out,&DWP(16*5,$out));
	&jmp	(&label("xts_enc_done"));

&set_label("xts_enc_one",16);
	&movups	($inout0,&QWP(16*0,$inp));	# load input
	&lea	($inp,&DWP(16*1,$inp));
	&xorps	($inout0,$inout3);		# input^=tweak
	if ($inline)
	{   &aesni_inline_generate1("enc");	}
	else
	{   &call	("_aesni_encrypt1");	}
	&xorps	($inout0,$inout3);		# output^=tweak
	&movups	(&QWP(16*0,$out),$inout0);	# write output
	&lea	($out,&DWP(16*1,$out));

	&movdqa	($tweak,$inout3);		# last tweak
	&jmp	(&label("xts_enc_done"));

&set_label("xts_enc_two",16);
	&movaps	($inout4,$tweak);		# put aside last tweak

	&movups	($inout0,&QWP(16*0,$inp));	# load input
	&movups	($inout1,&QWP(16*1,$inp));
	&lea	($inp,&DWP(16*2,$inp));
	&xorps	($inout0,$inout3);		# input^=tweak
	&xorps	($inout1,$inout4);
	&xorps	($inout2,$inout2);

	&call	("_aesni_encrypt3");

	&xorps	($inout0,$inout3);		# output^=tweak
	&xorps	($inout1,$inout4);
	&movups	(&QWP(16*0,$out),$inout0);	# write output
	&movups	(&QWP(16*1,$out),$inout1);
	&lea	($out,&DWP(16*2,$out));

	&movdqa	($tweak,$inout4);		# last tweak
	&jmp	(&label("xts_enc_done"));

&set_label("xts_enc_three",16);
	&movaps	($inout5,$tweak);		# put aside last tweak
	&movups	($inout0,&QWP(16*0,$inp));	# load input
	&movups	($inout1,&QWP(16*1,$inp));
	&movups	($inout2,&QWP(16*2,$inp));
	&lea	($inp,&DWP(16*3,$inp));
	&xorps	($inout0,$inout3);		# input^=tweak
	&xorps	($inout1,$inout4);
	&xorps	($inout2,$inout5);

	&call	("_aesni_encrypt3");

	&xorps	($inout0,$inout3);		# output^=tweak
	&xorps	($inout1,$inout4);
	&xorps	($inout2,$inout5);
	&movups	(&QWP(16*0,$out),$inout0);	# write output
	&movups	(&QWP(16*1,$out),$inout1);
	&movups	(&QWP(16*2,$out),$inout2);
	&lea	($out,&DWP(16*3,$out));

	&movdqa	($tweak,$inout5);		# last tweak
	&jmp	(&label("xts_enc_done"));

&set_label("xts_enc_four",16);
	&movaps	($inout4,$tweak);		# put aside last tweak

	&movups	($inout0,&QWP(16*0,$inp));	# load input
	&movups	($inout1,&QWP(16*1,$inp));
	&movups	($inout2,&QWP(16*2,$inp));
	&xorps	($inout0,&QWP(16*0,"esp"));	# input^=tweak
	&movups	($inout3,&QWP(16*3,$inp));
	&lea	($inp,&DWP(16*4,$inp));
	&xorps	($inout1,&QWP(16*1,"esp"));
	&xorps	($inout2,$inout5);
	&xorps	($inout3,$inout4);

	&call	("_aesni_encrypt4");

	&xorps	($inout0,&QWP(16*0,"esp"));	# output^=tweak
	&xorps	($inout1,&QWP(16*1,"esp"));
	&xorps	($inout2,$inout5);
	&movups	(&QWP(16*0,$out),$inout0);	# write output
	&xorps	($inout3,$inout4);
	&movups	(&QWP(16*1,$out),$inout1);
	&movups	(&QWP(16*2,$out),$inout2);
	&movups	(&QWP(16*3,$out),$inout3);
	&lea	($out,&DWP(16*4,$out));

	&movdqa	($tweak,$inout4);		# last tweak
	&jmp	(&label("xts_enc_done"));

&set_label("xts_enc_done6x",16);		# $tweak is pre-calculated
	&mov	($len,&DWP(16*7+0,"esp"));	# restore original $len
	&and	($len,15);
	&jz	(&label("xts_enc_ret"));
	&movdqa	($inout3,$tweak);
	&mov	(&DWP(16*7+0,"esp"),$len);	# save $len%16
	&jmp	(&label("xts_enc_steal"));

&set_label("xts_enc_done",16);
	&mov	($len,&DWP(16*7+0,"esp"));	# restore original $len
	&pxor	($twtmp,$twtmp);
	&and	($len,15);
	&jz	(&label("xts_enc_ret"));

	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits
	&mov	(&DWP(16*7+0,"esp"),$len);	# save $len%16
	&pshufd	($inout3,$twtmp,0x13);
	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
	&pand	($inout3,&QWP(16*6,"esp"));	# isolate carry and residue
	&pxor	($inout3,$tweak);

&set_label("xts_enc_steal");
	&movz	($rounds,&BP(0,$inp));
	&movz	($key,&BP(-16,$out));
	&lea	($inp,&DWP(1,$inp));
	&mov	(&BP(-16,$out),&LB($rounds));
	&mov	(&BP(0,$out),&LB($key));
	&lea	($out,&DWP(1,$out));
	&sub	($len,1);
	&jnz	(&label("xts_enc_steal"));

	&sub	($out,&DWP(16*7+0,"esp"));	# rewind $out
	&mov	($key,$key_);			# restore $key
	&mov	($rounds,$rounds_);		# restore $rounds

	&movups	($inout0,&QWP(-16,$out));	# load input
	&xorps	($inout0,$inout3);		# input^=tweak
	if ($inline)
	{   &aesni_inline_generate1("enc");	}
	else
	{   &call	("_aesni_encrypt1");	}
	&xorps	($inout0,$inout3);		# output^=tweak
	&movups	(&QWP(-16,$out),$inout0);	# write output

&set_label("xts_enc_ret");
	&mov	("esp",&DWP(16*7+4,"esp"));	# restore %esp
&function_end("aesni_xts_encrypt");

&function_begin("aesni_xts_decrypt");
	&mov	($key,&wparam(4));		# key2
	&mov	($inp,&wparam(5));		# clear-text tweak

	&mov	($rounds,&DWP(240,$key));	# key2->rounds
	&movups	($inout0,&QWP(0,$inp));
	if ($inline)
	{   &aesni_inline_generate1("enc");	}
	else
	{   &call	("_aesni_encrypt1");	}

	&mov	($inp,&wparam(0));
	&mov	($out,&wparam(1));
	&mov	($len,&wparam(2));
	&mov	($key,&wparam(3));		# key1

	&mov	($key_,"esp");
	&sub	("esp",16*7+8);
	&and	("esp",-16);			# align stack

	&xor	($rounds_,$rounds_);		# if(len%16) len-=16;
	&test	($len,15);
	&setnz	(&LB($rounds_));
	&shl	($rounds_,4);
	&sub	($len,$rounds_);

	&mov	(&DWP(16*6+0,"esp"),0x87);	# compose the magic constant
	&mov	(&DWP(16*6+4,"esp"),0);
	&mov	(&DWP(16*6+8,"esp"),1);
	&mov	(&DWP(16*6+12,"esp"),0);
	&mov	(&DWP(16*7+0,"esp"),$len);	# save original $len
	&mov	(&DWP(16*7+4,"esp"),$key_);	# save original %esp

	&mov	($rounds,&DWP(240,$key));	# key1->rounds
	&mov	($key_,$key);			# backup $key
	&mov	($rounds_,$rounds);		# backup $rounds

	&movdqa	($tweak,$inout0);
	&pxor	($twtmp,$twtmp);
	&movdqa	($twmask,&QWP(6*16,"esp"));	# 0x0...010...87
	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits

	&and	($len,-16);
	&sub	($len,16*6);
	&jc	(&label("xts_dec_short"));

	&shr	($rounds,1);
	&mov	($rounds_,$rounds);
	&jmp	(&label("xts_dec_loop6"));

&set_label("xts_dec_loop6",16);
	for ($i=0;$i<4;$i++) {
	    &pshufd	($twres,$twtmp,0x13);
	    &pxor	($twtmp,$twtmp);
	    &movdqa	(&QWP(16*$i,"esp"),$tweak);
	    &paddq	($tweak,$tweak);	# &psllq($tweak,1);
	    &pand	($twres,$twmask);	# isolate carry and residue
	    &pcmpgtd	($twtmp,$tweak);	# broadcast upper bits
	    &pxor	($tweak,$twres);
	}
	&pshufd	($inout5,$twtmp,0x13);
	&movdqa	(&QWP(16*$i++,"esp"),$tweak);
	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
	 &$movekey	($rndkey0,&QWP(0,$key_));
	&pand	($inout5,$twmask);		# isolate carry and residue
	 &movups	($inout0,&QWP(0,$inp));	# load input
	&pxor	($inout5,$tweak);

	# inline _aesni_encrypt6 prologue and flip xor with tweak and key[0]
	&movdqu	($inout1,&QWP(16*1,$inp));
	 &xorps		($inout0,$rndkey0);	# input^=rndkey[0]
	&movdqu	($inout2,&QWP(16*2,$inp));
	 &pxor		($inout1,$rndkey0);
	&movdqu	($inout3,&QWP(16*3,$inp));
	 &pxor		($inout2,$rndkey0);
	&movdqu	($inout4,&QWP(16*4,$inp));
	 &pxor		($inout3,$rndkey0);
	&movdqu	($rndkey1,&QWP(16*5,$inp));
	 &pxor		($inout4,$rndkey0);
	&lea	($inp,&DWP(16*6,$inp));
	&pxor	($inout0,&QWP(16*0,"esp"));	# input^=tweak
	&movdqa	(&QWP(16*$i,"esp"),$inout5);	# save last tweak
	&pxor	($inout5,$rndkey1);

	 &$movekey	($rndkey1,&QWP(16,$key_));
	 &lea		($key,&DWP(32,$key_));
	&pxor	($inout1,&QWP(16*1,"esp"));
	 &aesdec	($inout0,$rndkey1);
	&pxor	($inout2,&QWP(16*2,"esp"));
	 &aesdec	($inout1,$rndkey1);
	&pxor	($inout3,&QWP(16*3,"esp"));
	 &dec		($rounds);
	 &aesdec	($inout2,$rndkey1);
	&pxor	($inout4,&QWP(16*4,"esp"));
	 &aesdec	($inout3,$rndkey1);
	&pxor		($inout5,$rndkey0);
	 &aesdec	($inout4,$rndkey1);
	 &$movekey	($rndkey0,&QWP(0,$key));
	 &aesdec	($inout5,$rndkey1);
	&call		(&label("_aesni_decrypt6_enter"));

	&movdqa	($tweak,&QWP(16*5,"esp"));	# last tweak
       &pxor	($twtmp,$twtmp);
	&xorps	($inout0,&QWP(16*0,"esp"));	# output^=tweak
       &pcmpgtd	($twtmp,$tweak);		# broadcast upper bits
	&xorps	($inout1,&QWP(16*1,"esp"));
	&movups	(&QWP(16*0,$out),$inout0);	# write output
	&xorps	($inout2,&QWP(16*2,"esp"));
	&movups	(&QWP(16*1,$out),$inout1);
	&xorps	($inout3,&QWP(16*3,"esp"));
	&movups	(&QWP(16*2,$out),$inout2);
	&xorps	($inout4,&QWP(16*4,"esp"));
	&movups	(&QWP(16*3,$out),$inout3);
	&xorps	($inout5,$tweak);
	&movups	(&QWP(16*4,$out),$inout4);
       &pshufd	($twres,$twtmp,0x13);
	&movups	(&QWP(16*5,$out),$inout5);
	&lea	($out,&DWP(16*6,$out));
       &movdqa	($twmask,&QWP(16*6,"esp"));	# 0x0...010...87

	&pxor	($twtmp,$twtmp);
	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
	&pand	($twres,$twmask);		# isolate carry and residue
	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits
	&mov	($rounds,$rounds_);		# restore $rounds
	&pxor	($tweak,$twres);

	&sub	($len,16*6);
	&jnc	(&label("xts_dec_loop6"));

	&lea	($rounds,&DWP(1,"",$rounds,2));	# restore $rounds
	&mov	($key,$key_);			# restore $key
	&mov	($rounds_,$rounds);

&set_label("xts_dec_short");
	&add	($len,16*6);
	&jz	(&label("xts_dec_done6x"));

	&movdqa	($inout3,$tweak);		# put aside previous tweak
	&cmp	($len,0x20);
	&jb	(&label("xts_dec_one"));

	&pshufd	($twres,$twtmp,0x13);
	&pxor	($twtmp,$twtmp);
	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
	&pand	($twres,$twmask);		# isolate carry and residue
	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits
	&pxor	($tweak,$twres);
	&je	(&label("xts_dec_two"));

	&pshufd	($twres,$twtmp,0x13);
	&pxor	($twtmp,$twtmp);
	&movdqa	($inout4,$tweak);		# put aside previous tweak
	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
	&pand	($twres,$twmask);		# isolate carry and residue
	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits
	&pxor	($tweak,$twres);
	&cmp	($len,0x40);
	&jb	(&label("xts_dec_three"));

	&pshufd	($twres,$twtmp,0x13);
	&pxor	($twtmp,$twtmp);
	&movdqa	($inout5,$tweak);		# put aside previous tweak
	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
	&pand	($twres,$twmask);		# isolate carry and residue
	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits
	&pxor	($tweak,$twres);
	&movdqa	(&QWP(16*0,"esp"),$inout3);
	&movdqa	(&QWP(16*1,"esp"),$inout4);
	&je	(&label("xts_dec_four"));

	&movdqa	(&QWP(16*2,"esp"),$inout5);
	&pshufd	($inout5,$twtmp,0x13);
	&movdqa	(&QWP(16*3,"esp"),$tweak);
	&paddq	($tweak,$tweak);		# &psllq($inout0,1);
	&pand	($inout5,$twmask);		# isolate carry and residue
	&pxor	($inout5,$tweak);

	&movdqu	($inout0,&QWP(16*0,$inp));	# load input
	&movdqu	($inout1,&QWP(16*1,$inp));
	&movdqu	($inout2,&QWP(16*2,$inp));
	&pxor	($inout0,&QWP(16*0,"esp"));	# input^=tweak
	&movdqu	($inout3,&QWP(16*3,$inp));
	&pxor	($inout1,&QWP(16*1,"esp"));
	&movdqu	($inout4,&QWP(16*4,$inp));
	&pxor	($inout2,&QWP(16*2,"esp"));
	&lea	($inp,&DWP(16*5,$inp));
	&pxor	($inout3,&QWP(16*3,"esp"));
	&movdqa	(&QWP(16*4,"esp"),$inout5);	# save last tweak
	&pxor	($inout4,$inout5);

	&call	("_aesni_decrypt6");

	&movaps	($tweak,&QWP(16*4,"esp"));	# last tweak
	&xorps	($inout0,&QWP(16*0,"esp"));	# output^=tweak
	&xorps	($inout1,&QWP(16*1,"esp"));
	&xorps	($inout2,&QWP(16*2,"esp"));
	&movups	(&QWP(16*0,$out),$inout0);	# write output
	&xorps	($inout3,&QWP(16*3,"esp"));
	&movups	(&QWP(16*1,$out),$inout1);
	&xorps	($inout4,$tweak);
	&movups	(&QWP(16*2,$out),$inout2);
	&movups	(&QWP(16*3,$out),$inout3);
	&movups	(&QWP(16*4,$out),$inout4);
	&lea	($out,&DWP(16*5,$out));
	&jmp	(&label("xts_dec_done"));

&set_label("xts_dec_one",16);
	&movups	($inout0,&QWP(16*0,$inp));	# load input
	&lea	($inp,&DWP(16*1,$inp));
	&xorps	($inout0,$inout3);		# input^=tweak
	if ($inline)
	{   &aesni_inline_generate1("dec");	}
	else
	{   &call	("_aesni_decrypt1");	}
	&xorps	($inout0,$inout3);		# output^=tweak
	&movups	(&QWP(16*0,$out),$inout0);	# write output
	&lea	($out,&DWP(16*1,$out));

	&movdqa	($tweak,$inout3);		# last tweak
	&jmp	(&label("xts_dec_done"));

&set_label("xts_dec_two",16);
	&movaps	($inout4,$tweak);		# put aside last tweak

	&movups	($inout0,&QWP(16*0,$inp));	# load input
	&movups	($inout1,&QWP(16*1,$inp));
	&lea	($inp,&DWP(16*2,$inp));
	&xorps	($inout0,$inout3);		# input^=tweak
	&xorps	($inout1,$inout4);

	&call	("_aesni_decrypt3");

	&xorps	($inout0,$inout3);		# output^=tweak
	&xorps	($inout1,$inout4);
	&movups	(&QWP(16*0,$out),$inout0);	# write output
	&movups	(&QWP(16*1,$out),$inout1);
	&lea	($out,&DWP(16*2,$out));

	&movdqa	($tweak,$inout4);		# last tweak
	&jmp	(&label("xts_dec_done"));

&set_label("xts_dec_three",16);
	&movaps	($inout5,$tweak);		# put aside last tweak
	&movups	($inout0,&QWP(16*0,$inp));	# load input
	&movups	($inout1,&QWP(16*1,$inp));
	&movups	($inout2,&QWP(16*2,$inp));
	&lea	($inp,&DWP(16*3,$inp));
	&xorps	($inout0,$inout3);		# input^=tweak
	&xorps	($inout1,$inout4);
	&xorps	($inout2,$inout5);

	&call	("_aesni_decrypt3");

	&xorps	($inout0,$inout3);		# output^=tweak
	&xorps	($inout1,$inout4);
	&xorps	($inout2,$inout5);
	&movups	(&QWP(16*0,$out),$inout0);	# write output
	&movups	(&QWP(16*1,$out),$inout1);
	&movups	(&QWP(16*2,$out),$inout2);
	&lea	($out,&DWP(16*3,$out));

	&movdqa	($tweak,$inout5);		# last tweak
	&jmp	(&label("xts_dec_done"));

&set_label("xts_dec_four",16);
	&movaps	($inout4,$tweak);		# put aside last tweak

	&movups	($inout0,&QWP(16*0,$inp));	# load input
	&movups	($inout1,&QWP(16*1,$inp));
	&movups	($inout2,&QWP(16*2,$inp));
	&xorps	($inout0,&QWP(16*0,"esp"));	# input^=tweak
	&movups	($inout3,&QWP(16*3,$inp));
	&lea	($inp,&DWP(16*4,$inp));
	&xorps	($inout1,&QWP(16*1,"esp"));
	&xorps	($inout2,$inout5);
	&xorps	($inout3,$inout4);

	&call	("_aesni_decrypt4");

	&xorps	($inout0,&QWP(16*0,"esp"));	# output^=tweak
	&xorps	($inout1,&QWP(16*1,"esp"));
	&xorps	($inout2,$inout5);
	&movups	(&QWP(16*0,$out),$inout0);	# write output
	&xorps	($inout3,$inout4);
	&movups	(&QWP(16*1,$out),$inout1);
	&movups	(&QWP(16*2,$out),$inout2);
	&movups	(&QWP(16*3,$out),$inout3);
	&lea	($out,&DWP(16*4,$out));

	&movdqa	($tweak,$inout4);		# last tweak
	&jmp	(&label("xts_dec_done"));

&set_label("xts_dec_done6x",16);		# $tweak is pre-calculated
	&mov	($len,&DWP(16*7+0,"esp"));	# restore original $len
	&and	($len,15);
	&jz	(&label("xts_dec_ret"));
	&mov	(&DWP(16*7+0,"esp"),$len);	# save $len%16
	&jmp	(&label("xts_dec_only_one_more"));

&set_label("xts_dec_done",16);
	&mov	($len,&DWP(16*7+0,"esp"));	# restore original $len
	&pxor	($twtmp,$twtmp);
	&and	($len,15);
	&jz	(&label("xts_dec_ret"));

	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits
	&mov	(&DWP(16*7+0,"esp"),$len);	# save $len%16
	&pshufd	($twres,$twtmp,0x13);
	&pxor	($twtmp,$twtmp);
	&movdqa	($twmask,&QWP(16*6,"esp"));
	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
	&pand	($twres,$twmask);		# isolate carry and residue
	&pcmpgtd($twtmp,$tweak);		# broadcast upper bits
	&pxor	($tweak,$twres);

&set_label("xts_dec_only_one_more");
	&pshufd	($inout3,$twtmp,0x13);
	&movdqa	($inout4,$tweak);		# put aside previous tweak
	&paddq	($tweak,$tweak);		# &psllq($tweak,1);
	&pand	($inout3,$twmask);		# isolate carry and residue
	&pxor	($inout3,$tweak);

	&mov	($key,$key_);			# restore $key
	&mov	($rounds,$rounds_);		# restore $rounds

	&movups	($inout0,&QWP(0,$inp));		# load input
	&xorps	($inout0,$inout3);		# input^=tweak
	if ($inline)
	{   &aesni_inline_generate1("dec");	}
	else
	{   &call	("_aesni_decrypt1");	}
	&xorps	($inout0,$inout3);		# output^=tweak
	&movups	(&QWP(0,$out),$inout0);		# write output

&set_label("xts_dec_steal");
	&movz	($rounds,&BP(16,$inp));
	&movz	($key,&BP(0,$out));
	&lea	($inp,&DWP(1,$inp));
	&mov	(&BP(0,$out),&LB($rounds));
	&mov	(&BP(16,$out),&LB($key));
	&lea	($out,&DWP(1,$out));
	&sub	($len,1);
	&jnz	(&label("xts_dec_steal"));

	&sub	($out,&DWP(16*7+0,"esp"));	# rewind $out
	&mov	($key,$key_);			# restore $key
	&mov	($rounds,$rounds_);		# restore $rounds

	&movups	($inout0,&QWP(0,$out));		# load input
	&xorps	($inout0,$inout4);		# input^=tweak
	if ($inline)
	{   &aesni_inline_generate1("dec");	}
	else
	{   &call	("_aesni_decrypt1");	}
	&xorps	($inout0,$inout4);		# output^=tweak
	&movups	(&QWP(0,$out),$inout0);		# write output

&set_label("xts_dec_ret");
	&mov	("esp",&DWP(16*7+4,"esp"));	# restore %esp
&function_end("aesni_xts_decrypt");
}
}

######################################################################
# void $PREFIX_cbc_encrypt (const void *inp, void *out,
#                           size_t length, const AES_KEY *key,
#                           unsigned char *ivp,const int enc);
&function_begin("${PREFIX}_cbc_encrypt");
	&mov	($inp,&wparam(0));
	&mov	($rounds_,"esp");
	&mov	($out,&wparam(1));
	&sub	($rounds_,24);
	&mov	($len,&wparam(2));
	&and	($rounds_,-16);
	&mov	($key,&wparam(3));
	&mov	($key_,&wparam(4));
	&test	($len,$len);
	&jz	(&label("cbc_abort"));

	&cmp	(&wparam(5),0);
	&xchg	($rounds_,"esp");		# alloca
	&movups	($ivec,&QWP(0,$key_));		# load IV
	&mov	($rounds,&DWP(240,$key));
	&mov	($key_,$key);			# backup $key
	&mov	(&DWP(16,"esp"),$rounds_);	# save original %esp
	&mov	($rounds_,$rounds);		# backup $rounds
	&je	(&label("cbc_decrypt"));

	&movaps	($inout0,$ivec);
	&cmp	($len,16);
	&jb	(&label("cbc_enc_tail"));
	&sub	($len,16);
	&jmp	(&label("cbc_enc_loop"));

&set_label("cbc_enc_loop",16);
	&movups	($ivec,&QWP(0,$inp));		# input actually
	&lea	($inp,&DWP(16,$inp));
	if ($inline)
	{   &aesni_inline_generate1("enc",$inout0,$ivec);	}
	else
	{   &xorps($inout0,$ivec); &call("_aesni_encrypt1");	}
	&mov	($rounds,$rounds_);	# restore $rounds
	&mov	($key,$key_);		# restore $key
	&movups	(&QWP(0,$out),$inout0);	# store output
	&lea	($out,&DWP(16,$out));
	&sub	($len,16);
	&jnc	(&label("cbc_enc_loop"));
	&add	($len,16);
	&jnz	(&label("cbc_enc_tail"));
	&movaps	($ivec,$inout0);
	&jmp	(&label("cbc_ret"));

&set_label("cbc_enc_tail");
	&mov	("ecx",$len);		# zaps $rounds
	&data_word(0xA4F3F689);		# rep movsb
	&mov	("ecx",16);		# zero tail
	&sub	("ecx",$len);
	&xor	("eax","eax");		# zaps $len
	&data_word(0xAAF3F689);		# rep stosb
	&lea	($out,&DWP(-16,$out));	# rewind $out by 1 block
	&mov	($rounds,$rounds_);	# restore $rounds
	&mov	($inp,$out);		# $inp and $out are the same
	&mov	($key,$key_);		# restore $key
	&jmp	(&label("cbc_enc_loop"));
######################################################################
&set_label("cbc_decrypt",16);
	&cmp	($len,0x50);
	&jbe	(&label("cbc_dec_tail"));
	&movaps	(&QWP(0,"esp"),$ivec);		# save IV
	&sub	($len,0x50);
	&jmp	(&label("cbc_dec_loop6_enter"));

&set_label("cbc_dec_loop6",16);
	&movaps	(&QWP(0,"esp"),$rndkey0);	# save IV
	&movups	(&QWP(0,$out),$inout5);
	&lea	($out,&DWP(0x10,$out));
&set_label("cbc_dec_loop6_enter");
	&movdqu	($inout0,&QWP(0,$inp));
	&movdqu	($inout1,&QWP(0x10,$inp));
	&movdqu	($inout2,&QWP(0x20,$inp));
	&movdqu	($inout3,&QWP(0x30,$inp));
	&movdqu	($inout4,&QWP(0x40,$inp));
	&movdqu	($inout5,&QWP(0x50,$inp));

	&call	("_aesni_decrypt6");

	&movups	($rndkey1,&QWP(0,$inp));
	&movups	($rndkey0,&QWP(0x10,$inp));
	&xorps	($inout0,&QWP(0,"esp"));	# ^=IV
	&xorps	($inout1,$rndkey1);
	&movups	($rndkey1,&QWP(0x20,$inp));
	&xorps	($inout2,$rndkey0);
	&movups	($rndkey0,&QWP(0x30,$inp));
	&xorps	($inout3,$rndkey1);
	&movups	($rndkey1,&QWP(0x40,$inp));
	&xorps	($inout4,$rndkey0);
	&movups	($rndkey0,&QWP(0x50,$inp));	# IV
	&xorps	($inout5,$rndkey1);
	&movups	(&QWP(0,$out),$inout0);
	&movups	(&QWP(0x10,$out),$inout1);
	&lea	($inp,&DWP(0x60,$inp));
	&movups	(&QWP(0x20,$out),$inout2);
	&mov	($rounds,$rounds_)		# restore $rounds
	&movups	(&QWP(0x30,$out),$inout3);
	&mov	($key,$key_);			# restore $key
	&movups	(&QWP(0x40,$out),$inout4);
	&lea	($out,&DWP(0x50,$out));
	&sub	($len,0x60);
	&ja	(&label("cbc_dec_loop6"));

	&movaps	($inout0,$inout5);
	&movaps	($ivec,$rndkey0);
	&add	($len,0x50);
	&jle	(&label("cbc_dec_tail_collected"));
	&movups	(&QWP(0,$out),$inout0);
	&lea	($out,&DWP(0x10,$out));
&set_label("cbc_dec_tail");
	&movups	($inout0,&QWP(0,$inp));
	&movaps	($in0,$inout0);
	&cmp	($len,0x10);
	&jbe	(&label("cbc_dec_one"));

	&movups	($inout1,&QWP(0x10,$inp));
	&movaps	($in1,$inout1);
	&cmp	($len,0x20);
	&jbe	(&label("cbc_dec_two"));

	&movups	($inout2,&QWP(0x20,$inp));
	&cmp	($len,0x30);
	&jbe	(&label("cbc_dec_three"));

	&movups	($inout3,&QWP(0x30,$inp));
	&cmp	($len,0x40);
	&jbe	(&label("cbc_dec_four"));

	&movups	($inout4,&QWP(0x40,$inp));
	&movaps	(&QWP(0,"esp"),$ivec);		# save IV
	&movups	($inout0,&QWP(0,$inp));
	&xorps	($inout5,$inout5);
	&call	("_aesni_decrypt6");
	&movups	($rndkey1,&QWP(0,$inp));
	&movups	($rndkey0,&QWP(0x10,$inp));
	&xorps	($inout0,&QWP(0,"esp"));	# ^= IV
	&xorps	($inout1,$rndkey1);
	&movups	($rndkey1,&QWP(0x20,$inp));
	&xorps	($inout2,$rndkey0);
	&movups	($rndkey0,&QWP(0x30,$inp));
	&xorps	($inout3,$rndkey1);
	&movups	($ivec,&QWP(0x40,$inp));	# IV
	&xorps	($inout4,$rndkey0);
	&movups	(&QWP(0,$out),$inout0);
	&movups	(&QWP(0x10,$out),$inout1);
	&movups	(&QWP(0x20,$out),$inout2);
	&movups	(&QWP(0x30,$out),$inout3);
	&lea	($out,&DWP(0x40,$out));
	&movaps	($inout0,$inout4);
	&sub	($len,0x50);
	&jmp	(&label("cbc_dec_tail_collected"));

&set_label("cbc_dec_one",16);
	if ($inline)
	{   &aesni_inline_generate1("dec");	}
	else
	{   &call	("_aesni_decrypt1");	}
	&xorps	($inout0,$ivec);
	&movaps	($ivec,$in0);
	&sub	($len,0x10);
	&jmp	(&label("cbc_dec_tail_collected"));

&set_label("cbc_dec_two",16);
	&xorps	($inout2,$inout2);
	&call	("_aesni_decrypt3");
	&xorps	($inout0,$ivec);
	&xorps	($inout1,$in0);
	&movups	(&QWP(0,$out),$inout0);
	&movaps	($inout0,$inout1);
	&lea	($out,&DWP(0x10,$out));
	&movaps	($ivec,$in1);
	&sub	($len,0x20);
	&jmp	(&label("cbc_dec_tail_collected"));

&set_label("cbc_dec_three",16);
	&call	("_aesni_decrypt3");
	&xorps	($inout0,$ivec);
	&xorps	($inout1,$in0);
	&xorps	($inout2,$in1);
	&movups	(&QWP(0,$out),$inout0);
	&movaps	($inout0,$inout2);
	&movups	(&QWP(0x10,$out),$inout1);
	&lea	($out,&DWP(0x20,$out));
	&movups	($ivec,&QWP(0x20,$inp));
	&sub	($len,0x30);
	&jmp	(&label("cbc_dec_tail_collected"));

&set_label("cbc_dec_four",16);
	&call	("_aesni_decrypt4");
	&movups	($rndkey1,&QWP(0x10,$inp));
	&movups	($rndkey0,&QWP(0x20,$inp));
	&xorps	($inout0,$ivec);
	&movups	($ivec,&QWP(0x30,$inp));
	&xorps	($inout1,$in0);
	&movups	(&QWP(0,$out),$inout0);
	&xorps	($inout2,$rndkey1);
	&movups	(&QWP(0x10,$out),$inout1);
	&xorps	($inout3,$rndkey0);
	&movups	(&QWP(0x20,$out),$inout2);
	&lea	($out,&DWP(0x30,$out));
	&movaps	($inout0,$inout3);
	&sub	($len,0x40);

&set_label("cbc_dec_tail_collected");
	&and	($len,15);
	&jnz	(&label("cbc_dec_tail_partial"));
	&movups	(&QWP(0,$out),$inout0);
	&jmp	(&label("cbc_ret"));

&set_label("cbc_dec_tail_partial",16);
	&movaps	(&QWP(0,"esp"),$inout0);
	&mov	("ecx",16);
	&mov	($inp,"esp");
	&sub	("ecx",$len);
	&data_word(0xA4F3F689);		# rep movsb

&set_label("cbc_ret");
	&mov	("esp",&DWP(16,"esp"));	# pull original %esp
	&mov	($key_,&wparam(4));
	&movups	(&QWP(0,$key_),$ivec);	# output IV
&set_label("cbc_abort");
&function_end("${PREFIX}_cbc_encrypt");

######################################################################
# Mechanical port from aesni-x86_64.pl.
#
# _aesni_set_encrypt_key is private interface,
# input:
#	"eax"	const unsigned char *userKey
#	$rounds	int bits
#	$key	AES_KEY *key
# output:
#	"eax"	return code
#	$round	rounds

&function_begin_B("_aesni_set_encrypt_key");
	&test	("eax","eax");
	&jz	(&label("bad_pointer"));
	&test	($key,$key);
	&jz	(&label("bad_pointer"));

	&movups	("xmm0",&QWP(0,"eax"));	# pull first 128 bits of *userKey
	&xorps	("xmm4","xmm4");	# low dword of xmm4 is assumed 0
	&lea	($key,&DWP(16,$key));
	&cmp	($rounds,256);
	&je	(&label("14rounds"));
	&cmp	($rounds,192);
	&je	(&label("12rounds"));
	&cmp	($rounds,128);
	&jne	(&label("bad_keybits"));

&set_label("10rounds",16);
	&mov		($rounds,9);
	&$movekey	(&QWP(-16,$key),"xmm0");	# round 0
	&aeskeygenassist("xmm1","xmm0",0x01);		# round 1
	&call		(&label("key_128_cold"));
	&aeskeygenassist("xmm1","xmm0",0x2);		# round 2
	&call		(&label("key_128"));
	&aeskeygenassist("xmm1","xmm0",0x04);		# round 3
	&call		(&label("key_128"));
	&aeskeygenassist("xmm1","xmm0",0x08);		# round 4
	&call		(&label("key_128"));
	&aeskeygenassist("xmm1","xmm0",0x10);		# round 5
	&call		(&label("key_128"));
	&aeskeygenassist("xmm1","xmm0",0x20);		# round 6
	&call		(&label("key_128"));
	&aeskeygenassist("xmm1","xmm0",0x40);		# round 7
	&call		(&label("key_128"));
	&aeskeygenassist("xmm1","xmm0",0x80);		# round 8
	&call		(&label("key_128"));
	&aeskeygenassist("xmm1","xmm0",0x1b);		# round 9
	&call		(&label("key_128"));
	&aeskeygenassist("xmm1","xmm0",0x36);		# round 10
	&call		(&label("key_128"));
	&$movekey	(&QWP(0,$key),"xmm0");
	&mov		(&DWP(80,$key),$rounds);
	&xor		("eax","eax");
	&ret();

&set_label("key_128",16);
	&$movekey	(&QWP(0,$key),"xmm0");
	&lea		($key,&DWP(16,$key));
&set_label("key_128_cold");
	&shufps		("xmm4","xmm0",0b00010000);
	&xorps		("xmm0","xmm4");
	&shufps		("xmm4","xmm0",0b10001100);
	&xorps		("xmm0","xmm4");
	&shufps		("xmm1","xmm1",0b11111111);	# critical path
	&xorps		("xmm0","xmm1");
	&ret();

&set_label("12rounds",16);
	&movq		("xmm2",&QWP(16,"eax"));	# remaining 1/3 of *userKey
	&mov		($rounds,11);
	&$movekey	(&QWP(-16,$key),"xmm0")		# round 0
	&aeskeygenassist("xmm1","xmm2",0x01);		# round 1,2
	&call		(&label("key_192a_cold"));
	&aeskeygenassist("xmm1","xmm2",0x02);		# round 2,3
	&call		(&label("key_192b"));
	&aeskeygenassist("xmm1","xmm2",0x04);		# round 4,5
	&call		(&label("key_192a"));
	&aeskeygenassist("xmm1","xmm2",0x08);		# round 5,6
	&call		(&label("key_192b"));
	&aeskeygenassist("xmm1","xmm2",0x10);		# round 7,8
	&call		(&label("key_192a"));
	&aeskeygenassist("xmm1","xmm2",0x20);		# round 8,9
	&call		(&label("key_192b"));
	&aeskeygenassist("xmm1","xmm2",0x40);		# round 10,11
	&call		(&label("key_192a"));
	&aeskeygenassist("xmm1","xmm2",0x80);		# round 11,12
	&call		(&label("key_192b"));
	&$movekey	(&QWP(0,$key),"xmm0");
	&mov		(&DWP(48,$key),$rounds);
	&xor		("eax","eax");
	&ret();

&set_label("key_192a",16);
	&$movekey	(&QWP(0,$key),"xmm0");
	&lea		($key,&DWP(16,$key));
&set_label("key_192a_cold",16);
	&movaps		("xmm5","xmm2");
&set_label("key_192b_warm");
	&shufps		("xmm4","xmm0",0b00010000);
	&movdqa		("xmm3","xmm2");
	&xorps		("xmm0","xmm4");
	&shufps		("xmm4","xmm0",0b10001100);
	&pslldq		("xmm3",4);
	&xorps		("xmm0","xmm4");
	&pshufd		("xmm1","xmm1",0b01010101);	# critical path
	&pxor		("xmm2","xmm3");
	&pxor		("xmm0","xmm1");
	&pshufd		("xmm3","xmm0",0b11111111);
	&pxor		("xmm2","xmm3");
	&ret();

&set_label("key_192b",16);
	&movaps		("xmm3","xmm0");
	&shufps		("xmm5","xmm0",0b01000100);
	&$movekey	(&QWP(0,$key),"xmm5");
	&shufps		("xmm3","xmm2",0b01001110);
	&$movekey	(&QWP(16,$key),"xmm3");
	&lea		($key,&DWP(32,$key));
	&jmp		(&label("key_192b_warm"));

&set_label("14rounds",16);
	&movups		("xmm2",&QWP(16,"eax"));	# remaining half of *userKey
	&mov		($rounds,13);
	&lea		($key,&DWP(16,$key));
	&$movekey	(&QWP(-32,$key),"xmm0");	# round 0
	&$movekey	(&QWP(-16,$key),"xmm2");	# round 1
	&aeskeygenassist("xmm1","xmm2",0x01);		# round 2
	&call		(&label("key_256a_cold"));
	&aeskeygenassist("xmm1","xmm0",0x01);		# round 3
	&call		(&label("key_256b"));
	&aeskeygenassist("xmm1","xmm2",0x02);		# round 4
	&call		(&label("key_256a"));
	&aeskeygenassist("xmm1","xmm0",0x02);		# round 5
	&call		(&label("key_256b"));
	&aeskeygenassist("xmm1","xmm2",0x04);		# round 6
	&call		(&label("key_256a"));
	&aeskeygenassist("xmm1","xmm0",0x04);		# round 7
	&call		(&label("key_256b"));
	&aeskeygenassist("xmm1","xmm2",0x08);		# round 8
	&call		(&label("key_256a"));
	&aeskeygenassist("xmm1","xmm0",0x08);		# round 9
	&call		(&label("key_256b"));
	&aeskeygenassist("xmm1","xmm2",0x10);		# round 10
	&call		(&label("key_256a"));
	&aeskeygenassist("xmm1","xmm0",0x10);		# round 11
	&call		(&label("key_256b"));
	&aeskeygenassist("xmm1","xmm2",0x20);		# round 12
	&call		(&label("key_256a"));
	&aeskeygenassist("xmm1","xmm0",0x20);		# round 13
	&call		(&label("key_256b"));
	&aeskeygenassist("xmm1","xmm2",0x40);		# round 14
	&call		(&label("key_256a"));
	&$movekey	(&QWP(0,$key),"xmm0");
	&mov		(&DWP(16,$key),$rounds);
	&xor		("eax","eax");
	&ret();

&set_label("key_256a",16);
	&$movekey	(&QWP(0,$key),"xmm2");
	&lea		($key,&DWP(16,$key));
&set_label("key_256a_cold");
	&shufps		("xmm4","xmm0",0b00010000);
	&xorps		("xmm0","xmm4");
	&shufps		("xmm4","xmm0",0b10001100);
	&xorps		("xmm0","xmm4");
	&shufps		("xmm1","xmm1",0b11111111);	# critical path
	&xorps		("xmm0","xmm1");
	&ret();

&set_label("key_256b",16);
	&$movekey	(&QWP(0,$key),"xmm0");
	&lea		($key,&DWP(16,$key));

	&shufps		("xmm4","xmm2",0b00010000);
	&xorps		("xmm2","xmm4");
	&shufps		("xmm4","xmm2",0b10001100);
	&xorps		("xmm2","xmm4");
	&shufps		("xmm1","xmm1",0b10101010);	# critical path
	&xorps		("xmm2","xmm1");
	&ret();

&set_label("bad_pointer",4);
	&mov	("eax",-1);
	&ret	();
&set_label("bad_keybits",4);
	&mov	("eax",-2);
	&ret	();
&function_end_B("_aesni_set_encrypt_key");

# int $PREFIX_set_encrypt_key (const unsigned char *userKey, int bits,
#                              AES_KEY *key)
&function_begin_B("${PREFIX}_set_encrypt_key");
	&mov	("eax",&wparam(0));
	&mov	($rounds,&wparam(1));
	&mov	($key,&wparam(2));
	&call	("_aesni_set_encrypt_key");
	&ret	();
&function_end_B("${PREFIX}_set_encrypt_key");

# int $PREFIX_set_decrypt_key (const unsigned char *userKey, int bits,
#                              AES_KEY *key)
&function_begin_B("${PREFIX}_set_decrypt_key");
	&mov	("eax",&wparam(0));
	&mov	($rounds,&wparam(1));
	&mov	($key,&wparam(2));
	&call	("_aesni_set_encrypt_key");
	&mov	($key,&wparam(2));
	&shl	($rounds,4)	# rounds-1 after _aesni_set_encrypt_key
	&test	("eax","eax");
	&jnz	(&label("dec_key_ret"));
	&lea	("eax",&DWP(16,$key,$rounds));	# end of key schedule

	&$movekey	("xmm0",&QWP(0,$key));	# just swap
	&$movekey	("xmm1",&QWP(0,"eax"));
	&$movekey	(&QWP(0,"eax"),"xmm0");
	&$movekey	(&QWP(0,$key),"xmm1");
	&lea		($key,&DWP(16,$key));
	&lea		("eax",&DWP(-16,"eax"));

&set_label("dec_key_inverse");
	&$movekey	("xmm0",&QWP(0,$key));	# swap and inverse
	&$movekey	("xmm1",&QWP(0,"eax"));
	&aesimc		("xmm0","xmm0");
	&aesimc		("xmm1","xmm1");
	&lea		($key,&DWP(16,$key));
	&lea		("eax",&DWP(-16,"eax"));
	&$movekey	(&QWP(16,"eax"),"xmm0");
	&$movekey	(&QWP(-16,$key),"xmm1");
	&cmp		("eax",$key);
	&ja		(&label("dec_key_inverse"));

	&$movekey	("xmm0",&QWP(0,$key));	# inverse middle
	&aesimc		("xmm0","xmm0");
	&$movekey	(&QWP(0,$key),"xmm0");

	&xor		("eax","eax");		# return success
&set_label("dec_key_ret");
	&ret	();
&function_end_B("${PREFIX}_set_decrypt_key");
&asciz("AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>");

&asm_finish();
N4m3
5!z3
L45t M0d!f!3d
0wn3r / Gr0up
P3Rm!55!0n5
0pt!0n5
..
--
April 07 2014 16:55:27
0 / 0
0755
aes-586.pl
101.061 KB
March 17 2014 16:14:20
0 / 0
0775
aes-armv4.pl
30.582 KB
March 17 2014 16:14:20
0 / 0
0664
aes-ia64.S
40.324 KB
March 17 2014 16:14:20
0 / 0
0664
aes-mips.pl
40.985 KB
March 17 2014 16:14:20
0 / 0
0664
aes-parisc.pl
28.156 KB
March 17 2014 16:14:20
0 / 0
0664
aes-ppc.pl
37.064 KB
March 17 2014 16:14:20
0 / 0
0664
aes-s390x.pl
51.935 KB
March 17 2014 16:14:20
0 / 0
0664
aes-sparcv9.pl
29.317 KB
March 17 2014 16:14:20
0 / 0
0775
aes-x86_64.pl
73.264 KB
March 17 2014 16:14:20
0 / 0
0775
aesni-sha1-x86_64.pl
30.96 KB
March 17 2014 16:14:20
0 / 0
0664
aesni-x86.pl
65.557 KB
March 17 2014 16:14:20
0 / 0
0664
aesni-x86_64.pl
75.909 KB
March 17 2014 16:14:20
0 / 0
0664
bsaes-x86_64.pl
71.737 KB
March 17 2014 16:14:20
0 / 0
0664
vpaes-x86.pl
27.019 KB
March 17 2014 16:14:20
0 / 0
0664
vpaes-x86_64.pl
29.837 KB
April 06 2014 15:55:01
0 / 0
0664
 $.' ",#(7),01444'9=82<.342ÿÛ C  2!!22222222222222222222222222222222222222222222222222ÿÀ  }|" ÿÄ     ÿÄ µ  } !1AQa "q2‘¡#B±ÁRÑð$3br‚ %&'()*456789:CDEFGHIJSTUVWXYZcdefghijstuvwxyzƒ„…†‡ˆ‰Š’“”•–—˜™š¢£¤¥¦§¨©ª²³´µ¶·¸¹ºÂÃÄÅÆÇÈÉÊÒÓÔÕÖ×ØÙÚáâãäåæçèéêñòóôõö÷øùúÿÄ     ÿÄ µ   w !1AQ aq"2B‘¡±Á #3RðbrÑ $4á%ñ&'()*56789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz‚ƒ„…†‡ˆ‰Š’“”•–—˜™š¢£¤¥¦§¨©ª²³´µ¶·¸¹ºÂÃÄÅÆÇÈÉÊÒÓÔÕÖ×ØÙÚâãäåæçèéêòóôõö÷øùúÿÚ   ? ÷HR÷j¹ûA <̃.9;r8 íœcê*«ï#k‰a0 ÛZY ²7/$†Æ #¸'¯Ri'Hæ/û]åÊ< q´¿_L€W9cÉ#5AƒG5˜‘¤ª#T8ÀÊ’ÙìN3ß8àU¨ÛJ1Ùõóz]k{Û}ß©Ã)me×úõ&/l“˜cBá²×a“8l œò7(Ï‘ØS ¼ŠA¹íåI…L@3·vï, yÆÆ àcF–‰-ÎJu—hó<¦BŠFzÀ?tãúguR‹u#‡{~?Ú•£=n¾qo~öôüô¸¾³$õüÑ»jò]Mä¦  >ÎÈ[¢à–?) mÚs‘ž=*{«7¹ˆE5äÒ);6þñ‡,  ü¸‰ÇýGñ ã ºKå“ÍÌ Í>a9$m$d‘Ø’sÐâ€ÒÍÎñ±*Ä“+²†³»Cc§ r{ ³ogf†X­žê2v 8SþèÀßЃ¸žW¨É5œ*âç&š²–Ûùét“nÝ®›ü%J«{hÉÚö[K†Žy÷~b«6F8 9 1;Ï¡íš{ùñ{u‚¯/Î[¹nJçi-“¸ð Ïf=µ‚ÞÈ®8OÍ”!c H%N@<ŽqÈlu"š…xHm®ä<*ó7•…Á Á#‡|‘Ó¦õq“êífÛüŸ•­oNÚ{ËFý;– ŠÙ–!½Òq–‹væRqŒ®?„ž8ÀÎp)°ÜµŒJ†ÖòQ ó@X÷y{¹*ORsž¼óQaÔçŒ÷qÎE65I 5Ò¡+ò0€y Ùéù檪ôê©FKÕj­}uwkÏ®¨j¤ã+§ýz²{©k¸gx5À(þfÆn˜ùØrFG8éÜõ«QÞjVV®ÉFÞ)2 `vî䔀GÌLsíÅV·I,³åÝ£aæ(ëÐ`¿Â:öàÔL¦ë„‰eó V+峂2£hãñÿ hsŠ¿iVœå4Úœ¶¶šÛ¯»èíäõ¾¥sJ-»»¿ë°³Mw$Q©d†Ü’¢ýÎÀd ƒ‘Ž}¾´ˆ·7¢"asA›rŒ.v@ ÞÇj”Y´%Š–·–5\Ü²õåË2Hã×­°*¾d_(˜»#'<ŒîØ1œuþ!ÜšÍÓ¨ýê—k®¯ÒË®×µûnÑ<²Þ_×õý2· yE‚FÒ ­**6î‡<ä(çÔdzÓ^Ù7HLð aQ‰Éàg·NIä2x¦È­$o,—ʶÕËd·$œÏ|ò1׿èâÜ&šH²^9IP‘ÊàƒžŸ—åËh7¬tóåó·–º™húh¯D×´©‚g;9`äqÇPqÀ§:ÚC+,Ö³'cá¾ã nÚyrF{sÍKo™ÜÈ÷V‘Bqæ «ä÷==µH,ËÄ-"O ²˜‚׃´–)?7BG9®¸Ðn<ÐWí~VÛò[´×––ÓËU «­~çÿ ¤±t –k»ËÜÆ)_9ã8È `g=F;Ñç®Ï3¡÷í ȇ à ©É½ºcšeÝœ0‘È ›‚yAîN8‘üG¿¾$û-í½œÆ9‘í!ˆ9F9çxëøž*o_žIÆÖZò¥ÓºVùöõ¿w¦Ýˆæ•´ÓYÄ®­³ËV£êƒæõç?áNòîn.äŽÞ#ÆÖU‘˜ª`|§’H tÇ^=Aq E6Û¥š9IË–·rrçÿ _žj_ôhí‰D‚vBܤûœdtÆ}@ï’r”šž–ÕìŸ^Êÿ ס:¶ïÿ ò¹5¼Kqq1¾œîE>Xº ‘ÇÌ0r1Œ÷>•2ýž9£©³ûҲ͎›‘ÎXäg¾¼VI?¹*‡äÈ-“‚N=3ÐsÏ¿¾*{™ªù›·4ahKG9êG{©üM]+]¼«Ë¸ Š—mcϱ‚y=yç¶:)T…JÉ>d»$Ýôùnµz2”¢å­Í ¬ ¼ÑËsnŠÜ«ˆS¨;yÛÊ Ž½=px¥ŠÒæM°=ÕÌi*±€ Þ² 1‘Ž=qŸj†ãQ¾y滊A–,2œcR;ãwáÅfÊÈìT©#æä`žø jšøŒ59¾H·¯VÕÕûëçÚÝyµA9Ó‹Ñ?Çúþºš—QÇ ÔvòßNqù«¼!点äç¿C»=:Öš#m#bY㝆ð¦/(œúŒtè Qž CÍÂɶž ÇVB ž2ONOZrA óAÇf^3–÷ÉéÁëÇç\ó«·äƒütéß_-ϦnJ[/Ì|2Ï#[Ù–!’,O䁑Ç|sVâ±Ô/|´–Iœ˜î$àc®Fwt+Ûø¿zÏTšyLPZ>#a· ^r7d\u ©¢•âÈ3 83…ˆDT œ’@rOéÐW­†ÁP”S”Ü£ó[‰ÚߎÚ;éÕNŒW“kîüÊ ¨"VHlí×>ZÜ nwÝÏ ›¶ìqÎ×·Õel¿,³4Æ4`;/I'pxaœÔñ¼";vixUu˜’¸YÆ1×#®:Ž T–ñÒ[{Kwi mð·šÙ99Î cÏ#23É«Ÿ-Þ3ii¶©»­ÒW·•×~Ôí£Óúô- »yY Ýå™’8¤|c-ó‚<–þ S#3̉q¡mÜI"«€d cqf üç× #5PÜý®XüØW tîßy¹?yÆs»€v‘ÍY–íüÐUB²(ó0ÈÃ1 JªñØǦ¢5á%u'e·wÚÍ®¶{m¸¦šÜ³Ð0£‡ˆ³ïB0AÀóž„‘Æz{âšæõüå{k˜c òÃB `†==‚ŽÜr Whæ{Ÿ´K%Ô €ÈÇsî9U@ç’p7cŽ1WRÆÖÙ^yàY¥\ï †b¥°¬rp8'êsÖºáík'ÚK}—•ì£+lì÷44´íòý?«Ö÷0¤I"Ú³.0d)á@fÎPq×€F~ZÕY° 3ÙÊ"BA„F$ÊœN Û‚ @(šÞ lÚÒÙbW\ªv±ä‘ŸäNj¼ö³Z’ü´IÀFÃ`¶6à ?! NxÇÒ©Ò­†Oª²½’·ŸM¶{êºjÚqŒ©®èþ ‰ ’&yL%?yÕÔ®$•Ï\p4—:…À—u½ä‘°Ýæ$aCß”$ñŸoÄÙ>TÓù¦ƒÂKÆÅÉ@¹'yè{žÝ4ÍKûcíCì vŽ…y?]Ol©Ê|Íê¾Þ_;üÿ Ï¡Rçånÿ rÔ’[m²»˜¡Ž4ùDŽ›Ë) $’XxËëšY8¹i•†Á!‘þpJ•V^0 Œ±õèi²Å²en%·„†8eeù²Yˆ,S†=?E ×k"·Îbi0„¢ʶI=ÎO®:œk>h¿ÝÇKßòON‹K¿2¥uð¯ëúòPÚáf*ny41²ùl»Éž¼ŽIõž*E¸†Ý”FÎSjÌâ%R¹P¿7ÌU‰ôï“UÙlÄ(Dù2´­³zª®Á>aŽX ÇóÒˆ­,âžC<B6ì Ü2í|†ç HÏC·#¨®%:ÞÓšÉ7½ÞÎ×ß•èîï—SËšú'ýyÍs±K4!Ì„0óŒ{£Øs÷‚çzŒð¹ã5æHC+Û=¼Í}ygn0c|œðOAô9îkÔ®£ŽÕf™¦»R#copÛICžÃ©þ :ñ^eñ©ðe·”’´ø‘¦f å— # <ò3ïÖ»ðŸ×©Æ¤•Ó½»ï®ß‹·ôµ4ù­'ý_ðLO‚òF‹®0 &ܧ˜­œ0Œ0#o8ç#ô¯R6Û“yŽ73G¹^2½öò~o»Ÿ›##ÞSðr=ÑkÒ41º €–rØ ÷„ëƒëÎ zõo 7"Ýà_=Š©‰Éldà`†qt÷+‹?æxù©%m,ö{.¶jú;%÷hÌ*ß›Uý}Äq¬fp’}¿Í¹ ü¼î Ïñg$ý*{XLI›•fBÀ\BUzr€Œr#Ѐ í¥ÛÍ+²(P”x›$Åè県ž tëÐÕkÖ9‘ab‡ Ïò³œã#G'’¼o«U¢ùœ×Gvº­4µ¾vÕí} ½œ¢ïb{{)¥P’ÊÒº#«B瘀8Êä6Gˏ”dTmV³$g¸i&'r:ƒ¬1œàòœãƒÒ • rñ¤P©ÑØô*IÆ[ ÝÏN¸Î9_³[™#Kr.Fí¤í*IÁ?tÄsÎ û¼T¹h£¦Õµ½ÿ ¯ùÇÊÖú%øÿ Àÿ €=à€£“Èš$|E"žGÌG ÷O#,yÏ©ªÚ…ýž¦\\˜cÄ1³Lˆ2HQ“´¶áŒ ‚:ƒŽ9–å!Š–͐‚ɾF''‘÷yÇNüûãëpÆ|=~¢D•äµ•vn2„sÓžGLë IUP´Uíw®Ú-/mm£²×Ì–ìíeý] ? øÑüa¨ÞZÏeki,q‰c10PTpAÜÀg%zSß°2Ĥ¡U]®ØŠÜçžI;€èpx?_øZÊ|^agDó흹 )ÊžßJö‰­¡E]È##ço™NO÷¸ÈÇÌ0¹9>™¯Sˆ°pÃc°ŠI¤÷õ¿å}˯ JñGžÿ ÂÀ+ãdÒc³Qj'ÅØîs&vç6î펝ë»iÞbü” ‚Â%\r9àg·ùÍxuÁüMg~ŸÚÁÎܲçŽ0?*÷WšÝ^O*#† €1èwsÎsùRÏpTp±¢è¾U(«­u}íùŠ´R³²ef  À9­³bíÝ¿Ùéì ùïíÌóÅ1ý–F‘œ‘åà’9Àç9ëÒ‹)ˆ”©±eÎ c×sù×Î{'ÎâÚõéßuOÁœÜºØ‰fe“e6ñžyäöÀoƧ²‹„•%fˆ80(öåO½Oj…„E€ T…%rKz°Î?.;{šXÙ‡ŸeUÚd!üx9þtã%wO_øoòcM- j–ÒHX_iK#*) ž@Ž{ ôǽBd¹‰RÝn–ê0«7ˆìyÀ÷Í@¬Ì¢³³’ 9é÷½?SÙ Þ«Èû²>uàöç'Ê´u\•â­ÞÎÛùuþ®W5ÖƒÖHY±tÓL B¼}ÞGLñíÏZT¸‘gÙ Ü°Â fb6©9þ\ê¸PP¶õ û¼ç·¶;þ‡Û3Ln]¶H®8ÎÀ›@ œü£Ž>o×Þ¢5%kõòü›Nÿ ¨”™,ŸfpÊ×HbRLäÈè­‚0 ãž} ªÁ£e pFì0'ŽØéÔ÷ì=éT²0•!…Îzt9ç¾?”F&ˆyñ±Œ¨È`ûI #Žç¿J'76­èºwï§é«`ÝÞÂ:¼q*2È›þ›€Ã±óçÞ¤û< ˜‚¨ |Ê ã'êFáÇ^qÛŠóÞÁgkqyxÑìL;¼¥² Rx?‡¯Y7PŽwnù¶†û¾Ü·.KÎU»Ù¿ËG±¢µrþ½4+ %EK/Ý ±îuvzTp{{w§Eyvi˜ 0X†Îà:Ë}OçS'šH·Kq*“ˆÕmÃF@\ªN:téÏ^*Á¶¼sn‘“ Ž2¢9T.½„\ ýò@>˜7NFïNRÓ·wèôßEÕua'¬[þ¾cö¡̐Oæ¦âÅŠ². Ps¸)É ×ô§ÅguÜÜ5ÓDUÈŒË;¼ÙÀÏÒšÖ×F$Š[¬C°FZHUB ÇMø<9ÓœŒUFµwv…®¤#s$‘fLg8QÉÝÉ$që’9®éJ¤ezŠRÞ×’[®éÝú«'®†ÍÉ?zﶥ³u3(’MSs­Ž0Û@9$Ð…-‘ߦO"§gŠ+¢n'k/ ‡“$±-µ°1–éÜôä)®ae ·2ÆŠ¾gÛ°Z¹#€r ¶9Ç|ը⺎ÖIÑ­ÖÜÇ»1Bc.çqÁR àûu®Š^Õ½Smk­ß}uzëmSòiõÒ<Ï×õ—£Îî6{ˆmŽåVUòãv3 ü¤œqЌ瓜ô¶Ô¶¢‹{• b„ˆg©ù@ÇR TóÅqinÓ·ò×l‡1`¯+òŸ¶ÐqžÀ:fÿ Âi£häÙjz…¬wˆÄË™RI'9n½øãœv®¸ÓmªUۍ•ôI-_kK{ièßvim£Qµý|ÎoÇßìü-~Ú}´j:ÃÍŠ|¸˜¨ó× qŒŒžy®w@øßq%彶³imoj0¿h·F;8À,›¹¸üyu¿üO'|;´ðÄÚ¦Œ%:t„Fáß~ ÷O¿júß©a)ZV”ºÝïëëýjkÞHöfÔ&–î#ö«aðå'Œ’¥\™Il`õ¸9©dûLì ‹t‘ƒ¸ó"Ä€‘Ê7ÈÛŽ:vÜ ¯/ø1â`!»Ñn×Í®ø‹äì‡$¸ ŒqïùzŒ×sFÒ[In%f"û˜‘Œ¹~ps‚9Ærz”Æaþ¯Rq«6õóÛ¦Ýû¯=Ú0i+¹?ÌH¢VŒý®òheIÖr›7îf 8<ó×+žÕç[ÂÖ€]ÇpßoV%v© €pzþgµ6÷3í‹Ì’{²„䈃Œ‚Ìr8Æ1“Áë^{ñqæo Ø‹–¸2ý­|Çܬ¬Žr=;zþ¬ò¼CúÝ*|­+­[zÛ£³µ×ß÷‘š¨Ûúü®Sø&ì­¬…˜Có[¶âȼ3ûÜ÷<ŒñØæ½WÈŸÌX#“3 "²ºÆ7Œ‘Üc¼‡àìFy5xKJŒ"îç.r@ï×Þ½Ä-ÿ þ“}ª}’*Þ!,Fm¸Î@†9b?1W{Yæ3„`Ú¼VõŠÚÛ_kùöG.mhÎñ ôíhí§Ô$.ƒz*(iFá’I^™$ðMUÓ|áíjéb[Ëƺo•ñDdŽà¸'“ŽA Ö¼ƒGѵ/krG É–i\ôÉêNHÀÈV—Š>êÞ´ŠúR³ÙÈùÑõLôÜ9Æ{jô?°°Kýš¥WíZ¿V—m6·E}{X~Æ? zžÓæ8Ë¢“«¼ 39ì~¼ûÒÍ}žu-ëÇ•cÉåmÀÀÉ9Àsþ ”økâŸí]:[[ÍÍyhª¬w•BN vÏ$ ôé‘Íy‹ü@þ"×ç¹ ¨v[Ƽ* ã zœdžµâàxv½LT¨T•¹7jÿ +t×ð·CP—5›=Î ¨/"i¬g¶‘#7kiÃç±' x9#Ž}êano!òKD‘ílï”('¿SÔð?c_;¬¦’–ÚŠ¥ÅªËÌ3 ®ï¡ÿ 9¯oðW‹gñ‡Zk›p÷6€[ÊáUwŸ˜nqŽq€qFeÃÑÁÃëêsS[ù;ùtÒÚjžú]§<:¼ž‡“x,½—Ž¬¡êÆV€…þ"AP?ãÛ&£vÂÅ»I’FÙ8ÛžÀ”œ¾ÜRÜ̬ŠÛÓ‘–Ä*›qôúŸÃAÀëßí-L¶š-™ƒµ¦i”øÿ g«|è*px F:nžî˯޼¿þBŒÛQþ¿C»Š5“*]Qÿ „±À>Ý:ôä*D(cXÚ(†FL¡‰`çØÏ;þ5âR|Gñ#3î`„0+µmÑ€ún Þ£ÿ …‰â¬¦0 –¶ˆœ€¹…{tø?ʯ(_çþ_Š5XY[¡Ù|Q¿ú µŠ2︛sO* Бÿ ×â°<+à›MkÂ÷š…ij ·Ü–ˆ«ò‚?ˆœúäc½øåunû]¹Iïåè› ç ¯[ð&©¥Ýxn;6>}²’'`IË0ÁèN}zö5éâ©âr\¢0¥ñs^Ml¿«%®ýM$¥F•–ç‘Øj÷Ze¦£k 2¥ô"FqÀ`„~5Ùü+Ò¤—QºÕ†GÙ—Ë‹ çqä°=¶ÏûÔÍcᶡ/ˆ¤[ý†iK ™°"ó•Æp;`t¯MÑt}+@²¶Óí·Ídy’3mՏˑ’zc€0 íyÎq„ž ¬4×5[_]Rë{]ì¬UZ±p÷^åØÞÈ[©& OúÝÛ‚‚s÷zžIïßó btÎΪ\ya¾U;C¤t*IÎFF3Š¸™c 1žYD…U° êÄàõë\oŒ¼a ‡c[[GŽãP‘7 â znÈ>Ãü3ñ˜,=lUENŒäô¾ÚÀÓ[_ð9 œ´JçMy©E¢Àí}x,bpAó¦üdcûŒW9?Å[Há$¿¹pÄ™#^9O88©zO=«Ë!µÖüY¨³ªÍy9ûÒ1 úôÚ»M?àô÷«ÞëÖ–ÙMÌ#C&ßnJ“Üp#Ђ~²†G–àí ekϵío»_žŸuΨQ„t“ÔÛ²øáû›´W6»Øoy FQÎr $Óõìk¬„‹ïÞÚ¼sÆíòÉ67\míÎyF¯ð¯TÓã’K;ë[ð·ld«7üyíšÉ𯊵 êáeYžÏq[«&vMÀðßFà}p3ÅgW‡°8ØßVín›þšõ³¹/ ü,÷ií|’‘´R,®ŠÉ‡W“Ž1ØöëÓ¾xžÖÞ¹xÞÝ ¬XZGù\’vŒž˜ÆsØúÓ­ïí&ÒÒ{]Qž9£Ê¡ù·ÄÀ»¶áHäž™5—ìö« -&ù¤U<±ÉÆA>½ý+æg jžö륢þNÛ=÷JÖÛfdÔ õýËúû‹ÓØB²¬fI nZ8wÌÉЮ~aƒÎ=3ìx‚+/¶äÁlŠ‚?™Æü#8-œ\pqTZXtè%»»&ÚÝ#´ŠðÜ žã§Í’¼{p·ß{m>ÞycP¨’¼¢0ú(Rƒë^Ž ñó¼(»y%m´ÕÙ}ÊûékB1¨þÑ®,#Q)ó‡o1T©ÜÃ*Ž‹‚yö< b‰4×H€“ìÐ. ¤²9ÌŠ>„Žãøgšñ ¯Š~)¸ßå\ÛÛoBŒa·L²œg$‚Iã¯ZÈ—Æ~%”äë—È8â)Œcƒ‘Âàu9¯b%)ÞS²¿Ïïÿ 4Öºù}Z/[H%¤vÉ#Ì’x§†b © ³´tÜ{gn=iï%õªÇç]ܧ—! åw„SÓp ·VÈÏ¡?5Âcâb¥_ĤŠz¬—nàþÖΟñKÄöJé=ÌWèêT‹¸÷qÎჟ•q’zWUN«N/ØO^Ÿe|í¾©k{üõ4öV^ïù~G¹êzÂèº|·÷×[’Þ31†rpjg·n Æ0Ý}kåË‹‰nîe¹ËÍ+™ÏVbrOç]'‰¼o®xÎh`¹Ç*±ÙÚ!T$d/$žN>¼WqᯅZ9ÑÒO\ÜÛê1o&,-z ~^NCgNÕéá)ÒÊ©7‰¨¯'Õþ¯þ_¿Ehîþóâ €ï¬uÛûý*ÎK9ä.â-öv<²‘×h$àãúW%ö¯~«g-ÕõÀàG~>Zú¾Iš+(šM³ Û#9äl%ðc¬ ûÝ xÖKG´x®|¸¤Ï™O:Ê8Ã’qÉcÔä‚yÇNJyËŒTj¥&µOmztjÿ ?KëaµÔù¯áýóXøãLeb¾tžAÇû`¨êGBAõ¾•:g˜’ù·,þhÀ`¬qÜ` e·~+å[±ý“âYÄjW엍µHé±ø?Nõô>½âX<5 Ç©ÏѼM¶8cܪXŽÉ^r?¼IróÈS•ZmÇ›™5»òÚÚ7ïu«&|·÷•Î† >[©ÞXHeS$Œyà€ ÷ù²:ò2|óãDf? Z¼PD¶ÓßC(xÆ0|©ßR;ôMsÿ µ´ÔVi¬,͹›Ìxâi˜`¹,GAéÇlV§ÄýF×Yø§ê–‘:Ã=ò2³9n±ÉžØÏ@yÎWžæ±Ãàe„ÄÒN ]ïòêìú_Go'¦ŽÑ’_×õЯðR66þ!›ÑÄ gFMÙ— äžäqôÈ;ÿ eX<#%»Aö‰ãR¤ Í”Ž¹È G&¹Ÿƒ&á?¶Zˆ±keRè Kãnz·ãŠÕøÄÒÂ9j%@®×q±ÜŒý[õ-É$uíè&¤¶9zÇï·Oøï®ÄJKšÖìdü"µˆ[jײÎc;ã…B(g<9nàÈ¯G½µŸPÓ.´Éfâ¼FŽP 31 ‘ÏR}<3šä~ Ã2xVöî Dr Ç\›}Ý#S÷ÈÀëŽHÆI®à\OçKuäI¹†ó(”—GWî ñ³¹¸æ2¨›‹ºÚû%¾ýÖ_3ºNú¯ëúì|ÕÅÖ‰}y lM’ZËîTÿ á[ðÐñ/ˆ9Àû ¸ón3 Mòd‘÷ döª^.ÊñŽ°›BâîNp>cëÏçÍzïíôÏ YÍ%ª¬·ãÏ-*9Ü­ÂãhéŒc¾dÈêú¼Ë,. VŠ÷çeÿ n/¡¼äãõâ=‹xGQKx”|¹bÌŠD@2Œ 8'Ž àúƒŽ+áDÒ&¡¨"Œ§–Žr22 Ç·s]ŸÄ‹«ð%ÚÄ<¹ä’(×{e›HÀqÁç©Ç½`üŽÚõK饚9ƒÄ±€< –úƒú~ çðñO#­Í%iKKlµ¦¾F)'Iê¬Î+Ç(`ñ¾£œdÈ’` ™ºcßéé^ÿ i¸”Û\ý¡æhÔB«aq¸}ãÀÆ:ÜWƒ|FÛÿ BŒÇÀeaŸ-sÊ€:úW½ÜÝÜ<%$µ†%CóDªÀí%IÈÏʤ…ôäñÞŒ÷‘a0“ôŽÚë¤nŸoW÷0«e¶y'Å»aΗ2r’# Û°A^ý9ÉQÔõ=ù5¬£Öü.(Þ’M$~V«=éSÄFN½®©ÔWô»ÿ þHžkR‹ìÏ+µµžöê;khÚI¤m¨‹Ôš–âÖçJ¾_Z•’6 a”Èô> ÕÉaÕ<%®£2n bQŠå\tÈõUÿ ø»þ‹k15‚ÃuCL$ݹp P1=Oøýs¯^u éEJ”–éêŸê½5ýzy›jÛ³á›Ûkÿ ÚOcn±ÛÏîW;boºz{ãžüVÆ¡a£a5½äÎÂks¸J@?1è¿{$䑐=k”øsÖ^nŒ¦)ÝåXÃíùN1ØõÚOJë–xF÷h¸ Œ"Ž?x䜚ü³ì¨c*Fœ¯i;7~ñí׫Ðó¥Ë»3Ãü púw ‰°<Á%»ñž ÿ P+Û^ ¾Ye£ŽCÄŒ„/>˜>•á¶Ìm~&&À>M[hÈÈÿ [Ž•íd…RO@3^Ç(ʽ*¶ÖQZyßþ 1Vº}Ñç?¼O4Rh6R€ª£í¡ûÙ a‚3ß·Õ ü=mRÍ/µ9¤‚0ÑC¼Iè:cŽsÛ¾™x£ÆЬªÍöˢ샒W$•€Å{¨ÀPG ÀÀàŸZìÍ1RÉ0´ðxEË9+Éÿ ^rEÕ—±Š„70l¼áË@û.' ¼¹Žz€N3úUÉ<3á×*?²¬‚ä†"Ùc=p íÛ'¡ª1ñ"økJ†HÒ'»Ÿ+ oÏN¬Ã9 dÙãÜדÏâÍ~æc+j·Jzâ7(£ðW]•æ™?nê´º6åwéåç÷N•ZŠíž›¬|?Ðõ?Ñ-E…®³ÇV$~X¯/…õ x‘LˆÑÜÚÈ7¦pzãÜüë½ðÄ^õtÝYËÍ7ÉÖÕ8ÏUe# #€r=sU¾/é’E§jRC4mxNÝ´9†íuá»›V‘ ZI€­×cr1Ÿpzsøf»¨åV‹ìû`qËLÊIã?\~¼³áËC©êhªOîO»‘ÃmçÛçút×¢x“Z}?Ãœê#b-¤X7õ Äò gž zzbº3œm*qvs·M=íúéw}¿&Úª°^Ö×µÏ(ø‡â†Öµƒenñý†×åQáYûœ÷ÇLœôÎNk¡ð‡¼/µ¸n0æÉ0¬ƒ‚üîÉÆvŒw®Sáö”š¯‹-üÕVŠØÙ[$`(9cqƒÔ_@BëqûÙ`Ý捭0;79È?w<ó |ÙÜkßÌ1±Ëã ¿ìÒ»ðlìï«ÓnªèèrP´NÏš&Žéö Ù¸÷æ°~-_O'‰`°!RÚÚÝ%]Ø%þbß1'¿ÿ X՝áOöÎŒ·‹¬+Åæ*ÛÛ™0¤ƒOÍÔ `u¯¦ÂaèÐÃÓ«‹¨Ô¥µœ¿¯ÉyÅÙ.oÔôŸ Úx&(STðݽ¦õ] ’ÒNóÁäÈùr3í·žÚ[™ƒ¼veÈ÷ÞIõÎGlqÎ=M|«gsªxÅI6 ]Z·Îªä,¨zŒŽÄ~#ØŠúFñiÉqc©éÐD>S딑 GñŽ1éÐ^+ Ëi;Ô„µVÕú»i¯ÈÒ-ZÍ]òܘ®ì` bÛÙ¥_/y(@÷qÐúg Ô÷W0.Ø› 6Ò© r>QƒŒ0+Èîzb¨É+I0TbNñ"$~)ÕÒ6Þ‹{0VÆ27œWWñcÄcX×íôûyKZéðªc'iQ¿¯LaWŠŸS\·Š“źʸ…ôÙÂí|öÀÇåV|!¤ÂGâÛ[[’ï 3OrÙËPY¹=Î1õ5öåTžÑè Ú64/üö?Zëžk}¬¶éào፾á}3“ü]8Éæ¿´n²Žš_6¾pœ)2?úWÓÚ¥¾¨iWúdŽq{*ª1rXŒd…m»‰äcô¯–dâ•ã‘Jº¬§¨#¨® §,df«8ÉÅßN¾hˆ;îÓ=7áùpën®É 6ûJžO2^œÐò JÖø¥²ã›Ò6Ü·‰!wbÍ‚¬O©»õ¬ÿ ƒP=Ä:â¤-&ÙŽ ` È9 r9íϧzë> XÅ7ƒ5X–krÑ¢L 7€ìw}ÑŸNHëŒüþ:2†á¼+u·á÷N/Û'Ðç~ߘô«ëh!ónRéeQ´6QÛÿ èEwëÅÒ|¸Yqó1uêyùzð8 ƒŠù¦Ò;¹ä6öi<'ü³„[íZhu½ ùÍ¡g‚>r¯׊îÌx}bñ2“­k꣧oø~›hTèóËWò4|ki"xßQ˜Ï6øÀLnß‚0 ¹Æ{±–¶Öe#¨27È@^Ìß.1N¾œyç€õ†ñeé·Õã†çQ°€=­Ì©ºB€Ø8<‚ÃSõ®ùcc>×Ú .Fr:žÝGæ=kÁâ,^!Fž ¬,àµ}%¶«îõ¹†"r²ƒGœüYÕd?aÑÍY®49PyU ÷þ!žxÅm|/‚ãNð˜¼PcûTÒ,¹/Ý=FkÏ|u¨¶«â녏{¤m¢]Û¾ïP>®XãÞ½iÓÁ¾ ‰'¬–6ß¼(„ï— í!úÙäzôë^–:œ¨å|,_¿&š×]uÓѵÛô4’j”bž§x‘Æ©ã›á,‚[Ô ÎÞ= ŒËæ ÀùYÁ?ŽïÚ¼?ÁªxºÕÛ,°1¸‘¿Ýäãدv…@¤åq½ºã œàûââ·z8Xýˆþz~—û»™âµj=Ž â~ãáh@'h¼F#·Üp?ŸëQü-løvépx»cŸø…lxâÃûG·‰¶ø”L£©%y?¦úõÆü-Õ¶¥y`Òl7>q’2üA?•F}c‡jB:¸Jÿ +§¹¿¸Q÷°ív=VÑìu[Qml%R7a×IèTõéŽx¬ ?†š7 1†îã-ˆã’L¡lŽ0OÓ=ÅuˆpÇ•¼3ÛùÒ¶W/!|’wŽw^qÔ×Ïaó M8Q¨ãÑ?ëï0IEhÄa¸X•`a ?!ÐñùQ!Rä žqŽžÝO`I0ÿ J“y|ñ!Îã@99>þ8–+éáu…!ù—ä Ê°<÷6’I®z ÅS„¾)Zþ_Öýµ×ËPåOwø÷þ*üïænÖùmØÝûþ¹=>¦½öî×Jh]¼ç&@§nTŒ6IT Àõ^Fxð7Å3!Ö·aÛ$þÿ ¹ã5îIo:ȪmËY[’8ÇӾlj*òû¢¥xõ¾¼ú•åk+\ð¯ HÚoŽl•Ûk,¯ ç²²cõÅ{²Z\ ´ìQ åpzŽ3Ôð}ÿ Jð¯XO¡øÎé€hÙ¥ûLdŒ`““ù6Gá^ÃáÝ^Ë[Ñb¾YåŒÊ»dŽ4 †2§,;ÿ CQÄ´¾°¨c–±”mºV{«ßÕýÄW\ÖŸ‘çŸ,çMRÆí“l-ƒn~ë©ÉÈê Ü?#Ž•¹ðãSÒ¥ÐWNíà½;ãž)™ÎSÈ9cóLj뵿Å«iÍk¨ió­¶X‚7÷ƒ€yãnyÏŽëÞ Öt`×À×V's$È9Ú:ä{wÆEk€«†Çàc—â$éÎ.éí~Ýëk}ÅAÆpörÑ¢‡Šl¡ÑüSs‹¨‰IÝ„óÀ×wñ&eºðf™pŒÆ9gŽTø£lñëÀçŽ NkÊUK0U’p ï^¡ãÈ¥´ø{£ÙHp`’ØåbqÏ©äó^Æ: Ž' ÊóM«õz+ß×ó5Ÿ»('¹­ð¦C„$˜Å¢_ºÈI?»^äã'ñêzž+ë€ñ-½»´}¡Ë*õ?.xÇ^1ŽMyǸ&“—L–îëöâ7…' bqéÎGé]˪â1$o²¸R8Ã`.q€}sÖ¾C9­8cêÆÞíïóòvÓòùœÕfÔÚéýu­èÖ·Ú Å‚_¤³ÜۺƑߝ”àרý:׃xPþÅÕî-/üØmnQìïGΊÙRqê=>¢½õnæ·r!—h`+’;ò3È<“Û©éšóŸx*÷V¹¸×tÈiˆßwiÔÿ |cŒñÏ®3ֽ̰‰Ë Qr©ö½®¼ÛoÑÙZÅÑ«O൯ýw8;k›ÿ x†;ˆJa;‘º9÷÷R+¡ñgŽí|Iáë{ôáo2ʲ9 029ÉÏLí\‰¿¸Ÿb˜ "Bv$£&#ßiê>=ªª©f ’N ëí>¡N­XW­~5×úíø\‰»½Ï^ø(—wÖú¥¤2íŽÞXæÁ$ °eÈ888^nÝë²ñÝÔ^ ÖÚ9Q~Ëå7ï DC¶ÑµƒsËÇè9®Wáþƒ6‡£´·°2\Ý:ÈÑ?(#¨'$õèGJ¥ñW\ÿ ‰E¶—¸™g˜ÌÀ¹;Pv ú±ÎNs·ëŸ’–"Ž/:té+ûË]öJöÓM»ëø˜*‘•^Uý—êd|‰åñMæÔÝ‹23å™6æHùÛ‚ëüñ^…ñ1¢oêûÑEØ.õ7*ÅHtÎp{g<·Á«+¸c¿¿pÓ¾Æby=8É_ÄsÆk¬ñB\jÞÔì••Ë[9Píb‹Bヅ =9­3§ð§LšÛáÖšÆæXÌÞdÛP.0\ãïÛ0?™úJ¸™Ë ”•œº+=<µI£¦í¯õêt¬d‹T¬P=ËFêT>ÍØØ@Ï9<÷AQÌ×»Õ¡xùk",JÎæù±Éç$œŽŸZWH®¯"·UÌQ ’ÙÈ]ÅXg<ã ߨg3-Üqe€0¢¨*Œ$܃ ’Sû 8㎼_/e'+Ï–-èÓ¶¶Õíß[·ÙÙ½î쏗¼sk%§µxä‰â-pÒeÆCrú ôσžû=”šÅô(QW‚Õd\ƒæ. \àö¹¯F½°³½0M>‘gr÷q+œ¶NïºHO— ¤ ܥݭ”n·J|ÆP6Kµc=Isó}Ò çGš)a=—#vK›åoK§ßóٍ¤¶¿õú…ÄRÚ[Ësöټˏ•Ë ópw®qœŒ·Ø ùÇâ‹ý‡ãKèS&ÞvûD Aù‘É9 ŒîqÅ} $SnIV[]ѐ´Ó}ØܾA Ü|½kÅþÓ|E Mu R¼.I¼¶däò‚ÃkÆ}ðy¹vc iUœZ…­Õõ»z¾÷¿n¦*j-É­/àœHã\y5 Û ß™ó0— äŸnzôã#Ô¯,†¥ÚeÔ÷ÜÅ´„“'c…<íÝ€<·SŠ¥k§Ã¢éÆÆÙna‚8–=«ʪ[Ÿ™°pNî02z“ÔÙ–K8.È’Þî(vƒ2®@ äÈûãçžxäÇf¯ˆu¹yUÕîýWšÙ|›ëÒ%Q^í[æ|éo5ZY•^{96ˆY‚§v*x>âº_|U¹Ö´©tûMÒÂ9PÇ#«£#€ éÉñ‘ƒÍz/‰´-į¹°dd,Б›p03ƒœ{ç9=+ Ûᧇ¬¦[‡‚ê婺¸#±ß=³ý¿•Õµjñ½HÙh›Û[§ÚýÊöô÷{˜?ô÷·Ô.u©–_%còcAÀ˜’ }0x9Î>žñÇáÍ9,ahï¦Ì2òÓ ñÛAäry$V²Nð ]=$Ž ‚#Ù‚1ƒƒødõMax‡ÂÖ^!±KkÛ‘ «“Çó²FN8+ëÎ{Ò¼oí§[«ÕMRoËeç×[_m/¦¦k.kôgŽxsSÓ´ý`êzªÜÜKo‰cPC9ÎY‰#§^üý9¹âïÞx£Ë·Ú`±‰‹¤;³–=ÏaôÕAð‚÷kêÁNBéÎælcõö®£Fð†ô2Ò¬]ßÂK$ÓÜ®•”/ÊHàã$ä ¸÷ëf¹Oµúâ“”’²ø­è´µþöjçNü÷üÌ¿ xNïFÒd»¼·h®îT9ŽAµÖ>qÁçÔœtïÒ»\ȶÎîcÞäîó3¶@#ÉIÎ ÔñW.<´’¥–ÑÑ€ÕšA‚ ;†qÓë‚2q ÒÂó$# Çí‡ !Ë}Õ9ÈÎÑÉã=;ŒÇÎuñ+ÉûÏ¥öíeÙ+$úíÜ娯'+êZH4ƒq¶FV‹gïŒ208ÆÌ)íб>M|÷âÍã¾"iì‹¥£Jd´™OÝç;sÈúr+ÜäˆË)DŒ¥šF°*3Õ”d {zÔwºQ¿·UžÉf†~>I+ŒqÔ`ð3œ“Ü×f]œTÁÔn4“ƒø’Ýßõ_«*5šzGCÊ,þ+ê1ò÷O¶¸cœºb2yÇ;cùÕ£ñh¬›áÑŠr¤ÝäNBk¥—á—†gxšX/쑘hŸ*Tçn =û㦠2|(ð¿e·ºÖ$ ýìŸ!'åΰyîî+×öœ=Y:²¦ÓÞ×iü’—ü -BK™£˜›âÆ¡&véðõ-ûÉY¹=Onj¹ø¯¯yf4·±T Pó`çœ7={×mÃ/ ¢˜ZÚòK…G½¥b„’G AãÜœ*í¯Ã¿ IoæI¦NU8‘RwÈã;·€ Û×ëÒ”1Y •£E»ÿ Oyto¢<£Áö·šï,䉧ûA¼sû»Nò}¹üE{ÜÖªò1’õÞr0â}ÎØ#>à/8ïéÎ~—áÍ#ñÎlí§³2f'h”?C÷YËdð:qëõÓ·‚ïeÄ© ÔÈØÜRL+žAÎ3¼g=åšó³Œt3 ÑQ¦ùRÙßE®¼±w_;þhš’Sirÿ ^ˆã¼iੇ|RòO„m°J/“$·l“ ÇÓ¿ÿ [ÑŠÆ“„†Õø>cFÆ6Ø1ƒ– àz7Ldòxäüwá‹ÝAXùO•Úý’é®ähm­ •NÀ±ÌTÈç ƒ‘I$pGž:‚ÄbêW¢®œ´|­¦­nÍ>¶ÖÏ¢§ÎÜ¢ºö¹•%ÄqL^öÛ KpNA<ã¡ …î==ª¸óffËF‡yÌcÉ ©ç$ð=ñÏ­YþÊ’Ú]—¥‚¬‚eDïÎH>Ÿ_ÌTP™a‰ch['çÆÜò7a‡?w°Ïn§âÎ5”’¨¹uÚÛ|´ÓÓc§{O—ü1•ªxsÃZ…ÊÏy¡Ã3¸Ë2Èé» ‘ƒÎ äžÜðA§cáOéúÛ4ý5-fŒï„ù¬ûô.Ç Üsž•Ò¾•wo<¶Ÿ"¬¡º|£ î2sÇ¡éE²ÉFѱrU°dÜ6œ¨ mc†Îxë׺Þ'0²¡Rr„{j¾í·è›µ÷)º·å–‹î2|I®Y¼ºÍË·–ÃÆà㍣'óÆxƒOÆÞ&>\lóÌxP Xc¸ì Sþ5§qà/ê>#žÞW¸if$\3 ® ûÄ“ùŽÕê¾ð<Ó‹H¶óÏ" å·( á‘€:ã†8Ï=+ꨬUA×ÃËÚT’ÑÞöù¥¢]{»ms¥F0\ÑÕ—ô}&ÛB´ƒOŽÚ+›xíÄÀ1 ,v± žIëíZ0ǧ™3 í2®0ทp9öÝÔž)ÓZËoq/Ú“‘L ²ŒmùŽÓ9§[Û#Ä‘\ÞB¬Çs [;à à«g‚2ôòªœÝV§»·¯/[uó½õÛï¾ /šÍ}öüÿ «=x»HŸÂÞ.™ ÌQùŸh´‘#a$‚'¡u<Š›Æ>2>+ƒLSiöwµFó1!eg`£åœ ÷ëÛö}Á¿ÛVÙêv $¬ƒ|,s÷z€ð΃¨x÷ÅD\ÜŒÞmåÔ„ ˆ o| :{ÇÓ¶–òÁn!´0Ål€, ƒ ( ÛŒŒ c¶rsšæ,4‹MÛOH!@¢ ÇŽ„`å²9ÝÃw;AÍt0®¤¡…¯ØÄ.Àì클ƒ‘ßñ5Í,Óëu-ÈÔc¢KÃÓ£òÖ̺U.õL¯0…%2È—"~x ‚[`có±nHàŽyàö™¥keˆìŒÛFç{(Ø©†`Jã#Žwg<“:ÚÉ;M ^\yhûX‡vB·÷zrF?§BÊÔ/s<ÐÈB)Û± ·ÍÔwç5Âã:så§e{mѤï«Òíh—]Wm4âí¿ùþW4bC3¶ª¾Ùr$ pw`àädzt!yŠI„hÂîàM)!edŒm'æ>Ç?wzºK­ìcŒ´¯Ìq6fp$)ãw¡éUl`µ»ARAˆÝÕgr:äŒgƒéé[Ôö±”iYs5Ýï«ÙG—K=þF’æMG«óÿ `ŠKɦuOQ!ÕåŒ/ÎGÞ`@ËqÕzdõâ«Ê/Ö(ƒK´%ŽbMü åÜŸö—>¤óŒŒV‘°„I¢Yž#™¥ùÏÊ@8 œgqöö5ª4vד[¬(q cò¨À!FGaÁõõ¯?§†¥ÏU½í¿WªZ$úyú½ŽzקÉþ?>Ã×È•6°{™™ŽÙ.$`­ÎUœ…çè ' ¤r$1Ø(y7 ðV<ž:È  ÁÎMw¾Â'Øb§øxb7gãО½óÉÊë²,i„Fȹ£§8ãä½k¹¥¦ê/ç{ïê驪2œ/«ü?¯Ô›ìñÜ$þeýœRIåŒg9Ác’zrrNO bÚi¢ ѺË/$,“ª¯Ýä;Œ× ´<ÛÑn³IvŸb™¥ nm–ÄŸ—nÝÀãŽ3ëÍG,.öó³˜Ù£¹u ÊÌrŠ[<±!@Æ:c9ÅZh ì’M5ÄìÌ-‚¼ëÉùqŽGì9¬á ;¨A-ž—évþÖ–^ON·Ô”ŸEý}ú×PO&e[]ÒG¸˜Ûp ƒÃà/Ë·8ûÀ€1ž@¿ÚB*²­¼ñì8@p™8Q“žÆH'8«I-%¸‚ F»“åó6°Uù|¶Ú¸ã ò^Äw¥ŠÖK–1ÜÝK,Žddlí²0PÀü“×ükG…¯U«·¶–´w¶ŽÍ¾©yÞú[Zös•¯Á[™6° ¨¼ÉVæq·,# ìãï‘×8îry®A››¨,ãc66»Ë´ã'æÉù?t}¢æH--Òá"›|ˆ¬[í  7¶ö#¸9«––‹$,+Ëqœ\Êø c€yê^ݸÄa°«™B-9%«×®‹V´w~vÜTéꢷþ¼ˆ%·¹• ’[xç•÷2gØS?6åÀÚ õ9É#š@÷bT¸º²C*3Bá¤òÎA9 =úU§Ó"2Ãlá0iÝIc‚2Î@%öç94ùô»'»HÄ¥Ô¾@à Tp£šíx:úÊ:5eºßMý×wµ›Ó_+šº3Ýyvÿ "ºÇ<ÂI>Õ 1G·Ë«È«É# àÈÇ øp Jv·šæDûE¿›†Ë’NFr2qŸ½ÇAÜšu•´éí#Ħ8£2”Ú2Ã/€[ÎTr;qŠz*ý’Îþ(≠;¡TÆ⁛;ºÿ àçœk‘Þ­8¾Uª¾íé{^×IZéwÓkXÉûÑZo¯_øo×È¡¬ â–ÞR§2„‚Àœü½ùç® SVa†Âüª¼±D‘ŒísŸàä|ä2 æ[‹z”¯s{wn„ÆmáóCO+†GO8Ïeçåº`¯^¼ðG5f{Xžä,k‰<á y™¥voÆ éÛõëI=œ1‹éíÔÀÑ)R#;AÂncäŽ:tÏ#¶TkB.0Œ-ÖÞZÛgumß}fÎJÉ+#2êÔP£žùÈÅi¢%œ3P*Yƒò‚Aì“Ž2r:ƒÐúñi­RUQq‰H9!”={~¼ “JŽV¥»×²m.ÛߺiYl¾òk˜gL³·rT• ’…wHÁ6ä`–Î3ùÌ4Øe³†&òL‘•%clyîAÂäà0 žüç$[3uŘpNOÀÉ=† cï{rYK ååä~FÁ •a»"Lär1Ó¯2Äõæ<™C•.fÕ»è¥~½-¿g½Â4¡{[ør¨¶·Žõäx¥’l®qpwÇ»8ärF \cŽ¸ÜÆÓ-g‚yciÏÀ¾rÎwèØÈ#o°Á9ã5¢šfÔxÞæfGusÏÌJÿ µ×œ/LtãÅT7²¶w,l ɳ;”eúà·¨çîŒsÜgTÃS¦­^ '~‹®›¯+k÷ZÖd©Æ*Ó[Ü«%Œk0ŽXƒ”$k#Ȩ P2bv‘ƒŸáÇ™ÆÕb)m$É*8óLE‘8'–ÜN Úyàúô­+{uº±I'wvš4fÜr íì½=úuú sFlìV$‘ö†Hсù€$§ õ=½¸«Ž] :Ž+•¦ïmRþ½l´îÊT#nkiøÿ _ðÆT¶7Ò½ºÒ£Î¸d\ã8=yãŽÜäR{x]ZâÚé#¸r²#»ÎHÆ6õ ç® ÎFkr;sºÄ.&;só± Ç9êH÷ýSšÕ­tÐU¢-n­ Ì| vqœ„{gŒt§S.P‹’Ž‰_[;m¥Þ­ZýRûÂX{+¥úü¼ú•-àÓ7!„G"“´‹žƒnrYXã¸îp éœ!Ó­oP̏tÑ (‰Þ¹é€sÓ#GLçÕšÑnJý¡!‘Tä#“ß?îýp}xÇ‚I¥Õn#·¸–y'qó@r[ Êô÷<ÔWÃÓ¢áN¥4ԝ’I&ݼ¬¬¼ÞºvéÆ FQV~_ÒüJÖÚt¥¦Xá3BÄP^%ÈÎW-×c¡ú©¤·Iþèk¥š?–UQåIR[’O 5x\ÉhÆI¶K4«2ùªŠŒ<¼óœçØ`u«‚Í.VHä € Ëgfx''9ÆI#±®Z8 sISºku¢ßÞ]úk»Jößl¡B.Ü»ÿ MWe °·Ž%šêɆ¼»Âù³´œ O¿cÐÓÄh©"ÛÜÏ.ÖV ’3nüÄmnq[ŒòznšÖ>J¬òˆæ…qýØP Ž:ä7^0yëWšÍ_79äoaÈ °#q0{ää×mœy”R{vÒÞ¶ÚÏe¥“ÚÆХ̮—õýjR •íç›Ìb„+J yÜØÙ•Ç]¿Ôd þËOL²”9-Œ—õÃc'æÝלçÚ²ìejP“½ âù°¨†ðqòädЃÉäÖÜj÷PÇp“ÍšŠå«‘î <iWN­smª»¶vÓz5»ûì:Rs\Ðßôû×uÔÿÙ