view release on metacpan or  search on metacpan

xs/sc_ssl/openssl/source/crypto/aes/asm/aes-x86_64.pl  view on Meta::CPAN

#!/usr/bin/env perl
#
# ====================================================================
# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
# project. Rights for redistribution and usage in source and binary
# forms are granted according to the OpenSSL license.
# ====================================================================
#
# Version 1.2.
#
# aes-*-cbc benchmarks are improved by >70% [compared to gcc 3.3.2 on
# Opteron 240 CPU] plus all the bells-n-whistles from 32-bit version
# [you'll notice a lot of resemblance], such as compressed S-boxes
# in little-endian byte order, prefetch of these tables in CBC mode,
# as well as avoiding L1 cache aliasing between stack frame and key
# schedule and already mentioned tables, compressed Td4...
#
# Performance in number of cycles per processed byte for 128-bit key:
#
#		ECB		CBC encrypt
# AMD64		13.7		13.0(*)
# EM64T		20.2		18.6(*)
#
# (*)	CBC benchmarks are better than ECB thanks to custom ABI used
#	by the private block encryption function.

$verticalspin=1;	# unlike 32-bit version $verticalspin performs
			# ~15% better on both AMD and Intel cores
$output=shift;
open STDOUT,"| $^X ../perlasm/x86_64-xlate.pl $output";

$code=".text\n";

$s0="%eax";
$s1="%ebx";
$s2="%ecx";
$s3="%edx";
$acc0="%esi";
$acc1="%edi";
$acc2="%ebp";
$inp="%r8";
$out="%r9";
$t0="%r10d";
$t1="%r11d";
$t2="%r12d";
$rnds="%r13d";
$sbox="%r14";
$key="%r15";

sub hi() { my $r=shift;	$r =~ s/%[er]([a-d])x/%\1h/;	$r; }
sub lo() { my $r=shift;	$r =~ s/%[er]([a-d])x/%\1l/;
			$r =~ s/%[er]([sd]i)/%\1l/;
			$r =~ s/%(r[0-9]+)[d]?/%\1b/;	$r; }
sub _data_word()
{ my $i;
    while(defined($i=shift)) { $code.=sprintf".long\t0x%08x,0x%08x\n",$i,$i; }
}
sub data_word()
{ my $i;
  my $last=pop(@_);
    $code.=".long\t";
    while(defined($i=shift)) { $code.=sprintf"0x%08x,",$i; }
    $code.=sprintf"0x%08x\n",$last;
}

sub data_byte()
{ my $i;
  my $last=pop(@_);
    $code.=".byte\t";
    while(defined($i=shift)) { $code.=sprintf"0x%02x,",$i&0xff; }
    $code.=sprintf"0x%02x\n",$last&0xff;
}

sub encvert()
{ my $t3="%r8d";	# zaps $inp!

$code.=<<___;
	# favor 3-way issue Opteron pipeline...
	movzb	`&lo("$s0")`,$acc0
	movzb	`&lo("$s1")`,$acc1
	movzb	`&lo("$s2")`,$acc2
	mov	0($sbox,$acc0,8),$t0
	mov	0($sbox,$acc1,8),$t1
	mov	0($sbox,$acc2,8),$t2

	movzb	`&hi("$s1")`,$acc0
	movzb	`&hi("$s2")`,$acc1
	movzb	`&lo("$s3")`,$acc2
	xor	3($sbox,$acc0,8),$t0
	xor	3($sbox,$acc1,8),$t1
	mov	0($sbox,$acc2,8),$t3

	movzb	`&hi("$s3")`,$acc0
	shr	\$16,$s2
	movzb	`&hi("$s0")`,$acc2
	xor	3($sbox,$acc0,8),$t2
	shr	\$16,$s3
	xor	3($sbox,$acc2,8),$t3

	shr	\$16,$s1
	lea	16($key),$key
	shr	\$16,$s0

	movzb	`&lo("$s2")`,$acc0
	movzb	`&lo("$s3")`,$acc1
	movzb	`&lo("$s0")`,$acc2
	xor	2($sbox,$acc0,8),$t0
	xor	2($sbox,$acc1,8),$t1
	xor	2($sbox,$acc2,8),$t2

	movzb	`&hi("$s3")`,$acc0