#! /usr/bin/env perl
# Copyright 2009-2016 The OpenSSL Project Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.


# ====================================================================
# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
# project.
# ====================================================================
#
# 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.

# November 2015
#
# Add aesni_ocb_[en|de]crypt. [Removed in BoringSSL]

######################################################################
# Current large-block performance in cycles per byte processed with
# 128-bit key (less is better).
#
#		CBC en-/decrypt	CTR	XTS	ECB	OCB
# Westmere	3.77/1.37	1.37	1.52	1.27
# * Bridge	5.07/0.98	0.99	1.09	0.91	1.10
# Haswell	4.44/0.80	0.97	1.03	0.72	0.76
# Skylake	2.68/0.65	0.65	0.66	0.64	0.66
# Silvermont	5.77/3.56	3.67	4.03	3.46	4.03
# Goldmont	3.84/1.39	1.39	1.63	1.31	1.70
# Bulldozer	5.80/0.98	1.05	1.24	0.93	1.23

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

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

$output = pop;
open OUT,">$output";
*STDOUT=*OUT;

&asm_init($ARGV[0]);

&preprocessor_ifdef("BORINGSSL_DISPATCH_TEST")
&external_label("BORINGSSL_function_hit");
&preprocessor_endif();
&static_label("key_const");

if ($PREFIX eq $AESNI_PREFIX)	{ $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 extension
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,@_); }