#!/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 3.6.
#
# You might fail to appreciate this module performance from the first
# try. If compared to "vanilla" linux-ia32-icc target, i.e. considered
# to be *the* best Intel C compiler without -KPIC, performance appears
# to be virtually identical... But try to re-configure with shared
# library support... Aha! Intel compiler "suddenly" lags behind by 30%
# [on P4, more on others]:-) And if compared to position-independent
# code generated by GNU C, this code performs *more* than *twice* as
# fast! Yes, all this buzz about PIC means that unlike other hand-
# coded implementations, this one was explicitly designed to be safe
# to use even in shared library context... This also means that this
# code isn't necessarily absolutely fastest "ever," because in order
# to achieve position independence an extra register has to be
# off-loaded to stack, which affects the benchmark result.
#
# Special note about instruction choice. Do you recall RC4_INT code
# performing poorly on P4? It might be the time to figure out why.
# RC4_INT code implies effective address calculations in base+offset*4
# form. Trouble is that it seems that offset scaling turned to be
# critical path... At least eliminating scaling resulted in 2.8x RC4
# performance improvement [as you might recall]. As AES code is hungry
# for scaling too, I [try to] avoid the latter by favoring off-by-2
# shifts and masking the result with 0xFF<<2 instead of "boring" 0xFF.
#
# As was shown by Dean Gaudet <dean@arctic.org>, the above note turned
# void. Performance improvement with off-by-2 shifts was observed on
# intermediate implementation, which was spilling yet another register
# to stack... Final offset*4 code below runs just a tad faster on P4,
# but exhibits up to 10% improvement on other cores.
#
# Second version is "monolithic" replacement for aes_core.c, which in
# addition to AES_[de|en]crypt implements AES_set_[de|en]cryption_key.
# This made it possible to implement little-endian variant of the
# algorithm without modifying the base C code. Motivating factor for
# the undertaken effort was that it appeared that in tight IA-32
# register window little-endian flavor could achieve slightly higher
# Instruction Level Parallelism, and it indeed resulted in up to 15%
# better performance on most recent µ-archs...
#
# Third version adds AES_cbc_encrypt implementation, which resulted in
# up to 40% performance imrovement of CBC benchmark results. 40% was
# observed on P4 core, where "overall" imrovement coefficient, i.e. if
# compared to PIC generated by GCC and in CBC mode, was observed to be
# as large as 4x:-) CBC performance is virtually identical to ECB now
# and on some platforms even better, e.g. 17.6 "small" cycles/byte on
# Opteron, because certain function prologues and epilogues are
# effectively taken out of the loop...
#
# Version 3.2 implements compressed tables and prefetch of these tables
# in CBC[!] mode. Former means that 3/4 of table references are now
# misaligned, which unfortunately has negative impact on elder IA-32
# implementations, Pentium suffered 30% penalty, PIII - 10%.
#
# Version 3.3 avoids L1 cache aliasing between stack frame and
# S-boxes, and 3.4 - L1 cache aliasing even between key schedule. The
# latter is achieved by copying the key schedule to controlled place in
# stack. This unfortunately has rather strong impact on small block CBC
# performance, ~2x deterioration on 16-byte block if compared to 3.3.
#
# Version 3.5 checks if there is L1 cache aliasing between user-supplied
# key schedule and S-boxes and abstains from copying the former if
# there is no. This allows end-user to consciously retain small block
# performance by aligning key schedule in specific manner.
#
# Version 3.6 compresses Td4 to 256 bytes and prefetches it in ECB.
#
# Current ECB performance numbers for 128-bit key in CPU cycles per
# processed byte [measure commonly used by AES benchmarkers] are:
#
#		small footprint		fully unrolled
# P4		24			22
# AMD K8	20			19
# PIII		25			23
# Pentium	81			78

push(@INC,"perlasm","../../perlasm");
require "x86asm.pl";

&asm_init($ARGV[0],"aes-586.pl",$ARGV[$#ARGV] eq "386");

$s0="eax";
$s1="ebx";
$s2="ecx";
$s3="edx";
$key="edi";
$acc="esi";

$compromise=0;		# $compromise=128 abstains from copying key
			# schedule to stack when encrypting inputs
			# shorter than 128 bytes at the cost of
			# risksing aliasing with S-boxes. In return
			# you get way better, up to +70%, small block
			# performance.
$small_footprint=1;	# $small_footprint=1 code is ~5% slower [on
			# recent µ-archs], but ~5 times smaller!
			# I favor compact code to minimize cache
			# contention and in hope to "collect" 5% back
			# in real-life applications...
$vertical_spin=0;	# shift "verticaly" defaults to 0, because of
			# its proof-of-concept status...

# Note that there is no decvert(), as well as last encryption round is
# performed with "horizontal" shifts. This is because this "vertical"
# implementation [one which groups shifts on a given $s[i] to form a
# "column," unlike "horizontal" one, which groups shifts on different
# $s[i] to form a "row"] is work in progress. It was observed to run
# few percents faster on Intel cores, but not AMD. On AMD K8 core it's
# whole 12% slower:-( So we face a trade-off... Shall it be resolved
# some day? Till then the code is considered experimental and by
# default remains dormant...

sub encvert()
{ my ($te,@s) = @_;
  my $v0 = $acc, $v1 = $key;

	&mov	($v0,$s[3]);				# copy s3
	&mov	(&DWP(4,"esp"),$s[2]);			# save s2
	&mov	($v1,$s[0]);				# copy s0
	&mov	(&DWP(8,"esp"),$s[1]);			# save s1

	&movz	($s[2],&HB($s[0]));
	&and	($s[0],0xFF);
	&mov	($s[0],&DWP(0,$te,$s[0],8));		# s0>>0
	&shr	($v1,16);
	&mov	($s[3],&DWP(3,$te,$s[2],8));		# s0>>8
	&movz	($s[1],&HB($v1));
	&and	($v1,0xFF);
	&mov	($s[2],&DWP(2,$te,$v1,8));		# s0>>16
	 &mov	($v1,$v0);
	&mov	($s[1],&DWP(1,$te,$s[1],8));		# s0>>24

	&and	($v0,0xFF);
	&xor	($s[3],&DWP(0,$te,$v0,8));		# s3>>0
	&movz	($v0,&HB($v1));
	&shr	($v1,16);
	&xor	($s[2],&DWP(3,$te,$v0,8));		# s3>>8
	&movz	($v0,&HB($v1));
	&and	($v1,0xFF);
	&xor	($s[1],&DWP(2,$te,$v1,8));		# s3>>16
	 &mov	($v1,&DWP(4,"esp"));			# restore s2
	&xor	($s[0],&DWP(1,$te,$v0,8));		# s3>>24

	&mov	($v0,$v1);
	&and	($v1,0xFF);
	&xor	($s[2],&DWP(0,$te,$v1,8));		# s2>>0
	&movz	($v1,&HB($v0));
	&shr	($v0,16);
	&xor	($s[1],&DWP(3,$te,$v1,8));		# s2>>8
	&movz	($v1,&HB($v0));
	&and	($v0,0xFF);
	&xor	($s[0],&DWP(2,$te,$v0,8));		# s2>>16
	 &mov	($v0,&DWP(8,"esp"));			# restore s1
	&xor	($s[3],&DWP(1,$te,$v1,8));		# s2>>24

	&mov	($v1,$v0);
	&and	($v0,0xFF);
	&xor	($s[1],&DWP(0,$te,$v0,8));		# s1>>0
	&movz	($v0,&HB($v1));
	&shr	($v1,16);
	&xor	($s[0],&DWP(3,$te,$v0,8));		# s1>>8
	&movz	($v0,&HB($v1));
	&and	($v1,0xFF);
	&xor	($s[3],&DWP(2,$te,$v1,8));		# s1>>16
	 &mov	($key,&DWP(12,"esp"));			# reincarnate v1 as key
	&xor	($s[2],&DWP(1,$te,$v0,8));		# s1>>24
}

sub encstep()
{ my ($i,$te,@s) = @_;
  my $tmp = $key;
  my $out = $i==3?$s[0]:$acc;

	# lines marked with #%e?x[i] denote "reordered" instructions...
	if ($i==3)  {	&mov	($key,&DWP(12,"esp"));		}##%edx
	else        {	&mov	($out,$s[0]);
			&and	($out,0xFF);			}
	if ($i==1)  {	&shr	($s[0],16);			}#%ebx[1]
	if ($i==2)  {	&shr	($s[0],24);			}#%ecx[2]
			&mov	($out,&DWP(0,$te,$out,8));

	if ($i==3)  {	$tmp=$s[1];				}##%eax
			&movz	($tmp,&HB($s[1]));
			&xor	($out,&DWP(3,$te,$tmp,8));

	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],&DWP(4,"esp"));	}##%ebx
	else        {	&mov	($tmp,$s[2]);
			&shr	($tmp,16);			}
	if ($i==2)  {	&and	($s[1],0xFF);			}#%edx[2]
			&and	($tmp,0xFF);
			&xor	($out,&DWP(2,$te,$tmp,8));

	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],&DWP(8,"esp"));	}##%ecx
	elsif($i==2){	&movz	($tmp,&HB($s[3]));		}#%ebx[2]
	else        {	&mov	($tmp,$s[3]); 
			&shr	($tmp,24)			}
			&xor	($out,&DWP(1,$te,$tmp,8));
	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
	if ($i==3)  {	&mov	($s[3],$acc);			}
			&comment();
}

sub enclast()
{ my ($i,$te,@s)=@_;
  my $tmp = $key;
  my $out = $i==3?$s[0]:$acc;

	if ($i==3)  {	&mov	($key,&DWP(12,"esp"));		}##%edx
	else        {	&mov	($out,$s[0]);			}
			&and	($out,0xFF);
	if ($i==1)  {	&shr	($s[0],16);			}#%ebx[1]
	if ($i==2)  {	&shr	($s[0],24);			}#%ecx[2]
			&mov	($out,&DWP(2,$te,$out,8));
			&and	($out,0x000000ff);

	if ($i==3)  {	$tmp=$s[1];				}##%eax
			&movz	($tmp,&HB($s[1]));
			&mov	($tmp,&DWP(0,$te,$tmp,8));
			&and	($tmp,0x0000ff00);
			&xor	($out,$tmp);

	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],&DWP(4,"esp"));	}##%ebx
	else        {	mov	($tmp,$s[2]);
			&shr	($tmp,16);			}
	if ($i==2)  {	&and	($s[1],0xFF);			}#%edx[2]
			&and	($tmp,0xFF);
			&mov	($tmp,&DWP(0,$te,$tmp,8));
			&and	($tmp,0x00ff0000);
			&xor	($out,$tmp);

	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],&DWP(8,"esp"));	}##%ecx
	elsif($i==2){	&movz	($tmp,&HB($s[3]));		}#%ebx[2]
	else        {	&mov	($tmp,$s[3]);
			&shr	($tmp,24);			}
			&mov	($tmp,&DWP(2,$te,$tmp,8));
			&and	($tmp,0xff000000);
			&xor	($out,$tmp);
	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
	if ($i==3)  {	&mov	($s[3],$acc);			}
}

sub _data_word() { my $i; while(defined($i=shift)) { &data_word($i,$i); } }

&public_label("AES_Te");
&function_begin_B("_x86_AES_encrypt");
	if ($vertical_spin) {
		# I need high parts of volatile registers to be accessible...
		&exch	($s1="edi",$key="ebx");
		&mov	($s2="esi",$acc="ecx");
	}

	# note that caller is expected to allocate stack frame for me!
	&mov	(&DWP(12,"esp"),$key);		# save key

	&xor	($s0,&DWP(0,$key));		# xor with key
	&xor	($s1,&DWP(4,$key));
	&xor	($s2,&DWP(8,$key));
	&xor	($s3,&DWP(12,$key));

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

	if ($small_footprint) {
	    &lea	($acc,&DWP(-2,$acc,$acc));
	    &lea	($acc,&DWP(0,$key,$acc,8));
	    &mov	(&DWP(16,"esp"),$acc);	# end of key schedule
	    &align	(4);
	    &set_label("loop");
		if ($vertical_spin) {
		    &encvert("ebp",$s0,$s1,$s2,$s3);
		} else {
		    &encstep(0,"ebp",$s0,$s1,$s2,$s3);
		    &encstep(1,"ebp",$s1,$s2,$s3,$s0);
		    &encstep(2,"ebp",$s2,$s3,$s0,$s1);
		    &encstep(3,"ebp",$s3,$s0,$s1,$s2);
		}
		&add	($key,16);		# advance rd_key
		&xor	($s0,&DWP(0,$key));
		&xor	($s1,&DWP(4,$key));
		&xor	($s2,&DWP(8,$key));
		&xor	($s3,&DWP(12,$key));
	    &cmp	($key,&DWP(16,"esp"));
	    &mov	(&DWP(12,"esp"),$key);
	    &jb		(&label("loop"));
	}
	else {
	    &cmp	($acc,10);
	    &jle	(&label("10rounds"));
	    &cmp	($acc,12);
	    &jle	(&label("12rounds"));

	&set_label("14rounds");
	    for ($i=1;$i<3;$i++) {
		if ($vertical_spin) {
		    &encvert("ebp",$s0,$s1,$s2,$s3);
		} else {
		    &encstep(0,"ebp",$s0,$s1,$s2,$s3);
		    &encstep(1,"ebp",$s1,$s2,$s3,$s0);
		    &encstep(2,"ebp",$s2,$s3,$s0,$s1);
		    &encstep(3,"ebp",$s3,$s0,$s1,$s2);
		}
		&xor	($s0,&DWP(16*$i+0,$key));
		&xor	($s1,&DWP(16*$i+4,$key));
		&xor	($s2,&DWP(16*$i+8,$key));
		&xor	($s3,&DWP(16*$i+12,$key));
	    }
	    &add	($key,32);
	    &mov	(&DWP(12,"esp"),$key);	# advance rd_key
	&set_label("12rounds");
	    for ($i=1;$i<3;$i++) {
		if ($vertical_spin) {
		    &encvert("ebp",$s0,$s1,$s2,$s3);
		} else {
		    &encstep(0,"ebp",$s0,$s1,$s2,$s3);
		    &encstep(1,"ebp",$s1,$s2,$s3,$s0);
		    &encstep(2,"ebp",$s2,$s3,$s0,$s1);
		    &encstep(3,"ebp",$s3,$s0,$s1,$s2);
		}
		&xor	($s0,&DWP(16*$i+0,$key));
		&xor	($s1,&DWP(16*$i+4,$key));
		&xor	($s2,&DWP(16*$i+8,$key));
		&xor	($s3,&DWP(16*$i+12,$key));
	    }
	    &add	($key,32);
	    &mov	(&DWP(12,"esp"),$key);	# advance rd_key
	&set_label("10rounds");
	    for ($i=1;$i<10;$i++) {
		if ($vertical_spin) {
		    &encvert("ebp",$s0,$s1,$s2,$s3);
		} else {
		    &encstep(0,"ebp",$s0,$s1,$s2,$s3);
		    &encstep(1,"ebp",$s1,$s2,$s3,$s0);
		    &encstep(2,"ebp",$s2,$s3,$s0,$s1);
		    &encstep(3,"ebp",$s3,$s0,$s1,$s2);
		}
		&xor	($s0,&DWP(16*$i+0,$key));
		&xor	($s1,&DWP(16*$i+4,$key));
		&xor	($s2,&DWP(16*$i+8,$key));
		&xor	($s3,&DWP(16*$i+12,$key));
	    }
	}

	if ($vertical_spin) {
	    # "reincarnate" some registers for "horizontal" spin...
	    &mov	($s1="ebx",$key="edi");
	    &mov	($s2="ecx",$acc="esi");
	}
	&enclast(0,"ebp",$s0,$s1,$s2,$s3);
	&enclast(1,"ebp",$s1,$s2,$s3,$s0);
	&enclast(2,"ebp",$s2,$s3,$s0,$s1);
	&enclast(3,"ebp",$s3,$s0,$s1,$s2);

	&add	($key,$small_footprint?16:160);
	&xor	($s0,&DWP(0,$key));
	&xor	($s1,&DWP(4,$key));
	&xor	($s2,&DWP(8,$key));
	&xor	($s3,&DWP(12,$key));

	&ret	();

&set_label("AES_Te",64);	# Yes! I keep it in the code segment!
	&_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6);
	&_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591);
	&_data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56);
	&_data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec);
	&_data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa);
	&_data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb);
	&_data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45);
	&_data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b);
	&_data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c);
	&_data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83);
	&_data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9);
	&_data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a);
	&_data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d);
	&_data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f);
	&_data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df);
	&_data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea);
	&_data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34);
	&_data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b);
	&_data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d);
	&_data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413);
	&_data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1);
	&_data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6);
	&_data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972);
	&_data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85);
	&_data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed);
	&_data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511);
	&_data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe);
	&_data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b);
	&_data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05);
	&_data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1);
	&_data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142);
	&_data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf);
	&_data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3);
	&_data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e);
	&_data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a);
	&_data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6);
	&_data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3);
	&_data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b);
	&_data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428);
	&_data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad);
	&_data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14);
	&_data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8);
	&_data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4);
	&_data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2);
	&_data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda);
	&_data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949);
	&_data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf);
	&_data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810);
	&_data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c);
	&_data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697);
	&_data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e);
	&_data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f);
	&_data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc);
	&_data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c);
	&_data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969);
	&_data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27);
	&_data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122);
	&_data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433);
	&_data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9);
	&_data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5);
	&_data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a);
	&_data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0);
	&_data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e);
	&_data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c);
#rcon:
	&data_word(0x00000001, 0x00000002, 0x00000004, 0x00000008);
	&data_word(0x00000010, 0x00000020, 0x00000040, 0x00000080);
	&data_word(0x0000001b, 0x00000036, 0, 0, 0, 0, 0, 0);
&function_end_B("_x86_AES_encrypt");

# void AES_encrypt (const void *inp,void *out,const AES_KEY *key);
&public_label("AES_Te");
&function_begin("AES_encrypt");
	&mov	($acc,&wparam(0));		# load inp
	&mov	($key,&wparam(2));		# load key

	&mov	($s0,"esp");
	&sub	("esp",24);
	&and	("esp",-64);
	&add	("esp",4);
	&mov	(&DWP(16,"esp"),$s0);

	&call   (&label("pic_point"));          # make it PIC!
	&set_label("pic_point");
	&blindpop("ebp");
	&lea    ("ebp",&DWP(&label("AES_Te")."-".&label("pic_point"),"ebp"));

	&mov	($s0,&DWP(0,$acc));		# load input data
	&mov	($s1,&DWP(4,$acc));
	&mov	($s2,&DWP(8,$acc));
	&mov	($s3,&DWP(12,$acc));

	&call	("_x86_AES_encrypt");

	&mov	("esp",&DWP(16,"esp"));

	&mov	($acc,&wparam(1));		# load out
	&mov	(&DWP(0,$acc),$s0);		# write output data
	&mov	(&DWP(4,$acc),$s1);
	&mov	(&DWP(8,$acc),$s2);
	&mov	(&DWP(12,$acc),$s3);
&function_end("AES_encrypt");

#------------------------------------------------------------------#

sub decstep()
{ my ($i,$td,@s) = @_;
  my $tmp = $key;
  my $out = $i==3?$s[0]:$acc;

	# no instructions are reordered, as performance appears
	# optimal... or rather that all attempts to reorder didn't
	# result in better performance [which by the way is not a
	# bit lower than ecryption].
	if($i==3)   {	&mov	($key,&DWP(12,"esp"));		}
	else        {	&mov	($out,$s[0]);			}
			&and	($out,0xFF);
			&mov	($out,&DWP(0,$td,$out,8));

	if ($i==3)  {	$tmp=$s[1];				}
			&movz	($tmp,&HB($s[1]));
			&xor	($out,&DWP(3,$td,$tmp,8));

	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],$acc);		}
	else        {	&mov	($tmp,$s[2]);			}
			&shr	($tmp,16);
			&and	($tmp,0xFF);
			&xor	($out,&DWP(2,$td,$tmp,8));

	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],&DWP(8,"esp"));	}
	else        {	&mov	($tmp,$s[3]);			}
			&shr	($tmp,24);
			&xor	($out,&DWP(1,$td,$tmp,8));
	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
	if ($i==3)  {	&mov	($s[3],&DWP(4,"esp"));		}
			&comment();
}

sub declast()
{ my ($i,$td,@s)=@_;
  my $tmp = $key;
  my $out = $i==3?$s[0]:$acc;

	if($i==3)   {	&mov	($key,&DWP(12,"esp"));		}
	else        {	&mov	($out,$s[0]);			}
			&and	($out,0xFF);
			&movz	($out,&DWP(2048,$td,$out,1));

	if ($i==3)  {	$tmp=$s[1];				}
			&movz	($tmp,&HB($s[1]));
			&movz	($tmp,&DWP(2048,$td,$tmp,1));
			&shl	($tmp,8);
			&xor	($out,$tmp);

	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],$acc);		}
	else        {	mov	($tmp,$s[2]);			}
			&shr	($tmp,16);
			&and	($tmp,0xFF);
			&movz	($tmp,&DWP(2048,$td,$tmp,1));
			&shl	($tmp,16);
			&xor	($out,$tmp);

	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],&DWP(8,"esp"));	}
	else        {	&mov	($tmp,$s[3]);			}
			&shr	($tmp,24);
			&movz	($tmp,&DWP(2048,$td,$tmp,1));
			&shl	($tmp,24);
			&xor	($out,$tmp);
	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
	if ($i==3)  {	&mov	($s[3],&DWP(4,"esp"));		}
}

&public_label("AES_Td");
&function_begin_B("_x86_AES_decrypt");
	# note that caller is expected to allocate stack frame for me!
	&mov	(&DWP(12,"esp"),$key);		# save key

	&xor	($s0,&DWP(0,$key));		# xor with key
	&xor	($s1,&DWP(4,$key));
	&xor	($s2,&DWP(8,$key));
	&xor	($s3,&DWP(12,$key));

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

	if ($small_footprint) {
	    &lea	($acc,&DWP(-2,$acc,$acc));
	    &lea	($acc,&DWP(0,$key,$acc,8));
	    &mov	(&DWP(16,"esp"),$acc);	# end of key schedule
	    &align	(4);
	    &set_label("loop");
		&decstep(0,"ebp",$s0,$s3,$s2,$s1);
		&decstep(1,"ebp",$s1,$s0,$s3,$s2);
		&decstep(2,"ebp",$s2,$s1,$s0,$s3);
		&decstep(3,"ebp",$s3,$s2,$s1,$s0);
		&add	($key,16);		# advance rd_key
		&xor	($s0,&DWP(0,$key));
		&xor	($s1,&DWP(4,$key));
		&xor	($s2,&DWP(8,$key));
		&xor	($s3,&DWP(12,$key));
	    &cmp	($key,&DWP(16,"esp"));
	    &mov	(&DWP(12,"esp"),$key);
	    &jb		(&label("loop"));
	}
	else {
	    &cmp	($acc,10);
	    &jle	(&label("10rounds"));
	    &cmp	($acc,12);
	    &jle	(&label("12rounds"));

	&set_label("14rounds");
	    for ($i=1;$i<3;$i++) {
		&decstep(0,"ebp",$s0,$s3,$s2,$s1);
		&decstep(1,"ebp",$s1,$s0,$s3,$s2);
		&decstep(2,"ebp",$s2,$s1,$s0,$s3);
		&decstep(3,"ebp",$s3,$s2,$s1,$s0);
		&xor	($s0,&DWP(16*$i+0,$key));
		&xor	($s1,&DWP(16*$i+4,$key));
		&xor	($s2,&DWP(16*$i+8,$key));
		&xor	($s3,&DWP(16*$i+12,$key));
	    }
	    &add	($key,32);
	    &mov	(&DWP(12,"esp"),$key);	# advance rd_key
	&set_label("12rounds");
	    for ($i=1;$i<3;$i++) {
		&decstep(0,"ebp",$s0,$s3,$s2,$s1);
		&decstep(1,"ebp",$s1,$s0,$s3,$s2);
		&decstep(2,"ebp",$s2,$s1,$s0,$s3);
		&decstep(3,"ebp",$s3,$s2,$s1,$s0);
		&xor	($s0,&DWP(16*$i+0,$key));
		&xor	($s1,&DWP(16*$i+4,$key));
		&xor	($s2,&DWP(16*$i+8,$key));
		&xor	($s3,&DWP(16*$i+12,$key));
	    }
	    &add	($key,32);
	    &mov	(&DWP(12,"esp"),$key);	# advance rd_key
	&set_label("10rounds");
	    for ($i=1;$i<10;$i++) {
		&decstep(0,"ebp",$s0,$s3,$s2,$s1);
		&decstep(1,"ebp",$s1,$s0,$s3,$s2);
		&decstep(2,"ebp",$s2,$s1,$s0,$s3);
		&decstep(3,"ebp",$s3,$s2,$s1,$s0);
		&xor	($s0,&DWP(16*$i+0,$key));
		&xor	($s1,&DWP(16*$i+4,$key));
		&xor	($s2,&DWP(16*$i+8,$key));
		&xor	($s3,&DWP(16*$i+12,$key));
	    }
	}

	&declast(0,"ebp",$s0,$s3,$s2,$s1);
	&declast(1,"ebp",$s1,$s0,$s3,$s2);
	&declast(2,"ebp",$s2,$s1,$s0,$s3);
	&declast(3,"ebp",$s3,$s2,$s1,$s0);

	&add	($key,$small_footprint?16:160);
	&xor	($s0,&DWP(0,$key));
	&xor	($s1,&DWP(4,$key));
	&xor	($s2,&DWP(8,$key));
	&xor	($s3,&DWP(12,$key));

	&ret	();

&set_label("AES_Td",64);	# Yes! I keep it in the code segment!
	&_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a);
	&_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b);
	&_data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5);
	&_data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5);
	&_data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d);
	&_data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b);
	&_data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295);
	&_data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e);
	&_data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927);
	&_data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d);
	&_data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362);
	&_data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9);
	&_data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52);
	&_data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566);
	&_data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3);
	&_data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed);
	&_data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e);
	&_data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4);
	&_data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4);
	&_data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd);
	&_data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d);
	&_data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060);
	&_data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967);
	&_data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879);
	&_data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000);
	&_data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c);
	&_data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36);
	&_data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624);
	&_data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b);
	&_data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c);
	&_data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12);
	&_data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14);
	&_data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3);
	&_data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b);
	&_data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8);
	&_data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684);
	&_data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7);
	&_data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177);
	&_data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947);
	&_data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322);
	&_data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498);
	&_data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f);
	&_data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54);
	&_data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382);
	&_data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf);
	&_data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb);
	&_data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83);
	&_data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef);
	&_data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029);
	&_data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235);
	&_data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733);
	&_data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117);
	&_data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4);
	&_data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546);
	&_data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb);
	&_data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d);
	&_data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb);
	&_data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a);
	&_data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773);
	&_data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478);
	&_data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2);
	&_data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff);
	&_data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664);
	&_data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0);
#Td4:
	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
&function_end_B("_x86_AES_decrypt");

# void AES_decrypt (const void *inp,void *out,const AES_KEY *key);
&public_label("AES_Td");
&function_begin("AES_decrypt");
	&mov	($acc,&wparam(0));		# load inp
	&mov	($key,&wparam(2));		# load key

	&mov	($s0,"esp");
	&sub	("esp",24);
	&and	("esp",-64);
	&add	("esp",4);
	&mov	(&DWP(16,"esp"),$s0);

	&call   (&label("pic_point"));          # make it PIC!
	&set_label("pic_point");
	&blindpop("ebp");
	&lea    ("ebp",&DWP(&label("AES_Td")."-".&label("pic_point"),"ebp"));

	# prefetch Td4
	&lea	("ebp",&DWP(2048+128,"ebp"));
	&mov	($s0,&DWP(0-128,"ebp"));
	&mov	($s1,&DWP(32-128,"ebp"));
	&mov	($s2,&DWP(64-128,"ebp"));
	&mov	($s3,&DWP(96-128,"ebp"));
	&mov	($s0,&DWP(128-128,"ebp"));
	&mov	($s1,&DWP(160-128,"ebp"));
	&mov	($s2,&DWP(192-128,"ebp"));
	&mov	($s3,&DWP(224-128,"ebp"));
	&lea	("ebp",&DWP(-2048-128,"ebp"));

	&mov	($s0,&DWP(0,$acc));		# load input data
	&mov	($s1,&DWP(4,$acc));
	&mov	($s2,&DWP(8,$acc));
	&mov	($s3,&DWP(12,$acc));

	&call	("_x86_AES_decrypt");

	&mov	("esp",&DWP(16,"esp"));

	&mov	($acc,&wparam(1));		# load out
	&mov	(&DWP(0,$acc),$s0);		# write output data
	&mov	(&DWP(4,$acc),$s1);
	&mov	(&DWP(8,$acc),$s2);
	&mov	(&DWP(12,$acc),$s3);
&function_end("AES_decrypt");

# void AES_cbc_encrypt (const void char *inp, unsigned char *out,
#			size_t length, const AES_KEY *key,
#			unsigned char *ivp,const int enc);
{
# stack frame layout
# -4(%esp)	0(%esp)		return address
# 0(%esp)	4(%esp)		tmp1
# 4(%esp)	8(%esp)		tmp2
# 8(%esp)	12(%esp)	key
# 12(%esp)	16(%esp)	end of key schedule
my $_esp=&DWP(16,"esp");	#saved %esp
my $_inp=&DWP(20,"esp");	#copy of wparam(0)
my $_out=&DWP(24,"esp");	#copy of wparam(1)
my $_len=&DWP(28,"esp");	#copy of wparam(2)
my $_key=&DWP(32,"esp");	#copy of wparam(3)
my $_ivp=&DWP(36,"esp");	#copy of wparam(4)
my $_tmp=&DWP(40,"esp");	#volatile variable
my $ivec=&DWP(44,"esp");	#ivec[16]
my $aes_key=&DWP(60,"esp");	#copy of aes_key
my $mark=&DWP(60+240,"esp");	#copy of aes_key->rounds

&public_label("AES_Te");
&public_label("AES_Td");
&function_begin("AES_cbc_encrypt");
	&mov	($s2 eq "ecx"? $s2 : "",&wparam(2));	# load len
	&cmp	($s2,0);
	&je	(&label("enc_out"));

	&call   (&label("pic_point"));		# make it PIC!
	&set_label("pic_point");
	&blindpop("ebp");

	&pushf	();
	&cld	();

	&cmp	(&wparam(5),0);
	&je	(&label("DECRYPT"));

	&lea    ("ebp",&DWP(&label("AES_Te")."-".&label("pic_point"),"ebp"));

	# allocate aligned stack frame...
	&lea	($key,&DWP(-64-244,"esp"));
	&and	($key,-64);

	# ... and make sure it doesn't alias with AES_Te modulo 4096
	&mov	($s0,"ebp");
	&lea	($s1,&DWP(2048,"ebp"));
	&mov	($s3,$key);
	&and	($s0,0xfff);		# s = %ebp&0xfff
	&and	($s1,0xfff);		# e = (%ebp+2048)&0xfff
	&and	($s3,0xfff);		# p = %esp&0xfff

	&cmp	($s3,$s1);		# if (p>=e) %esp =- (p-e);
	&jb	(&label("te_break_out"));
	&sub	($s3,$s1);
	&sub	($key,$s3);
	&jmp	(&label("te_ok"));
	&set_label("te_break_out");	# else %esp -= (p-s)&0xfff + framesz;
	&sub	($s3,$s0);
	&and	($s3,0xfff);
	&add	($s3,64+256);
	&sub	($key,$s3);
	&align	(4);
	&set_label("te_ok");

	&mov	($s0,&wparam(0));	# load inp
	&mov	($s1,&wparam(1));	# load out
	&mov	($s3,&wparam(3));	# load key
	&mov	($acc,&wparam(4));	# load ivp

	&exch	("esp",$key);
	&add	("esp",4);		# reserve for return address!
	&mov	($_esp,$key);		# save %esp

	&mov	($_inp,$s0);		# save copy of inp
	&mov	($_out,$s1);		# save copy of out
	&mov	($_len,$s2);		# save copy of len
	&mov	($_key,$s3);		# save copy of key
	&mov	($_ivp,$acc);		# save copy of ivp

	&mov	($mark,0);		# copy of aes_key->rounds = 0;
	if ($compromise) {
		&cmp	($s2,$compromise);
		&jb	(&label("skip_ecopy"));
	}
	# do we copy key schedule to stack?
	&mov	($s1 eq "ebx" ? $s1 : "",$s3);
	&mov	($s2 eq "ecx" ? $s2 : "",244/4);
	&sub	($s1,"ebp");
	&mov	("esi",$s3);
	&and	($s1,0xfff);
	&lea	("edi",$aes_key);
	&cmp	($s1,2048);
	&jb	(&label("do_ecopy"));
	&cmp	($s1,4096-244);
	&jb	(&label("skip_ecopy"));
	&align	(4);
	&set_label("do_ecopy");
		&mov	($_key,"edi");
		&data_word(0xA5F3F689);	# rep movsd
	&set_label("skip_ecopy");

	&mov	($acc,$s0);
	&mov	($key,16);
	&align	(4);
	&set_label("prefetch_te");
		&mov	($s0,&DWP(0,"ebp"));
		&mov	($s1,&DWP(32,"ebp"));
		&mov	($s2,&DWP(64,"ebp"));
		&mov	($s3,&DWP(96,"ebp"));
		&lea	("ebp",&DWP(128,"ebp"));
		&dec	($key);
	&jnz	(&label("prefetch_te"));
	&sub	("ebp",2048);

	&mov	($s2,$_len);
	&mov	($key,$_ivp);
	&test	($s2,0xFFFFFFF0);
	&jz	(&label("enc_tail"));		# short input...

	&mov	($s0,&DWP(0,$key));		# load iv
	&mov	($s1,&DWP(4,$key));

	&align	(4);
	&set_label("enc_loop");
		&mov	($s2,&DWP(8,$key));
		&mov	($s3,&DWP(12,$key));

		&xor	($s0,&DWP(0,$acc));	# xor input data
		&xor	($s1,&DWP(4,$acc));
		&xor	($s2,&DWP(8,$acc));
		&xor	($s3,&DWP(12,$acc));

		&mov	($key,$_key);		# load key
		&call	("_x86_AES_encrypt");

		&mov	($acc,$_inp);		# load inp
		&mov	($key,$_out);		# load out

		&mov	(&DWP(0,$key),$s0);	# save output data
		&mov	(&DWP(4,$key),$s1);
		&mov	(&DWP(8,$key),$s2);
		&mov	(&DWP(12,$key),$s3);

		&mov	($s2,$_len);		# load len

		&lea	($acc,&DWP(16,$acc));
		&mov	($_inp,$acc);		# save inp

		&lea	($s3,&DWP(16,$key));
		&mov	($_out,$s3);		# save out

		&sub	($s2,16);
		&test	($s2,0xFFFFFFF0);
		&mov	($_len,$s2);		# save len
	&jnz	(&label("enc_loop"));
	&test	($s2,15);
	&jnz	(&label("enc_tail"));
	&mov	($acc,$_ivp);		# load ivp
	&mov	($s2,&DWP(8,$key));	# restore last dwords
	&mov	($s3,&DWP(12,$key));
	&mov	(&DWP(0,$acc),$s0);	# save ivec
	&mov	(&DWP(4,$acc),$s1);
	&mov	(&DWP(8,$acc),$s2);
	&mov	(&DWP(12,$acc),$s3);

	&cmp	($mark,0);		# was the key schedule copied?
	&mov	("edi",$_key);
	&mov	("esp",$_esp);
	&je	(&label("skip_ezero"));
	# zero copy of key schedule
	&mov	("ecx",240/4);
	&xor	("eax","eax");
	&align	(4);
	&data_word(0xABF3F689);	# rep stosd
	&set_label("skip_ezero")
	&popf	();
    &set_label("enc_out");
	&function_end_A();
	&pushf	();			# kludge, never executed

    &align	(4);
    &set_label("enc_tail");
	&push	($key eq "edi" ? $key : "");	# push ivp
	&mov	($key,$_out);			# load out
	&mov	($s1,16);
	&sub	($s1,$s2);
	&cmp	($key,$acc);			# compare with inp
	&je	(&label("enc_in_place"));
	&align	(4);
	&data_word(0xA4F3F689);	# rep movsb	# copy input
	&jmp	(&label("enc_skip_in_place"));
    &set_label("enc_in_place");
	&lea	($key,&DWP(0,$key,$s2));
    &set_label("enc_skip_in_place");
	&mov	($s2,$s1);
	&xor	($s0,$s0);
	&align	(4);
	&data_word(0xAAF3F689);	# rep stosb	# zero tail
	&pop	($key);				# pop ivp

	&mov	($acc,$_out);			# output as input
	&mov	($s0,&DWP(0,$key));
	&mov	($s1,&DWP(4,$key));
	&mov	($_len,16);			# len=16
	&jmp	(&label("enc_loop"));		# one more spin...

#----------------------------- DECRYPT -----------------------------#
&align	(4);
&set_label("DECRYPT");
	&lea    ("ebp",&DWP(&label("AES_Td")."-".&label("pic_point"),"ebp"));

	# allocate aligned stack frame...
	&lea	($key,&DWP(-64-244,"esp"));
	&and	($key,-64);

	# ... and make sure it doesn't alias with AES_Td modulo 4096
	&mov	($s0,"ebp");
	&lea	($s1,&DWP(2048+256,"ebp"));
	&mov	($s3,$key);
	&and	($s0,0xfff);		# s = %ebp&0xfff
	&and	($s1,0xfff);		# e = (%ebp+2048+256)&0xfff
	&and	($s3,0xfff);		# p = %esp&0xfff

	&cmp	($s3,$s1);		# if (p>=e) %esp =- (p-e);
	&jb	(&label("td_break_out"));
	&sub	($s3,$s1);
	&sub	($key,$s3);
	&jmp	(&label("td_ok"));
	&set_label("td_break_out");	# else %esp -= (p-s)&0xfff + framesz;
	&sub	($s3,$s0);
	&and	($s3,0xfff);
	&add	($s3,64+256);
	&sub	($key,$s3);
	&align	(4);
	&set_label("td_ok");

	&mov	($s0,&wparam(0));	# load inp
	&mov	($s1,&wparam(1));	# load out
	&mov	($s3,&wparam(3));	# load key
	&mov	($acc,&wparam(4));	# load ivp

	&exch	("esp",$key);
	&add	("esp",4);		# reserve for return address!
	&mov	($_esp,$key);		# save %esp

	&mov	($_inp,$s0);		# save copy of inp
	&mov	($_out,$s1);		# save copy of out
	&mov	($_len,$s2);		# save copy of len
	&mov	($_key,$s3);		# save copy of key
	&mov	($_ivp,$acc);		# save copy of ivp

	&mov	($mark,0);		# copy of aes_key->rounds = 0;
	if ($compromise) {
		&cmp	($s2,$compromise);
		&jb	(&label("skip_dcopy"));
	}
	# do we copy key schedule to stack?
	&mov	($s1 eq "ebx" ? $s1 : "",$s3);
	&mov	($s2 eq "ecx" ? $s2 : "",244/4);
	&sub	($s1,"ebp");
	&mov	("esi",$s3);
	&and	($s1,0xfff);
	&lea	("edi",$aes_key);
	&cmp	($s1,2048+256);
	&jb	(&label("do_dcopy"));
	&cmp	($s1,4096-244);
	&jb	(&label("skip_dcopy"));
	&align	(4);
	&set_label("do_dcopy");
		&mov	($_key,"edi");
		&data_word(0xA5F3F689);	# rep movsd
	&set_label("skip_dcopy");

	&mov	($acc,$s0);
	&mov	($key,18);
	&align	(4);
	&set_label("prefetch_td");
		&mov	($s0,&DWP(0,"ebp"));
		&mov	($s1,&DWP(32,"ebp"));
		&mov	($s2,&DWP(64,"ebp"));
		&mov	($s3,&DWP(96,"ebp"));
		&lea	("ebp",&DWP(128,"ebp"));
		&dec	($key);
	&jnz	(&label("prefetch_td"));
	&sub	("ebp",2048+256);

	&cmp	($acc,$_out);
	&je	(&label("dec_in_place"));	# in-place processing...

	&mov	($key,$_ivp);		# load ivp
	&mov	($_tmp,$key);

	&align	(4);
	&set_label("dec_loop");
		&mov	($s0,&DWP(0,$acc));	# read input
		&mov	($s1,&DWP(4,$acc));
		&mov	($s2,&DWP(8,$acc));
		&mov	($s3,&DWP(12,$acc));

		&mov	($key,$_key);		# load key
		&call	("_x86_AES_decrypt");

		&mov	($key,$_tmp);		# load ivp
		&mov	($acc,$_len);		# load len
		&xor	($s0,&DWP(0,$key));	# xor iv
		&xor	($s1,&DWP(4,$key));
		&xor	($s2,&DWP(8,$key));
		&xor	($s3,&DWP(12,$key));

		&sub	($acc,16);
		&jc	(&label("dec_partial"));
		&mov	($_len,$acc);		# save len
		&mov	($acc,$_inp);		# load inp
		&mov	($key,$_out);		# load out

		&mov	(&DWP(0,$key),$s0);	# write output
		&mov	(&DWP(4,$key),$s1);
		&mov	(&DWP(8,$key),$s2);
		&mov	(&DWP(12,$key),$s3);

		&mov	($_tmp,$acc);		# save ivp
		&lea	($acc,&DWP(16,$acc));
		&mov	($_inp,$acc);		# save inp

		&lea	($key,&DWP(16,$key));
		&mov	($_out,$key);		# save out

	&jnz	(&label("dec_loop"));
	&mov	($key,$_tmp);		# load temp ivp
    &set_label("dec_end");
	&mov	($acc,$_ivp);		# load user ivp
	&mov	($s0,&DWP(0,$key));	# load iv
	&mov	($s1,&DWP(4,$key));
	&mov	($s2,&DWP(8,$key));
	&mov	($s3,&DWP(12,$key));
	&mov	(&DWP(0,$acc),$s0);	# copy back to user
	&mov	(&DWP(4,$acc),$s1);
	&mov	(&DWP(8,$acc),$s2);
	&mov	(&DWP(12,$acc),$s3);
	&jmp	(&label("dec_out"));

    &align	(4);
    &set_label("dec_partial");
	&lea	($key,$ivec);
	&mov	(&DWP(0,$key),$s0);	# dump output to stack
	&mov	(&DWP(4,$key),$s1);
	&mov	(&DWP(8,$key),$s2);
	&mov	(&DWP(12,$key),$s3);
	&lea	($s2 eq "ecx" ? $s2 : "",&DWP(16,$acc));
	&mov	($acc eq "esi" ? $acc : "",$key);
	&mov	($key eq "edi" ? $key : "",$_out);	# load out
	&data_word(0xA4F3F689);	# rep movsb		# copy output
	&mov	($key,$_inp);				# use inp as temp ivp
	&jmp	(&label("dec_end"));

    &align	(4);
    &set_label("dec_in_place");
	&set_label("dec_in_place_loop");
		&lea	($key,$ivec);
		&mov	($s0,&DWP(0,$acc));	# read input
		&mov	($s1,&DWP(4,$acc));
		&mov	($s2,&DWP(8,$acc));
		&mov	($s3,&DWP(12,$acc));

		&mov	(&DWP(0,$key),$s0);	# copy to temp
		&mov	(&DWP(4,$key),$s1);
		&mov	(&DWP(8,$key),$s2);
		&mov	(&DWP(12,$key),$s3);

		&mov	($key,$_key);		# load key
		&call	("_x86_AES_decrypt");

		&mov	($key,$_ivp);		# load ivp
		&mov	($acc,$_out);		# load out
		&xor	($s0,&DWP(0,$key));	# xor iv
		&xor	($s1,&DWP(4,$key));
		&xor	($s2,&DWP(8,$key));
		&xor	($s3,&DWP(12,$key));

		&mov	(&DWP(0,$acc),$s0);	# write output
		&mov	(&DWP(4,$acc),$s1);
		&mov	(&DWP(8,$acc),$s2);
		&mov	(&DWP(12,$acc),$s3);

		&lea	($acc,&DWP(16,$acc));
		&mov	($_out,$acc);		# save out

		&lea	($acc,$ivec);
		&mov	($s0,&DWP(0,$acc));	# read temp
		&mov	($s1,&DWP(4,$acc));
		&mov	($s2,&DWP(8,$acc));
		&mov	($s3,&DWP(12,$acc));

		&mov	(&DWP(0,$key),$s0);	# copy iv
		&mov	(&DWP(4,$key),$s1);
		&mov	(&DWP(8,$key),$s2);
		&mov	(&DWP(12,$key),$s3);

		&mov	($acc,$_inp);		# load inp

		&lea	($acc,&DWP(16,$acc));
		&mov	($_inp,$acc);		# save inp

		&mov	($s2,$_len);		# load len
		&sub	($s2,16);
		&jc	(&label("dec_in_place_partial"));
		&mov	($_len,$s2);		# save len
	&jnz	(&label("dec_in_place_loop"));
	&jmp	(&label("dec_out"));

    &align	(4);
    &set_label("dec_in_place_partial");
	# one can argue if this is actually required...
	&mov	($key eq "edi" ? $key : "",$_out);
	&lea	($acc eq "esi" ? $acc : "",$ivec);
	&lea	($key,&DWP(0,$key,$s2));
	&lea	($acc,&DWP(16,$acc,$s2));
	&neg	($s2 eq "ecx" ? $s2 : "");
	&data_word(0xA4F3F689);	# rep movsb	# restore tail

    &align	(4);
    &set_label("dec_out");
    &cmp	($mark,0);		# was the key schedule copied?
    &mov	("edi",$_key);
    &mov	("esp",$_esp);
    &je		(&label("skip_dzero"));
    # zero copy of key schedule
    &mov	("ecx",240/4);
    &xor	("eax","eax");
    &align	(4);
    &data_word(0xABF3F689);	# rep stosd
    &set_label("skip_dzero")
    &popf	();
&function_end("AES_cbc_encrypt");
}

#------------------------------------------------------------------#

sub enckey()
{
	&movz	("esi",&LB("edx"));		# rk[i]>>0
	&mov	("ebx",&DWP(2,"ebp","esi",8));
	&movz	("esi",&HB("edx"));		# rk[i]>>8
	&and	("ebx",0xFF000000);
	&xor	("eax","ebx");

	&mov	("ebx",&DWP(2,"ebp","esi",8));
	&shr	("edx",16);
	&and	("ebx",0x000000FF);
	&movz	("esi",&LB("edx"));		# rk[i]>>16
	&xor	("eax","ebx");

	&mov	("ebx",&DWP(0,"ebp","esi",8));
	&movz	("esi",&HB("edx"));		# rk[i]>>24
	&and	("ebx",0x0000FF00);
	&xor	("eax","ebx");

	&mov	("ebx",&DWP(0,"ebp","esi",8));
	&and	("ebx",0x00FF0000);
	&xor	("eax","ebx");

	&xor	("eax",&DWP(2048,"ebp","ecx",4));	# rcon
}

# int AES_set_encrypt_key(const unsigned char *userKey, const int bits,
#                        AES_KEY *key)
&public_label("AES_Te");
&function_begin("AES_set_encrypt_key");
	&mov	("esi",&wparam(0));		# user supplied key
	&mov	("edi",&wparam(2));		# private key schedule

	&test	("esi",-1);
	&jz	(&label("badpointer"));
	&test	("edi",-1);
	&jz	(&label("badpointer"));

	&call	(&label("pic_point"));
	&set_label("pic_point");
	&blindpop("ebp");
	&lea	("ebp",&DWP(&label("AES_Te")."-".&label("pic_point"),"ebp"));

	&mov	("ecx",&wparam(1));		# number of bits in key
	&cmp	("ecx",128);
	&je	(&label("10rounds"));
	&cmp	("ecx",192);
	&je	(&label("12rounds"));
	&cmp	("ecx",256);
	&je	(&label("14rounds"));
	&mov	("eax",-2);			# invalid number of bits
	&jmp	(&label("exit"));

    &set_label("10rounds");
	&mov	("eax",&DWP(0,"esi"));		# copy first 4 dwords
	&mov	("ebx",&DWP(4,"esi"));
	&mov	("ecx",&DWP(8,"esi"));
	&mov	("edx",&DWP(12,"esi"));
	&mov	(&DWP(0,"edi"),"eax");
	&mov	(&DWP(4,"edi"),"ebx");
	&mov	(&DWP(8,"edi"),"ecx");
	&mov	(&DWP(12,"edi"),"edx");

	&xor	("ecx","ecx");
	&jmp	(&label("10shortcut"));

	&align	(4);
	&set_label("10loop");
		&mov	("eax",&DWP(0,"edi"));		# rk[0]
		&mov	("edx",&DWP(12,"edi"));		# rk[3]
	&set_label("10shortcut");
		&enckey	();

		&mov	(&DWP(16,"edi"),"eax");		# rk[4]
		&xor	("eax",&DWP(4,"edi"));
		&mov	(&DWP(20,"edi"),"eax");		# rk[5]
		&xor	("eax",&DWP(8,"edi"));
		&mov	(&DWP(24,"edi"),"eax");		# rk[6]
		&xor	("eax",&DWP(12,"edi"));
		&mov	(&DWP(28,"edi"),"eax");		# rk[7]
		&inc	("ecx");
		&add	("edi",16);
		&cmp	("ecx",10);
	&jl	(&label("10loop"));

	&mov	(&DWP(80,"edi"),10);		# setup number of rounds
	&xor	("eax","eax");
	&jmp	(&label("exit"));
		
    &set_label("12rounds");
	&mov	("eax",&DWP(0,"esi"));		# copy first 6 dwords
	&mov	("ebx",&DWP(4,"esi"));
	&mov	("ecx",&DWP(8,"esi"));
	&mov	("edx",&DWP(12,"esi"));
	&mov	(&DWP(0,"edi"),"eax");
	&mov	(&DWP(4,"edi"),"ebx");
	&mov	(&DWP(8,"edi"),"ecx");
	&mov	(&DWP(12,"edi"),"edx");
	&mov	("ecx",&DWP(16,"esi"));
	&mov	("edx",&DWP(20,"esi"));
	&mov	(&DWP(16,"edi"),"ecx");
	&mov	(&DWP(20,"edi"),"edx");

	&xor	("ecx","ecx");
	&jmp	(&label("12shortcut"));

	&align	(4);
	&set_label("12loop");
		&mov	("eax",&DWP(0,"edi"));		# rk[0]
		&mov	("edx",&DWP(20,"edi"));		# rk[5]
	&set_label("12shortcut");
		&enckey	();

		&mov	(&DWP(24,"edi"),"eax");		# rk[6]
		&xor	("eax",&DWP(4,"edi"));
		&mov	(&DWP(28,"edi"),"eax");		# rk[7]
		&xor	("eax",&DWP(8,"edi"));
		&mov	(&DWP(32,"edi"),"eax");		# rk[8]
		&xor	("eax",&DWP(12,"edi"));
		&mov	(&DWP(36,"edi"),"eax");		# rk[9]

		&cmp	("ecx",7);
		&je	(&label("12break"));
		&inc	("ecx");

		&xor	("eax",&DWP(16,"edi"));
		&mov	(&DWP(40,"edi"),"eax");		# rk[10]
		&xor	("eax",&DWP(20,"edi"));
		&mov	(&DWP(44,"edi"),"eax");		# rk[11]

		&add	("edi",24);
	&jmp	(&label("12loop"));

	&set_label("12break");
	&mov	(&DWP(72,"edi"),12);		# setup number of rounds
	&xor	("eax","eax");
	&jmp	(&label("exit"));

    &set_label("14rounds");
	&mov	("eax",&DWP(0,"esi"));		# copy first 8 dwords
	&mov	("ebx",&DWP(4,"esi"));
	&mov	("ecx",&DWP(8,"esi"));
	&mov	("edx",&DWP(12,"esi"));
	&mov	(&DWP(0,"edi"),"eax");
	&mov	(&DWP(4,"edi"),"ebx");
	&mov	(&DWP(8,"edi"),"ecx");
	&mov	(&DWP(12,"edi"),"edx");
	&mov	("eax",&DWP(16,"esi"));
	&mov	("ebx",&DWP(20,"esi"));
	&mov	("ecx",&DWP(24,"esi"));
	&mov	("edx",&DWP(28,"esi"));
	&mov	(&DWP(16,"edi"),"eax");
	&mov	(&DWP(20,"edi"),"ebx");
	&mov	(&DWP(24,"edi"),"ecx");
	&mov	(&DWP(28,"edi"),"edx");

	&xor	("ecx","ecx");
	&jmp	(&label("14shortcut"));

	&align	(4);
	&set_label("14loop");
		&mov	("edx",&DWP(28,"edi"));		# rk[7]
	&set_label("14shortcut");
		&mov	("eax",&DWP(0,"edi"));		# rk[0]

		&enckey	();

		&mov	(&DWP(32,"edi"),"eax");		# rk[8]
		&xor	("eax",&DWP(4,"edi"));
		&mov	(&DWP(36,"edi"),"eax");		# rk[9]
		&xor	("eax",&DWP(8,"edi"));
		&mov	(&DWP(40,"edi"),"eax");		# rk[10]
		&xor	("eax",&DWP(12,"edi"));
		&mov	(&DWP(44,"edi"),"eax");		# rk[11]

		&cmp	("ecx",6);
		&je	(&label("14break"));
		&inc	("ecx");

		&mov	("edx","eax");
		&mov	("eax",&DWP(16,"edi"));		# rk[4]
		&movz	("esi",&LB("edx"));		# rk[11]>>0
		&mov	("ebx",&DWP(2,"ebp","esi",8));
		&movz	("esi",&HB("edx"));		# rk[11]>>8
		&and	("ebx",0x000000FF);
		&xor	("eax","ebx");

		&mov	("ebx",&DWP(0,"ebp","esi",8));
		&shr	("edx",16);
		&and	("ebx",0x0000FF00);
		&movz	("esi",&LB("edx"));		# rk[11]>>16
		&xor	("eax","ebx");

		&mov	("ebx",&DWP(0,"ebp","esi",8));
		&movz	("esi",&HB("edx"));		# rk[11]>>24
		&and	("ebx",0x00FF0000);
		&xor	("eax","ebx");

		&mov	("ebx",&DWP(2,"ebp","esi",8));
		&and	("ebx",0xFF000000);
		&xor	("eax","ebx");

		&mov	(&DWP(48,"edi"),"eax");		# rk[12]
		&xor	("eax",&DWP(20,"edi"));
		&mov	(&DWP(52,"edi"),"eax");		# rk[13]
		&xor	("eax",&DWP(24,"edi"));
		&mov	(&DWP(56,"edi"),"eax");		# rk[14]
		&xor	("eax",&DWP(28,"edi"));
		&mov	(&DWP(60,"edi"),"eax");		# rk[15]

		&add	("edi",32);
	&jmp	(&label("14loop"));

	&set_label("14break");
	&mov	(&DWP(48,"edi"),14);		# setup number of rounds
	&xor	("eax","eax");
	&jmp	(&label("exit"));

    &set_label("badpointer");
	&mov	("eax",-1);
    &set_label("exit");
&function_end("AES_set_encrypt_key");

sub deckey()
{ my ($i,$ptr,$te,$td) = @_;

	&mov	("eax",&DWP($i,$ptr));
	&mov	("edx","eax");
	&movz	("ebx",&HB("eax"));
	&shr	("edx",16);
	&and	("eax",0xFF);
	&movz	("eax",&BP(2,$te,"eax",8));
	&movz	("ebx",&BP(2,$te,"ebx",8));
	&mov	("eax",&DWP(0,$td,"eax",8));
	&xor	("eax",&DWP(3,$td,"ebx",8));
	&movz	("ebx",&HB("edx"));
	&and	("edx",0xFF);
	&movz	("edx",&BP(2,$te,"edx",8));
	&movz	("ebx",&BP(2,$te,"ebx",8));
	&xor	("eax",&DWP(2,$td,"edx",8));
	&xor	("eax",&DWP(1,$td,"ebx",8));
	&mov	(&DWP($i,$ptr),"eax");
}

# int AES_set_decrypt_key(const unsigned char *userKey, const int bits,
#                        AES_KEY *key)
&public_label("AES_Td");
&public_label("AES_Te");
&function_begin_B("AES_set_decrypt_key");
	&mov	("eax",&wparam(0));
	&mov	("ecx",&wparam(1));
	&mov	("edx",&wparam(2));
	&sub	("esp",12);
	&mov	(&DWP(0,"esp"),"eax");
	&mov	(&DWP(4,"esp"),"ecx");
	&mov	(&DWP(8,"esp"),"edx");
	&call	("AES_set_encrypt_key");
	&add	("esp",12);
	&cmp	("eax",0);
	&je	(&label("proceed"));
	&ret	();

    &set_label("proceed");
	&push	("ebp");
	&push	("ebx");
	&push	("esi");
	&push	("edi");

	&mov	("esi",&wparam(2));
	&mov	("ecx",&DWP(240,"esi"));	# pull number of rounds
	&lea	("ecx",&DWP(0,"","ecx",4));
	&lea	("edi",&DWP(0,"esi","ecx",4));	# pointer to last chunk

	&align	(4);
	&set_label("invert");			# invert order of chunks
		&mov	("eax",&DWP(0,"esi"));
		&mov	("ebx",&DWP(4,"esi"));
		&mov	("ecx",&DWP(0,"edi"));
		&mov	("edx",&DWP(4,"edi"));
		&mov	(&DWP(0,"edi"),"eax");
		&mov	(&DWP(4,"edi"),"ebx");
		&mov	(&DWP(0,"esi"),"ecx");
		&mov	(&DWP(4,"esi"),"edx");
		&mov	("eax",&DWP(8,"esi"));
		&mov	("ebx",&DWP(12,"esi"));
		&mov	("ecx",&DWP(8,"edi"));
		&mov	("edx",&DWP(12,"edi"));
		&mov	(&DWP(8,"edi"),"eax");
		&mov	(&DWP(12,"edi"),"ebx");
		&mov	(&DWP(8,"esi"),"ecx");
		&mov	(&DWP(12,"esi"),"edx");
		&add	("esi",16);
		&sub	("edi",16);
		&cmp	("esi","edi");
	&jne	(&label("invert"));

	&call	(&label("pic_point"));
	&set_label("pic_point");
	blindpop("ebp");
	&lea	("edi",&DWP(&label("AES_Td")."-".&label("pic_point"),"ebp"));
	&lea	("ebp",&DWP(&label("AES_Te")."-".&label("pic_point"),"ebp"));

	&mov	("esi",&wparam(2));
	&mov	("ecx",&DWP(240,"esi"));	# pull number of rounds
	&dec	("ecx");
	&align	(4);
	&set_label("permute");			# permute the key schedule
		&add	("esi",16);
		&deckey	(0,"esi","ebp","edi");
		&deckey	(4,"esi","ebp","edi");
		&deckey	(8,"esi","ebp","edi");
		&deckey	(12,"esi","ebp","edi");
		&dec	("ecx");
	&jnz	(&label("permute"));

	&xor	("eax","eax");			# return success
&function_end("AES_set_decrypt_key");

&asm_finish();